University of Southern California Dissertations and Theses

Earthquake-driven landsliding, erosion and mountain building: from the eastern Tibetan mountains towards global models

(USC Thesis Other)

Earthquake-driven landsliding, erosion and mountain building: from the eastern Tibetan mountains towards global models

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

EARTHQUAKE-DRIVEN LANDSLIDING, EROSION AND MOUNTAIN BUILDING:
FROM THE EASTERN TIBETAN MOUNTAINS TOWARDS GLOBAL MODELS

by

Gen Li

_______________________________

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(GEOLOGICAL SCIENCES)

December, 2017

Copyright 2017 Gen Li
i

Epigraph

“The progress of science has always been the result of a close interplay between our concepts of
the universe and our observations on nature. The former can only evolve out of the latter and yet
the latter is also conditioned greatly by the former. Thus in our exploration of nature, the
interplay between our concepts and our observations may sometimes lead to totally unexpected
aspects among already familiar phenomena. As in the present case, these hidden properties are
usually revealed only through a fundamental change in our basic concept concerning the
principles that underly natural phenomena. While all this is well-known, it is nevertheless an
extremely rich and memorable experience to be able to watch at close proximity in a single
instance the mutual influence and the subsequent growth of these two factors - the concept and
the observation.”

-Dr. Tsung Dao Lee, the Nobel Prize Lecture in Physics in 1957

ii

Acknowledgement
I would like to first thank my advisor Dr. A. Joshua West. It was a great fortune to have Josh as
my advisor and role model. I want to thank Josh for teaching me the philosophy and recipes of
doing good science from making solid observations to thinking about big picture questions, for
introducing me into the fascinating world of geomorphology, and for taking me to the mountains
and showing me the beauty of field work – those trips have been among the most unforgettable
days in my life. Six years are not a short period of time, but many times I thought it was still too
short for me to learn more from Josh. With that feeling, I often recalled a scene from one of my
favorite Chinese novels: “… later he realized that his master (‘shifu’ in Chinese) wanted to teach
him so many things in those stories, but there was not much time left…while telling the stories,
‘shifu’ had to keep counting how much time was remaining from second to second….to the end,
he had to leave”. I am looking forward towards future opportunities to learn from Josh again.
Thanks to Doug Hammond and Alexander Robinson for being my committee members and
for their input on my work. Doug’s scientific intuition and curiosity had much influence on me.
Thanks to Alex Densmore and Bob Hilton for sharing their wisdom and for hosting my visit to
Durham University. Thanks to Zhangdong Jin, Fei Zhang and Jin Wang from the Chinese
Academy of Sciences for helping making the Sichuan campaigns a great success. Thanks to
Deepak, Bibek and Suman for all the help in the Nepal field trips. Thanks to Marin Clark,
Dimitrios Zekkos and Billy Medwedeff for all the helpful discussions.
Many thanks to the lab members in Josh & Doug’s groups, past and present: Mark, Jotis
(awesome officemate!), Max, Paul, Kirstin, Eric, Joyce, Paulina, Audra, Jessica, Mathieu, and
Camilo, for all the help and discussions. Several USC undergraduates, Zichen, Renee and Ellie,
have helped with lab work and I owe them thanks as well. I also want to thank the faculty
members in the department. Will’s questions motivated me to think deeper. Yehuda’s spirit and
philosophy as a physicist always inspired me, though we still need to figure out that aftershock
problem. Thanks to Sarah for sharing her knowledge in organic geochemistry. Thanks to James
for teaching me faults and earthquake geology. Thanks to David Okaya for sharing the expertise
in seismic reflection profiles and for the many discussions. Thanks to Yong-Gang Li for all the
care. Thanks to the staff members, John McRaney, Cindy, Vardui, John Yu, Miguel, Barbara,
Karen and Deborah, for all the help throughout my study and life here. Thanks to James Polk for
being a great English teacher. My friends in and outside the department have also contributed to
those wonderful days and I want to thank them as well. Thank you, Hongrui, Feng, Hao, Nemo,
Junsong, Jie, Wenrong, Xin(s), Haoran(s), Jun(s), Lei, Sijia, Xueyao, Danielle, Guang-Sin, Yi,
Hannah, Xiao, Rudy, and the unmentioned many!
I am always grateful to the days when I was an undergraduate at Nanjing University. Many
thanks to Prof. Junfeng Ji for enrolling me into the Institute of Surficial Geochemistry where I
started my research life. Thanks to Prof. Jiedong Yang for introducing me to the exciting stories
of mountains, carbon and climate during the Cenozoic. Thanks to Liang, Gaojun and Tony for all
the support throughout the years.
Last but most importantly, I want to thank my family. Thank you, my parents and Christine,
for all your love and support!
iii

EARTHQUAKE-DRIVEN LANDSLIDING, EROSION AND MOUNTAIN BUILDING:
FROM THE EASTERN TIBETAN MOUNTAINS TOWARDS GLOBAL MODELS
by
Gen Li

Abstract
In tectonically active regions, earthquakes are a key driver of landscape evolution, erosional
effluxes, mountain building, and the carbon cycle. Seismic processes deform the lithosphere and
create permanent topographic features, like mountain belts, at Earth’s surface. However, strong
ground motion during large earthquakes also causes widespread mass wasting which collectively
generates large volumes of clastic sediment and enhances fluvial erosional fluxes out of
mountains. Due to the rarity of high magnitude earthquakes, limited studies have provided direct
observations on how earthquakes affect landscapes and mass fluxes into and out of mountain
ranges. Whereas numerous studies have looked into how earthquakes induce rock uplift, the
erosive power of earthquakes remains to be better understood, especially in the context of the
tectonic evolution of mountain belts. Considering those knowledge gaps, this thesis is aimed at
understanding the effects of large earthquakes on mountainous landscapes, with an emphasis on
earthquake-triggered landslides. Specifically, this thesis addressed the research question of how
earthquakes affect landscapes from three aspects: (1) the locations of earthquake-triggered
landslides in landscapes and relative to fluvial networks, as presented in Chapter 2 and the
Appendix; (2) the erosive power of earthquake cycles over tectonic timescales, as presented in
Chapter 3; and (3) orogenic growth under the competition between earthquake-triggered
landslide erosion and seismically caused rock uplift. I took the 2008 M
w
7.9 Wenchuan earthquake
and associated landsliding as a case study to conduct empirical observations and further
developed a generalized theoretical framework. In the last part, I discussed future research plans
on post-earthquake sediment transport, interactions between earthquakes and the carbon cycle,
and non-landsliding erosion processes.

Chapter 2 studied the location of landslides relative to the fluvial networks, which is expected to
have an important influence on post-earthquake sediment transport. Understanding sediment
transport is critical for evaluating how earthquakes affect erosion, landscape evolution and
sediment-related hazards. This chapter examined the position of the landslides triggered by the
Wenchuan earthquake in the context of the mountainous fluvial network in the steep Longmen
Shan mountain range. I developed a raster data-based approach to quantify landslide-channel
connectivity using a landslide inventory and analysis of digital topography. The landslide-channel
connectivity was presented in terms of landslide number, landslide area and landslide volume.
Across the Longmen Shan range, there was a significant spatial variability in landslide-channel
connectivity, with volumetric connectivity ranging from ~20% to ~90% for different catchments
(43+9/-7% for total), likely caused by variations in topographic, seismic and substrate conditions
that my affect both channelization and landslide size. After the Wenchuan earthquake, suspended
iv

sediment (mainly composed of fine grains with diameter < 0.25 mm) yield across the Longmen
Shan catchments correlated positively with catchment area-normalized landslide volume, but this
correlation was statistically indistinguishable from the correlation where only channel-connected
landslides are considered. This suggested that landslide-channel connectivity may have limited
influence on mobilizing fine-grained landslide debris in hillslope domains, but may have more
significant impact on the transport of the coarser material which comprised >90% of the landslide
sediment.

Chapter 3 focused on the erosive power of earthquake cycles across a wide range of timescales.
Previous studies have shown that earthquake-triggered landslides could dominate erosional
fluxes over decadal timescales, but the influence of earthquakes and associated landsliding on the
erosional buget over geological timescales is not clear. I examined the erosional budget related to
earthquake-triggered landslides in the steep Longmen Shan range at the eastern margin of the
Tibetan Plateau, using empirical observations from the M
w
7.9 2008 Wenchuan and M
w
6.6 2013
Lushan earthquakes. I used riverine sediment fluxes, cosmogenic nuclide chronology, low
temperature thermochronology and models of earthquake-triggered landslides to estimate
denudation rates over decadal, kyr, Myr and multiple earthquake cycle timescales, respectively.
The highest denudation rates across this wide range of timescales coincide spatially with the
region where most landslides occurred during the Wenchuan earthquake. Catchment-scale
post-Wenchuan denudation rates are closely correlated with seismic shaking and the volumes of
Wenchuan landslides, demonstrating a dominant control of seismic processes on denudation in
time periods shortly following earthquakes. Combining models predicting landslide volumes,
calibrated against the Wenchuan and the Lushan data, and the Longmen Shan seismic catalog, I
calculated a long-term, landslide-sustained “seismic erosion rate”, and found this theoretical
erosion rate has a similar magnitude to the measured regional long-term denudation rates (~0.5-1
mm yr
-1
). The similar magnitude and spatial coincidence suggest that earthquake-triggered
landslides are a major contributor to the long-term erosional budget in the study area, highliting
seismogenic faulting as an important mechanism causing focused denudation in fault-bounded
mountain ranges.

Chapter 4 developed a generalized modeling framework to probe the earthquake volume balance
problem, i.e., the competition between earthquake-triggered landslide erosion and seismically
induced rock uplift. The contribution of earthquakes to building mountainous topography remains
to be fully understood, particularly considering the erosive role of seismically-triggered
landslides. Combining geophysical solutions to fault-related deformations over earthquake cycles
and models of earthquake-triggered landslides, I simulated the 2-D surface deformation field
caused by co-seismic and inter-seismic processes including viscoelastic relaxation, landslide
erosion, and flexural-isostatic restoration. I developed a new metric: E, the efficiency of
earthquakes in building topography, defined as the volume of formed topography over one full
earthquake cycle relative to the co-seismically uplifted volume. For earthquake with given
magnitudes, E is primarily controlled by the flexural rigidity of the lithosphere. Even those
v

earthquakes that trigger significant landslide volumes do not cause net topographic destruction
over complete earthquake cycles, except over narrow spatial windows focused near the
seismogenic fault. While landslide erosion is concentrated in this narrow zone, uplift is
distributed over a wider area, and the effects of inter-seismic processes such as isostatic
compensation act to significantly offset mass loss due to landslides. The wavelength of
seismically constructed topography scales with earthquake magnitude, fault dimensions, and
lithospheric rigidity, such that outwardly propagating fault systems in major collisional mountain
ranges might be expected to produce low-relief interiors and high-relief margins where erosion is
focused.

The Appendix presented a comprehensive analysis of how landslides triggered by the Wenchuan
earthquake distribute across landscapes. I combined a Wenchuan landslide inventory map (Li et
al., 2014; Li et al., 2017) with analysis of digital topography, regional geology, and ground
motion data, to explore the controlling factors of the Wenchuan landslides. I examined the
hillslope aspects of landslides and discussed how the preferred facing directions reveal
information about the earthquake source, seismic ground motion, and rupture propagation. I
evaluated the distribution of landslides from hillslope tops to bases and discussed the
implications for long-term evolution of hillslope morphology. Assuming that the Wenchuan
seismogenic fault was a linear energy source, I successfully modeled the pattern of Wenchuan
landslides by adapting a functional form of the law of seismic wave attenuation which accounts
for both geometric spreading and quality decay. In conjunction with models predicting total
volumes of earthquake-triggered landslides, this approach has promise for predicting the
magnitude and pattern of landslides caused by earthquakes based on known characteristics of the
seismogenic faults and the seismotectonic setting.

vi

List of Figures
Figure 1.1 Schematic diagram of landslide-induced primary and secondary hazards…….2
Figure 1.2 Schematic diagram of landslide-induced channel aggradation and flooding….3
Figure 1.3 Schematic diagram of the effects of earthquakes on landscapes……………3

Figure 2.1 Maps of topography, seismology, and lithology of the study area…………….29
Figure 2.2 Map of the fluvial network and catchments in the study area………………...30
Figure 2.3 Typical gradient-area plots of the catchments in the study area………………31
Figure 2.4 Illustration of catchment-scale landslide location index………………………32
Figure 2.5 Statistical distributions of landslide volumes vs. upstream area………………33
Figure 2.6 Inverse gamma function fits of the Wenchuan landslide areas………………..34
Figure 2.7 Relationship between landslide location index and connectivity to channels...35
Figure 2.8 Spatial patterns of the Wenchuan landslides…………………………………..36
Figure 2.9 Controls on landslide-channel connectivity…………………………………...37
Figure 2.10 Post-Wenchuan earthquake sediment yields vs. landslide volumes………...…38
Figure 2.S1 Hydrological map and gradient-upstream area plots………………………….39
Figure 2.S2 Uncertainty caused by sampling rasters of different resolutions……………...40
Figure 2.S3 Landslide-channel connectivity as a function of channel threshold area……...41
Figure 2.S4 Uncertainties of connectivity vs. channel threshold area……………………...42
Figure 2.S5 Statistical distributions of landslide volumes vs. upstream area………………43
Figure 2.S6 Slope and seismic shaking effects on drainage density and landslide area…...44

Figure 3.1 Maps of topography, seismology, and denudation rates in the study area…….74
Figure 3.2 Swath profiles of topography, seismicity, and denudation rates………………76
Figure 3.3 Post-earthquake denudation rates vs. topographic and climatic metrics……...78
Figure 3.4 Post-earthquake denudation rates vs. landslide volumes and seismic shaking..79
Figure 3.5 Earthquake-triggered landslide volume models……………………………….80
Figure 3.6 Longmen Shan seismicity and seismic erosion rates………………………….81
Figure 3.7 Longmen Shan denudation rates across different timescales………………….82
Figure 3.S1 Slopes derived from SRTM30 vs. SRTM90 DEM data……………………….83
Figure 3.S2 Threshold runoff vs. the squares of correlation coefficient between denudation rate
and the proportion of total runoff from days > threshold value……………………….83
Figure 3.S3 Wenchuan landslide volumes vs. PGA and distance to fault………………….84
Figure 3.S4 Longmen Shan seismicity……………………………………………………..84
Figure 3.S5 PCA plot for the studied metrics………………………………………………85
Figure 3.S6 Relationships between PGA, slope and runoff………………………………..85
Figure 3.S7 Seismic erosion rates determined from different landslide volume models…..86

vii

Figure 4.1 The theoretical efficiency of topographic growth over a complete earthquake cycle
as a function of co-seismic and inter-seismic processes……………………………104
Figure 4.2 Patterns of seismic deformations, landsliding and isostacy…………………105
Figure 4.3 The efficiency of topographic growth vs. co-seismic volume ratio, effective elastic
thickness and landslide pattern……………………………………………...…….106
Figure 4.4 Wavelength of seismic topography and schematic illustration of plateau growth by
earthquake cycles……………………………………………………………………...107
Figure 4.S1 Schematic illustration of the fault setting……………………………………108
Figure 4.S2 Wenchuan landslide data explained by seismic wave attenuation…………...109
Figure 4.S3 Earthquake magnitude vs. volumes of co-seismically created topography and
co-seismic landslides……………………………………………………………………..110
Figure 4.S4 Variations of co-seismic volume ratio across earthquake magnitude for near field
and far field scenarios……………………………………………………………….111
Figure 4.S5 Spatial variations of co-seismic uplift and landslide erosion caused by the
Wenchuan earthquake…………………………………………………………………….112
Figure 4.S6 Mountain uplift rates over multiple earthquake cycles………………………113

Figure A.1 Maps of topography, seismology, and lithology of the study area…………...134
Figure A.2 Spatial patterns of the Wenchuan landslides…………………………………134
Figure A.3 Distribution of the Wenchuan landslides over topographic attributes and
lithologic units……………………………………………………………………………135
Figure A.4 Landslide-fluvial network connectivity……………………………………...136
Figure A.5 Landslide areal density vs. mean peak ground acceleration…………………137
Figure A.6 The product of landslide areal density and distance to seismic energy source vs.
distance to seismic energy source………………………………………………………...138
Figure A.7 Distribution of landslides induced by earthquakes and rainfall triggers across
hillslopes………………………………………………………………………………….139
Figure A.8 Hillslope scale local gradients vs. normalized distance to streams…………..140
Figure A.9 Schematic diagram illustrating the effect of aspect on earthquake-triggered
landslides………………………………………………………………………………….141
Figure A.10 Preferred hillslope aspect directions of the Wenchuan earthquake-triggered
landslides along fault strike………………………………………………………………141

viii

List of Tables
Table 2.1 Notation for symbols in Chapter 2…………………………………………….45
Table 2.2 Topographic and landslide parameters for studied catchments……………….46
Table 2.3 Post-earthquake sediment yields and landslide volumes……………………...48

Table 3.1 Notation for symbols in Chapter 3…………………………………………….87
Table 3.S1 Denudation rates and metrics of topography, hydrology, fluvial transport
capacity, and seismic processes in the study area…………………………………….88
Table 3.S2 Notation for symbols in Supplementary Materials in Chapter 3……………...90
Table 3.S3 Compiled total dissolved load : suspended load ratios………………………..91
Table 3.S4 Factor loading matrix from the PCA analysis…………………………………92
Table 3.S5 Total variance explained by the PCA components…………………………….92

ix

Table of Contents

Epigraph………………………………………………………………………………..……….i

Acknowledgement……………………………………………………………………..………ii

Abstract…………………………………………………………………………………..……iii

List of Figures…………………………………………………………………………..……..vi

List of Tables…………………………………………………………………………..…….viii

Table of Contents……………………………………………………………………………...ix

Chapter 1. Introduction………………………………………………………………………...1
1.1. Thesis outline……………………………………………………………………………...3
1.2. Landslide location in landscapes………………………………………………………….4
1.3. Erosive power of earthquake cycles………………………………………………………4
1.4. The earthquake volume balance problem…………………………………………………4

Chapter 2. Connectivity of earthquake-triggered landslides with the fluvial network:
implications for landslide sediment transport after the 2008 Wenchuan earthquake…………..5
2.1. Preamble……….………………………………………………………………………….5
2.2. Introduction………………………………………………………………………………..7
2.3. Setting……………………………………………………………………………………..8
2.4. Materials and approaches………………………………………………………………….9
2.5. Results and discussion…………………………………………………………………...15
2.6. Conclusions………………………………………………………………………………22
2.7. Supplementary materials to Chapter 2…………………………………………………...23

Chapter 3. Earthquakes drive focused denudation along a tectonically active mountain front.…51
3.1. Preamble……….………………………………………………………………………...51
3.2. Introduction………………………………………………………………………………53
3.3. Setting……………………………………………………………………………………54
3.4. Materials and approaches………………………………………………………………...55
3.5. Results and discussions…………………………………………………………………..61
3.6. Conclusions and implications……………………………………………………………67
3.7. Supplementary materials to Chapter 3…………………………………………………...68

x

Chapter 4. Mountain building over earthquake cycles considering erosion driven by
earthquake-triggered landslides………………………………………………………………94
4.1. Preamble……….………………………………………………………………………...94
4.2. Introduction………………………………………………………………………………95
4.3. The efficiency of earthquakes in building topography…………………………………..96
4.4. Model setup………………………………………………………………………………96
4.5. Patterns of seismically induced deformations……………………………………………98
4.6. The ratio between co-seismic uplift and landslide erosion………………………………98
4.7. Inter-seismic controls on topographic growth…………………………………………...99
4.8. Wavelength of seismic topography and implications for plateau construction…………..100
4.9. Supplementary materials to Chapter 4…………………………………………………...100

Chapter 5. Conclusions and Future Work…………………………………………………...115
5.1. Conclusions……………………………………………………………………………..115
5.2. Future plans……………………………………………………………………………..116

Appendix. Distribution of earthquake-triggered landslides across landscapes: Towards
understanding erosional agency and cascading hazards…………………………………120
A1. Preamble…….….………………………………………………………………………120
A2. Introduction….………………………………………………………………………….121
A3. Setting……….………………………………………………………………………….124
A4. Materials and methodology….………………………………………………………….125
A5. Distribution of the Wenchuan landslides across the Longmen Shan……………….…..125
A6. Seismic controls on the pattern of the Wenchuan landslides……………….…………..129
A7. Conclusions, implications and future directions……….………………….…………….132

Bibliography………………………………………………………………………………...143

1

Chapter 1 Introduction

Earthquakes represent a powerful tectonic forcing on Earth’s surface environments, driving
changes in mass fluxes into and out of active mountain ranges (Avouac, 2007). Earthquakes
deform the lithosphere, create topographic features, and accumulate small deformations over
multiple earthquake cycles to build large scale geological structures like mountain ranges (King
et al., 1988; Simpson, 2014). On the other side, during earthquakes, strong ground motion can
shatter rocks and fail hillslopes, causing landsliding (Keefer et al., 1984; Keefer, 1994; Harp and
Gibson, 1996; Malamud et al., 2004). Earthquake-triggered landslides are a major form in
mountainous communities, a dominant erosional agent, as well as a key driver of the
biogeochemical cycles of carbon (Hilton et al., 2011; Emberson et al., 2016; Jin et al., 2016;
Wang et al., 2016). Earthquake-triggered landslides collectively generate large volumes of clastic
sediment. Supply of landslide debris to river channels enhances the fluvial export of sediment,
causing mass loss from mountains (Hovius et al., 2011; Wang et al., 2015). Landslides not only
cause immediate damage during earthquakes, but also lead to prolonged, secondary hazards like
aggradation and flooding (Figure 1.1, Densmore, 2014; Figure 1.2, Wang et al., 2015) during
post-earthquake time periods. Landslides harvest organic carbon from bedrock, soil and
vegetation carbon reservoirs, thus enhancing erosion and export of organic carbon from
mountainous landscapes (Wang et al., 2016). Landslides also expose fresh mineral surfaces and
can potentially promote chemical weathering (Emberson et al., 2016; Jin et al., 2016).

Beyond these concepts, the effects of large earthquakes on landscapes remain to be fully
understood, especially considering the role of earthquake-triggered landslides. Specifically,
several key questions remain to be addressed. First, how do rivers mobilize and evacuate
landslide debris after earthquakes? This question is not only critical for managing landslide
hazards following earthquakes, but also important for quantifying erosional fluxes over
earthquake cycles (Hovius et al., 2011; Parker et al., 2011). The extent to which rivers can export
landslide debris before the next seismic event also determines whether landslide debris
accumulates in mountain valleys and thus the topographic structures of the mountain ranges.
Second, how do earthquake cycles contribute to the erosional budget of mountain ranges over
geological timescales? This question is focused on the erosive power of earthquake cycles, and
requires quantifying earthquake-induced landslide erosional fluxes over tectonic timescales
versus the observed long-term erosion rates measured from geochemical tools like low
temperature thermochronometer or cosmogenic nuclides. Third, how do earthquake cycles build
mountainous topography under the competition between earthquake-triggered landslide erosion
and seismically induced rock uplift (Figure 1.3)? Empirical observations have shown that, for
some large earthquakes, the amount of landslide erosion is comparable to seismically induced
rock uplift, meaning that seismic uplift is sufficiently counteracted by earthquake-induced
erosion (Hovius et al., 2011; Parker et al., 2011; Li et al., 2014). Thus single large earthquakes do
not necessarily build mountains, and a fundamental question about mountain building in

seism
small
struct
as the
place
et al.,
relaxa
et al.,
synth
do ea
scales
key to
and g
Engla

With
empir
questi
M
w
7.9
landsl
provid
landsc
theori
simila

Figur
and s
earthq
mically active
deformation
tures (King et
e cumulative
more rigorou
, 2016a), but
ation of co-se
2012; Molna
etic model co
arthquakes al
s? This quest
o resolving th
global cooling
and, 1990; Li
these questi
rical observa
ions and prov
9 Wenchuan
liding in the
ding a good n
capes. On the
ies about eart
ar settings.
e 1.1. Schem
secondary h
quakes (Dens
regions rema
ns of single
t al., 1988), b
effects of mu
us constraints
t lacked cons
eismic deform
ar et al., 2012
onsidering tho
lter chemical
tion is critica
he late Cenozo
g and their c
et al., 2013; T
ons in mind
ations and th
vided basic da
earthquake a
steep Longm
natural experi
e basis of the
thquake-drive
atic illustratio
azards (e.g.,
more, 2014)
ains to be an
earthquakes
but these stud
ultiple earthqu
s on earthqua
siderations of
mations and is
). Therefore,
ose unaccount
weathering
l for understa
oic conundrum
complex inter
Torres et al., 2
d, I designed
heory develo
atasets for fut
as a case stud
men Shan mou
iment to expl
e Wenchuan
en erosion an
on of primary
, aggradation
2
nswered. Prev
s can accum
dies had loose
uake cycles w
ake-triggered
f the effects
sostatic respo
to fully under
ted effects is
and organic
anding tecton
m: the tempor
ractions (Ray
2014).
d my thesis f
opment. This
ture reseach o
dy. The Wen
untains at the
ore how earth
case study, I
nd mountain
y hazards (e.g
n and flood
vious modelin
mulate to bui
e constraints o
with varying m
landslide ero
of inter-seism
onses to erosi
rstand seismi
needed. Four
carbon trans
nic controls o
ral coinciden
ymo and Rud
following a
thesis addr
on those left u
chuan earthq
e eastern mar
hquakes affec
developed g
building, wh
g., immediate
ding) caused
ng efforts hav
ild large sca
on the erosion
magnitudes. R
osion (Li et al
mic processe
ional unloadi
ic mountain b
rth, how and
sfer processe
on the carbon
nce between m
ddiman, 1989
strategy to c
ressed some
unresolved. I
quake triggere
rgin of the Ti
ct erosion and
generalized ap
hich are appli

e damage to in
by landslid
ve shown tha
ale geologica
n term as we
Recent studie
l., 2014; Mar
es, specificall
ng (Densmor
building, a new
to what exten
es at orogeni
n cycle, and i
mountain upli
9; Molnar an
combine soli
of the liste
took the 200
ed widesprea
ibetan Plateau
d mountainou
pproaches an
icable to othe
nfrastructures
des followin
at
al
ll
es
rc
ly
re
w
nt
ic
is
ft
nd
id
ed
08
ad
u,
us
nd
er
s)
ng

Figur
debris

Figur
earthq

1.1. T
In thi
relativ
relate
Wenc
cycles
gener
metric
comp
patter
e 1.2. Schem
s to river chan
e 1.3. Schem
quake-trigger
Thesis outline
s thesis, I pre
ve to the fluv
d influence o
chuan earthqu
s and associa
ralized model
c: the effici
rehensive ana
rn of the lands
matic illustrat
nnels
atic illustratio
ed landslides
e
esented four r
vial channels,
on sediment
uake and over
ated landslid
l accounting
ency of eart
alysis of the W
slides and the
tion of aggra
on of the com
and associate
research proj
the controllin
transport. C
r multiple ear
ding in the lo
for co-seismi
thquakes in
Wenchuan lan
e related contr
3
adation and f
mpetition betw
ed fluvial exp
ects. Chapter
ng factors on
hapter 3 qua
rthquake cycle
ong-term ero
ic and post-s
building mo
ndslide inven
rolling factors
flooding caus
ween seismica
port of landsli
r 2 examined
n landslide-ch
antified denu
es, and evalu
osional budge
eismic proce
ountains. Th
ntory, with mu
s.
sed by supply

ally induced r
ide debris
d the locations
hannel connec
udation rates
uated the role
et. Chapter 4
esses and intr
he Appendix
uch emphasis
y of landslid
rock uplift an
s of landslide
ctivity, and th
following th
of earthquak
4 developed
roduced a new
conducted
s on the spatia
de
nd
es
he
he
ke
a
w
a
al
4

1.2. Landslide location in landscapes
Chapter 2 and the Appendix investigated the locations of the Wenchuan landslides in landscapes
and relative to the fluvial network, by analyzing a Wenchuan co-seismic landslide inventory (Li
et al., 2014) and datasets of digital topography, lithology, and seismology. I mapped the fluvial
network and channels in the study area, and quantified the number, area and volume of landslides
connected to the channels versus deposited on hillslopes. I developed a new catchment-scale
landslide location index, which was independent of mapped channels, to validate estimates of
landslide-channel connectivity. I examined the relationships between landslide connectivity
between seismic and topographic factors, and lithologic units, to understand controls on landslide
locations relative to the fluvial network. I considered how landslide-channel connectivity affects
sediment transport using suspended sediment flux, landslide and connectivity data. I also
identified the preferred locations of landslides in landscapes and demonstrated dominant seismic
controls from fault zone scales to hillslope scales.

1.3. Erosive power of earthquake cycles
Chapter 3 evaluated the erosive power of earthquake cycles from decadal to Myr timescales. I
quantified denudation rates in the frontal Longmen Shan in the four years following the
Wenchuan earthquake and over multiple earthquakes. I examined the relationships between
post-earthquake denudation rates and metrics describing topography, hydrology,
earthquake-triggered landslides, and seismic processes to understand what controls
post-earthquake erosional processes. I used empirical and mechanistic models to predict volumes
of landslides triggered by earthquakes in the study area, with key parameters calibrated against
the landslide inventories associated with the Wenchuan earthquake and the adjacent 2013 M
w
6.6
Lushan earthquake. I calculated “seismic erosion rates” sustained by earthquake-triggered
landslides over multiple earthquake cycles, compiled regional long-term denudation rates
recorded by low temperature thermochronometer and cosmogenic nuclides, and compared the
observed long-term denudation rates to the theoretical seismic erosion rates.

1.4. The earthquake volume balance problem
Chapter 4 was focused on the earthquake volume balance problem (Figure 1.3) in the context of
a generalized modeling framework. Empirical observations demonstrate the significant erosive
power of earthquakes (Parker et al., 2011; Li et al., 2014; Li et al., 2017) and raise the earthquake
volume balance problem. I developed a generalized model to simulate the 2-D surface
deformation field caused by co-seismic and inter-seismic processes operated over full earthquake
cycles, combining geophysical solutions to seismic deformations and advances in predicting the
patterns and magnitudes of earthquake-triggered landslides. The modeling results are then
summarized in a new metric: E, the efficiency of earthquakes in building mountainous
topography. I further discussed the wavelength of topography created by earthquake cycles and
the implications for building orogenic plateau.

5

Chapter 2
Connectivity of earthquake-triggered landslides with the fluvial network:
implications for landslide sediment transport after the 2008 Wenchuan
earthquake

2.1. Preamble
Understanding erosional processes and sediment transport after earthquakes requires
quantification of post-earthquake sediment fluxes. Whereas suspended load can be determined
from hydrological gauging, bedload transport is a notoriously complex problem (Wong and
Parker, 2006; Ma et al., 2014). Post-earthquake sediment transport is further complicated by
sediment input from multiple landslide sources. Estimates of bedload fluxes rely on either field
gauging using sediment traps (e.g., a reservoir) or geochemical tools (e.g.,
10
Be), or empirical
relations with grain size, discharge and channel morphology as determined from flume
experiments. Moreover, evaluating the influence of bedload transport on hazard generation,
landforms and mountain belt evolution requires combination of hydraulic modeling efforts (e.g.,
Egholm et al., 2013) and empirical observations that give realistic input parameters for modeling.
Since landslides are a major sediment source after large earthquakes, the characteristics of
landslide debris are key parameters for modeling post-earthquake sediment transport that are
generally not well known. To describe landslide as sediment soruce, two key parameters are
needed: the grain size composition and the locations relative to the fluvial network. Many studies
have developed protocols for collecting grain size information (e.g., Attal et al., 2015), but
limited studies have attempted to develop a systematic approach to evaluate the extent to which
landslides connect to the fluvial channels and to understand the controlling factors behind this
connectivity (e.g., Meunier et al., 2008). This chapter aimed at developing a systematic approach
to quantify landslide-fluvial channel connectivity, applying the approach to the Wenchuan case,
understanding the controls on the landslide-fluvial connectivity, and evaluating the influence of
landslide-fluvial connectivity on sediment transport. Whereas the ultimate goal is to resolve
post-earthquake bedload transport, this study provides a basic dataset characterizing landslides as
sediment source with insights into hillslope-channel coupling relationships in
landsliding-dominated landscapes.

I was the main author, mapped the co-seismic landslides associated with the 2008 Wenchuan
earthquake, analyzed the landslides and the digital topography, and wrote the manuscript. Josh
West provided the main supervision. Specifically, he advised on the outline of the manuscript,
collected remote imagery and was the main editor. Alex Densmore helped with refining the
upstream area-gradient scaling relations, and provided suggestions on the interpretations of the
relations between landslide volumes and sediment fluxes. Doug Hammond, Zhangdong Jin, Fei
6

Zhang and Jin Wang assisted in the field work and participated in general discussions. Bob Hilton
provided general guidance. All authors contributed to paper revision.
This research was supported by the U.S. National Science Foundation (NSF-EAR/GLD grant
1053504 to Josh West) and the Chinese Academy of Sciences (YIS fellowship grant
2011Y2ZA04 to Josh West). I was supported by a USC college merit fellowship and a GSA
graduate research grant. We thank the NSF-supported Polar Geospatial Center for providing
DigitalGlobe imagery. The landslide inventory data are available upon request. Constructive
comments from the Associate Editor, Isaac Larsen, and two anonymous reviewers greatly helped
to improve an earlier version of the manuscript. This work also benefited from conversations
with Niels Hovius, Dimitri Lague, Patrick Meunier, Marc Odin, and Joel Scheingross.

This Chapter was published as:
Li, G., A. J. West, A. L. Densmore, D. E. Hammond, Z. Jin, F. Zhang, J. Wang and R. G. Hilton
(2016), Connectivity of earthquake-triggered landslides with the fluvial network: implications for
sediment transport after the 2008 Wenchuan earthquake. Journal of Geophysical Research –
Earth Surface. 121, 703–724, doi:10.1002/2015JF003718.

7

2.2. Introduction
High-magnitude earthquakes can cause widespread landslides that collectively generate large
volumes of clastic sediment (Keefer, 1984; Keefer, 1994), contributing significantly to erosion in
tectonically-active mountain ranges (Dadson et al., 2004; Korup et al., 2004; Malamud et al.,
2004; Yanites et al., 2010; Wang et al., 2015). Fluvial evacuation of landslide-derived sediment
removes mass from mountains, influencing landscape evolution (Pearce and Watson, 1986;
Malamud et al., 2004; Korup et al., 2007; Hovius et al., 2011; Parker et al., 2011; Egholm et al.,
2013; Li et al., 2014). Landslides also impact the terrestrial biosphere (Garwood et al., 1977;
Allen et al., 1999; Clark et al., 2016), and delivery of eroded material to river channels can
redistribute essential nutrient elements (e.g., carbon and nitrogen), contributing to tectonic
forcing of global biogeochemical cycles (Hilton et al., 2011; Ramos Scharrón et al., 2012; Jin et
al., 2016). Furthermore, sediment supply from landslides to rivers may cause prolonged
secondary natural hazards, via channel aggradation and enhanced flooding, and may reduce the
storage capacity of downstream reservoirs (Korup et al., 2004; Glade and Crozier, 2005; Huang
and Fan, 2013; Wang et al., 2015).

Several studies have quantified sediment mass flux and the associated residence times of
landslide material in mountain belts using hydrological gauging data (e.g., Pearce and Watson,
1986; Dadson et al., 2004; Korup et al., 2004; Hovius et al., 2011; Tsai et al., 2013; Wang et al.,
2015), topographic surveys of individual rivers (Liu et al., 2015; Yanites et al., 2010), and
geochemical measurements such as cosmogenic nuclide inventories (e.g., West et al., 2014;
McPhillips et al., 2014). Other studies have used numerical models to predict the entrainment,
transport, and deposition of sediment and to predict evacuation rates (e.g., Attal and Lave, 2006;
Cui et al., 2003; Sutherland et al., 2002; Ferguson et al., 2015). All of these approaches depend
on understanding the amount of sediment that landslides make available for fluvial transport,
which is determined by the number and size of landslides as well as by their location.
Earthquake-triggered landslides are not distributed evenly across landscapes: some are directly
connected to river channels and thus prone to fluvial transport, whereas others are sequestered on
hillslopes away from the river network, where they are expected to contribute less immediately to
the riverine sediment budget (Meunier et al., 2008; Dadson et al., 2004; Hovius et al., 2011;
Huang and Fan, 2013). Moreover, earthquake-triggered landslides have a range of sizes
(Malamud et al., 2004), potentially affecting their impact on river systems. The
“landslide-landscape relationship” (Dadson et al., 2004; Hovius et al., 2011; Tsai et al., 2013),
which is governed by factors such as the hydrological and topographic characteristics of the
landscape, the location of landslides, and the geometric properties (e.g., size and runout length)
of landslides, is expected to determine the magnitude and duration of associated sediment
transport. But key aspects of this relationship and how it controls sediment dynamics are not
completely understood.

Previous studies of landslide spatial distribution have measured landslide locations with respect
to channels versus ridges, demonstrating that landslides cluster in specific landscape positions
8

depending on hillslope topography and the landslide triggering mechanism, e.g., earthquake
versus rainstorm (Densmore and Hovius, 2000; Meunier et al., 2008; Huang and Montgomery,
2014). Other studies have distinguished landslides that are “visibly connected” to river channels
and presumably available for fluvial transport (Hilton et al., 2011; West et al., 2011; Clark et al.,
2016). For the 1999 M
w
7.6 Chi-Chi earthquake in Taiwan, an estimated 8% of the
earthquake-triggered landslide population was connected to channels (Dadson et al., 2004), and
this connectivity showed little spatial variability (Hovius et al., 2011). The 2008 M
w
7.9
Wenchuan earthquake and associated widespread landsliding in China provides an opportunity to
explore systematically how and why landslide-landscape relationships vary spatially, and to
assess how this variability might regulate sediment export.

In this paper, we characterize the spatial distribution of co-seismic landslides associated with the
Wenchuan earthquake, allowing us to evaluate the extent to which landslide debris is delivered
directly to channels, and we then explore how this distribution affects post-earthquake sediment
export by rivers. We map hillslope and channel domains using a digital elevation model of the
regional topography, and we identify landslide populations in each setting using a co-seismic
landslide inventory. We define and quantify landslide-channel volumetric connectivity ( ξ) for the
Wenchuan landslide inventory and use a new catchment-averaged metric, the landslide location
index ( ψ), as a reference for comparison to our calculation of ξ. We find that landslide-channel
connectivity varies across the Longmen Shan region, and we investigate the effects of seismology,
topography and geology on the spatial variability of ξ, providing general insight into the factors
that determine connectivity between landslides and river channels. Finally, we assess the role that
landslide-channel connectivity may play in determining river sediment yields in the years
following the earthquake.

2.3. Setting
2.3.1. Topography, hydrology and climate
The Longmen Shan mountain range defines the eastern margin of the Tibetan Plateau and marks
the steepest topographic gradient among modern-day plateau edges (Burchfiel et al., 1995;
Densmore et al., 2007a) (Figure 2.1a). This region is characterized by a steep, high-relief margin
on the east, with > 5 km elevation rise over 50 km horizontal distance from the Sichuan Basin
(Burchfiel et al., 2008). Relief decreases westwards towards the Tibetan Plateau (Figure 2.2b, c,
d). Several large rivers, including the Min Jiang, Tuo Jiang, Fu Jiang, Jialing Jiang, Qingyi Jiang,
and Dadu He, drain the Longmen Shan range. River suspended sediment flux data indicate a
denudation rate of ~0.5 mm/yr during the decades before the 1990s (Liu-Zeng et al., 2011),
comparable to
10
Be-derived millennial denudation rates and geological exhumation rates
(Liu-Zeng et al., 2011; Godard et al., 2010; Ouimet, 2009). The regional climate is dominated by
the East Asian monsoon, with average annual rainfall varying from ~1100 mm at the margin to
~600 mm on the plateau. About 70%-80% of the precipitation occurs from June to September
(Liu-Zeng et al., 2011).

9

2.3.2. Geology
The bedrock geology of the Longmen Shan is dominated by a Proterozoic basement of granitoid
and high-grade metamorphic rocks, a Paleozoic passive margin sequence of metasediments and
granitic intrusions, and a thick Mesozoic foreland basin succession composed of marine and
clastic sediments (Burchfiel, 1995; Li et al., 2003; Densmore et al., 2007a) (Figure 2.1b). Three
large faults, the Wenchuan-Maowen fault, the Yingxiu-Beichuan fault, and the Pengguan fault, all
strike parallel to the Longmen Shan margin (Figure 2.1b). These faults were reactivated in the
India-Asia collision and have been active as dextral-thrust oblique-slip faults during the late
Cenozoic (Burchfiel et al., 1995; Densmore et al., 2007a; Wang and Meng, 2009). Before 2008,
GPS measurements showed slow deformation rates across the Longmen Shan range, implying
limited strain accumulation and low perceived seismic hazard (Zhang et al., 2004; Meade, 2007;
Kirby et al., 2008). The recurrence time for large, catastrophic earthquakes like the 2008
Wenchuan event in the Longmen Shan range is estimated to be ~2000-4000 years based on
geodetic and paleoseismic observations (Densmore et al., 2007a; Shen et al., 2009).

2.3.3. Seismology and landsliding
The M
w
7.9 Wenchuan earthquake occurred on May 12th, 2008. Rupture initiated in the southern
Longmen Shan and propagated for ~270 km along segments of the Yingxiu-Beichuan and
Pengguan faults (Figure 2.1c) (Burchfiel et al., 2008; Shen et al., 2009). Fault displacements and
seismic moment release varied along the rupture trace but were highest around Yingxiu and
Beichuan (Xu et al., 2009; Shen et al., 2009; Liu-Zeng et al., 2011). The motion along the fault
changed from predominantly thrusting in the southwest, near the epicenter, to strike-slip in the
northeast (Shen et al., 2009; Xu et al., 2009). The strong ground motion and intensive seismic
shaking caused over 60,000 co-seismic and immediately post-earthquake (defined here as
occurring within 6 months) landslides (Figure 2.1c) (Dai et al., 2011; Parker et al., 2011; Li et al.,
2014). Large increases in sediment fluxes have been observed after the earthquake from
hydrometric gauging of rivers (Wang et al., 2015) and
10
Be measurements on detrital quartz from
riverbed sediments (West et al., 2014). These methods both average over the spatial scale of river
catchments that span areas > 1000 km
2
and include thousands of landslides.

2.4. Materials and approaches
2.4.1. Landslide inventory mapping and volume estimation
Li et al. (2014) produced a co-seismic and immediately post-earthquake landslide inventory map
(Figure 2.1c) using high-resolution images collected within six months of the Wenchuan
earthquake from SPOT and DigitalGlobe satellites. Post-earthquake images were compared with
those collected before the earthquake. The mapping technique combined automated algorithms
and manual screening, allowing removal of non-landslide objects and segmentation of
amalgamated landslides, which can significantly bias volume estimates (Li et al., 2014; Marc and
Hovius, 2015). Landslides were mapped at 10 m spatial resolution. At this resolution, it was not
possible to separate depletion zones (i.e., landslide scars) versus accumulation zones (i.e,
deposits), so mapped landslide polygons include both scars and deposits. For this study, we have
10

slightly expanded the mapped region reported previously (Li et al., 2014), based on newly
available imagery, and using identical techniques (Figure 2.1b).

Landslide volumes were calculated using empirical area-volume scaling relationships (Guzzetti
et al., 2009; Larsen et al., 2010; Yanites et al., 2010; Parker et al., 2011; Li et al., 2014):

(2.1)

and

(2.2)

where A
ls
i
and V
ls
i
are the area and the volume for one single landslide respectively, V
ls
is the total
landslide volume, and α and γ are empirical scaling factors. The scaling factors α and γ and
related uncertainties were determined in the Longmen Shan based on field measurements of the
depths of landslide scars (see supporting information for details) (e.g., Parker et al., 2011;
Whadcoat, 2011). Uncertainties reported with the original publication of landslide volumes
included propagation of uncertainty on the scaling parameters (Li et al., 2014) but did not
account for additional uncertainty resulting from applying area-volume calibrations defined
principally by scar areas to mapped landslide areas that include both scars and deposits (Cruden
and Varnes, 1996). To constrain this additional uncertainty, we used high-resolution (0.5 m)
WorldView images, which allowed identification of scars and deposits. We estimated scar area as
a proportion of total landslide area for >500 landslides. Our mapping approach may overestimate
landslide areas by ~15-30%, depending on assumptions about the proportion of scar areas
covered by landslide deposits (see supporting information). In any case, this potential uncertainty
is much smaller than the ~+260%/-70% uncertainty arising from the scaling parameters alone (Li
et al., 2014). Since we lack imagery at sufficiently high resolution to distinguish scars from
deposits across the entire Wenchuan study area, in this study we use the uncertainties from
scaling factors, consistent with Li et al. (2014), acknowledging that the definition of landslide
areas introduces a minor additional bias. We note that distinguishing scars from deposits and
carefully considering what areas have been used in calibration datasets (e.g., Larsen et al., 2010)
would help reduce uncertainties on volume estimates in future landslide studies.

We have considered only co-seismic and immediately post-earthquake landslides in this study.
Observations of enhanced post-seismic landslide rates have been attributed to rock weakening
during shaking (Dadson et al., 2004; Marc et al., 2015), and post-seismic landslides may deliver
additional sediment to river systems (Dadson et al., 2004; Hovius et al., 2011). However,
following earthquakes in Taiwan, Japan, and Papua New Guinea, post-seismic landslides added
only a very small proportion (~2-5%) to the volume of co-seismic landslides (Marc et al., 2015).

 ) (
i
ls
i
ls
A V 


i
i
ls ls
A V

 ) (
11

For the Wenchuan case, local studies (catchments < 4% of the total landslide-impacted area)
show that rainfall-triggered, post-seismic landslides add 51% to the landslide number, 30% to the
landslide area, and ~20% to the co-seismic landslide volume (volumetric addition estimated
based on Eqs. (1) and (2)) in 2008, and 5% to the landslide volume in 2010 (Tang et al., 2011;
Zhang et al., 2014). The areas covered in these studies are in the frontal Longmen Shan, with the
most favorable conditions for post-seismic landsliding (proximity to the faults, highest
co-seismic shaking, and most intense rainfall). For the whole Longmen Shan, we expect
post-seismic landsliding following the Wenchuan earthquake to be relatively less important,
perhaps more analogous to the Chi-Chi case (Marc et al., 2015). Further work mapping
post-Wenchuan landslides would be needed to fully resolve this problem, but we do not expect
these additional landslides to significantly change the results and conclusions of this study.

2.4.2. Topographic and hydrographic mapping
For topographic analysis, we used 87 m resolution post-processed SRTM data from the
Consultative Group for International Agricultural Research (CGIAR) that includes regional void
filling using other data products (local data sources and SRTM 30 data) and re-interpolation
algorithms (Jarvis et al., 2008). The DEM data provides meaningful information about landslide
locations in the study area, since regional hillslope length scales are around 1 km (Kirby et al.,
2003). This length is equivalent to >10 DEM cells, sufficient to define hillslope and channel
morphology. We calculated gradient (G) and upstream contributing area (A) for each DEM raster
cell using the Spatial Analyst toolbox in ArcGIS. Although calculated gradients are strongly
dependent on DEM resolution (e.g., Larsen et al., 2014), the biases are systematic and the relative
trends between different sites should not be influenced. Since we are most interested in the
relative values of gradients and general patterns in this study, we used uncorrected values derived
from the SRTM DEM.

To map hydrographic networks, we used a geographic dataset of ordered catchments and
drainage basins from the Chinese Lake and Watershed Data Center (http://lake.geodata.cn/). We
focused on the three main large river catchments draining the Longmen Shan: the Min Jiang, Tuo
Jiang and Fu Jiang, which together cover ~90% of the total area affected by Wenchuan
co-seismic landslides. We mapped 26 sub-catchments and tributary catchments with reference to
the geographic dataset (Figure 2.2a). Sub-catchments were defined as constituent segments of a
main catchment along the main stem (for example, Min Jiang Pengshan-Dujiangyan segment,
catchment L7 on Figure 2.2a), and tributary catchments were defined as those secondary
catchments contributing to a main stem (for example, the Yuzixi catchment, catchment L4 on
Figure 2.2a). All the catchments were analyzed independently. Parts of the studied catchments are
monitored by the hydrological gauging network of the Chinese Bureau of Hydrology (Wang et al.,
2015) (Table 2.2).

12

Using the landslide inventory map, we quantified landslide impact by calculating the landslide
areal density P
Als
(%) for each catchment:

(2.3)

where A is the area of a selected catchment and A
ls
is the total area of landslides in the catchment.
Seventeen of the 26 tributary and sub-catchments had substantial landslide impacts, with P
Als

ranging from 0.1% to 5% (catchments L1-L17, Figure 2.2a, Table 2.2). Similarly, we also defined
the landslide volumetric density P
Vls
(m
3
km
-2
) for each catchment:

A
V
P
ls
Vls
 (2.4)

where V
ls
is the total volume of landslides in the selected catchment.

Channels across the study area were derived from the DEM by using gradient-upstream area
relations (Dadson et al., 2004; Meunier et al., 2008; Huang and Montgomery, 2014). In this study,
we specifically refer to “fluvial” channels, as distinct from “colluvial” channels and other types
of hillslope areas upstream of the channel head, in order to characterize landslide distribution
within the landscape. For each tributary and sub catchment, we plotted the G-A relationship for
all raster cells (Figure 2.3) and defined fluvial channels based on the expected power-law G-A
relationship (e.g., Montgomery and Foufoula-Georgiou, 1993; Montogmery and Buffington,
1997; Sklar and Dietrich, 1998; Montgomery, 2001; Stock and Dietrich, 2003; Densmore et al.,
2007b) (e.g., Figure 2.3a). Specifically, we fit a group of linear relations to segments of the
logarithmic G-A plots and identified five major geomorphic process domains: (1) hillslope, (2)
valley head, (3) colluvial, (4) bedrock and (5) alluvial (cf. Montgomery, 2001; Brardinoni and
Hassan, 2006). Modifying the “pruning” approach for determining power-law fits on G-A plots
(Stock and Dietrich, 2003; Densmore et al., 2007b), we first calculated a linear fit on logarithmic
G-A plots at the smallest upstream areas. We then successively added larger upstream areas and
refit the linear relation until a local optimal fit was identified (based on correlation coefficient
and mean squared residuals). Using this procedure, we identified domains with linear behavior
and uniform power-law exponent on logarithmic G-A plots. We repeated this approach, together
with visual examination, to define successive domains with higher A values in the logarithmic
G-A plots (e.g., progressing from hillslopes to valley heads to colluvial to bedrock and finally to
alluvial domains). The A value defining the transition between the colluvial and bedrock domains
(determined as the transition between domains (3) and (4); cf. Montgomery, 2001) was then
selected as a threshold area, A
min
, to represent a minimum upstream area for channelization, thus
distinguishing fluvial channels from hillslopes (Figure 2.S3 L1-L17) (Montgomery, 2001;
% 100  
A
A
P
ls
Als
13

Dadson et al., 2004; Brardinoni and Hassan, 2006; Meunier et al., 2008). We found that most
catchments yielded A
min
~ 1 km
2
. Some catchments showed more scatter and less clear transitions
among domains on the G-A plots (L6, L8, L10 and L12, see supporting information Figure 2.S1),
and so for those catchments we used a regional channel threshold A
min
of 1 km
2
, as derived from
the G-A relation combining all Longmen Shan catchments (Figure 2.3b).

Parts of catchments with A > A
min
were defined broadly as channel domains (including
alluvial-bedrock channels and alluvial channels), while Parts with A < A
min
were defined broadly
as hillslope domains, including colluvial channels, valley heads and strictly-defined “hillslopes”.
The latter are often characterized by a positive G-A exponent indicating convexity (Montgomery,
2001; Brardinoni and Hassan, 2006), but our general definition of hillslope as used in this
analysis simply refers to those regions outside of the fluvial network, rather than regions with
convex morphology per se.

To quantify the extent of fluvial channelization in each catchment, we calculated the drainage
density ( ρ, km km
-2
) as:

(2.5)

where L (km) is the total length of fluvial channels within the catchment (Dingman, 1978). We
observe a decline in drainage density from the basin toward the plateau (Figure 2.2g).

We plotted 170 km-wide swath profiles (Figure 2.2b, c, d, e, f and g) along the trend of the
steepest topographic gradient (NW-SE, perpendicular to the Longmen Shan faults). We projected
mean elevation, relief, gradient and catchment-scale drainage density  onto the swath trend (A-A’
on Figure 2.2). Catchment-scale relief was defined as the range of elevations within a 2.5
km-radius circle (e.g., Montgomery and Brandon, 2002; DiBiase et al., 2010). The same
parameters for each sub-catchment and tributary catchment were also projected onto the swath
profile trend A-A’ (Figure 2.2e, f and g).

2.4.3. Characterizing landslide locations and constraining uncertainties
To determine the locations of landslides relative to the fluvial network, we compared the
maximum upstream contributing area value within each mapped landslide polygon with the
threshold A
min
for channels within that catchment. If the maximum A for a landslide was larger
than A
min
, the entire landslide was assigned as connected to or located within the channel domain,
while landslides with maximum A values smaller than A
min
were defined as located on hillslopes.
This approach provides an algorithm for estimating whether the toe of a landslide intersects what
we have classified as a channel (e.g., Dadson et al., 2004; Meunier et al., 2010; Huang and
Montgomery, 2014). We then calculated the proportions of all landslides that are connected to the
river system in terms of numbers of landslides (“population connectivity”), landslide area (“areal
A
L
 
14

connectivity”), and landslide volume (“volumetric connectivity”). To determine volumetric
connectivity, we combined the landslide geomorphic classification and the landslide area-volume
scaling relation, using a Monte Carlo random sampling method to estimate uncertainties sourced
from the landslide area-volume scaling parameters α and γ in Eq. 2.2. We determined the
volumetric percentage of channel-connected landslides in the whole landslide inventory and for
landslides within individual catchments; we term this percentage the landslide-channel
volumetric connectivity, ξ (%), for each catchment (cf. Dadson et al., 2004; Meunier et al., 2008;
Hovius et al., 2011; Huang and Montgomery, 2014).

Whereas the landslides were mapped at 10 m resolution, the DEM had a coarser resolution of
~87 meters. The inconsistent resolutions between the landslide inventory and the DEM dataset
could introduce potential sampling bias: the actual upstream contributing area value of an
individual landslide cell may not be the same as the A value of the larger DEM cell. To constrain
the uncertainties that arise by extracting the landslide upstream area values from a coarser A
raster, we estimated the potential difference between sampling a 10 m-resolution raster dataset
and a 87 m-resolution raster dataset (details in supporting information). Our results show that the
maximum sampling error on A is around 0.01 km
2
, on the order of 1% of the threshold upstream
area for channels (~1 km
2
), and that the maximum error decays quickly as A grows (see
supporting information Figure 2.S2). For comparison to this theoretically predicted uncertainty,
we also considered the difference between calculations using 87 m vs. 30 m SRTM data. We
observed some spatial mismatch of channels extracted from these two DEMs in the same
catchment, and these differences introduced ~ 0.1 km
2
difference in calculated upstream areas
(see supporting information). This difference is ~10 times higher than the predicted uncertainty
arising from the difference in resolution alone, pointing to the importance of other factors such as
voids and DEM accuracy. We conclude that our evaluation of connectivity is not strongly biased
by this level of uncertainty.

To evaluate the sensitivity of connectivity to channel threshold A
min
, we calculated connectivity
for each catchment across a wide range of A
min
values as reported in the Longmen Shan and other
mountain belts (~0.3-3 km
2
) (Kirby et al., 2003; Montgomery, 2001; Dadson et al., 2004;
Meunier et al., 2008) (supporting information Figure 2.S3). We found that the connectivity
determined using variable A
min
differed by ~20% (relative percentage) when compared to using
the catchment-specific A
min
determined in this study (supporting information Figure 2.S4),
leading us to conclude that calculated connectivity is relatively insensitive to uncertainty in A
min
.

2.4.4. Catchment-scale landslide location index
To complement the gradient-upstream area approach, we propose a new metric that considers
landslide locations at the catchment scale. The “landslide location index” ( ψ, dimensionless) for
individual catchments characterizes how landslides are distributed relative to the background
landscapes in upstream contributing area (A) space, with no assumptions of channel threshold
A
min
. To calculate ψ for each catchment, we integrated below the cumulative A distribution (cf.
15

Figure 2.4a) to derive (i) an integrated area (IA
ls
) for the cumulative landslide volume-A
distribution curve, and (ii) an integrated area (IA
c
) for the cumulative catchment DEM cell-A
distribution curve. We then calculated each catchment’s landslide location index ( ψ) as (Figure
2.4a and 4b):

(2.6)

For a higher ψ in a given catchment, the landslide inventory is preferentially located at larger
upstream areas, and should have a higher potential to connect to channels (Figure 2.4c and 4d). A
medium ψ represents less potential of landslide-channel connection compared to a higher ψ
regime (Figure 2.4e and 4f). A lower ψ suggests lower potential to connect to channels (Figure
2.4g and 4h). Since ψ considers where landslides are located with respect to the distribution of
upstream contributing area in a catchment, higher values should reflect landslide positions
characterized by greater flow accumulation for fixed hydrologic conditions.

2.5. Results and Discussion
2.5.1. Landscape position and connectivity of Wenchuan-triggered landslides
2.5.1.1. Cumulative landslide volume curves
To gain a general perspective on the locations of landslides in relation to the morphology of the
Longmen Shan catchments, we report the cumulative volumetric fraction of landslides as a
function of upstream area (Figure 2.5 and Figure 2.S5). For all landslides in the study area, a high
proportion of total landslide volume is located in the hillslope domain (A < ~1 km
2
,
the threshold
A
min
for the Longmen Shan catchments), as indicated by the steep rise in cumulative volumetric
fraction as a function of A (Figure 2.5a). A lower proportion of total landslide volume (reflected
by more gentle rise in the cumulative curves on Figure 2.5a) is located in the channel domain (A >
1 km
2
). For the three main catchments, different proportions of landslides in different domains
lead to varied patterns in the cumulative volume curves (Figure 2.5b, c and d). Large landslides
introduce significant discontinuities to the distribution curves, because the non-linear
volume-area relationship (  > 1) means that they contribute disproportionately to total sediment
volume (Larsen et al., 2010; Guzetti et al., 2010; Li et al., 2014; Marc and Hovius, 2015). The
marked increase in landslide volume at A ~ 30 km
2
is caused by the Daguangbao landslide, the
largest landslide in the Wenchuan inventory, with an area of ~7.2 km
2
, around two orders of
magnitudes larger than the median area (Chen et al., 2014). The Daguangbao landslide was
located in the Kai Jiang tributary of the Fu Jiang catchment and causes discontinuities in the
distribution curves of these two catchments (Figure 2.5d and L10). For sub-catchments and
tributaries (L1-L17), two types of landslide distribution curves are observed (Figure 2.S5): some
catchments (e.g., L2 and L5) show similar patterns to the overall landslide inventory, while
others (e.g., L7 and L16) are influenced by very large landslides with sharp rises in
the V
ls
distribution curve in domains with higher upstream areas.
ls
c
IA
IA
 
16

2.5.1.2. Region-wide connectivity values
Based on the landslide upstream contributing area values and the threshold A
min
values, we
estimate that ~16% of the landslide population (in terms of number of landslides) is connected to
river channels. Estimates of the number of channel-connected landslides associated with the
Chi-Chi earthquake in Taiwan are lower (~8% of the total), while those triggered by Typhoon
Herb are higher (~24%), a difference that may be attributed at least in part to clustering of
co-seismic landslides at hillslope crests (Dadson et al., 2004). As far as we are aware, directly
comparable estimates of areal or volumetric connectivity are not available for the Chi-Chi
earthquake or for other events.

For the Wenchuan-triggered landslides, 30% of total landslide area is connected to rivers, higher
than the 16% population connectivity. The difference between these values indicates that larger
landslides (by area) are more likely to be connected to channels. The Wenchuan landslide
inventory follows a heavy-tailed distribution and can be fit by an inverse gamma function (Figure
2.6) (e.g., Malamud et al., 2004) defined by:

(2.7)

where A
ls
represents individual landslide areas, p is the probability density, and q, m and s are
inverse-gamma function parameters. Landslides in the hillslope and fluvial domains both follow
inverse-gamma distributions (Eq. 2.7), but with different parameters, confirming that landslides
in the channel domain on average have larger areas than those in the hillslope domain (Figure
2.6a). Corroborating this interpretation, we group landslide populations based on areas and
observe a well-defined positive correlation between average landslide area and average
landslide-channel connectivity (Figure 2.6b). The influence of landslide area on connectivity is
consistent with the expectation that, due to self-similar properties, larger landslides generally
have longer lengths (L ~ A
ls
1/2
) (Hovius et al., 1997; Bellugi et al., 2015), and are more likely to
reach hillslope bases and connect to channels.

The volumetric connectivity of the Wenchuan-triggered landslides is 43+9/-7% (median ± 5
th
/95
th

percentiles from 1,000 Monte Carlo simulations, propagating uncertainties from landslide
area-volume scaling; Figure 2.5a). Volumetric connectivity is higher than areal connectivity
because of the non-linear area-volume scaling relationships and the above-mentioned pattern of
higher connectivity for larger area landslides. The remaining 57+7/-9% of the total landslide
volume is located, at least temporarily, on hillslopes.















 


s A
m
s A
m
q m
s m q A p
ls
q
ls
ls
exp
) (
1
) , , ; (
1
17

2.5.2. Comparing landslide connectivity and catchment-scale location index
We calculated the catchment-scale landslide location index ( ψ) for the 17 sub-catchments and
tributaries impacted by landslides (Table 2.2). All ψ values, which range from 1.2 to 1.8, are
greater than one. The maximum ψ in the Wenchuan case is 1.8, for the Kai Jiang tributary, and is
caused by the large Daguangbao landslide, which is located in the fluvial domain and contributes
to the much lower IA
ls
(thus higher ψ) for this catchment.

Because landslide-channel connectivity ( ξ) and landslide location index ( ψ) both quantify
landslide distribution as functions of upstream area, ξ should theoretically correlate well with ψ.
Connectivity ( ξ) depends on the determination of channels, which might be complicated by
multiple factors (Montgomery, 2001). In contrast, location index ( ψ) does not rely on how
channels are defined and thus provides an independent representation of landslide location. For
our data, the mean ξ in a given catchment is positively correlated with ψ (r
2
= 0.78, p < 0.001)
(Figure 2.7), as we expect given that they represent similar characteristics of the landslide
distribution. A principal difference between the two metrics is that, unlike ξ, the value of ψ will
be influenced by the relative position of landslides with respect to distance from headwaters, and
thus to some extent with flow accumulation at the landslide locations. We may thus expect ψ to
more generally reflect the influence of landslide position on sediment transport, a question we
consider further in Section 2.5.4.

2.5.3. What determines the connectivity of Wenchuan-triggered landslides?
2.5.3.1. Spatial patterns of landslide-channel connectivity
A central observation from our analysis is that landslide-channel connectivity is not uniform
across the Wenchuan earthquake-affected region. In this section, we consider how and why
connectivity varies spatially across the Longmen Shan. To illustrate the spatial patterns of the
landslide inventory, we mapped areal densities for total landslides and channel-connected
landslides, as well as landslide-channel connectivity over the three main study catchments
(Figure 2.8). All landslides and channel-connected landslides have similar spatial distributions,
with higher areal densities in the hanging wall of the Yingxiu-Beichuan fault and lower densities
towards the Sichuan Basin and the plateau (Figures 8a,b; e.g., Dai et al., 2011; Gorum et al.,
2011). The spatial pattern of landslide-channel connectivity is less distinct, although a clustering
around the middle of the Yingxiu-Beichuan fault rupture is evident (Figure 2.8c).

When plotted in 5 km corridors along the steepest topographic gradient (A-A’), all landslides and
channel-connected landslides show clear clustering around the Yingxiu-Beichuan fault (Figures
8d,e) (e.g., Dai et al., 2011). Except for some statistically less significant landslide groups (with
landslide populations of < 20), landslide-channel connectivity shows a similar general trend,
peaking around the Yingxiu-Beichuan fault and decaying towards the Sichuan Basin and the
Tibetan Plateau (Figure 2.8f).

18

2.5.3.2. Controls on the landslide-channel connectivity
The spatial variability in landslide-channel connectivity for the Wenchuan earthquake may
provide general insight into what factors set the amount of landslide sediment delivered directly
to river systems. To achieve high connectivity, landslides need to reach the hillslope base, and
channel densities need to be high enough to sample a large number of landslides. Several factors
may determine connectivity by influencing landslide sizes (and thus likelihood of reaching rivers)
or regional channel densities. Here, we focus on how topography, seismic intensity, and lithology
influence landslide sizes, and on how topography and lithology influence channel densities. We
explore each factor independently, acknowledging that there will be coupling and
inter-correlation among them.

Topographic control on connectivity: To constrain the role of topography, we group landslides in
the Wenchuan inventory by bins of mean gradient, mean elevation, and mean relief for all DEM
cells within each landslide polygon extent, with bin sizes of 1° for gradient and 100 m for
elevation and relief. We calculate the landslide-channel connectivity ξ within each group (Figure
2.9a, b and c). Landslide-channel connectivity and landslide gradient (defined as the mean
gradient for all DEM cells within each landslide polygon) show a well-defined negative trend
(Figure 2.9a), except for a few landslides at the highest observed gradients. The general pattern of
lower ξ as gradients increase can be attributed to the negative relationship between mean
catchment gradient and drainage density (Figure 2.S6a), implying a geomorphic control on
channel distribution. High gradient areas have smaller upstream areas to support fluvial channels,
thus leading to lower channel densities and lower landslide-channel connectivity. Gradients also
have an effect on landslide size (Figure 2.S6b) with steeper gradients facilitating larger landslides
(meaning higher landslide-channel connectivity), competing with the gradient effect on drainage
density. The overall negative trend between ξ and gradient indicates that the effect on drainage
density is dominant. The effect of gradient on landslide size is most relevant for the steepest areas
(>60°) where a small fraction (<1%) of landslides are located (grey dots on Figure 2.S6b).
Elevation and relief couple tightly with gradient in the Longmen Shan. Low-elevation areas are
limited to proximal valley floors and the Sichuan Basin, dominated by fluvial processes with
dense fluvial networks (Figure 2.2g), leading to high probability for landslide-channel connection.
Medium- to high-elevation areas represent mountain and plateau regions which have lower
drainage density and are thus less prone to landslide-channel connection (Figure 2.2b and Figure
2.9b). Similarly, lower relief occurs in either basin or plateau areas, with higher relief in between.
Landslides in low-relief areas occur mostly in plateau regions where drainage density and
associated landslide-channel connectivity are low (Figure 2.9c). Medium-high relief areas are
mainly found in the eastern Longmen Shan, where most earthquake-triggered landslides occurred,
with medium-high drainage density, representing favorable conditions for landslide-channel
connection (Figure 2.2g and Figure 2.9c). Very high-relief areas may represent some local areas
with lower drainage density and consequently lower landslide-channel connectivity. Overall, the
observed correlations suggest that topography influences landslide-channel connectivity mainly
via controls on channel densities.
19

Lithological control on connectivity: We calculated ξ for different lithological units (China
Geological Survey, 2004) grouped both (i) as metamorphic, igneous and sedimentary rocks and
(ii) as the main sub-types outcropping in the Longmen Shan range. The metamorphic rocks in the
study area are mainly composed of low- to medium-grade units (e.g., slate, phyllite, schist,
meta-sandstone and marble). The major igneous rock is granitic. For sedimentary rocks, we
identify three sub-types: carbonate, mudstone (fine clastic material) and sandstone (coarser
clastic material). Across most lithologies, ξ is similar, but values are significantly higher for
sandstone and mudstone and lower for slate (Figure 2.9d). In conjunction with other factors like
precipitation, surface/subsurface hydrology and topography, lithology influences drainage
densities (Day, 1980; Tucker and Bras, 1998; Moglen et al., 1998; Duvall et al., 2004; Luo and
Stepinski, 2008) and thus landslide-channel connectivity. We observe varied drainage density
among different lithological units and a good correlation (r
2
= 0.80) between connectivity and
drainage density for each lithological unit, excluding the < 2% landslide population from the slate
unit (Figure 2.9e). Landslides in slate lithologies are mostly located at the northern end of the
main fault rupture, in an area that is characterized by dominantly strike-slip fault motion, low
topographic relief, and low PGA, potentially explaining their anomalously low connectivity.
Lithology-dependent rock strength might also influence landslide areas, and thus connectivity,
but we did not find a statistically significant correlation between landslide area and connectivity
among different lithologies. Our lithological classification is inevitably simplified and does not
account for the complex interactions between lithology, rock strength, erodibility and landsliding
(Montgomery, 2001; Chen et al., 2011; Gallen et al., 2015). Nonetheless, our observations point
to a potential role for substrate properties in modulating landslide-landscape relations,
particularly through setting drainage density.

Seismic control on connectivity: Previous studies have shown that the spatial distribution of
earthquake-triggered landslides is determined by the patterns of peak ground acceleration (PGA)
(e.g., Keefer, 1984; Jibson and Keefer, 1993; Meunier et al., 2007; Meunier et al., 2008; Kritikos
et al., 2015). Correlations between landslide densities and local PGA were also observed for the
Wenchuan earthquake (Dai et al., 2011; Yuan et al., 2013; Gallen et al., 2015). Here we find a
positive correlation not only between landslide occurrence and PGA, as observed previously, but
also between landslide-channel connectivity ξ and PGA (Figure 2.9f). This correlation can be
explained by the observed dependence of individual landslide area on PGA (Figure 2.S6c):
higher seismic intensity and stronger ground motions tend to trigger larger landslides, which are
more likely to reach channels. As expected, there is no correlation between PGA and drainage
density (Figure 2.S6d).

Spatial patterns of earthquake-triggered landslides are also sensitive to the sense of motion on the
fault (Meunier et al., 2008; Barlow et al., 2015; Gorum and Carranza, 2015). The Wenchuan
landslides occurred in areas of complex faulting, varying along strike from dextral-thrust to
nearly pure strike-slip motion (Xu et al., 2009; Liu-Zeng et al., 2011; Gorum and Carranza, 2015).
We classified the Wenchuan landslides following the approach of Gorum and Carranza (2015),
20

projecting landslides to the nearest segment of the fault rupture that was categorized based on the
predominant type of motion (thrust, oblique-slip and strike-slip) (Liu-Zeng et al., 2009). We
calculated the corresponding landslide-channel connectivity for the different types of faulting
along each segment (Figure 2.9g). We acknowledge that landslides at any given site may be
triggered by energy from different segments along the rupture, and not just by the nearest
segment, but PGA values show limited variation along the main fault rupture, so we do not
expect large biases.

The average landslide-channel connectivity for landslides in areas dominated by thrust and
oblique-slip segments ( ξ ~ 37+4/-3%, reported as median ± 5
th
/95
th
percentiles from 1000 Monte
Carlo simulations, propagating uncertainties from landslide area-volume scaling, and excluding
the anomalously large Daguangbao landslide) is very slightly higher than for landslides adjacent
to strike-slip segments ( ξ ~ 32±1%) (Figure 2.9k). This finding is consistent with the fact that the
mean landslide area near thrust and oblique-slip segments (~ 8300 m
2
, excluding the Daguangbao
landslide) is larger than that near strike-slip segments (~ 6300 m
2
), which may be explained by
the spatial clustering of the seismic moment release around Yingxiu (dominated by thrust faulting)
and Beichuan (Shen et al., 2009; Parker et al., 2011). Several modeling studies have shown that,
for the same magnitude of initial stress, thrust or reverse dip-slip faults can generate stronger
ground motion compared to strike-slip faults (e.g., Oblesby and Day, 2002; Gabuchian et al.,
2014). Such relationships would mean both a higher susceptibility to landsliding and larger
landslides, and consequently higher channel connectivity (Figure 2.6b), in thrust earthquakes
compared to strike-slip events, consistent with the variations of landslide-channel connection that
we observe.

Coupling of effects: In some cases, the factors discussed above will be inter-related; for example,
multiple factors could contribute to the observed lower connectivity for landslides in slate (Figure
2.9f), but their interrelations cannot be untangled with available data. Nonetheless, we do not
observe highly systematic spatial correlations among topography, lithology and seismic
parameters across the Longmen Shan, suggesting that our analysis still provides first-order
insights into the roles of these parameters in controlling landslide-channel connectivity.

2.5.4. Implications for post-Wenchuan sediment transport
For the Wenchuan case, 43+9/-7% of the co-seismic landslide volume is directly connected to
channels, representing ~1.4 km
3
of the ~3 km
3
total landslide volume in the Longmen Shan
catchments. The remaining 57+7/-9% of landslide sediment volume (~1.6 km
3
) resides higher on
hillslopes. This landslide-channel connectivity sets an initial condition for post-Wenchuan
sediment transport. Channel-connected landslides are expected to have high potential for fluvial
evacuation in the monsoonal climate (Liu-Zeng et al., 2011; Wang et al., 2015), whereas
landslide material in the hillslope domain should be less immediately available for river transport
(Meunier et al., 2008; Dadson et al., 2004; Hovius et al., 2011; Huang and Fan, 2013; Tsai et al.,
21

2013). These predictions can be tested by evaluating relationships between post-earthquake
sediment fluxes and the landslide inventory.

We currently lack constraints on bedload sediment transport following the Wenchuan earthquake,
but Wang et al. (2015) reported suspended sediment (predominantly material <0.25 mm diameter)
fluxes, allowing us to examine transport of the fine landslide material. We calculated the
differences in suspended sediment fluxes for nested catchments reported in Wang et al. (2015),
converting the total fluxes above each gauging station to a suspended sediment yield for
individual sub-catchments and tributary catchments (see details in supporting information). This
approach returns negative values for two out of the 16 catchments where data are available, likely
due to large sediment sinks such as reservoirs that are not accounted for in this analysis
(catchments labeled as “N.A.” in Figure 2.10a and Table 2.3). These two catchments were
excluded from the following analysis. For the other catchments, we normalized the estimated
landslide density and sediment yield to the fraction of mountainous area (defined here as
elevation > 800 m, Table 2.3) in the catchment. Several gauging stations located at further
downstream sites also include large floodplain areas that contribute little to landsliding and
sediment export, so we excluded these areas by using a threshold elevation.

Across the Longmen Shan catchments, we observe a positive correlation (Figure 2.10b, r
2
= 0.40,
p < 0.05) between total landslide volumetric density P
Vls
and post-seismic (2008/6-2008/12)
suspended sediment yield. This relationship is consistent with landslides being a significant
source of sediment following the earthquake. A similar positive correlation (r
2
= 0.66, p < 0.05) is
observed between P
Vls
and suspended sediment yield over the following three years (2009-2012)
(Figure 2.10c). Although we do not account for post-seismic landslides in our correlation analysis,
we expect that the <~20% additional post-seismic landslide volume (see above, and Tang et al.,
(2011)) would not significantly affect our conclusions.

To examine the role of connectivity, we regressed P
Vls
of channel-connected landslides with
suspended sediment yield (Figure 2.10d and 10e). We find that there is no statistically significant
difference in the correlation coefficients between total landslide density and suspended sediment
yield on the one hand, and between channel-connected landslide density and suspended sediment
yield on the other (examined by Meng’s Z test (p>0.6) (Meng et al., 1992)). The correlation
coefficients are statistically indistinguishable for total and channel-connected landslides whether
considering the 2008 sediment flux data or the 2009-2012 data.

The normalized residuals from the regression between total landslide density and suspended
sediment yield provide further insight into the possible role of connectivity in sediment transport.
If landslide locations relative to channels play an important role in explaining post-earthquake
sediment fluxes, we should find that these residuals are positively related to connectivity or
location index. Indeed we find a weak but statistically significant relationship (r
2
= 0.26, p<0.1)
between residuals and location index ( ψ) for the sediment yield measured in 2008 (Figure 2.10f).
22

Connectivity ( ξ) shows no similar correlation. By using a constant threshold area, the
connectivity index does not account for the greater efficacy of sediment transport in higher-order
channels, which is encapsulated in the location index, possibly explaining why ψ might better
describe the potential for sediment transport. Nonetheless, the relationship between the residuals
and ψ disappears for the sediment yield measured between 2009 and 2012 (Figure 2.10g). We
speculate that this observation can be explained by an initially weak influence of landslide
location on fine sediment fluxes (as observed in the 2008 data), with the influence of landslide
locations fading over time. This could be because fine-grained landslide material is relatively
readily entrained from landslide deposits even in the hillslope domain, for example via overland
flow during monsoonal storms. It is also possible that any signal in the 2009-2012 data is
convoluted by other effects, including post-seismic landslides, although we expect these to make
a relatively minor contribution as discussed above.

Fine sediment only represents a small proportion (<10 weight %) of the total volume of material
from Wenchuan earthquake-triggered landslides, leaving coarse material (>0.25 mm) as the
dominant component (Wang et al., 2015). We expect that the supply of coarse material to river
channels may be more affected by landslide-channel connectivity than the supply of fine-grained
material, since coarse material is likely to be less easily mobilized from hillslopes. We currently
do not have data constraining coarse sediment fluxes from multiple catchments in the Longmen
Shan, but such information would be valuable for more completely testing conceptual models for
landslide sediment transport.

2.6. Conclusions
We have systematically explored the locations of landslides triggered by the 2008 Wenchuan
earthquake within the fluvial network across the Longmen Shan range at the eastern edge of the
Tibetan Plateau. We quantified landslide-channel connectivity in terms of landslide volume, area
and number, and we examined how volumetric connectivity ξ varies spatially in order to
understand what controls the supply of sediment for fluvial evacuation. Finally, we have
considered our connectivity results in the context of sediment transport following the Wenchuan
earthquake. Several key findings contribute to better understanding of co-seismic landslides as
sediment sources:

(1) For the co-seismic landslide inventory within the three main catchments draining the
Longmen Shan (covering over 90% of the total landslide-impacted area), 16% of the total
landslide number, 30% of the total landslide area, and 43+9/-7% of the total landslide volume
(~1.4 km
3
) is directly connected to fluvial channels and thus prone to entrainment and transport
by rivers. The remaining 57+7/-9% by volume (~1.6 km
3
) was deposited higher on hillslopes,
beyond the immediate extent of the fluvial channel network. If connectivity plays an important
role in sediment dynamics, we would expect this material to be unavailable for immediate
transport.

23

(2) The catchment landslide location index ψ, which describes the relative distribution of
landslides versus catchment topography as a function of upstream contributing area, provides an
additional constraint on landslide locations in landscapes independent of channel definition. We
find a positive correlation across different catchments between ψ and our determination of
landslide-channel connectivity ( ξ) based on gradient-upstream area relations, suggesting these
metrics are consistent. We suggest that ψ may provide a complementary index that also reflects
upstream area and thus may capture the more general influence of landslide position on sediment
transport, as hinted by sediment flux data following the Wenchuan earthquake.

(3) Landslide-channel connectivity is linked to topographic parameters, specifically to gradient
and drainage density, which are themselves correlated. Lithology is also an important control on
connectivity, with higher volumetric landslide-channel connectivity ( ξ ~ 60-80%) for clastic
sedimentary bedrock and lower (10-20%) for high-grade metamorphic bedrock. Our analysis also
suggests higher landslide-channel connectivity in areas with higher PGA and near fault segments
dominated by thrust and oblique slip during the Wenchuan earthquake. The coupling of these
factors may also contribute to the observed landslide-channel connectivity pattern.

(4) The correlation between suspended sediment yield and the volumetric density of
channel-connected landslides is statistically indistinguishable from the correlation with the
volumetric density of all landslides. Residuals from the correlation between sediment yield and
the density all landslides are weakly related to location index ψ for data from 2008, immediately
following the earthquake, but not to location index for data from the ensuing years. This suggest
a weak initial influence of connectivity on fine sediment fluxes that decreases over time. This
observation may be related to relatively rapid mobilization of fine-grained material from
hillslopes, and future work needs to establish whether the fluvial transport of coarser fractions of
landslide debris is affected more significantly by landslide locations.

Overall, our results shed light on landslide locations in landscapes and how landslide-channel
connection regulates sediment transport after large earthquakes. This work provides an important
database for future research on sediment dynamics following the Wenchuan earthquake. The
framework and methodology developed in this study are also applicable to other earthquakes in
similar settings, promising greater understanding of the role of rare, high-magnitude seismic
events in regulating sediment transport processes, in influencing associated sediment-related
hazards, and in the long-term tectonic and topographic evolution of mountain belts.

2.7. Supplementary materials to Chapter 2
2.7.1. Uncertainty on landslide volume arising from mapping approach
Because of limited image resolution, our mapping did not separate landslide scars from deposits.
Potential bias in the landslide volume estimates may therefore result when applying empirical
areal-volume scaling relations. To constrain the resulting uncertainty, we mapped a subset (n=505)
of co-seismic landslides using high resolution (~0.5 m) WorldView images, which allow
24

differentiation of landslide scars versus deposits. We determined the fraction, f
scar
, of the total
mapped area for each landslide polygon that is landslide scar. Because scars possibly overlap
with deposits, the scar mapping approach may fail to capture parts of scars under deposits. Thus
the estimated f
scar
is a lower bound on the scar fraction. Based on the newly-mapped,
scar-differentiated landslides, we estimated f
scar
for the Wenchuan landslides as 70 ± 6% (mean ±
one standard deviation).

The volume of one landslide can be estimated from the geometry of the scar:

V
scar
= f
scar
× A
total
× d
scar
(2.S1)

where V
scar
is the scar-based landslide volume, A
total
is the total area of a landslide from mapping,
and d
scar
is the scar’s depth. Whadcoat (2011) calibrated field-measured landslide scar depths to
the mapped total landslide area as d
scar
× A
total
= αA
total
γ
, which becomes:

d
scar
= αA
total
γ-1
(2.S2)

where α and γ are scaling factors (values reported in Parker et al., 2011; Whadcoat, 2011; Li et al.,
2014). Combining (S2) and (S3), we have:

V
scar
= f
scar
× α × A
total
γ
(2.S3)

For comparison, landslide volume calculated from total mapped landslide area is:

V
total
= αA
total
γ
(2.S4)

Comparing Eq. 2.S4 to Eq. 2.S3, we find that the total landslide area-based volume may have
systematically over-estimated the landslide volume by a maximum of ~30%, considering that our
estimate of f
scar
is a lower bound on the actual scar fraction. If we assume a uniform distribution
of scar areas covered by deposits, we can correct for this effect as:

f
scar
* = (1+f
scar
)/2 (2.S5)

and in this case, the overestimation of landslide areas is 15±6% of landslide volume. In
comparison, the uncertainties associated with the scaling parameters α and γ are ~+260%/-70%
from a Monte Carlo-based approach [Li et al., 2014], much larger than the potential error
introduced by the mapping that did not distinguish scars from deposits.

2.7.2. Uncertainty arising from differences in raster resolution
In defining landslide position with respect to upstream contributing area, a potential bias may be
caused by the fact that, for a low-resolution raster a (e.g., at the DEM resolution) one large cell
25

with a given A value will overlap with several small cells with variable ‘true’ A values in a
high-resolution raster b (e.g., at the resolution of satellite imagery). If we sample one point from
the many smaller cells in b bounded by the extent of the a-cell, we may get various results
depending on which cell is chosen. As rasters a and b describe the same variable in the same
coordinate system, the high-resolution raster b should cover the information carried by raster a,
and thus the raster a-cell value should be within the range of the values from the corresponding
b-cells.

On a GIS platform, the upstream contributing area for each cell is calculated by multiplying the
flow accumulation number (N) and the cell area (W
2
, W is the cell width). Within the boundary of
one raster a-cell with width W
a
, the approximate number of raster b-cells (cell width W
b
) can be
calculated as:

(2.S6)

If we sample one point from one a-cell with flow accumulation number N
a
, the upstream
contributing area for this point is always determined as N
a
(W
a
)
2
, independent of where this point
is taken within the area bounded by this raster a-cell. Ideally, the corresponding flow
accumulation number in raster b, N
ba
, which is within the raster a-cell and gives the exact same
upstream contributing area, can be calculated as:

(2.S7)

and

(2.S8)

Practically, the sampling point in raster b may capture a different flow accumulation number N
b
.
The difference in upstream contributing areas calculated from N
b
and N
ba
is the sampling
uncertainty. As flow accumulation number is essentially a count of relevant cells, the difference
between N
b
and N
ba
can be constrained considering two end-member cases.

2









b
a
W
W
n
2 2
) ( ) (
b ba a a
W N W N   
a ba
nN N 
26

In one case, all the other b-cells within the a-cell-sized area contribute downstream flow to the
b-cell valued N
b
, and the b-cell N
ba
is the most upstream cell with no input from other cells in this
region. In this case:

N
b
-N
ba
= n-1 (2.S9)

where n is the total number of b-cells within one a-cell-size area.

Similarly we can consider the other case, where all the other b-cells bounded by the a-cell are the
relevant upstream cells for the N
ba
b-cell and N
b
has no other upstream cells in the large a-cell.
We then get:

N
b
-N
ba
= -(n-1) (2.S10)

These two end-member cases bound the range of N
b
with a given N
ba
as:

(2.S11)

Next we derive the uncertainties for upstream contributing area (A). The A values derived from
raster a and b can be written as:

(2.S12)

and

(2.S13)

Here we define the sampling uncertainty of A as:

(2.S14)

)] 1 ) (( ) ( ), 1 ) (( ) ( [
2 2 2 2
    
b
a
b
a
a
b
a
b
a
a b
W
W
W
W
N
W
W
W
W
N N
2
) (
a a a
W N A  
2
) (
b b b
W N A  
% 100 (%) 


a
b a
A
A A

27

Taking into account Eq. 2.S6, the maximum uncertainty is:

(2.S15)

Combining Eq. 2.S6 and Eq. 2.S15, we have:

(2.S16)

When A
a
= A
b
, we have the minimum uncertainty:
(2.S17)

Taking W
a
= 87 m and W
b
= 10 m, the resolutions of the DEM and the images used for mapping,
respectively, we estimate the theorectical uncertainty arising from the different resolutions as a
function of upstream contributing area A
a
(Figure 2.S2a).

Note that the above analysis represents a theoretical evaluation of the uncertainty associated with
the sampling approach and assumes idealized DEM data (i.e., no sinks or voids and no spatial
uncertainty). However, our estimate is not equivalent to the actual sampling difference between
different DEM data. In practice, the difference from sampling resolution-differed DEM data can
be largely influenced by the quality and uncertainty of actually adopted DEM data. Also,
high-resolution DEM data likely better captures complex local features than a low-resolution
DEM. Thus the channels extracted from the high-resolution DEM may not match those defined
by a low-resolution DEM, leading to sampling errors.

By visual examination of the channels extracted from SRTM 30 data versus SRTM 90 data for a
studied catchment (L14), we clearly observe mismatches between some extracted channels
(Figure 2.S2b). For all channel cells in this catchment, we overall calculate a median sampling
error on upstream contributing area of around 0.1 km
2
. This means that the median sampling
uncertainty decays from ~10% at 1 km
2
to ~0.1% at 10 km
2
(Figure 2.S2c), which is relatively
small but higher than our theoretical estimate (<0.01 km
2
, Eq. 2.S15). For all DEM cells within
the catchment, 85% of the cells show sampling error within 0.1 km
2
, and 90% show sampling
error within 0.2 km
2
. We attribute this observed sampling uncertainty to a combination of the
resolution influence on channel delineation and the accuracy and potential defect in the DEM
data. Nonetheless, our evaluation of connectivity is not strongly biased by this level of
uncertainty, as discussed in the main text. The approach presented in this study can be used
elsewhere but the results may be influenced by the quality of DEM data used.
% 100
) 1 (
(%)
2
max



a
b
A
W n

% 100 (%)
2 2
max



a
b a
A
W W

0 (%)
min
 
28

2.7.3. Mass balance approach for deriving sediment yield in sub-catchments and tributaries
In Wang et al (2015), total sediment fluxes above gauging stations are reported, representing
fluxes from assemblages of upstream sub-catchments and tributaries. Here we partition the total
sediment fluxes to each individual sub-catchment and tributaries based on mass balance
principles. For tributary catchments where gauging data is available (e.g., G2 Guojiaba),
sediment yields are directly calculated as sediment flux divided by corresponding upstream
contributing area. For sub-catchments of larger, higher-order rivers, sediment yields are derived
from differences between sediment fluxes at downstream gauging stations minus fluxes at
upstream stations (e.g., sediment flux in G4 Lower Zagunao is equal to the difference between
sediment fluxes gauged at the downstream site Sangping minus the flux at upstream site Zagunao,
and then converted to yield).

Figur

Figur

Figur
Sichu
three
litholo
(p Є)
Ordov
(inclu
seque
2,500
mecha
2011)
Wenc

res and Table
e 2.1
e 2.1. Maps
uan Basin. (a)
large river c
ogical units a
granitoids a
vician, S: S
uding T: Tria
ences, and li
,000 China G
anism, afters
), co-seismic
chuan earthqu
es
of the Long
), shaded reli
atchments (M
and faults at
nd high-grad
ilurian, D: D
assic, J: Jur
mited Cenoz
Geological B
shocks, PGA
landslides (L
uake (Liu-Zen
gmen Shan r
ief map (from
Min Jiang, Tu
the eastern T
de metamorp
Devonian, C
assic and K
zoic (includin
Base Map (Ch
contours (U
Li et al., 2014
ng et al., 2009

29
region of the
m SRTM-der
uo Jiang and
Tibetan Platea
phic rocks, P
C: Carbonifer
K: Cretaceous
ng Q: Quate
hina Geologi
USGS, 2008)
4), and fault r
9).
e eastern Tib
rived DEM)
Fu Jiang) ou
au region, inc
Paleozoic (in
rous and P:
s) passive m
ernary) sedim
ical Survey, 2
, surface def
rupture assoc
betan Plateau
of the study
utlined in bla
cluding main
ncluding Є:
Permian) a
margin and f
ment, modifi
2004). (c), ep
formation (F
ciated with th
u and adjacen
area with th
ack. (b), majo
nly Proterozoi
Cambrian, O
and Mesozoi
foreland basi
ed from a 1
picenter, foca
Fielding et al
he 2008 M
w
7.
nt
he
or
ic
O:
ic
in
1:
al
l.,
.9

Figur

Figur
netwo
the bo
three
areal
signif
landsl
dashe
in the
the st
deviat
km-ra
deviat
the tr
backg
indica
and tr
and tr
the L
Tibeta

e 2.2
e 2.2. Hydro
ork in the stud
oundaries of
main catchm
density in e
ficant landsli
lide impact. A
ed lines) along
e study area, s
tudy area, sh
tion. (d), pro
adius circular
tion. Panels (
rend line A-
ground showi
ate the 1 σ ran
ributaries, wi
ributaries. Th
Longmen Sha
an Plateau.
logical map a
dy area with
sub-catchme
ments (Min Jia
each sub-cat
ide impact (
A-A’ represen
g the steepes
showing max
howing mean
ofile of relief
r window (b
e)-(g) are plo
-A’. (e), mea
ng the mean,
nge of elevatio
th error bars
he fluvial chan
an and the S
and geomorp
the distributi
ents and tribu
ang, Fu Jiang
tchment and
L1-L17, Tab
nts the trend
t topographic
ximum, mean
n gradient an
f in the stud
lack line) an
otted by projec
an elevations
, maximum a
ons in each su
representing
nnels have hi
ichuan Basin

30
phic swath pr
ion of areal la
utaries, and b
g and Tuo Jia
tributary, w
ble 2.2) and
of 170 km-w
c gradient sho
and minimum
nd the range
dy area, show
nd the range
cting the calc
s for sub-cat
and minimum
ub-catchment
1 σ range. (g
gher densities
n, and lower
rofiles of the
andslide dens
black lines sh
ang). Color sh
with 17 catch
the remaini
wide swath p
own in (b)-(g
m elevations.
(grey area)
wing mean re
(grey area)
culated param
tchments and
m elevations in
t. (f), mean gr
g), channel de
s (~0.9 km km
densities (~
study area. (
sity P
Als
. Yello
how the bou
hading repres
hments defin
ing (S1-S9)
profiles (boun
g). (b), profile
. (c), profile o
bounded by
elief calculat
bounded by
meters for sub-
d tributaries,
n the study a
radients for su
ensities for su
m
-2
) on the ea
~0.6 km km
-2
(a), the fluvia
ow lines show
undaries of th
sents landslid
ned as havin
as negligibl
nded by whit
e of elevation
of gradients i
one standar
ed with a 2.
one standar
-catchments t
, with shade
area; error bar
ub-catchment
ub-catchment
astern flank o
2
) towards th
al
w
he
de
ng
le
te
ns
in
rd
.5
rd
to
ed
rs
ts
ts
of
he

Figur

Figur
comb
logari
by dif
in eac
linear
simila
define
diagra
logari
region
clear
suppo

e 2.3
e 2.3. Gradien
ined Longme
ithmic G-A di
fferent geomo
ch A bin ( δlo
r regression o
ar studies (e.g
ed as the chan
am), valley h
ithmic G-A d
nal channel th
transitions th
orting informa
nt-upstream c
en Shan catc
iagram from t
orphic proces
g
10
A = 0.1), a
on logarithm
g., Dadson et
nnel domain,
head and coll
diagram comb
hreshold of A
m
han shown in
ation.
contributing a
chments (com
the Yuzixi cat
sses following
and the powe
plots) are re
t al., 2004; M
and hillslope
luvial areas a
bining all Lo
min
~ 1 km
2
is
n these examp

31
area diagrams
mpilation of L
tchment, with
g Montgomer
er-law expone
eported for e
Meunier et al.
(strictly defin
are grouped t
ongmen Shan
used for catc
ples. G-A plo
s of the Yuzix
L1-L17, Tab
h characteriza
ry (2001). Me
ents between
each landscap
., 2008), bed
ned by the po
together as th
n catchments
chments wher
ots for all cat
xi catchment (
ble 2.2). (a),
ation of doma
ean gradients
G and A (i.e
pe zone. In t
drock and allu
ositive expone
he hillslope d
(L1-L17, Ta
re data are no
tchments are

(L4) and of th
example of
ains dominate
are calculate
e., the slope o
this and othe
uvial areas ar
ent on the G-A
domain. (b),
able 2.2). Th
oisier, with les
shown in th
he
a
ed
ed
of
er
re
A
a
he
ss
he

Figur

Figur
illustr
the ca
and th
lines
upstre
value
landsc
fairly
as see
case
comp
affect

e 2.4
e 2.4. Illustr
rating landslid
atchment cum
he landslide v
and red line
eam area, res
s of upstream
cape. (d), car
evenly acros
en in this stud
where most
ared to the la
ted by Wench
ration of cat
de location in
mulative distr
volumetric dis
es represent
spectively. (c
m contributin
rtoon illustrat
ss the catchm
dy) compared
landslides o
andscape. (h)
huan earthquak
tchment-scale
ndex ψ, which
ribution curve
stribution cur
catchment a
c) shows a c
ng area A (e.g
tion of case (c
ment, similar to
d to the landsc
ccur at relat
, cartoon illu
ke-triggered l

32
e landslide lo
h is defined a
e as a functio
ve (red). (b),
and landslide
ase where m
g., ψ ~ 1.8 a
c). (e) shows
o the catchme
cape. (f), cart
tively low va
ustration of ca
landslides are
ocation index
as the ratio o
on of upstrea
compiled dat
e distribution
most landslide
as seen in th
s a case wher
ent-scale dist
toon illustrati
alues of upst
ase (g). The L
e similar to (c
x. (a), schem
of the integrat
am contributin
ta from the st
n curves as
es occur at r
his study) com
re landslides a
tribution of A
ion of case (e
tream contrib
Longmen Sh
c) and (e).
matic diagram
ted area below
ng area (blue
tudy area; blu
a function o
relatively hig
mpared to th
are distribute
A (e.g., ψ ~ 1.
e). (g) shows
buting area A
an catchment
m
w
e)
ue
of
gh
he
ed
.2
a
A
ts

Figur

Figur
contri
distrib
curve
histog
thresh
Carlo
grey b

e 2.5
e 2.5. Cumul
ibuting area.
bution curve
and histogra
gram for land
holds (determ
simulations
bands are the
lative distribu
(a), distributi
and histogram
am for landsl
dslides within
mined from Fig
propagating
90% Monte C
ution curves
ion curve and
m for landslid
lides within t
the Fu Jiang
gure 3); red c
uncertainties
Carlo envelop

33
and histogram
d histogram f
des within th
the Tuo Jiang
catchment. T
curves represe
s in landslide
pe (5
th
-95
th
pe
ms of landsli
for all landsl
e Min Jiang
g catchment.
The dashed bl
ent the media
e area-volum
ercentiles).
ide volumes o
lides in the st
catchment. (c
(d), distribut
lack lines rep
an results from
me scaling pa
over upstream
tudy area. (b
c), distributio
tion curve an
resent channe
m 1000 Mont
arameters; an
m
),
on
nd
el
te
nd

Figur

Figur
contro
and t
densit
catchm
hillslo
to all
0.87),
-1.83×
10.91
landsl
Mean
size o

e 2.6
e 2.6. Statist
ol on landslid
the best-fit th
ty distribution
ments, red s
ope domain-in
landslides (b
, channel-con
×10
3
m
2
, with
×10
3
m
2
, and
lide-channel
n ξ values are
of δlog
10
A
ls
=
tical distribut
de-channel co
hree-paramet
ns. Grey sym
symbols show
nventory. Bla
best fit param
nnected lands
h r
2
= 0.84) an
d s = -1.81×1
connectivity,
e calculated f
0.1.
tion of the W
onnectivity. (a
er inverse g
mbols show th
w the chann
ack, red and b
eters: q = 1.8
slides (best fi
nd hillslope-d
0
3
m
2
, with r
showing a
from landslid

34
Wenchuan co
a), landslide p
amma functi
he overall lan
nel domain-in
blue curves in
81, m = 7.62×
it parameters
domain lands
r
2
= 0.83), res
positive corr
e populations
o-seismic lan
probability de
ions for the
ndslide inven
nventory, and
ndicate best f
×10
3
m
2
, and
s: q = 1.51, m
slides (best fit
spectively. (b
relation betw
s in each lan

ndslides and
ensity versus l
landslide ar
ntory within t
d blue symb
fit inverse gam
s = -1.31×10
m = 9.71×10
t parameters:
b), landslide s
ween ξ and l
ndslide area b
landslide are
landslide area
ea probabilit
the three mai
bols show th
mma function
0
3
m
2
, with r
2
0
3
m
2
, and s
q = 2.32, m
size control o
landslide area
bin, with a bi
ea
a,
ty
in
he
ns
=
=
=
on
a.
in

Figur

Figur
conne
repres

e 2.7
e 2.7. Relatio
ectivity ξ. The
sent 95% con
onship betwe
e solid line re
nfidence band
een landslide
epresents the
s.

35
location ind
best fit from
dex ψ and lan
linear regres
ndslide-chann
sion, and the
nel volumetri
grey shadow
ic
ws

Figur

Figur
volum
P
Als
in
(Liu-Z
landsl
distrib
area-v
corrid
maxim
densit
mean
Red d
show
Yingx

e 2.8
e 2.8. Spat
metric landslid
n 5 x 5 km
Zeng et al.,
lides in 5 x
bution of lan
volume scalin
dors along sw
mum and m
ty in 5 km cor
values of lan
dots represent
populations
xiu-Beichuan
ial patterns
de-channel co
windows. Th
2009). (b),
5 km wind
ndslide-chann
ng parameter
wath A-A’, s
minimum elev
rridors along
ndslide area-v
t landslide po
s of < 20.
fault.
of co-seism
onnectivity. (
he black lines
distribution
ows. A-A’ in
nel connectiv
rs) in 5 x 5
superimposed
vations). (e),
swath A-A’.
volume scalin
opulations of
The black

36
mic landslide
(a), distributio
s show the e
of landslide
ndicates the
vity ξ (calcu
km windows
d on the Lon
distribution
(f), landslide-
ng parameter
f numbers > 8
dashed line
es, channel-c
on of co-seism
extent of the
e areal dens
swath trend
ulated from
s. (d), landsl
ngmen Shan
of channel-
-channel conn
rs) in 5 km co
80 in each 5
e in (d), (
connected la
mic landslide
Wenchuan su
sity for chan
defined in
mean values
lide areal den
swath topog
-connected la
nectivity ξ (ca
orridors alon
km corridor
e) and (f)
andslides an
e areal densit
urface ruptur
nnel-connecte
Figure 2. (c
s of landslid
nsity in 5 km
graphy (mean
andslide area
alculated from
ng swath A-A
and grey dot
indicates th
nd
ty
re
ed
),
de
m
n,
al
m
A’.
ts
he

Figur

Figur
landsl
( δ) of
( δ= 1
indica
less-re
local
the sw
are ta
Resul
from
draina
total l
dot. N
correl
PGA
increm
type c
fault s
anom
conne
5
th
-95
e 2.9
e 2.9. Contr
lide-channel
f each parame
00 m); (c), ξ
ate results fro
epresented gr
maximum re
wath profile).
aken from the
lts are present
1,000 Mont
age density fo
landslide pop
Numbers asid
lation betwee
and landslid
ments along t
control on lan
segments, the
malously large
ectivity. Resu
5
th
percentiles
rols on land
connectivity.
eter, defined
versus relief
om the very la
roups (<1% o
lief not revea
. (d), litholog
e 1: 2,500,000
ted as median
te Carlo simu
or landslides
pulation. The
de each dot
en peak grou
de-channel co
the swath pro
ndslide-chann
e landslide-ch
Daguangbao
ults are prese
) from 1,000
dslide-channe
Mean ξ valu
as follows: (
f within a 2.5
arge Daguang
of total lands
aled in Figure
gical control
0 China Geo
n values (bars
ulations. (e),
within each l
black solid l
denote corre
und accelerati
onnectivity r
ofile A-A’. T
nel connectiv
hannel connec
o landslide du
ented as med
Monte Carlo
37
el connectivi
ues are calcul
a), ξ versus g
5 km-radius c
gbao landslid
slide number)
e 2d (average
on landslide-
logical Base
s) and 90% en
, dependence
lithological u
line represent
esponding lit
ion (PGA) an
represent mea
The solid line
vity. For lands
ctivity is also
ue to its signi
dian values (
simulations.
ity ξ. (a)-(c)
lated from lan
gradient ( δ=
circular windo
de, while grey
). Note that t
ed relief ± on
-channel conn
Map (China
nvelopes (erro
e of landslid
unit. The grey
ts the best lin
thology as s
and landslide-
an values ca
represents th
slides nearest
o reported for
ificant, dispro
(bars) and 90

), topographi
ndslide popu
1°); (b), ξ ve
ow ( δ= 100 m
y dots represe
the relief in
e standard de
nectivity; Lit
Geological S
or bars, 5
th
-95
de-channel co
y dot represen
near fit, exclu
shown in (d)
-channel con
alculated from
he best linear
t to thrust an
r populations
oportional inf
0% envelope
ic control o
lations in bin
ersus elevatio
m). Black dot
ent statisticall
(c) may show
eviation acros
thological dat
Survey, 2004
5
th
percentiles
onnectivity o
nts <2% of th
uding the gre
). (f), positiv
nnectivity. Th
m 5 km-wid
r fit. (g), fau
nd oblique-sli
excluding th
fluence on th
es (error bars
on
ns
on
ts
ly
w
ss
ta
).
s)
on
he
ey
ve
he
de
ult
ip
he
he
s,

Figur

Figur
co-sei
sedim
2008/
m), w
indica
reserv
(norm
landsl
determ
2009-
volum
from
the fit
index
the be

e 2.10
e 2.10. Spat
ismic landslid
ment yield (t
/6-2008/12, b
with color co
ate negative-v
voirs (see su
malized to one
lide volumetr
mined from l
-2012 plotted
metric density
least-squares
t between tota
( ψ) using the
est fit between
tial pattern o
de volumetric
t km
-2
) in c
ased on data
oding of the
valued calcula
upporting in
e year) plotted
ric density (
east-squares
against total
y (e) for each
fit in logarit
al landslide d
e 2008 data (f
n the normali
of post-Wenc
c density (Eq
catchments a
from Wang e
derived susp
ated sedimen
nformation).
d vs. total lan
d) for each
fit in the log
landslide vo
h catchment;
thmic space.
density and se
f) and 2009-2
zed residual a

38
chuan suspen
. 2.4). (a), dis
across the L
et al. 2015, no
pended sedim
nt yield due to
(b and d),
ndslide volum
catchment; s
garithm space
lumetric dens
solid lines s
(f and g), rel
diment yield
2012 data (g)
and location i
nded sedime
stribution of
Longmen Sha
ormalized for
ment yield. C
o unaccounte
sediment yi
metric density
solid lines sh
e. (c and e), a
sity (c) and ch
show power
lations of no
vs. the corres
respectively.
index using th
ent yield and
post-earthqua
an range (in
r areas with el
Catchments la
ed large sedim
ield over 20
y (b) and chan
how power l
annual sedim
hannel-conne
law best-fit
rmalized resi
sponding land
The solid lin
he 2008 data.

d relations t
ake suspende
ntegrated ove
levation > 80
abeled “N.A.
ment sinks lik
008/6-2008/1
nnel-connecte
aw best-fit a
ment yield ove
ected landslid
as determine
idual (P) from
dslide locatio
ne on (f) show

to
ed
er
00
.”
ke
2
ed
as
er
de
ed
m
on
ws

Figur

Figur
area (
the d
sub-ca
(Min
sub-ca
(L1-L
surrou
landsl
determ
chann
comb

e 2.S1
e 2.S1. Hydr
(MJ: Min Jian
distribution o
atchments an
Jiang, Fu J
atchment and
L17, Table 2.2
unding pane
lide-impacted
mined from t
nel domain. D
ining all Lon
rologic map a
ng; FJ: Fu Jia
of areal lan
d tributaries,
Jiang and Tu
d tributary, w
2). (b), examp
els show g
d sub-catchm
he G-A meth
Dashed black
gmen Shan ca
and gradient
ang; TJ: Tuo J
ndslide dens
and black lin
uo Jiang). C
with 17 catchm
ple of a logar
gradient -
ments L1-L17
hod. Solid gre
lines represe
atchments in

39
– upstream c
Jiang). (a), Th
ity P
Als
. Ye
nes show the b
Color shading
ments defined
rithmic G-A d
upstream c
7. Solid bla
ey lines repre
ent inferred c
the study.
contributing a
he fluvial netw
ellow lines
boundaries of
g represents
d as having s
diagram from
contributing
ack lines rep
esent the best
channel thresh
area diagram
work in the s
show the b
f the three ma
landslide de
significant lan
m the Yuzixi c
area diagra
present chan
t fit for the fl
hold from the

ms of the stud
tudy area wit
boundaries o
ain catchment
ensity in eac
ndslide impac
atchment. Th
ams for th
nnel threshol
fluvial-bedroc
e regional A
m
dy
th
of
ts
ch
ct
he
he
ld
ck
min

Figur

Figur
sampl
this c
shows
value
chann
(Eq. 2
media
area. T

e 2.S2
e 2.S2. Unce
ling uncertain
case, 10 m vs
s rapid decay
s near the s
nels extracted
2.S13). The g
an values of o
The red lines
ertainty of sa
nty ε
max
(%) i
s. 87 m resol
y as the samp
elected chan
from SRTM
grey area repr
observed sam
represent sam
ampling raste
introduced by
lution) as a fu
pled upstream
nnel threshold
30 data versu
resents 5
th
-95
mpling uncerta
mpling error o

40
ers of differe
y sampling A
function of sa
m contributing
d A
min
(~1 k
us SRTM 90
th
percentiles
ainty as a fun
of 0.01 km
2
, 0
ent resolution
from rasters
ampled upstre
g area increa
km
2
). (b), ob
data. (c), obs
of all data p
nction of sam
0.1 km
2
and 0

ns. (a), inferr
of different r
eam contribu
ses, and is w
bserved mism
served sampli
points. The bl
mpled upstream
0.5 km
2
, respe

red maximum
resolutions (i
uting area. ε
ma
within 1% at A
match betwee
ng uncertaint
lack shows th
m contributin
ectively.
m
in
ax

A
en
ty
he
ng

Figur

Figur
In all
propa
Mont

e 2.S3
e 2.S3. Calcu
l panels, red
agating uncert
e Carlo envel
ulated landslid
d curves repr
tainties in lan
lope (5
th
-95
th
p
de-channel co
resent the m
ndslide area-v
percentiles).

41
onnectivity vs
median results
volume scalin
s. channel thr
s from 1000
ng parameters
reshold area A
Monte Carl
; and grey ba

A
min
(L1-L17
lo simulation
ands are the 9
).
ns
90%

Figur

Figur
chann
contri
(43%
repres
averag
respec
thresh

e 2.S4
e 2.S4. Norm
nel threshold
ibuting area m
, in the main
sent the mean
ging all the
ctively. The
hold areas is t
malized diffe
areas and th
method, norm
n text). The
n positive and
positive and
uncertainty
thus within 20
erence (%) b
he catchment
malized by the
red solid lin
d negative nor
d negative d
(normalized
0%.

42
between the
t-specified co
e mean conne
ne represents
rmalized diffe
ifferences ov
difference) o
calculated c
onnectivity fr
ctivity of all
zero differen
erences (+10%
ver the selec
of connectiv

onnectivity u
from the grad
Longmen Sh
nce. The red
% and -9% re
cted A range
vity from var
using variabl
dient-upstream
han catchment
d dashed line
espectively) b
e (0.3-3 km
2
riable channe
le
m
ts
es
by
),
el

Figur

Figur
L1-L1
from
propa
Mont

e 2.S5
e 2.S5. Panel
17 (Table 2.2)
Figure 3); re
agating uncert
e Carlo envel
ls show distri
). In all panel
ed curves rep
tainties in lan
lope (5
th
-95
th
p
ibution curve
ls, the dashed
present the m
ndslide area-v
percentiles).

43
es and histogr
d black lines re
median result
volume scalin
rams for land
epresent chan
ts from 1000
ng parameters
dslide-impact
nnel threshold
0 Monte Carl
; and grey ba
ed catchment
ds (determine
lo simulation
ands are the 9
ts
ed
ns
90%

Figur

Figur
contro
draina
calcul
depen
dots
landsl
profil
obvio

e 2.S6
e 2.S6. Gradi
ols on draina
age density; (
lated for eac
ndence of land
represent lan
lide areas rep
e A-A’. (d),
ous correlation
ient and PGA
age density an
(b), weak dep
ch gradient
dslide area on
ndslides from
present mean
drainage den
n is observed.
A effects on d
nd landslide
pendence of l
bin of 2°, a
n PGA; red d
m the foot w
n values calcu
nsity versus m
.

44
drainage dens
area: (a), neg
landslide area
and grey do
ots represent
wall of the Y
ulated from 5
mean PGA o
sity and land
gative correla
a on gradient
ots indicate s
landslides on
Yingxiu-Beich
5 km-wide in
f each sub-ca

slide area. (a
ation between
t. Mean lands
smaller samp
n the hanging
huan fault. T
ncrements alo
atchment and
a)-(b), gradien
n gradient an
slide areas ar
ple sizes. (c
g wall and blu
The PGA an
ong the swat
d tributary; n
nt
nd
re
),
ue
nd
th
no
45

Table 2.1. Notation for symbols
Symbol Notation Unit
Equation
introduced
A
ls
Landslide area km
2
or m
2
1
V
ls
Landslide volume km
3
or m
3
1
γ
Landslide area-volume scaling factor,
m
(3-2 α)
1
γ=1.388±0.087 (Li et al., 2014)
α
Landslide area-volume scaling factor,
Dimensionless 1
Log
10
α=-0.974±0.366 (Li et al., 2014)
A Upstream contributing area km
2
3
P
Als
Landslide areal density % 3
P
Vls
Landslide volumetric density m
3
km
-2
4
L Channel length km 5
ρ Drainage density km km
-2
5
ψ Landslide location index Dimensionless 6
IA
c

Integrated area for catchment landscapes on
distribution plots for A
Dimensionless 6
IA
ls
Integrated area for landslides on distribution plots for A Dimensionless 6
m Inverse-gamma function parameter m
2
7
s Inverse-gamma function parameter m
2
7
q Inverse-gamma function parameter Dimensionless 7
G Gradient ° or m m
-1

ξ Landslide-channel volumetric connectivity %
PGA Peak ground acceleration g

46

Table 2.2. Topographic and landslide parameters for the catchments in the study area
Large
catchment
ID Catchment name Controlling
hydrological
station
Catchment
type
Area
(km
2
)
Elevation
(mean±1 σ)
(m)
Slope
(mean±1
σ) (°)
Drainage
density
(km
-1
)
P
Als (
%)
1
Min Jiang
L1
Upper Zagunao above
Zagunao
Zagunao Tributary 2397 3864±634 30±10 0.60 0.220
L2
Lower Zagunao above
Sangping
Sangping Tributary 2220 3350±786 31±10 0.60 0.351
L3
Min Jiang Dujiangyan
to Zhenjiangguan
Dujiang Sub-catchment 4341 2780±834 30±10 0.61 5.184
L4 Yuzixi N.A.
4
Tributary 1735 3546±877 31±10 0.58 3.395
L5 Guojiaba Guojiaba Tributary 575 2253±713 28±10 0.61 1.802
L7
Min Jiang Pengshan to
Dujiangyan
Pengshan Sub-catchment 3980 690±460 5±10 0.92 0.209
S3
Upper Heishui He
above Heishui
Heishui Tributary 1716 3852±527 27±9 0.61 0.003
S4
Lower Heishui He
above Shaba
Shaba Tributary 5484 3541±592 26±10 0.62 0.003
S5 Xiaoxinggou N.A.
4
Tributary 1700 3540±313 23±9 0.66 N.A.
5

S6
Min Jiang above
Zhenjiangguan
Zhenjiangguan Sub-catchment 2770 3677±394 23±9 0.65 N.A.
5

Tuo Jiang
L8
Tuo Jiang above
Dengyinyan
Dengyingyan Sub-catchment 7321 607±425 6±8 0.87 0.411
L9 Jian Jiang N.A.
4
Tributary 2859 1256±988 14±16 0.88 4.065
S1 Qiuxi He N.A.
4
Tributary 2497 428±54 5±4 0.81 <0.001
S2 Zishui He N.A.
4
Tributary 1959 435±30 6±4 0.81 <0.001
Fu Jiang
L6 Danan He N.A.
4
Tributary 3620 753±401 8±10 0.84 0.032
L10 Kai Jiang above Santai Santai Tributary 2560 611±355 7±9 0.81 0.764
L11 Anchang He N.A.
4
Tributary 957 935±462 15±13 0.71 3.846
L12
Fu Jiang Fujiangqiao
to Jiangyou
Fujiangqiao Sub-catchment 707 723±357 8±9 0.81 0.094
L13 Tongkou He N.A.
4
Tributary 4217 2016±825 28±10 0.62 0.970
L14
Pingtong He above
Ganxi
Ganxi Tributary 1062 1526±426 27±9 0.62 0.135
L15
Fu Jiang Jiangyou to
Pingwu
Jiangyou Sub-catchment 1581 1341±464 23±11 0.69 0.245
L16
Fu Jiang above
Pingwu
Pingwu Sub-catchment 2808 2729±932 29±10 0.63 0.102
L17 Huoxi He N.A.
4
Tributary 1494 2815±679 30±10 0.64 0.150
S7
Fu Jiang Shehong to
Fujiangqiao
Shehong Sub-catchment 3088 480±66 7±6 0.80 N.A.
5

S8 Lower Zitong Jiang N.A.
4
Tributary 3505 503±81 10±6 0.76 N.A.
5

S9
Upper Zitong Jiang
above Zitong
Zitong Tributary 1546 770±311 11±10 0.74 N.A.
5

1
P
Als
(%): landslide areal density calculated as landslide area/catchment area (Eq. 2.3)
2
ξ: landslide-channel volumetric connectivity; results are reported as the medians and the 5th and 95th percentiles from 1000 Monte Carlo simulations propagatin
area-volume scaling
3
ψ: catchment landslide location index, as calculated from Eq. 2.6 and Figure 4, see more details in supporting information
4
Not available
5
Landslide coverage is not available due to lack of satellite imagery coverage. These catchments are beyond the 0.2 g PGA contour (USGS, 2008) (http://earthqua
low relief, suggesting low landslide susceptibility
6
Landslide-channel connectivity was not calculated for catchments with P
Als
< 0.01% and areas with no imagery coverage
7
Landslide location index was not calculated for catchments with P
Als
< 0.01% and areas with no imagery coverage

47

Table 2.3. Compiled data for post-earthquake suspended sediment yields and landslide
volumetric densities

Large
catch
ment
I
D
Catchment
notation
Catch
ment
type
Fractio
n of
mounta
inous
area

(elevati
on >
800 m)
in total
catchm
ent
area
Controlli
ng
hydrolog
ical
station
1
Post-Wenc
huan
(2008/6-20
08/12)
suspended
sediment
yield
2

(t km
-2
)
Post-Wen
chuan
(2009-201
2) annual
suspended
sediment
yield
3

(t km
-2

yr
-1
)
Total
landslide
volumetric
density
4

(×10
3
m
3

km
-2
)
Channel-co
nnected
landslide
volumetric
density
5

(×10
3
m
3

km
-2
)
Land
slide
locati
on
index
ψ
Min
Jiang
G
1
Min Jiang
Pengshan
to Dujiangyan
Sub-
catchm
ent
0.19
Pengsha
n
1194±133 1494±389
27.83+70.
87/
-19.55
12.46+36.5
1/
-9.08
1.31
G
2
Guojiaba
Tributa
ry
1
Guojiab
a
195±19 1202±120
61.74+127
.16/
-42.10
12.23+25.5
0/
-8.43
1.17
G
3
Min Jiang
Dujiangyan
to
Zhenjiangguan
Sub-
catchm
ent
1 Dujiang N.A.
6
N.A.
6

251.37+58
5.53/
-180.94
76.20+182.
57/
-55.47
1.28
G
4
Lower Zagunao
Tributa
ry
1
Sangpin
g
108±17 242±59
14.80+37.
90/
-10.31
3.44+9.15/
-2.42
1.29
G
5
Upper Zagunao
Tributa
ry
1 Zagunao 50±5 259±26
10.61+25.
88/
-7.43
4.59+11.46
/
-3.25
1.48
G
6
Lower Heishui
Tributa
ry
1 Shaba 56±7 87±17
0.11+0.24/
-0.07
0.02+0.05/-
0.01
1.17
G
7
Upper Heishui
Tributa
ry
1 Heishui 27±3 210±21
0.11+0.25/
-0.08
0.09+0.21/-
0.06
1.74
G
8
Min Jiang above
Zhenjiangguan
Sub-
catchm
ent
1
Zhenjian
gguan
19±2 173±17 N.A.
7
N.A.
7
N.A.
8
Tuo
Jiang
G
9
Tuo Jiang main
Main
catchm
ent
0.14
Dengyin
gyan
1227±123 2658±266
500.72+13
71.63/-359
.46
214.87+63
7.43/
-158.08
1.32
Fu
Jiang
G
1
0
Kai Jiang above
Santai
Tributa
ry
0.10 Santai 715±71 2164±216
1826.49+6
103.69/-14
22.71
1655.16+5
774.35/
-1303.56
1.79
G
1
1
Fu Jiang
Shehong
to Fujiangqiao
Sub-
catchm
ent
0.16 Shehong N.A.
6
N.A.
6

376.03+94
2.51/
-264.55
128.77+33
6.19/
-92.08
1.26
G
1
2
Fu Jiang
Fujiangqiao
to Jiangyou
Sub-
catchm
ent
0.85
Fujiangq
iao
3496±475 2343±318
54.74+132
.72/
-38.84
33.85+85.7
0/
-24.56
1.33
G
1
3
Pingtong He
above Ganxi
Tributa
ry
0.98 Ganxi 1641±164 1226±123
5.96+13.9
7/
-4.14
0.87+2.02/-
0.60
1.12
G
1
4
Upper Zitong
Jiang
above Zitong
Tributa
ry
0.26 Zitong 3210±321 1537±154 N.A.
7
N.A.
7
N.A.
8

G
1
5
Fu Jiang
Jiangyou
to Pingwu
Sub-
catchm
ent
0.93 Jiangyou 1504±226 N.A.
6

12.03+28.
81/
-8.46
4.13+9.86/-
2.91
1.19
48

G
1
6
Fu Jiang above
Pingwu
Sub-
catchm
ent
1 Pingwu 231±23 708±71
7.34+20.3
9/
-5.26
0.86+1.93/-
0.59
1.20
1
Detailed information on hydrological stations are reported in Wang et al. 2015
2
Post-Wenchuan (2008/6-2008/12) suspended sediment yield is the total suspended sediment yield during 2008/6-2008/12 based on
the dataset from the hydrological stations [Wang et al., 2015]; 1 σ uncertainty is propagated from uncertainties in original hydrological
dataset
3
Post-Wenchuan (2009-2012) annual suspended sediment yield is the annual mean sediment yield during 2009-2012 based on the
dataset from the hydrological stations [Wang et al., 2015]; 1 σ uncertainty is propagated from uncertainties in original hydrological
dataset
4
Total landslide density is calculated as landslide volume/catchment area (Eq. 2.4); Monte Carlo simulations are run for propagating
uncertainties from parameters in landslide area-volume scaling and reported as the medians and the 16th and the 84th percentiles [Li et
al., 2014]
5
Channel-connected landslide volumteric density is calculated as channel-connected landslide volume/catchment area (Eq. 2.4);
Monte Carlo simulations are run for propagating uncertainties from parameters in landslide area-volume scaling and reported as the
medians and the 16th and the 84th percentiles (Li et al., 2014)
6
Sediment yield is not available because the calculations return negative values; see main text
7
Landslide data are not available due to no coverage of satellite imagery
8
Landslide location index is not available due to no landslide data

49

This page is left blank intentionally

50

Chapter 3
Earthquakes drive focused denudation along a tectonically
active mountain front

3.1. Preamble
Chapter 2 evaluated landslide-channel connectivity shortly after the Wenchuan earthquake.
Within the short time period window (half a year to four years), the connectivity study of Chapter
2 has major implications for understanding hazard generation and short-term sediment transport,
but is disconnected from longer geological timescales relevant to orogenic evolution. Landslides
are known to be a major erosional agent in mountain ranges, and observations have shown that,
over decadal to centennial timescales, landslide fluxes to first order match the observed erosional
fluxes (Keefer, 1994; Hovius et al., 1997). However, limited studies have considered specifically
the role of earthquake-triggered landslides over longer kyr-Myr timescales, largely due to limited
observations of both earthquake-triggered landslides and long-term erosion. Chapter 3 was
designed to extrapolate earthquake-induced landsliding erosion to timescales of multiple
earthquake cycles. With sufficient measurements of regional long-term denudation rates and high
quality landslide inventory maps, the Wenchuan case provides a good setting to test the
hypothesis that earthquake-triggered landslides may sustain mountain erosion over geological
timescales. In this work, I first calculated denudation rates over the four years following the
earthquake using hydrological gauging data, and evaluated controls from topographic,
hydrological and seismic factors. I then calculated theoretical “seismic erosion rates” caused by
earthquake-triggered landsliding in the study area combining seismology data and different
landslide volume models with key parameters calibrated using the Wenchuan and the Lushan
landslide data. The seismic erosion rates were then compared to the regional long-term
denudation rates recorded by cosmogenic nuclides and low temperature chronometer, to evaluate
how earthquake-triggered landslides contribute to long-term erosion.

I was the main author, analyzed the hydrological data, the seismic catalog data, and the digital
topography, modeled landslide volumes, and wrote the manuscript. Josh West was the main editor,
and suggested approaches for models to estimate landslide volumes. Zhangdong Jin, Fei Zhang
and Jin Wang helped on compiling the hydrological data. All other authors contributed to paper
revision.

This research was supported by the U.S. National Science Foundation (NSF-EAR/GLD grant
1053504 to Joshua West) and the Chinese Academy of Sciences (YIS fellowship grant
2011Y2ZA04 to Joshua West). I was supported by the USC Dornsife College Merit Fellowship.
This work benefited from conversations with Sean Gallen, Niels Hovius, Odin Marc, Haoran
Meng, Joel Scheingross, James Dolan and Seulgi Moon. We thank constructive comments from
51

an anonymous reviewer that greatly helped to improve an earlier version of the manuscript and
An Yin for editorial handling.

This work was published as:
Li, G., A. J. West, A. L. Densmore, Z. Jin, F. Zhang, J. Wang and R. G. Hilton (2017),
Earthquakes drive focused denudation along a tectonically active mountain front. Earth and
Planetary Science Letters. 472, 253-265, doi: 10.1016/j.epsl.2017.04.040.

52

3.2. Introduction
Mountain erosion affects rates and patterns of crustal deformation including seismogenic faulting
(e.g., Steer et al., 2014) and flexural-isostatic responses (e.g., Molnar and England, 1990), and
influences the geological carbon cycle and consequently the climate system (e.g., Raymo et al.,
1988; Wang et al., 2016). Large earthquakes are thought to play an important role in the
denudation of tectonically-active mountain ranges because they cause widespread landslides that
generate large volumes of clastic sediment (Keefer, 1994; Larsen et al., 2010; Hovius et al., 2011;
Parker et al., 2011; Wang et al., 2015a). Delivery of landslide debris to rivers and the subsequent
fluvial evacuation can increase erosion rates over years to decades (e.g., Hovius et al., 2011;
Wang et al., 2015a). However, over longer timescales relevant to orogenic evolution (10
4
-10
6
yr),
the role of earthquakes in denudation remains less well constrained, even though the volume of
seismically triggered landslides may be sufficient to partly or wholly counteract seismically
induced rock uplift (Parker et al., 2011; Hovius et al., 2011; Li et al., 2014; Marc et al., 2016a).

Detailed mapping of landslides (e.g., Keefer, 1994; Parker et al., 2011; Li et al., 2014; Xu et al.,
2015) and hydrological gauging of sediment fluxes (e.g., Hovius et al., 2011; Wang et al., 2015a)
capture the aftermath of individual events. Across multiple events, landslide volume scales with
earthquake magnitude (Keefer, 1994; Malamud et al., 2004; Marc et al., 2016b). Combined with
return time statistics for earthquakes, this scaling relationship can yield an estimate of long-term
landslide rate that should reflect a “seismic erosion rate” associated with repeated earthquakes,
assuming fluvial evacuation of landslide debris (Keefer, 1994; Malamud et al., 2004; Lavé and
Burbank, 2004; Li et al., 2014; Marc et al., 2016a). Keefer (1994) found that seismic erosion
rates are comparable to fluvial sediment yields measured in several regions. Cosmogenic nuclide
and thermochronology datasets allow us to expand this approach to consider denudation rates
measured over longer timescales that encompass multiple earthquakes and that are more relevant
to mountain belt evolution. In this study, we focus on the Longmen Shan region of central China,
where the 2008 M
w
7.9 Wenchuan and 2013 M
w
6.6 Lushan earthquakes allow us to make
estimates of seismic erosion rates. We evaluate both the spatial distribution and magnitude of
these rates in the context of datasets from fluvial sediment fluxes, cosmogenic nuclides, and
low-temperature thermochronology (e.g., Kirby et al., 2002; Ouimet et al., 2009; Godard et al.,
2010; Liu-Zeng et al., 2011; Wang et al., 2015a).

The steep Longmen Shan mountain range defines the eastern margin of the Tibetan Plateau. This
region has been at the nexus of contentious debates over the importance of motion along shallow
faults versus ductile flow of lower crust for collisional mountain building (e.g., Clark and Royden,
2000; Hubbard and Shaw, 2009). Focused denudation along the steep topographic front of such
plateau margins may exert an important influence on deformation (e.g., Beaumont et al., 2001).
However, the relative roles of tectonic and climatic drivers of denudation – and thus the link
between climate and the geodynamic processes – remain unresolved, both for the Longmen Shan
(e.g., Ouimet et al., 2009; Godard et al., 2010; Liu-Zeng et al., 2011) and elsewhere. We aim to
53

gain new general insight into the long-term role of seismic erosion in tectonically active
mountains and how it may contribute to focused denudation along the eastern margin of the
Tibetan plateau.

3.3. Setting
With elevations rising to higher than 5 km over a 50 km horizontal distance, the eastern Longmen
Shan flank represents one of Earth’s steepest plateau margins (Clark and Royden, 2000;
Densmore et al., 2007; Burchfiel et al., 2008). Several Yangtze headwater rivers (mainly the Min
Jiang, Fu Jiang, Tuo Jiang, Qingyi Jiang and Dadu He) drain from the Longmen Shan into the
Sichuan Basin (Figure 3.1a). A series of dextral-thrusting, oblique-slip faults bound the mountain
front and comprise the Longmen Shan fault system (Densmore et al., 2007; Burchfiel et al.,
2008). The bedrock geology consists mainly of Proterozoic basement granitoids and high-grade
metamorphic rocks, metamorphosed sedimentary rocks of a Paleozoic passive margin sequence,
unmetamorphosed sedimentary rocks associated with a Mesozoic foreland-basin succession, and
limited Cenozoic sediments (Burchfiel et al., 2008). Climatically, the Longmen Shan range is
located at the transition between the domains dominated by the east Asian monsoon and the
westerlies. Across the Longmen Shan, average annual rainfall decreases from the margin (~1100
mm yr
-1
) towards the plateau (as low as ~ 600 mm yr
-1
) (Liu-Zeng et al., 2011). This regional
climate pattern is largely determined by the high topography, which acts as an orographic barrier
and may also affect atmospheric circulation by heating of the atmosphere (Molnar et al., 2010
and references therein). Precipitation is highly seasonal, with most rainfall during the wet season
from June to September.

The M
w
7.9 Wenchuan earthquake on May 12
th
, 2008 initiated in the southern Longmen Shan,
near the town of Yingxiu, and ruptured northeastward for ~270 km along the Longmen Shan fault
system (Figure 3.1a) (Burchfiel et al., 2008; Shen et al., 2009). The strong ground motion
triggered > 56,000 landslides in the steep mountainous topography (Figure 3.1a) (Parker et al.,
2011; Li et al., 2014; Xu et al., 2014). These seismically induced landslides introduced large
volumes of clastic sediment into the fluvial system, estimated to total ~3 km
3
(Li et al., 2014).
Prior work has aimed to understand the effects on sediment transport. Li et al. (2014)
documented the spatial pattern and volume of landsliding, and Li et al. (2016) assessed the
connectivity of these landslides to the river network as a means of understanding their behavior
as sediment sources. Wang et al. (2015a) used data from the Chinese Hydrology Bureau to
quantify suspended sediment transport rates. After the Wenchuan earthquake (2008-2012),
suspended sediment fluxes from the Min Jiang, Fu Jiang and Tuo Jiang catchments increased by
3 to 7 times compared to pre-earthquake levels (2006-2007). Based on
10
Be concentrations in
quartz from Min Jiang riverbed sands, West et al. (2014) suggested that bedload transport rates
had increased by a similar order of magnitude to those of suspended load. The present study takes
advantage of this prior work, including the landslide inventory and sediment fluxes, in order to
compare spatial patterns of denudation across a range of timescales.

54

We use the M
w
6.6 Lushan event as an additional constraint on the magnitude of seismic erosion
rates. The Lushan earthquake occurred on April 20
th
, 2013 in the southern Longmen Shan, 80 km
south of the Wenchuan epicenter (Figure 3.1a). This event initiated on a ramp in the range-front
blind thrust fault, in the footwall of the Wenchuan rupture (Wang et al., 2014). As in the
Wenchuan event, widespread landsliding occurred in the southern Longmen Shan range during
the Lushan earthquake. Xu et al. (2015) reported more than 20,000 co-seismic landslides, with a
total area of 18.88 km
2
and an estimated volume of 0.042 km
3
across the region of the Lushan
earthquake.

3.4. Materials and Approaches
3.4.1. Landslide inventory
For the Wenchuan earthquake, co-seismic and immediately post-seismic landslides (within six
months after the earthquake) were mapped by Li et al. (2014). Landslide volumes were
calculated from empirical landslide area-volume scaling relations (e.g., Larsen et al., 2010). We
assume that mapped landslides mainly resulted from the Wenchuan mainshock because we find
that aftershocks contributed <5% of the total seismic moment release across the Longmen Shan,
based on the seismic catalog spanning over six months following the mainshock (CSN Catalog,
2015). This finding is consistent with observations from other earthquakes that suggest most
landslides occur during the main shock (e.g., Roback et al., 2017). Additional volume associated
with post-seismic (e.g., storm-triggered) landslides is likely to be on the order of a few percent of
the total landslide volume (Li et al., 2016 and references therein).

For the landsides triggered by the Lushan earthquake, we refer to the landslide inventory
compiled by Xu et al. (2015), who also used empirical scaling relationships reported in Larsen et
al. (2010) to estimate volumes from a landslide map based on satellite imagery. Xu et al. (2015)
mapped Lushan landslides using images collected from April – May 2013, around five years after
the Wenchuan earthquake but immediately after the Lushan event. Also, there is not much
overlap between the mapping extents and the intensive shaking zones for the Lushan and
Wenchuan events (Li et al., 2014; Xu et al., 2015). Thus we expect limited influence on the
mapped Lushan landslides from the Wenchuan earthquake.

3.4.2. Geomorphic characterization
In evaluating the spatial patterns of denudation in the Wenchuan region, we focus on three main
catchments (Min Jiang, Fu Jiang and Tuo Jiang), comprising sixteen sub-catchments as
delineated by Li et al. (2016) (Figure 3.1b and Table 3.S1). SRTM30 digital elevation model
(DEM) data have incomplete coverage of the study region, so we used void-filled SRTM90 DEM
data for topographic analysis (Jarvis et al., 2008). Slopes were calculated using standard
algorithms provided in the ArcGIS platform (Figure 3.1d). Although the derived slopes vary as a
function of DEM resolution (e.g., Larsen et al., 2014), the biases are systematic (e.g., Figure 3.S1)
so we expect little influence on the relative trends between different catchments. Relief was
determined as the ranges of elevations in 2.5 km-radius and 5 km-radius circular windows. We
55

also calculated the volumetric density of landslides (m
3
km
-2
, landslide volume per unit
catchment area) within each studied catchment (Table 3.S1) and along a swath profile A-A’
(Figure 3.1f). We derived channel steepness indexes (k
sn
, normalized to θ
ref
= 0.45, cf. Ouimet et
al., 2009; symbol notation listed in Table 3.1) using the Stream Profiler toolbox
(http://www.geomorphtools.org). DEM cells with drainage area < 1 km
2
were excluded to
remove colluvial landscapes (Li et al., 2016). To characterize ground motion associated with the
Wenchuan earthquake, we used gridded peak ground acceleration (PGA) data obtained from the
USGS ShakeMap (USGS Hazard Program, http://earthquake.usgs.gov/earthquakes).

3.4.3. Hydrological data
Wang et al. (2015a) analyzed data from the Chinese Hydrology Bureau and calculated total
suspended load fluxes and runoff from 16 gauging stations across the Longmen Shan catchments
(Figure 3.1a). This data covered both the pre-Wenchuan (2006-2007) and the post-Wenchuan
(2008-2012) time periods. Using this dataset, Li et al. (2016) derived the suspended load fluxes
for sub-catchments, by taking the differences between fluxes gauged at one station and all
neighboring upstream stations, following a mass balance principle (Li et al., 2016). Three
sub-catchments yielded negative sediment fluxes, attributed to large sedimentary sinks such as
reservoirs. As in Li et al. (2016), these were excluded from analysis (catchments labeled as “N.A.”
in Figure 3.1b).

Based on the gauged discharge and catchment area, we derived annual runoff for sub-catchments
using an analogous mass balance approach to that for calculating sediment fluxes. Wang et al.
(2015a) suggested that high magnitude runoff events play an important role in post-Wenchuan
suspended sediment transport. We explored different thresholds for high magnitude runoff and
found that a 6 mm day
-1
threshold best correlated with denudation rates (Figure 3.S2 in
Supplementary Materials), close to the 5 mm day
-1
used by Wang et al. (2015a). We also
calculated specific stream power ( ω) adopting the approach of Burbank et al. (2003) (details in
Supplementary Materials) using our compiled runoff and topography data.

3.4.4. Calculation of sediment flux-derived (“short term”) denudation rates
The total mass flux from a river catchment (i.e., the denudation rate) includes suspended, bedload,
and dissolved load. To calculate total denudation rates, we adopted the approach of Liu-Zeng et
al. (2011), who determined a pre-Wenchuan earthquake ratio of dissolved load to suspended load
of 19±6%, and a ratio of bedload to suspended load of 25±15%. After the earthquake
(2009-2012), Jin et al. (2016) measured solute fluxes at two sites (Zhenjiangguan and Weizhou,
Figure 3.1a) in the Min Jiang catchment, yielding a post-earthquake dissolved:suspended load
ratio of 19±7% (Table 3.S3 and text in Supplementary Materials), similar to pre-earthquake
estimates. Measurements of
10
Be concentrations in detrital quartz from bed sediments collected in
2009-2010 (West et al., 2014) indicate that, after the earthquake, bedload increased by a similar
factor as suspended load. Thus the bedload:suspended load ratio is likely to be similar to that
reported prior to the earthquake, i.e., 25±15%. Using these ratios between the dissolved load,
56

bedload and suspended load, we calculated total denudation fluxes (t yr
-1
) from the suspended
sediment fluxes and converted these to denudation rates (mm yr
-1
), assuming material density of
2.65×10
3
kg m
-3
(cf. Liu-Zeng et al., 2011). To account for low-relief, frontal plains, which are
expected to contribute little to the denudation flux, we normalized the calculated denudation rates
to the fraction of mountainous area (defined as areas > 800 m elevation) in each catchment
(Figure 3.1b and Table 3.S1). We have examined relationships between these denudation rates
and various hydrological and topographic metrics we have calculated for the Longmen Shan
(Section 3.4.2, 3.4.3).

3.4.5. Compilation of long-term (kyr to Myr) denudation rates
To characterize the denudation of the Longmen Shan over longer timescales, we compiled
10
Be-derived catchment-scale millennial denudation rates (Ouimet et al., 2009; Godard et al.,
2010; Ansberque et al., 2015) and refer to a data set of bedrock cooling ages and corresponding
exhumation rates across the mountain range from low-temperature thermochronology studies
(apatite fission track (AFT), apatite (U-Th)/He (AHe), zircon fission track (ZFT) and zircon
(U-Th)/He (ZHe)), compiled by Tian et al. (2013) (data from Arne et al., 1997; Kirby et al., 2002;
Richardson et al., 2008; Godard et al., 2009; Wang et al., 2012; Tian et al., 2013). Tian et al.
(2013) converted the cooling ages to time-averaged exhumation rates assuming a
one-dimensional, steady state upper crustal section and taking into account the effects of cooling
rate on closure temperature together with heat advection during exhumation, following Reiners
and Brandon (2006). For geothermal gradient, Tian et al. (2013) assumed a pre-exhumation
geothermal gradient of 20 °C km
-1
, which yields a syn-exhumation gradient of 23-30°C km
-1
,
consistent with the present geothermal gradient in the Longmen Shan determined from thermal
logging of local boreholes (> 4.5 km deep) and numerical modeling (Tian et al., 2013).
Uncertainties on the exhumation rates were propagated from uncertainties on thermochronology
measurements (Tian et al., 2013).

3.4.6. Calculation of seismic erosion rate over multiple earthquake cycles
3.4.6.1. Approach to calculating seismic erosion rate
Over multiple recurrence cycles, earthquakes of various magnitudes occur at different frequency.
To characterize the cumulative effect, we defined a seismic erosion rate (mm yr
-1
) as the total
volume of landslides triggered over multiple earthquake cycles and over a specified area,
following Keefer (1994):

A t
M V M N
e
w
M
w L w




) ( ) (

(3.1)

where t represents the total time (yr) over which repeated earthquakes are integrated, A is the area
of the region of landslide occurrence, N(M
w
) is the number of landslide-triggering earthquakes in
57

the magnitude bin (M
w
, M
w
+0.1), and V
L
(M
w
) refers the corresponding landslide volume triggered
by an earthquake of magnitude M
w
.

Based on the Wenchuan landslide data, we assumed that all landslides have occurred within an
intensive erosion zone along the frontal Longmen Shan, with an area of 170 km×80 km (A). This
area, to first order, matches the areal extent of landslide occurrence predicted for a M
w
8 event by
Keefer (1994), and the length of this region also approximates the rupture length of the
Wenchuan earthquake (Burchfiel et al., 2008). By distributing the total volume of
earthquake-triggered landslides over area A and time period t, we obtain a spatially and
temporally averaged rate of seismic erosion.

To calculate a seismic erosion rate from Eq. 3.1, we adopted a numerical integration approach
(after Keefer, 1994; see details in Supplementary Materials). This approach combines (1) a
scaling relationship between earthquake magnitude and the volume of earthquake-triggered
landslides (Section 3.6.2) and (2) a statistical description of earthquake magnitude and
corresponding frequency throughout an earthquake sequence (Section 3.6.3). Summing the
landslide volumes throughout a full earthquake sequence, we determined the total volume of
landslides occurring over a time period t that captures multiple earthquake cycles, yielding a
long-term seismic erosion rate (Eq. 3.1).

3.4.6.2. Earthquake magnitude-landslide volume scaling relations: predictive models of the
total volume of earthquake-triggered landslides
Descriptions of the landslide volume associated with earthquakes range from simple empirical
regression of volume versus earthquake magnitude (e.g., Keefer, 1994; Malamud et al., 2004) to
models that seek to capture the mechanics of landslide triggering, including slope stability as it
relates to hillslope angles and near-surface rock strength (e.g., Gallen et al., 2015; Marc et al.,
2016b). Large uncertainties plague the empirical regressions, whereas the more mechanistic
models – although able to reproduce global patterns (e.g., Marc et al., 2016b) – include
parameters that are often not precisely known, for example those describing rock strength,
earthquake asperity depth, and ground motion attenuation. To capture this range of approaches,
we estimated seismic erosion rates using both empirical regression and the model of Marc et al.
(2016b). We also calculated rates with a Longmen Shan-specific landslide volume model based
on locally-calibrated parameters and ground motion equations. To most accurately estimate
seismic erosion rates for the Longmen Shan region, we evaluated the predictions of these
different models with reference to the observed landslide volumes from the Wenchuan and
Lushan cases (see Section 3.5.1). The global empirical regression model, global
seismologically-based model, and Longmen Shan-specific model are summarized as follows.

58

Global empirical regression: Keefer (1994) and Malamud et al. (2004) compiled a global dataset
of landslide inventories triggered by large earthquakes. They reported a logarithmic scaling
relation between the total volume of landslides triggered by an earthquake, V
L
, and earthquake
magnitude, M
w
(Malamud et al., 2004):

Log
10
V
L
= 1.42M
w
– 11.26 (± 0.52) (3.2)

Global seismologically-based landslide model: Marc et al. (2016b) developed an approach to
predict the total volume of earthquake-triggered landslides, taking into account seismogenic
characteristics (e.g., seismic moment and asperity depth), landscape steepness, and material
sensitivity (rock strength and pore pressure). We adopted this model using seismogenic and
topographic parameters appropriate for the Wenchuan and Lushan earthquakes (see discussion in
Section 3.5.2 and more details in Supplementary Materials).

Longmen Shan-specific landslide model: Using local observations of ground motion attenuation,
we derived an earthquake magnitude-landslide volume scaling relation specific for the Longmen
Shan (details in Supplementary Materials). In brief, we first defined a landslide volume-PGA
relation using the Wenchuan data (following Meunier et al., 2007; Figure 3.S3a in Supplementary
Materials). We combined this with a locally-calibrated equation describing ground motion
attenuation in the Longmen Shan and neighboring areas (Cui et al., 2012; Wang et al., 2015b; and
references therein):

Log
10
PGA = c
1
+ c
2
M
w
+ c
3
Log
10
(D + c
4
) (3.3)

where D represents distance to fault trace and c
1
, c
2
, c
3
and c
4
are empirical parameters,
determined empirically from the Wenchuan data and other earthquakes in the Longmen Shan and
neighboring region (see Figure 3.S3b in Supplementary Materials). Combining the landslide
volume-PGA relation and the PGA-M
w
relation allows us to calculate the landslide volume for
earthquakes across a range of earthquake magnitudes. The relation between earthquake
magnitude and landslide volume can be well described by a logarithmic fit (r
2
= 0.99):

Log
10
V
L
= 23.77Log
10
M
w
– 11.91(±0.07) (3.4)

Slope angles also influence where landslides occur during earthquakes (Gallen et al., 2015; Marc
et al., 2016b). The Newmark model framework adopted by Gallen et al. (2015) accounts for slope
angles but depends on assumptions about landslide geometry, complicating its application in this
case. However, consistent with studies of other earthquakes (Meunier et al., 2007), we find that in
the case of the Longmen Shan PGA provides a good first-order empirical prediction of regional
patterns in landslide occurrence without considering differences in slope angle (Figure 3.S3a),
perhaps because regional variability in slopes is relatively small when compared to PGA (Figure
3.1).
59

3.4.6.3. Longmen Shan earthquake sequence
Inferring a long-term seismic erosion rate from landslide volume predictions requires
assumptions about the earthquake population over the timescales of multiple earthquake cycles.
We simulated a sequence of earthquakes with various magnitudes using available seismological
data from the Longmen Shan region. Because the Wenchuan earthquake ruptured almost the full
length of the Longmen Shan frontal fault system (Burchfiel et al., 2008), M
w
~8 represents a
reasonable upper bound for earthquake magnitudes in the study area. We chose M
w
~5 as a
minimum magnitude for landslide triggering (Marc et al., 2016b). The occurrence times of
earthquakes of various magnitudes were determined using an earthquake frequency-magnitude
distribution, with reference to regional historic seismicity data (China Earthquake Networks
Center, 1657-2013) (Wang et al., 2015b) and results from paleoseismological and geodetic
studies that suggest a recurrence interval (T) for Wenchuan-like events of 500 to 4000 years
(Densmore et al., 2007; Shen et al., 2009; Thompson et al., 2015). Across this range, the
frequency-magnitude distribution of the Longmen Shan earthquakes could be well described
using a truncated G-R function (Utsu, 1999 and references therein). For the shortest estimated T
(~500 years) (Thompson et al., 2015), Longmen Shan earthquake occurrence follows a classical,
linear G-R relation.

To estimate seismic erosion rates following Eq. 3.1, we integrated predictions of the landslide
volumes across the simulated earthquake sequences. Predicted landslide volumes are sensitive to
the source depth of each simulated earthquake (see details in Supplementary Materials). For the
global seismologically-based model (Marc et al., 2016b), we assumed a scaling relation between
earthquake magnitude and focal depth, calibrated using the Wenchuan and the Lushan data. For
the Longmen Shan-specific model, we used the scaling between earthquake magnitude and focal
depth to define a characteristic landslide-triggering depth for each earthquake magnitude. We
assumed that only earthquakes shallower than this depth cause landslides. For a given magnitude,
we estimated the proportion of events with a focal depth shallower than this threshold based on a
local seismic catalog (CSN Catalog, 2015; see Figure 3.S4).

3.4.6.4. Failure and resetting of hillslopes over earthquake cycles
In using scaling relationships to calculate seismic erosion rate over multiple earthquake cycles,
we have assumed that for each event there are sufficient hillslopes that are prone to fail.
Following one earthquake cycle, failed hillslopes need to be re-weakened and re-steepened to
initiate new landsliding in the following earthquake cycle. We estimate that for the Longmen
Shan, the pace of hillslope resetting is capable of keeping up with earthquake recurrence. For
example, if we assume that the landscape fails following a patchwork fashion, then each
earthquake triggers landslides on a different part of the unfailed landscape, allowing the failed
hillslopes time to recover (e.g., Parker et al., 2015). During the Wenchuan earthquake, around 1%
of the high PGA-area (>0.2 g) was impacted by landsliding (Li et al., 2014). Thus to fail the full
landsliding-susceptible landscape would take ~100 earthquake cycles, or 50-400 kyrs (given an
estimated return time of Wenchuan-like events of ~500-4000 yrs, see Section 3.6.3). If
60

landscapes are steepened by river incision, the steepness resetting time (i.e., time required for
resetting the failed landscape to pre-Wenchuan steepness) can be approximated as the ratio of the
landslide depth versus the channel incision rate. We estimate an average steepness resetting time
of ~6-26 kyr, or ~2-50 earthquake cycles for the Longmen Shan region (based on a mean
Wenchuan landslide depth of ~6-13 m (Gallen et al., 2015) and a regional incision rate of ~0.5-1
mm yr
-1
(Tian et al., 2015 and references therein)). This resetting time is much shorter than the
50-400 kyr to fully fail the landscape. Similarly, for a Longmen Shan chemical denudation rate of
~0.1-0.2 mm yr
-1

(Liu-Zeng et al., 2011, and references therein)

we anticipate that re-weakening
of hillslope material to a depth of ~6-13 m would take 30-130 kyr. Therefore, regional channel
incision and chemical weathering propagation rates should be fast enough to re-steepen and
re-weaken failed landscapes, rejuvenating hillslopes for landsliding over multiple earthquake
cycles.

3.5. Results and Discussion
3.5.1. Intensive and persistent denudation along the frontal Longmen Shan across
timescales, and its relation to seismically triggered landsliding
3.5.1.1. Zone of focused denudation
Profiles of short-term denudation rates calculated from hydrological gauging and longer-term
denudation rates from cosmogenic nuclides and thermochronology are shown in Figures 1 and 2,
respectively. Rates are highest along the eastern Longmen Shan front and decrease towards the
Tibetan Plateau along the A-A’ trend. Based on these observations, we delineate an 80 km-wide
zone of intensive erosion perpendicular to the strike of the mountain range (bounded by dashed
blue lines in Figure 3.1c). This distinct zone features steep topography, maximum PGA, and the
highest concentration of Wenchuan-triggered landslides (Figure 3.1c and Figure 3.1e). This
region of intensive denudation was also identified by Liu-Zeng et al. (2011) based on denudation
rates calculated from 1960s-1980s hydrological gauging data, though their rates were slightly
higher (up to 0.5-0.8 mm yr
-1
) than our pre-Wenchuan (2006-2007) estimates (0.24±0.04 mm
yr
-1
). The difference is consistent with background declines in both water discharge and sediment
flux in the Yangtze basin over the past 50 years, attributed in part to decreased precipitation as
well as to human activities like dam building (Yang et al., 2015).

Long-term denudation rates (Section 3.5) along the A-A’ trend (Figure 3.2a, b, c, d and e) are also
higher (> 0.5 mm yr
-1
) along the front of the eastern Longmen Shan, compared to the western
(plateau) side. The similarity to sediment flux-derived rates suggests a persistent denudation
pattern across both modern and longer-term timescales. Both patterns are, to first order, similar to
the distribution of the Wenchuan co-seismic landslides (Figure 3.2b, c, d and e). We suggest that
seismicity associated with the Longmen Shan fault system, which runs along the front of the
range, provides a mechanism for generating repeated landslides in this zone, as seen during the
Wenchuan earthquake. The spatial coincidence of these landslides with measured denudation
rates is consistent with landslides sustaining denudation fluxes across a wide range of timescales.

61

Additional second-order features in the exhumation data from the Longmen Shan include two
local high rates to the west (Figure 3.2e), beyond the region of most active landsliding associated
with the Wenchuan earthquake. The extent to which these high rates are also explained by
earthquake-triggered landsliding is unclear, but some evidence suggests that they might be. The
high exhumation rates around 160 km distance along the A-A’ transect (Figure 3.2e) are located
along the Wenchuan-Maowen fault (WMF, Figure 3.2f), another major thrust fault within the
Longmen Shan. Rapid exhumation in this region has been attributed to active thrusting of the
WMF in the late Cenozoic (Tian et al., 2013 and references therein). Accompanying seismic
activity could have triggered earthquake-triggered landsliding, converting uplifted mass to clastic
sediment, enhancing denudation fluxes, and leaving imprints in the exhumation rates seen today.
The other zone of locally elevated rates, closer to the plateau (Figure 3.2e), is less well-defined:
moderately high values are recorded by the cosmogenic nuclides and seen in part of the AFT data
from Arne et al. (1997) but not in other data (e.g., ZFT). A potential local maximum in
denudation rate coincides with a local peak of seismic moment release (Figure 3.2f, A-A’ distance
around 200 km) as calculated from short-term historic seismicity (1970-2015) (CSN Catalog,
2015). Local clustering of seismicity might indicate the potential for landslide-triggering
earthquakes in this region. Further low-temperature thermochronology would help to better
constrain the exhumation pattern across the Longmen Shan and the linkage between exhumation
and co-seismic landsliding.

3.5.1.2. Seismic control on focused denudation of the Longmen Shan
In addition to seismicity, several other features also vary along the profiles shown in Figures 1
and 2, potentially influencing spatial patterns of denudation (e.g., Burbank et al., 2003; Ouimet et
al., 2009). To distinguish these effects, we have examined the relationships between denudation
rates inferred from post-earthquake gauging data (2008-2012) and a group of metrics of
topography (slope, relief), hydrology (mean annual runoff, proportion of runoff from high
intensity runoff events), fluvial erosion potential (stream power and normalized channel
steepness index), seismic shaking (PGA, distance to fault rupture as a metric for seismic energy
release), and the density of earthquake-triggered landslides (Figures 3 and 4). Using principle
component analysis (PCA), these metrics cluster into two statistically distinct groups that reflect
mechanistically distinct processes: (1) a “seismic component”, comprising PGA, landslide
volumetric density and distance to fault rupture; and (2) a “non-seismic component,” comprising
slope angles, relief, steepness index, and the hydrological metrics (for detailed PCA results, see
Figure 3.S5, Table 3.S4 and Table 3.S5 and text in Supplementary MaterialsSupplementary
Materials).

Across the non-seismic metrics, post-Wenchuan denudation rates correlate positively but
moderately (r
2
= 0.36, P < 0.05, Figure 3.3a) with the proportion of catchment runoff from high
intensity runoff events (> 6 mm day
-1
), consistent with the findings of Wang et al. (2015a).
Pre-Wenchuan denudation rates also show a moderate, positive correlation with intense runoff
events (r
2
= 0.33, P < 0.05, Figure 3.3a), but with a shallower slope than post-earthquake rates.
62

The steeper slope of the post-earthquake data indicates that the denudation rate has become more
sensitive to hydrological conditions under enhanced sediment supply following the earthquake.
We find no correlation between denudation rates and mean annual catchment runoff. There are
also no statistically significant correlations between the post-Wenchuan denudation rates and
other non-seismic metrics including channel steepness index and stream power (Figure 3.3b, c, d,
e and f), perhaps because landscape steepness exceeds the threshold where these relationships are
easily observed (e.g., Ouimet et al., 2009).

For seismic metrics (Figure 3.4), we find statistically significant correlations between
post-Wenchuan denudation rates and catchment-scale mean PGA (r
2
= 0.61, p < 0.002, Figure
3.4a), catchment-scale maximum PGA (r
2
= 0.64, p < 0.001, Figure 3.4b), distance to the fault
rupture (r
2
= 0.67, p < 0.001, Figure 3.4c), and the volumetric density of earthquake-triggered
landslides (r
2
= 0.66, p < 0.01, Figure 3.4d). Although USGS ShakeMap PGA data do not include
local site effects and topographic amplification that may affect landslide occurrence, we expect
these to have limited influence on the first-order spatial patterns that indicate a seismic control on
denudation rates.

Since the principal component analysis separates these seismic parameters from the non-seismic
metrics, we do not expect cross-correlation between these two groups (e.g., between PGA and
catchment slope or runoff, Figure 3.S6) to bias our interpretations. Seismic intensity and
denudation rates co-vary both across the plateau margin (i.e., along A-A’), and also along strike
of the range. While the former gradient coincides to some extent with changes in relief, slope,
and runoff, the latter does not, emphasizing the seismic role in denudation.

The spatial coverage of the longer-term denudation rate data is not sufficient to conduct a similar
analysis, but the correlations described here show that seismicity in the Longmen Shan is not
inextricably coupled to other parameters that influence denudation rates. The correlation between
sediment flux-derived denudation rates and seismic parameters suggests that the coincidence of
high denudation rates and intensive landsliding along the frontal Longmen Shan reflects a
seismic driver of denudation, rather than a coincidental relationship with an underlying control
by topographic or other non-seismic parameters. We expect that the seismic control on
post-earthquake denudation rates as observed via sediment fluxes would recur for repeated
earthquakes, providing a mechanism for seismicity to influence the longer-term pattern and rate
of denudation.

We next consider the theoretical magnitude of denudation rate sustained by earthquake-triggered
landsliding over multiple earthquake cycles (Section 3.5.2). We then compare these estimated
rates of seismic erosion with the rates measured across timescales (Section 3.5.3).

63

3.5.2. Quantifying seismic erosion rates
3.5.2.1. Predictions of landslide volumes
Calculating a seismic erosion rate following Eq. 3.1 depends on predicting landslide volumes
associated with earthquakes of varying magnitude. In Figure 3.5, we show results from the three
predictive landslide volume models considered here: the global empirical regression of Malamud
et al. (2004), the global seismologically-based model of Marc et al. (2016b), and the Longmen
Shan-specific model. We compare these predictions to the volumes of landslides triggered by the
Wenchuan and Lushan earthquakes, as determined from landslide mapping (Li et al., 2014; Xu et
al., 2015). The global empirical regression systematically underestimates the volumes of the
Wenchuan and the Lushan landslides (Figure 3.5a). The Longmen Shan model accurately
predicts both the Wenchuan and the Lushan landslide volumes (Figure 3.5b). The global
seismologically-based model of Marc et al. (2016b) fits the observations if adjustable parameters
in the model are tuned (Figures 5c,d).

In more detail, the results of the global model of Marc et al. (2016b) are sensitive to the adopted
parameters, including landscape steepness, mean asperity depth and hillslope material sensitivity.
Whereas slope and asperity depth can be determined from DEM and seismological data,
respectively, we lack independent constraints on the term describing material sensitivity, which is
related to rock strength and pore pressure. Using a global average material sensitivity (as reported
by Marc et al., 2016b) and local parameters describing seismology and topography, this model
under-predicts the landslide volumes for both the Wenchuan and the Lushan earthquakes (Figure
3.5c). The model fits the Wenchuan and the Lushan observations for increases in the material
sensitivity term of 6 times and 4.3 times, respectively. Model results for a 5× increase in material
sensitivity (determined by the minimization of the sum of the squared residuals) closely
approximate both volumes (Figure 3.5d).

Differences in curvature between the two seismologically-based V
L
-M
w
relations (Figure 3.5b vs.
5d) derive from assumptions about ground motion attenuation and the landslide volume-ground
motion scaling relation. Marc et al. (2016b) assumed (i) a linear relationship between landslide
volume and ground motion, in contrast to the non-linear relationship from Wenchuan-specific
observations (Figure 3.S3), and (ii) a significant “saturation” effect of ground motion at high
earthquake magnitude, thought to describe attenuation of high-frequency (e.g., 1 Hz) spectral
accelerations (e.g., Boore and Atkinson, 2008). Thus the Marc et al. (2016b) model predicts that
ground motion and landslide volumes should increase only slightly with magnitude for large
(M
w
> ~6.5) earthquakes. Since the magnitude dependence is small, a ~5 km shallower asperity
depth is needed in this model to explain the much greater landslide volume from the Wenchuan
earthquake (M
w
=7.9; V
L
=2.7-4.4 km
3
) vs. the Lushan event (M
w
=6.6; V
L
= 0.042 km
3
from Xu et
al., 2015, although Marc et al., 2016b quoted a lower volume for this event), assuming similar
material sensitivity.

64

3.5.2.2. Seismic erosion rates
The Longmen Shan earthquake frequency-magnitude distribution depends on the recurrence
interval for Wenchuan-like events, T (Section 3.6.3, Figure 3.6a); across the range of plausible
estimates for T (~500-4000 years), we find a lower seismic erosion rate with longer T (Figure
3.5b). For a given T, the corresponding seismic erosion rate also differs depending on the
landslide volume model (Figure 3.S7). Using the Longmen Shan seismologically-based landslide
volume model (Figure 3.5b), we calculate an erosion rate of 0.44-0.96 mm yr
-1
(with a central
estimate of 0.51-0.81 mm yr
-1
), which reduces to 0.34-0.81 mm yr
-1
(with a central estimate of
0.40-0.69 mm yr
-1
) taking into account focal depth (Figure 3.6b). The rates calculated using the
global empirical regression (Keefer, 1994) and the global seismologically-based model (Marc et
al., 2016b) are lower than the result using our Longmen Shan-specific model. This outcome is not
surprising, since the two former models underestimate the observed landslide volumes for
Wenchuan and Lushan, a discrepancy that we attribute to uncertainties in applying
globally-calibrated parameters to a specific region (see Section 4.3.1, above). On the other hand,
using the global seismologically-based model with the ~5× increase in material sensitivity that
captures the observed Longman Shan volumes (Figure 3.5d) yields a seismic erosion rate of
0.37-1.68 mm yr
-1
, with a central estimate of 0.73-0.99 mm yr
-1
, close to that calculated from the
Longmen Shan-specific model although slightly higher because of the strong non-linearity of the
global model. These comparisons emphasize the sensitivity of landslide erosion calculations to
model assumptions, and particularly the importance of using locally calibrated parameters since
global average values may not accurately reflect a specific region (as suggested by Keefer, 1994).

3.5.3. Comparing magnitudes of denudation rates across different timescales
3.5.3.1. Average denudation rate of the frontal Longmen Shan
The catchment area-weighted means (±1 standard deviation) of pre-Wenchuan and
post-Wenchuan denudation rates inferred from hydrological gauging are 0.24±0.04 mm yr
-1
and
1.09±0.13 mm yr
-1
respectively. The equivalent average kyr-timescale cosmogenic
nuclide-derived denudation rate is 0.55±0.05 mm yr
-1
. For denudation over Myr timescales, we
interpolate the exhumation rates in the intensive erosion zone and run 1,000,000 Monte Carlo
random simulations to account for the uncertainties from individual exhumation rate estimates.
The resulting Myr denudation rate is 0.61+0.14/-0.08 mm yr
-1
(uncertainties indicate the 16
th
and
84
th
percentiles of the Monte Carlo results). The inferred seismic erosion rates over multiple
earthquake cycles range from 0.34 to 0.81 mm yr
-1
across the range of plausible T values
(500-4000 years), using the Longmen Shan-specific seismologically-based V
L
-M
w
relation.

3.5.3.2. Pre- and Post-earthquake sediment fluxes versus long-term denudation rate
The 2006-2007 pre-Wenchuan denudation rate calculated from sediment fluxes is lower than the
long-term denudation rate. The rate reported from the 1960s to 1980s gauging data is higher than
that calculated from the 2006-2007 data, but still slightly lower than the long-term rate. In
contrast, the post-Wenchuan sediment flux-derived denudation rate is the highest amongst all
observed denudation rates (Figure 3.7). The long-term average denudation rates (i.e., from
65

cosmogenics and thermochronology) thus fall between the immediate pre- and post-earthquake
values derived from sediment fluxes. In general terms, this pattern is consistent with a conceptual
model for erosional dynamics over one complete earthquake cycle in which the pre-earthquake
rates reflect below-average values and the post-earthquake rates reflect an immediate
post-seismic denudational pulse (Ouimet, 2010). However, given the large variability in
pre-earthquake rates (comparing 2006-2007 data vs. 1960s-1980s), it is difficult to use these data
to evaluate quantitatively the extent to which the post-earthquake pulse contributes to the
long-term budget; considering the 2006-2007 data alone, the post-earthquake pulse would need to
play a major role to make the long-term value, but considering the data from the 1960s-1980s,
this role would need to be relatively small over the long term.

3.5.3.3. Seismic erosion rate vs. long-term denudation rate
Our best estimate of the seismic erosion rate (0.34-0.81 mm yr
-1
, using the Longmen
Shan-specific landslide volume model) is comparable to the measured long-term denudation rates
of the frontal Longmen Shan (kyr denudation rate: 0.55±0.05 mm yr
-1
; Myr exhumation rate:
0.61+0.14/-0.08 mm yr
-1
) (Figure 3.7). Using the global seismologically-based model for
landslide volumes yields a slightly higher calculated denudation rate via earthquake-induced
landslides (0.37-1.68 mm yr
-1
, with central estimate of 0.73-0.99 mm yr
-1
; Figure 3.S7).
Denudation rates determined either from gauging data and long-term chronometers include mass
loss via dissolved load in addition to via physical erosion. If dissolved load is mainly derived
from weathering of landslide material (e.g., Emberson et al., 2016), the calculated seismic
erosion rate should be directly comparable to the denudation rate. Significant non-landslide
solute sources, such as from ground water release (e.g., Jin et al., 2016), would increase the
denudation rate compared to the calculated seismic erosion rates, although such effects should be
small since dissolved load represents a relatively low proportion of the total denudation flux
(~20%, Section 3.4). Either way, additional dissolved load contributions would not explain a
lower measured denudation rate compared to the calculated seismic erosion rate.

We recognize that not all landslide material is necessarily evacuated from a mountain belt by
rivers within one earthquake cycle. Landslide erosion rates could be higher than actual measured
long-term denudation rates if new seismically triggered landslides re-mobilize debris associated
with prior earthquakes. Given the relatively rapid rates of sediment evacuation observed in the
Longmen Shan region compared to the long earthquake recurrence times (Liu et al., 2013) and
the lack of extensive storage of intra-montane landslide debris in this setting (Parker et al., 2011),
we view near-complete evacuation as a reasonable first-order assumption for comparing rates. If
this is the case, the seismic erosion rates calculated from the Longmen Shan-specific landslide
volume model are consistent with the measured long-term denudation rates.

Whether the seismic erosion rate is similar in magnitude or slightly higher than the measured
long-term rate (Figure 3.7), our results demonstrate that, at least in this region,
earthquake-triggered landslides are capable of sustaining denudation rates that are comparable to
66

long-term averages at the mountain belt scale and over timescales of multiple seismic cycles. A
corollary is that the observed erosional fluxes are likely to be dominated by material derived from
earthquake-triggered landslides, with implications not only for sediment dynamics and landscape
evolution, but also for biogeochemical cycles (e.g., Jin et al., 2016; Emberson et al., 2016).

Over longer timescales, climate fluctuations may influence the capacity of the sediment routing
systems for removing landslide debris and maintaining river incision and hillslope weathering
rates required to sustain landslides. Indeed, we find that high intensity runoff events help to
explain the observed short-term denudation rates in the Longmen Shan (Fig. 3a), pointing to the
importance of climate variability as an erosional agent. Thus we expect that climate – which may
have changed in response to topographic evolution (e.g., Molnar et al., 2010) – interacts with
seismic events to influence the pace of surface processes in the Longmen Shan, even though a
primary control may be related to the seismotectonic processes that trigger landslides and thus
convert rocks to erodible sediment. Our results suggest that earthquake-triggered landslides can
focus denudation in mountains where rivers have the capacity to transport and export the excess
sediment supplied from hillslopes (e.g. Fig. 3a). In these settings the denudation can be
considered as tectonically limited (Montgomery and Brandon, 2002), yet fully understanding
topographic evolution will require better understanding how landslide frequency and magnitude,
and the associated fluvial export of sediment, respond to long-term climatic fluctuations.

3.6. Conclusions and Implications
By considering the effects of multiple earthquakes, we calculate that co-seismic landslides can
sustain denudation rates that are similar to the rates recorded by chronometers over timescales of
thousands to millions of years (Figure 3.7). In addition to similar magnitudes, we find a
first-order spatial coincidence of long-term denudation and Wenchuan-triggered landslides
(Section 4.1, Figure 3.2), and a correlation between the sediment-derived denudation rates and
co-seismic landsliding (Figure 3.4d). Over the long term, the location of earthquake-triggered
landslides should reflect earthquake sources and ground motion attenuation, with landslide
density decreasing away from seismogenic faults (Meunier et al., 2007), as we also observe for
the Wenchuan case.

Our observations suggest that the long-term location and rate of denudation in the Longmen Shan
is consistent with the focusing of seismic energy release along range-bounding faults, with the
resulting landslides serving as a primary mechanism sustaining denudation fluxes. Landslides can
only continue to operate as effective denudation mechanisms if hillslopes are sufficiently steep. It
is likely that the combination of uplift, river incision, and fluvial evacuation of sediment (e.g.,
Burbank et al., 1996; Bennett et al., 2016) is capable of continually re-steepening and
re-weakening failed landscapes between earthquake cycles. Earthquakes should increase the
efficiency of landslide generation for a given steepness and hillslope strength, such that the
location of most intense erosion at the orogenic scale may be determined by seismic energy
release. These effects could be enhanced by rock weakening resulting from greater deformation
67

and orographic rainfall close to mountain fronts (Gallen et al., 2015; Vanmaercke et al., 2017).
We thus suggest that focused denudation along a high-relief plateau margin may be regulated at
least in part by the location and activity of seismogenic faulting, and specifically by the resulting
earthquake-triggered landslides.

3.7. Supplementary materials to Chapter 3
3.7.1. Calculation of specific stream power
Specific stream power is calculated using the formula in Burbank et al. (2003):

ω = ρ
w
gQG/W (3.S1)

where ρ
w
is the density of water, g is the gravitational acceleration, Q is water discharge (the
product of upstream area and catchment area-normalized runoff), G is the channel gradient, and
W is the channel width (calculated from an empirical scaling relation: W = 10
-2
Q
0.4
(Burbank et a.,
2003; Liu-Zeng et al., 2011; Ansberque et al., 2015)). Specific stream power is calculated for
catchments with drainage area > 1 km
2
, thought to be the regional threshold for fluvial channels
(Li et al., 2016).

3.7.2. The ratio between dissolved load and suspended load
Jin et al. (2016) reported the chemical composition of the dissolved load together with discharge
at weekly frequency following the Wenchuan earthquake from two sites located on the Min Jiang
river. Following Liu-Zeng et al. (2011), we calculate the daily dissolved load by multiplying the
total concentrations of dissolved solids (mg L
-1
of Ca
2+
, Mg
2+
, K
+
, Na
+
, SiO
2
, Cl
-
, SO
4
2-
, and
HCO
3
-
) and the corresponding discharge (m
3
day
-1
). We interpolate the daily dissolved load and
integrate the daily fluxes to get annual fluxes. The annual fluxes of the dissolved load are then
compared to the fluxes of the suspended sediment. For the downstream Min Jiang site at Weizhou,
the ratio between the total dissolved load and the suspended load is 30%, similar to the ratio
reported before the Wenchuan earthquake (~35%, Chen et al., 2002). For the upstream Min Jiang
site at Zhenjiangguan, the ratio is 75%, much higher than the downstream ratio. However, the
drainage area upstream Zhenjiangguan only represents 6% of the total area of the studied
catchments, whereas the drainage area upstream of Weizhou represents 28% of the total area
(including areas above Zhenjiangguan), so we view the ratio at Weizhou as likely to be more
representative. In a comprehensive compilation, Liu-Zeng et al. (2011) found a dissolved
load:suspended load ratio of 35±15%, similar to our estimate for Weizhou. Combining these
values, we calculate a mean ratio of 38±14% (Table 3.S3).

The total dissolved load is corrected for atmospheric and anthropogenic inputs to obtain the
denudation-derived dissolved load (i.e., from chemical weathering of rocks) using a coefficient of
49.5±2.5% (ratio of denudation-dissolved load to total dissolved load), as suggested by Liu-Zeng
et al. (2011). This coefficient for the denudation-dissolved load has been estimated independently
for the Min Jiang, Jialing Jiang (a neighboring river of the Min Jiang), Yangtze (Chang Jiang) and
68

a compiled dataset of the global rivers, and the results from different studies are consistent
around 50% (Summerfield and Hulton, 1994; Chetelat et al., 2008; Liu-Zeng et al., 2011 and
references therein).

Finally, the ratio between the denudation-dissolved load and the suspended load is the product of
the coefficient for the denudation component (49.5±2.5%) and the total dissolved load:suspended
load ratio (38±14%), yielding a dissolved:suspended ratio for denudation of 19±7% as reported
in the main text.

3.7.3. Threshold runoff for high magnitude runoff events
We examine the relationships between denudation rate and an index of high magnitude runoff
events (the proportion of total runoff from events above a specified threshold runoff) across a
wide range of runoff values (Figure 3.S2). We find best correlations at a threshold runoff of 6 mm
day
-1
for data from both before and after the Wenchuan earthquake.

3.7.4. Principal component analysis (PCA) of studied metrics
The result of the PCA analysis is reported in Figure 3.S5 and Table 3.S4 and Table 3.S5. All
metrics cluster into two groups: seismic (metrics with loading factor > 0.9) versus non-seismic
(Table 3.S4). These two groups explain > 85% of the total variance carried by the dataset (Table
3.S5). The high magnitude runoff metric shows moderate correlation with both groups, but
considering the physical meaning of this metric, we attribute it to the non-seismic group.

3.7.5. Applying a global seismologically-based model to the Longmen Shan
Marc et al. (2016) derived a seismologically-based expression for the total volume of
earthquake-triggered landslides, as:
) ( ), exp( ) ( ) 1 (
0
mod 2
0
2
0
R a S b
T
S
I
L
a R
S b
A R a V
c
SV asp c
topo c V L
    
(3.S2)
where V
L
is the predicted landslide volume,
V
 is an average material sensitivity term (global
average = 4174±212 m
3
km
-2
), a
c
is a critical acceleration to trigger landsliding (0.15±0.02 g), R
0

is the mean asperity depth (in the absence of data from rupture inversions, R
0
equals focal depth
H for M
w
<7.5 and half of focal depth H for M
w
≥7.5), L is the fault rupture length (calculated
using an empirical scaling relation reported below), I
asp
is the characteristic length of an asperity
(I
asp
= 3km), S
mod
is a modal slope angle determined from the slope histogram for specific cases,
A
topo
is the proportion of areas with sufficient steepness for landslide failure (slope angle > ~10°),
T
SV
is a steepness scaling constant (global average = 11.6±0.6°), b is the inferred source
acceleration at a 1km distance from the source generating 1Hz seismic waves, and
S
is an
average amplification factor of seismic acceleration due to site effects. The term
S b
is calculated
using equations Eq. 3.S4 and Eq. 3.S5, below (cf. Marc et al., 2016).

69

The fault rupture length is determined using an empirical scaling relation (Leonard, 2010):

Log
10
L = m
1
M
w
+ m
2
(3.S3)

where m
1
and m
2
are scaling constants determined from the regression of a global dataset (m
1
=
0.599,m
2
= -2.497) (Leonard, 2010).

The term
S b
is determined as (cf. Marc et al., 2016):

M M M M e M M e S b S b
h w h w h w sat
) ( ), ) ( ) ( exp(
2
6 5
     (3.S4)

M M M M e S b S b
h w h w sat
) ( )), ( exp(
7
   (3.S5)

where M
h
is a “hinge” magnitude (acceleration above this hinge magnitude saturates at b
sat
, with
M
h
= 6.75±0.1), b
sat
is the saturated value for acceleration at 1Hz, and e
5
, e
6
and e
7
are empirical
constants for 1Hz seismic waves (empirically determined from 58 worldwide earthquakes; e
5
=
0.6728, e
6
= -0.1826, e
7
= 0.054) (Boore and Atkinson, 2008; Marc et al., 2016). S b
sat
is chosen
as 4000±400 m (Marc et al., 2016).

To apply this landslide volume model to specific cases, seismological parameters (L, R
0
) and
topographic parameters (S
mod
, A
topo
) need to be constrained independently, for example using
local seismological and digital topographic data, respectively, whereas other parameters can be
inferred from globally-averaged values (Marc et al., 2016), or if possible, calibrated for specific
region.

To apply the model to the Longmen Shan region, we use the Wenchuan and the Lushan cases
(Wenchuan: focal depth H = 19km, R
0
= 9.5km, L = 170km, S
mod
= 31±4°, A
topo
= 1; Lushan:
focal depth H = R
0
= 14km, L = 20km, S
mod
= 19±2.5°, A
topo
= 0.8) (Burchfiel et al., 2008; Zhang
et al., 2014; Marc et al., 2016). The uncertainties are calculated from 1,000,000 Monte Carlo
random sampling simulations for each earthquake with specified magnitude. For each set of
parameters, we calculate the results for 1000 earthquakes across a continuous range of
magnitudes (M
w
= 5-8). The results are reported as the medians (the solid curves connecting 1000
data points on Figure 3.5b), and the 16
th
and 84
th
percentiles (the dashed curves on Figure 3.5b)
of the Monte Carlo results.

70

Implementing this model across multiple earthquakes (i.e., to calculate a seismic erosion rate
over simulated earthquake cycles) requires some estimate of the focal depth of each synthetic
earthquake event. Using the Wenchuan and Lushan data, we fit a relation between M
w
and focal
depth of the form (following Leonard, 2010):

Log
10
H = m
3
M
w
+ m
4
(3.S6)

where m
3
and m
4
are empirical scaling constants. This M
w
-focal depth scaling relation is used for
determining the focal depth and associated mean asperity depth when applying the landslide
volume model of Marc et al. (2016) over multiple synthetic earthquakes.

S6. Development of a seismologically-based model for the Longmen Shan
Using the Wenchuan ground motion and landslide dataset, we first establish the relationships
between landslide volumetric density (volume of landslides in per unit area, m
3
km
-2
) and peak
ground accelerations (PGA). The landslide volumetric density is calculated every 2 km increment
along the swath profile A-A’. PGA is calculated as the mean of the PGA data in the
corresponding increment.

For the hanging wall (where >90% of the Wenchuan landslide volume is located), we find the
relation best described using a power-law function (r
2
= 0.82) (Figure 3.S3a):

Log
10
α
V
= 7.24(PGA – 0.10)
0.28
, (PGA > 0.1g) (3.S7)

where α
V
is the landslide volumetric density.

For the footwall, we find the relationship is best described using a logarithmic function (r
2
= 0.95)
(Figure 3.S3a):

Log
10
α
V
= 13.79PGA – 4.24, (PGA > 0.3g) (3.S8)

We then refer to a ground motion attenuation law to link PGA to earthquake magnitude and
distance to the fault (Figure 3.S3b) (Cui et al., 2012; Wang et al., 2015a):

Log
10
PGA = c
1
+ c
2
M
w
+ c
3
Log
10
(D + c
4
) (3.S9)

where D is the distance to the fault rupture, and c
1
, c
2
, c
3
and c
4
are empirical constants. Note that
(1) this function is determined based on the ground motion data from seismic events in the
Longmen Shan region and the neighboring areas (Cui et al., 2012; Wang et al., 2015a and
references therein), and (2) we slightly modify the original law using the distance to fault rupture,
replacing this term with the distance to hypocenter, assuming linear sources. This simplification
still returns good fits for the Wenchuan ground motion data (Figure 3.S3b). In applying Eq. 3.S8,
71

we thus use c
2
=0.21 as determined from regional regression (cf. Cui et al., 2012) and define c
1
, c
3

and c
4
using the Wenchuan ground motion data (USGS, 2008).

We next simulate ground motion for earthquakes occurring in the Longmen Shan region with
specified magnitude using Eq. 3.S8 and derive associated landslide volumes using Eq. 3.S6 and
Eq. 3.S7. We calculate the volumes of landslides triggered by 20 earthquakes with different
magnitudes M
w
=5 to M
w
=8. We find the resulting M
w
-landslide volume relation can be well
described using a power-law function (Eq. 3.4 and Figure 3.5c in the main text).

S7. Estimating the probability of landslide-triggering for earthquakes
Some earthquakes may be too deep to trigger landslides. To evaluate how this effect influences
seismic erosion rates calculated using the Longmen Shan specific model, we have excluded a
proportion of earthquakes from the calculation. We determine probable earthquake depths with
reference to the seismic catalog collected by the China Seismograph Network stations
(1970-2015) (CSN catalog, 2015), which reports earthquake location, surface wave magnitude,
local magnitude, and focal depth. We extract the earthquakes in the Longmen Shan region and
convert all earthquake magnitudes to moment magnitudes (M
w
) using the scaling relations
reported in Wang et al. (2015a) and Scordilis (2006).

To assess the probability that an earthquake of a given magnitude would trigger landslides in this
region, we use the Wenchuan and Lushan cases as defining the limiting conditions for landslide
triggering. With reference to the earthquake magnitude-focal depth scaling relation calibrated by
the Wenchuan and the Lushan data (Eq. 3.S12), we see the determined focal depth H a
characteristic depth for landslide-triggering. Earthquakes with focal depth deeper than this depth
are assumed to generate little landslides, so the landslide-triggering probability of earthquakes is
defined as the proportion of earthquakes with focal depth shallower than the corresponding
characteristic depth (red curve in Figure 3.S4a). Note that different thresholds for landslide
triggering (i.e., different parameters in Eq. 3.S12) would have limited affect on the conclusions of
our study, since this correction for earthquake depth is small.

All the recorded earthquakes in the Longmen Shan region are then grouped by magnitude. The
proportions of surface rupturing earthquakes are calculated for each magnitude bin containing >
40 events, which includes earthquakes with M
w
< ~5.6, and we fit a non-linear relationship
between M
w
and the proportion of surface rupturing earthquakes (Figure 3.S4b), assuming
Wenchuan-like events have 100% probability triggering landslides. We use this non-linear
relationship as an estimate of the probability of earthquakes that can trigger landslides for a given
magnitude and consequently adjust the seismic erosion rate. This approach provides an upper
estimate for the magnitude of correction, since some earthquakes occurring at deeper depth may
also trigger co-seismic landslides.

72

S8. Simulating earthquake sequences using a truncated Gutenburg-Richter relation
We use the Caputo equation (Utsu, 1999) as a truncated Gutenburg-Richter function to generate
earthquake sequences:

Log
10
n(M
w
) = g
1
– g
2
M
w
– Log
10
(1-10
-g2(g3-Mw)
), (M
w
≤ g
3
and M
w
≤ M
wmax
) (3.S10)

and

n(M
w
) = 0, (M
w
> M
wmax
or M
w
> g
3
) (3.S11)

where n(M
w
) is an exceedance frequency or the frequency (yr
-1
) of earthquakes with magnitude ≥
M
w
, g
1
, g
2
and g
3
are empirical constants determined from regional seismicity data, and M
wmax
is
the maximum earthquake magnitude in the study area (M
w
= 7.9 for the Longmen Shan).

The number of earthquakes with specified magnitudes over time t, N(M
w
), are calculated
numerically as:

N(M
w
) = t × n(M
w
) – t × n(M
w
+ ΔM
w
) (3.S12)

where ΔM
w
is set as 0.1 (cf. Keefer, 1994).

Figur

Figur
M
w
6.6
from
landsl
topog
earthq
earthq
conto
river w
and t
e 3.1
e 3.1. Maps o
6 2013 Lusha
the USGS h
lides and p
graphy, denud
quake and t
quake-trigger
urs (dashed l
water sample
the regional
of topography
an earthquake
azards progra
ost-Wenchua
dation rates,
the M
w
6.6 L
ed landslides
lines), the tre
es (Jin et al., 2
context of th
y of the study
es, the Wench
am, http://ear
n gauging-d
PGA, and l
Lushan earth
s (yellow po
end of the 17
2016) (square
he study are
73
area, location
huan peak gro
rthquake.usgs
derived denu
andslides. (a
hquake epice
olygons) ove
70 km-wide s
es), Wenchuan
a (inset pane
ns of the M
w
7
ound accelera
s.gov/earthqu
udation rates
a) Map of th
enters (red s
er shaded rel
swath profile
n earthquake
el); (b) map
7.9 2008 Wen
ations (PGA)
uakes), Wench
s, and swath
he M
w
7.9 20
stars), mappe
lief map, We
e (A-A’), sam
surface ruptu
of post-seis

nchuan and th
) (gridded dat
huan-triggere
h profiles o
008 Wenchua
ed Wenchua
Wenchuan PGA
mpling sites o
ure (red lines
smic sedimen
he
ta
ed
of
an
an
A
of
),
nt
74

flux-derived denudation rates upstream of hydrological gauging stations (circles); denudation
rates are normalized to areas with elevation > 800 m to account for limited contribution from flat
frontal plains; (c) swath profile of the topography projected along A-A’ showing the mean (white
line) and maximum and minimum elevations (grey area); blue dashed lines delimit the zone with
intensive denudation between profile distances of 120 km and 200 km); (d) swath profile of
slopes (°) projected along A-A’ with mean slope (white line) ±1 standard deviation on the mean
slope (grey area); (e) swath profile of pre-Wenchuan earthquake (2006-2007, red circles) and
post-Wenchuan earthquake (2008-2012, grey squares) denudation rates (y-axis error bars = ±1 s.d.
uncertainties in denudation rates, x-axis error bars = square root of catchment area), the
decreasing trend of post-seismic denudation rates along the swath is shown by least-squares
fitting of denudation vs. distance along A-A’ (solid black line); for denudation in the frontal
Sichuan basin, we divide the estimated denudation fluxes by the total catchment area including
areas with elevation < 800 m, and this provides an upper limit; (f) swath profiles sampled at 5
km-intervals along A-A’ of PGA (grey) and the volumetric density for all landslides (red, m
3
km
-2
,
landslide volume over the specified area).

Figur

Figur
rates
Longm
elevat
(b) sw
s.d. u
apatit
are ±
(ZFT)
comp
with d
(c), (d
e 3.2
e 3.2. Swath
and seismicit
men Shan ra
tion envelop;
wath profile o
uncertainties,
te fission track
1 s.d. uncerta
)-determined
iled exhumat
distance < 25
d) and (e) sho
h profiles of
ty across the
ange (A-A’);
blue dashed
of
10
Be-derive
and on the x
k (AFT) and
ainties; (d) sw
exhumation
tion rates (red
0 km, data po
ow the Wench
topography,
Longmen Sh
white line –
lines bound th
ed millennial
-axis the squ
apatite (U-Th
wath profile
rates with ±1
d bars: range
oints are binn
huan landslid
75
millennial de
han range. (a)
mean elevat
he intensive d
denudation ra
uare roots of t
h)/He (AHe)-
of zircon (U-
1 s.d. uncerta
of exhumatio
ned in 5 km-w
de distribution

enudation rat
) Swath profi
tions; grey a
denudation zo
ates; error bar
the catchmen
-determined e
-Th)/He (ZH
ainties (error
on rates inclu
wide incremen
n, as in Figure
tes, geologica
ile of topogra
area – minim
one as defined
rs on the y-ax
nt area; (c) sw
exhumation ra
He) and zircon
bars); (e) sw
uding uncertai
nts); the grey
e 3.1f; (f) sw
al exhumatio
aphy along th
mum-maximum
d in Figure 3.
xis indicate ±
wath profile o
ates; error bar
n fission trac
wath profile o
inties; for dat
curves on (b
wath profiles o
on
he
m
.1;
±1
of
rs
ck
of
ta
),
of
76

topography (black solid curve: mean elevations; grey solid curves: minimum and maximum
elevations), historic seismicity (1970-2015, grey circles sized by the estimated magnitude; star:
Wenchuan earthquake) (CSN Catalog, 2015) and seismic moment release (red curve, data binned
in 5 km-wide increments), and a simplified sketch of the Longmen Shan fault system (WMF:
Wenchuan-Maowen fault; YBF: Yingxiu-Beichuan fault; PGF: Pengxian-Guanxian fault)
(Liu-Zeng et al., 2011; Ansberque et al., 2015 and references therein).

Figur

Figur
Shan
pre-W
with r
(c), an
5 km-
stream
(f), P
Stream
repres
e 3.3
e 3.3. Sedim
plotted vers
Wenchuan (squ
runoff > 6 mm
nd relief (d);
-radius (diam
m power calcu
Post-Wenchua
m Profiler (h
sent ±1 s.d. un
ment flux-deri
us hydrologi
uares) denuda
m day
-1
; post-
relief is calcu
monds) circula
ulated using t
an denudation
http://www.ge
ncertainties.
ived denudati
ical and topo
ation rate as a
-Wenchuan de
ulated as the
ar windows.
the approach
n rate vs. no
eomorphtools
77
ion rates for
ographic met
a function of
enudation rat
ranges of elev
(e), Post-Wen
in Burbank et
ormalized cha
.org) and no
the catchme
trics. (a), Po
the proportio
te plotted vers
vations over
nchuan denud
t al., (2003) a
annel steepne
ormalized to

ents draining
ost-Wenchuan
on of total run
sus annual ru
2.5 km-radiu
dation rate ag
and Ansberqu
ess index cal
θ
ref
= 0.45. A

the Longme
n (circles) an
noff from day
unoff (b), slop
us (circles) an
gainst specifi
ue et al. (2015
lculated usin
All error bar
en
nd
ys
pe
nd
ic
5).
ng
rs

Figur

Figur
Post-W
maxim
(d) la
regres
uncer
bars f
densit
(i.e., r
rando
e 3.4
e 3.4. Relatio
Wenchuan de
mum PGA; (c
andslide volu
ssion (perfor
rtainties of th
for distance to
ties are repor
ranges of ±1
om sampling s
onships betwe
enudation rat
c) mean dista
umetric dens
rmed in the
e fits; error b
o the fault rup
rted as the m
s.d. in a stand
simulations.
een sediment f
te as a func
ance of each c
sity. Red sol
logarithmic
bars for denu
pture are the
median values
dard normal d
78
flux-derived d
ction of: (a)
catchment to t
lid lines sho
space for
udation and P
square roots
and the 16
th
distribution) f
denudation ra
catchment m
the Yingxiu-B
ow best fits
d); grey da
PGA indicate
of catchment
h
and 84
th
per
from the resu

ates and seism
mean PGA;
Beichuan fau
from linear
ashed lines s
±1 s.d. uncer
t area; landsli
rcentiles of th
ults from 1000
mic parameter
(b) catchmen
ult rupture; an
r least-square
show ±1 s.d
rtainties; erro
ide volumetri
he distributio
0 Monte Carl
rs.
nt
nd
es
d.
or
ic
on
lo

Figur

Figur
betwe
2004)
uncer
Lusha
mode
from
of ear
uncer
mode
using
range
(mean
freque
sensit
accou
Inform
sensit

e 3.5
e 3.5. Earth
een earthquak
); solid line r
rtainties (resid
an data, respe
l; the crosses
5 to 8; the so
rthquake mag
rtainties (resid
l (Marc et al
the Wenchua
at 98% of m
n asperity dep
ency defined
tivity; dashed
unting for unc
mation); (d)
tivity.
quake-trigger
ke magnitude
represents the
dual errors) o
ectively; (b)
s represent th
olid curve repr
gnitude (detai
dual errors)
l., 2016b); th
an parameters
modal slope fre
pth = 14 km,
d in Marc et
d curves sh
certainties of
the modeled
red landslide
and associat
e logarithmic
of the fit; th
Longmen Sh
he modeled la
resents the be
ils in Supplem
of the fit; (c
he blue and g
s (mean asper
equency defin
modal slope
t al., 2016b)
how 68% co
the relevant p
results from

79
e volume mo
ted landslide
c least square
he blue and t
han-specific, s
andslide volum
est fit of the m
mentary Mat
c) a global s
grey curves r
rity depth = 9
ned in Marc e
= 19±2.5°, un
, respectively
onfidence int
parameters (M
m the model o
odels. (a) Gl
volume (Kee
es linear fit;
the grey circ
seismological
mes for 20 ea
modeled land
terials); the d
seismological
refer to the m
.5 km, modal
et al., 2016b)
ncertainty: ra
y, both using
terval from
Marc et al., 20
of Marc et a

lobal empiric
efer, 1994; M
dashed lines
les are Wenc
lly-based lan
arthquakes w
slide volumes
dashed curves
lly-based land
modeled land
l slope = 31±4
and the Lush
ange at 98% o
g global ave
Monte Carl
016b; details
al. 2016b wit
cal regressio
Malamud et al
show ±1 s.d
chuan and th
ndslide volum
with magnitud
s as a functio
s show ±1 s.d
dslide volum
dslide volume
4°, uncertaint
han parameter
of modal slop
erage materia
o simulation
in Supportin
th 5× materia
on
l.,
d.
he
me
de
on
d.
me
es
ty:
rs
pe
al
ns
ng
al

Figur

Figur
calcul
descri
(white
from
G-R r
T ~ 5
Shan
interv
assum
thresh

e 3.6
e 3.6. Earthq
lated seismic
ibed by a tru
e circles) (W
paleoseismol
relation for T
500 years (Th
seismologica
val for Wench
ming landslid
hold focal dep
quake magnitu
c erosion rate
uncated Guten
Wang et al., 2
ogical/geodet
= 3000 years
hompson et a
ally-based lan
huan-alike ev
es are trigge
pth for earthqu
ude-frequency
e. (a) Longm
nberg-Richter
015b) and th
tic studies (gr
s (Shen et al.
al., 2015); (b
ndslide volum
vents, T; shad
ered by all si
uake triggerin

80
y distribution
men Shan ea
r relation (Ut
he recurrence
rey-shaded zo
, 2009); dash
) calculated s
me model as
ded area show
imulated eart
ng.
n of the Long
arthquake ma
tsu, 1999) usi
e interval (T)
one); red dot
hed line – line
seismic erosi
a function o
ws ±1 s.d; the
thquakes; the

gmen Shan re
agnitude-frequ
ing historic s
for Wenchu
and solid cur
ear G-R relati
ion rate with
of the estimat
e solid curve
e dashed cur
egion, and th
uency relatio
seismicity dat
uan-like event
rve – truncate
ion, which fit
the Longme
ted recurrenc
shows result
rve includes
he
on
ta
ts
ed
ts
en
ce
ts
a

Figur

Figur
circle
hydro
rates
determ
AHe)
the ca
uncer
denud
Tian
exhum
across
red ba
black

e 3.7
e 3.7. Denud
s show pre-W
ological gaug
reported in
mined from
1
. For catchme
atchment area
rtainty. For
dation rate is
et al. (2013)
mation rate es
s the range o
ar for all eart
ticks for note
dation rates of
Wenchuan an
ging; the squ
Liu-Zeng et
0
Be measurem
ent-scale den
a-weighted m
the low te
reported as
, with uncert
stimate (Tian
of estimated W
thquakes, a d
ed T values; d
f the frontal L
nd post-Wenc
are shows pr
t al. (2011);
ments and low
nudation (rive
means, and the
emperature th
the average
tainties propa
et al., 2013).
Wenchuan re
dashed red ba
dashed grey li

81
Longmen Sha
chuan earthqu
re-Wenchuan
; yellow circ
w temperatur
r gauging and
e error bars re
hermochrono
after interpol
agated from t
Red bars wit
currence inte
ar considering
ine = 1 s.d. un
an across diff
uake denudat
n earthquake
cles show lo
re thermochro
d
10
Be studie
epresent catch
ology-based
lating the exh
the reported
th black ticks
ervals (T=500
g focal depth
ncertainties.

fferent timesc
tion rates det
(1960s-1980
ong-term den
onology analy
s), the denud
hment area-w
denudation
humation dat
uncertainties
show seismi
0-4000 years)
threshold (se

ales. The blu
termined from
0s) denudatio
nudation rate
ysis (AFT an
dation rates ar
weighted 1 s.d
estimate, th
ta compiled i
s in individua
c erosion rate
), with a soli
ee Fig. 6), an
ue
m
on
es
nd
re
d.
he
in
al
es
id
nd

Figur

Figur
DEM
red do
regres

Figur

Figur
denud
indica
(2006

e 3.S1
e 3.S1. Corr
data for the s
ots represent
ssion. The gre
e 3.S2
e 3.S2. Thr
dation rate an
ate post-Wen
6-2007). Runo
elations betw
seven studied
slope data. T
ey dashed line
eshold runof
nd the propor
nchuan data
off data from
ween slopes d
d catchments w
he black solid
e indicates 1:
ff versus the
rtion of total
a (2008-2012
Wang et al. (2
82
derived from
which have c
d line denotes
1 line.
e squares o
l runoff from
2). The blu
2015b).
m SRTM30 D
complete cove
s the best fit
f correlation
m days > thre
ue dots indi
DEM data ve
erage of SRTM
from the line

n coefficient
eshold value.
icate pre-We
ersus SRTM9
M30 data. Th
ar least squar
(r
2
) betwee
The red dot
Wenchuan dat
90
he
re
en
ts
ta

Figur

Figur
earthq
volum
Yingx
wall a

Figur

Figur
surfac
earthq
crosse
scalin
earthq
black
propo
assum
trigge

e 3.S3
e 3.S3. Relat
quake, and b
metric densit
xiu-Beichuan
and from the f
e 3.S4
e 3.S4. Earth
ce in the st
quakes from
es refer to th
ng relation ca
quakes that ru
bars represe
ortions from t
mption that ea
er landslides.
tionships bet
between PGA
ty and PG
fault rupture
footwall, resp
quake magni
tudy area. (a
the China Se
e seismic cat
alibrated by
upture the sur
ent calculated
the least squa
arthquakes w
tween landsli
A and distan
GA; (b) rela
e. Red crosses
pectively.
tude, focal de
a) Earthquak
eismograph N
talog data; th
the Wenchu
rface for magn
d proportions
ared fit of th
with magnitud

83
ide volumetri
nce to fault.
ationships b
s and blue cr
epth, and pro
ke magnitude
Network catal
e red curve r
an and the L
nitude-binned
s from (a); t
he observed p
de of the We
ic density an
. (a) Relatio
between PGA
rosses are for
oportions of e
e and corres
log (1970-20
represents a M
Lushan earth
d earthquakes
the red dashe
proportions (M
enchuan even

nd PGA for t
onships betw
A and dist
r the data from

arthquakes ru
sponding foc
15) in the stu
M
w
-surface ru
hquakes; (b)
s in the seism
ed lines repr
M
w
=4.5-5.6)
nt (M
w
=7.9)
the Wenchua
ween landslid
tance to th
m the hangin
upturing to th
cal depth fo
udy area; gre
upturing dept
proportion o
mic catalog; th
resent inferre
and from th
would alway
an
de
he
ng
he
or
ey
th
of
he
ed
he
ys

Figur

Figur
comp

Figur

Figur
mean
signif

e 3.S5
e 3.S5. PCA
onent. PC2 co
e 3.S6
e 3.S6. Relat
slope; (b) C
ficant correlat
plot for the s
orresponds to
ions between
Catchment m
tions are foun
studied metric
o the non-seis
n PGA, slope
mean PGA v
nd between PG

84
cs. Principle c
mic compone
and runoff. (
vs. catchment
GA and slope
component (P
ent.
(a) Catchmen
t annual mea
e and runoff.
PC) 1 indicat
nt mean PGA
an runoff. N
tes the seismi

vs. catchmen
No statisticall
ic
nt
ly

Figur

Figur
seism
the Lo
regres
using
using
increa
Mont
the Su
the Lo
(T); t
proba
correc

e 3.S7
e 3.S7. Seism
mologically-ba
ongmen Shan
ssion; the sha
the global s
the global av
ase in the ma
e Carlo simu
upplementary
ongmen Shan
the solid red
ability of land
ction; the sha
mic erosion r
ased model of
n-specific mod
aded area rep
seismological
verage mater
aterial sensiti
ulations accou
y Information
n landslide vo
d curve den
dslide-triggeri
ded area repr
rates calculate
f Marc et al. (
del. (a) Seism
present ±1 s.d
ly-based mod
rial sensitivity
ivity; the sha
unting for unc
text (Marc et
olume model
otes the resu
ing; the dashe
esent ±1 s.d.

85
ed using the
(2016) with a
mic erosion ra
d. uncertainti
del vs. T; th
y; the dashed
aded area rep
certainties of
t al., 2016); (
vs. the recur
ults without
ed curve refe
uncertainties.
global empir
an adjusted m
ates calculated
ies; (b) seism
he solid blue
d blue curve d
presents 68%
f the relevant
c) seismic ero
rrence interva
focal depth
ers to the resu
.
rical regressi
material sensiti
d using the gl
mic erosion ra
curve refers
denotes the re
% confidence
t parameters,
osion rates ca
al of Wenchu
h-based corre
ults with foca
on, the globa
ivity term, an
obal empirica
ates calculate
to the result
esults with 5
interval from
as reported i
alculated usin
uan-like event
ection for th
al depth-base
al
nd
al
ed
ts
×
m
in
ng
ts
he
ed
86

Table 3.1. Notation for symbols

Symbol Notation Unit
A Area of landslide occurrence over earthquake cycles km
2

c
1
c
2
c
3
c
4
Parameters of ground motion attenuation equation, Longmen Shan
D Distance to fault rupture km

Seismic erosion rate mm yr
-1

k
sn
Normalized channel steepness index m
0.9

M
w
Earthquake moment magnitude -
N(M
w
) Number of earthquakes in the magnitude bin [M
w
, M
w
+ 0.1]
R
0
Mean asperity depth km
S
mod
Modal slope °
t Time period of multiple earthquake cycles yr
T Recurrence interval for Wenchuan-like earthquakes yr
V
L
Landslide volume m
3

θ
ref
Reference concavity index -
ω Specific stream power
J m
-2

yr
-1

e 
87

Table 3.S1. Denudation rates and metrics of topography, hydrology, fluvial transport capacity,
and seismic processes associated with the Wenchuan earthquake of the studied catchments
draining the Longmen Shan mountain range

Main
catchment

ID
1

Catchment
notation
Controlling
hydrological
stations
Pre-
Wenchuan
(2006-2007)
denudation
rate (mm
yr
-1
)
Post-
Wenchuan
(2008-2012)
denudation
rate (mm
yr
-1
)
Pre-
Wenchuan
(2006-2007)
proportion of
runoff from
high
magnitude
runoff events
(> 6 mm
day
-1
)
Post-
Wenchuan
(2008-2012)
proportion of
runoff from
high
magnitude
runoff events
(> 6 mm
day
-1
)
Runoff
(2008-2012,
mean±1s.d.)
(mm yr
-1
)

G1
Min Jiang
Pengshan
to Dujiangyan
Pengshan 0.26±0.04 0.74±0.15 33.46 46.74 440±340
G2 Guojiaba Guojiaba 0.09±0.01 0.52±0.07 8.46 34.57 663±184

G3
Min Jiang
Dujiangyan
to
Zhenjiangguan
Dujiangyan N.A.
3
N.A.
3
0.00 0.00 N.A.
3

Min Jiang G4
Lower
Zagunao
Sangping 0.04±0.02 0.13±0.03 1.90 3.64 527±178
G5 Upper Zagunao Zagunao 0.14±0.02 0.12±0.02 8.08 12.17 808±83
G6 Lower Heishui Shaba 0.12±0.02 0.04±0.01 0.87 1.38 471±109
G7 Upper Heishui Heishui 0.09±0.01 0.09±0.01 6.81 12.02 773±84

G8
Min Jiang
above
Zhenjiangguan
Zhenjiangguan 0.05±0.01 0.07±0.01 0.00 0.54 350±76
Tuo Jiang G9 Tuo Jiang main Dengyinyan 0.42±0.06 1.20±0.16 5.86 18.83 583±113

G10
Kai Jiang
above Santai
Santai 0.14±0.02 0.96±0.13 21.11 27.78 442±114

G11
Fu Jiang
Shehong
to Fujiangqiao
Shehong N.A.
3
N.A.
3
42.12 52.66 N.A.
3

G12
Fu Jiang
Fujiangqiao
to Jiangyou
Fujiangqiao 0.19±0.04 1.36±0.21 12.59 22.60 828±278
Fu Jiang G13
Pingtong He
above Ganxi
Ganxi 0.14±0.02 0.69±0.09 16.29 25.60 509±102

G14
Upper Zitong
Jiang
above Zitong
Zitong 0.36±0.02 0.98±0.12 64.95 64.18 512±196

G15
Fu Jiang
Jiangyou
to Pingwu
Jiangyou N.A.
3
N.A.
3
65.78 60.97 N.A.
3

G16
Fu Jiang
above Pingwu
Pingwu 0.17±0.02 0.32±0.04 0.98 8.21 764±121
1
All metrics are calculated for mountainous areas with
elevation > 800 m;
2
Landslide results are reported as the medians and the 16th and the 84th
percentiles of the results from 1000 Monte Carlo simulations propagating
uncertainties from parameters in landslide area-volume scaling [Li et al.,
2014];

3
Sediment/discharge is not available because the calculations return negative
values; see main text;
4
Landslide data are not available due to no coverage of
satellite imagery.

88

Table 3.S1. Continued

Slope
(mean
±1 s.d.)
(°)
Relief (2.5
km-radius,
mean±1s.d.
) (mm yr
-1
)
Relief (5
km-radius,
mean±1s.d
.) (mm
yr
-1
)
Log
10
specific
stream
power (J
m-2 yr-1)
(mean±1s.
d.)
Normalize
d
steepness
index
Log
10
ksn
(mean±1s.
d.)
Catchment
mean PGA
(g)
Catchm
ent
maximu
m PGA
(g)
Catchment
mean
distance to
the
Yingxiu-Beic
huan fault
rupture (km)
Landslide volumetric
density (10
3
m
3
km
-2
)
2

23±10 1082±342 1507±401 9.06±0.40 2.08±0.31 0.54±0.19 0.94 10 28.36+72.21/-19.92
28±10 1476±318 2049±364 9.39±0.41 2.28±0.20 0.54±0.11 0.86 12 62.91+129.55/-42.89
30±10 1622±351 2204±439 N.A.
3
2.35±0.24 0.37±0.19 1.08 40 256.11+596.58/-184.35
31±10 1691±265 2299±341 9.43±0.46 2.37±0.23 0.23±0.07 0.54 67 15.07+38.61/-10.50
30±10 1635±338 2176±424 9.49±0.44 2.33±0.23 0.16±0.04 0.25 87 10.81+26.37/-7.57
26±10 1223±412 1608±532 9.25±0.52 2.20±0.32 0.12±0.04 0.28 121 0.11+0.24/-0.07
27±9 1281±250 1711±305 9.40±0.44 2.25±0.24 0.10±0.01 0.14 137 0.12+0.26/-0.08
23±9 927±255 1226±336 9.04±0.45 2.07±0.25 0.11±0.03 0.19 129 N.A.
4

27±12 1390±556 1931±705 9.22±0.45 2.16±0.37 0.68±0.12 0.91 3 510.17+1397.51/-366.24
28±10 1232±232 1656±271 9.13±0.40 2.13±0.26 0.65±0.10 0.91 4 1860.96+6218.85/-1449.55
26±10 1085±249 1389±281 N.A.
3
2.10±0.26 0.61±0.08 0.78 1 383.13+960.30/-269.54
29±10 1373±337 1808±429 9.43±0.43 2.25±0.21 0.36±0.17 0.79 37 55.78+135.23/-39.57
27±9 1119±204 1399±235 9.22±0.40 2.13±0.17 0.38±0.12 0.73 26 6.08+14.23/-4.21
23±11 967±278 1227±296 8.93±0.40 1.97±0.38 0.39±0.07 0.58 14 N.A.
4

24±11 1040±271 1385±344 N.A.
3
2.13±0.22 0.48±0.13 0.77 18 12.26+29.36/-8.62
30±10 1470±351 1981±429 9.42±0.45 2.29±0.20 0.17±0.05 0.32 82 7.48+20.77/-5.36

89

Table 3.S2. Notation for symbols (including the ones used in supplementary materials)
Symbol Notation Unit
A Area of landslide occurrence over earthquake cycles km
2

a
c
Critical acceleration to trigger landslides g
A
topo
Proportion of steep landscape %
b
Inferred source acceleration at 1km
from the source with a given frequency (1 Hz)
m
b
sat
saturation acceleration for the scaling of b m
c
1
c
2
c
3
c
4
Parameters of ground motion attenuation law, Longmen Shan
D Distance to fault rupture km

Seismic erosion rate mm yr
-1

e
5
e
6
e
7
Parameters of ground motion attenuation law, global
G Channel gradient m m
-1

g Gravitational acceleration
g
1
g
2
g
3
Parameters of truncated Guthenberg-Richter distribution
H Focal depth km
I
asp
Characteristic length of asperity km
k
sn
Normalized channel steepness index m
0.9

L Fault rupture length km
m
1
m
2
Parameters of M
w
-fault rupture length scaling relation
m
3
m
4
Parameters of M
w
-focal depth scaling relation
M
h
Hinge magnitude
M
w
Earthquake moment magnitude
n(Mw) Exceedance frequency of earthquakes with magnitude > M
w
yr
-1

N(M
w
) Number of earthquakes in magnitude bin [M
w
, M
w
+ 0.1]
Q Water discharge m
3
year
-1

R
0
Mean asperity depth km

Average site effects on the amplification of acceleration
S
mod
Modal slope °
t Time period of multiple earthquake cycles Myr
T Recurrence inteval for Wenchuan-like events year
T
SV
Landscape steepness scaling constant °
V
L
Landslide volume m
3

W Channel width m
δ
V
Material sensitivity to landsliding m
3
km
-2

θ
ref
Referene concavity index
ρ
w
Density of water
ω Specific stream power J m
-2
yr
-1

e 
S
90

Table 3.S3. Compiled total dissolved load:suspended load ratios. Note only the two data from this
study (Weizhou and Zhenjiangguan) are from after the 2008 Wenchuan earthquake. Other data
are all from before the Wenchuan earthquake.

River Station
Mean
annual
suspended
load
(10
4
ton
year
-1
)
Mean
annual
dissolved
load
(10
4
ton
year
-1
)
Dissolved
:
suspended
ratio
Reference
Weizhou 807 245 0.30 Jin et al., 2016; this study (post-Wenchuan)
Zhenjiangguan 64 48 0.75 Jin et al., 2016; this study (post-Wenchuan)
Min Pengshan 1030 360 0.35 Liu-Zeng et al., 2011; Chen et al., 2002
Jiang Duoyingping 935 210 0.22 Liu-Zeng et al., 2011; Chen et al., 2002
Duoyingping 935 220 0.24 Liu-Zeng et al., 2011; Qin et al., 2006
Jiajiang 1034 297 0.29 Liu-Zeng et al., 2011; Chen et al., 2002
Shimian 1498 715 0.48 Liu-Zeng et al., 2011; Chen et al., 2002
Yanrun 239 50 0.21 Liu-Zeng et al., 2011; Chen et al., 2002
Wutongqiao 3470 1499 0.43 Liu-Zeng et al., 2011; Chen et al., 2002
Qingshuixi 145 73 0.50 Liu-Zeng et al., 2011; Chen et al., 2002
Gaochang 4950 1580 0.32 Liu-Zeng et al., 2011; Chen et al., 2002
Xuankou 905 320 0.35 Liu-Zeng et al., 2011; Qin et al., 2006
Sangping 172 67 0.39 Liu-Zeng et al., 2011; Qin et al., 2006
Yibing 4950 1936 0.39 Liu-Zeng et al., 2011; Chetelat et al., 2008
Gaocheng 4950 2202 0.44 Liu-Zeng et al., 2011; Qin et al., 2006
Dadu He Luding 970 557 0.57 Liu-Zeng et al., 2011; Qin et al., 2006
Tuo Jiang Sanhuangmiao 595 199 0.33 Liu-Zeng et al., 2011; Chen et al., 2002
Fu Jiang Fujiangqiao 1086 195 0.18 Liu-Zeng et al., 2011; Chen et al., 2002
Yangtze Datong 50000 17422 0.35 Liu-Zeng et al., 2011; Zhang et al., 1990
main Wuhan 50000 22600 0.45 Liu-Zeng et al., 2011; Hu et al., 1982
mean 0.38±0.14

91

Table 3.S4. Factor loading matrix from the PCA analysis (first two principle components or PCs
explain > 85% of the total variance)

Metric PC1 PC2
Slope -0.0930.945
Relief 5 km-radius 0.023 0.962
Relief 2.5 km-radius -0.063 0.984
Log
10
k
sn
-0.376 0.899
Mean runoff -0.321 0.572
Proportion of total
runoff from high
magnitude runoff events
(> 5 mm day
-1
)
0.705 -0.463
mean PGA 0.944 -0.246
max PGA 0.942 -0.256
distance to fault -0.977 0.140
Log
10
landslide
volumetric density
0.943 0.183

Table 3.S5. Total variance explained by the components

Component
% of
Variance Cumulative %
1 57.598 57.598
2 29.330 86.929
3 7.166 94.095
4 2.925 97.020
5 1.607 98.628
6 0.848 99.475
7 0.280 99.755
8 0.212 99.967
9 0.025 99.992
10 0.008 100.000

92

This page is left blank intentionally

93

Chapter 4
Mountain building over earthquake cycles considering erosion driven by
earthquake-triggered landslides

4.1. Preamble
Chapter 3 demonstrated that in seismically active mountains, earthquake-triggered landslides
could be a major contributor to the long-term erosional budget. This lays the foundation for
constraining the erosion term in mountain building on a seismological basis. If rock uplift is also
dominated by seismic processes (e.g., in collisional plate boundaries), it is promising to probe
mountain belt evolution by combining observations and theories of seismic deformations and
earthquake-triggered landslides. In this context, a primary question is the earthquake volume
balance problem, i.e., what is the net effect of the competition between seismically induced rock
uplift and earthquake-triggered landsliding erosion. Previous studies modeling surface
deformations over earthquake cycles had loose constraints on the erosion term (King et al., 1988).
Empirical observations were focused on single events (Hovius et al., 2011; Parker et al., 2011)
but did not consider how volume balance varies over earthquake magnitudes. Other studies
investigated co-seismic volume balance across earthquake magnitudes but have not accounted for
the effects of inter-seismic processes: post-seismic relaxation and isostatic compensation to
erosional unloading (Li et al., 2014; Marc et al., 2016a; Molnar, 2012). In this chapter, I
developed a generalized framework to investigate mountain building over earthquake cycles,
synthesizing co-seismic and inter-seismic processes, and landslide erosion. I adopted a modeling
approach, developed a model to simulate the 2-D surface deformation produced over full
earthquake cycles, and discussed how and to what extent earthquake cycles build or destroy
mountainous topography, in the context of a new metric: the efficiency of earthquakes in building
topography. It is also promising to apply this new metric to real cases to better understand how
earthquakes doing geomorphic work.

I was the main author, developed the 2-D surface deformation model to simulate topographic
growth over earthquake cycles, and wrote the manuscript. Josh West was the main editor, and
suggested considering broad geodynamic implications of the model. Hongrui Qiu joined in
building the post-seismic relaxation component and general discussions. We thank funding from
the U.S. National Science Foundation (NSF-EAR/GLD grant 1053504 to Joshua West). This
manuscript benefited from conversations with Josh Roering, Yehuda Ben-Zion, James Dolan,
Niels Hovius, Haoran Meng, Zhigang Peng, Mong-Han Huang and Joel Scheingross.

This manuscript is in preparation as:
Li, G., A. J. West, and H. R. Qiu, Mountain building over earthquake cycles considering erosion
driven by earthquake-triggered landslides.
94

4.2. Introduction
Mountain ranges are among the most conspicuous landforms, and they have global-scale effects
on the Earth system (e.g., Molnar and England, 1990). Thrust faulting-earthquakes are important
drivers of mountain uplift via repeated vertical displacement (e.g., Avouac, 2007). However, large
earthquakes can also cause widespread landslides that facilitate export of sediment from
mountains (Keefer et al., 1994; Hovius et al., 2011; Wang et al., 2015), effectively destroying the
seismically uplifted topography. The significant erosive power of large earthquakes raises
fundamental questions about how earthquakes build mountainous topography.

The volume of earthquake-triggered landslides can be comparable to or even exceed that of
co-seismically induced rock uplift (Parker et al., 2011; Hovius et al., 2011; Li et al., 2014; Marc
et al., 2016a). However, this “co-seismic” volume balance is complicated because removal of
landslide material does not actually occur co-seismically (Hovius et al., 2011; Wang et al., 2015).
Moreover, over the earthquake cycle, evacuation of landslide debris is accompanied by
flexural-isostatic compensation to erosional unloading (King et al., 1988; Densmore et al., 2012;
Molnar, 2012). And, although co-seismic deformation causes focused uplift along fault zones, the
deformed lithosphere gradually relaxes during inter-seismic time periods, distributing co-seismic
deformation through rheologically-dependent post-seismic relaxation (King et al., 1988, Huang et
al., 2014).

Models representing the combined effect of these multiple processes can explain first-order
topographic forms, but prior analyses have adopted a simplified representation of erosion (King
et al., 1988). Advances in understanding the total volume and pattern of earthquake-triggered
landslides (Malamud et al., 2004; Meunier et al., 2007; Marc et al., 2016b), together with data
showing that these landslides can account for the erosional flux from tectonically active
mountains (Keefer, 1994; Hovius et al., 1997; Li et al., 2017), lay the foundation for constraining
the erosional term on a seismological basis – providing the opportunity to more completely
address the problem of earthquake volume balance as well as shedding new light on how fault
properties influence topography.

In this study, we develop a generalized model to simulate the 2-D topographic structure created
over earthquake cycles, by combining geophysical solutions to co-seismic and inter-seismic
deformation and predictions for the magnitude and pattern of earthquake-triggered landslides.
This model allows us to more completely evaluate the earthquake volume balance problem and to
consider how earthquake cycles might produce simplified orogenic structures, including how
fault properties influence topographic wavelengths and the focusing of denudation.

95

4.3. The efficiency of earthquakes in building topography
We define the efficiency of earthquakes in building topography (E, %) as the ratio of the volume
of the topography remaining after one full earthquake cycle versus the volume of co-seismically
uplifted topography, within a specified spatial window:
seismic co
isostacy seismic post ls seismic co
V
V V V V
E

 
   

(4.1)
where V
co-seismic
represents the volume from co-seismic deformation (i.e., associated with fault
slip), ΔV
post-seismic
is the change in V
co-seismic
during post-seismic relaxation, V
ls
is the volume of
earthquake-triggered landslides (as an approximation of the total eroded volume during the
earthquake cycle), and V
isostacy
represents isostatic compensation to erosion. V
co-seismic
and V
isostacy

both represent the sum of local uplift and subsidence, with positive values meaning net uplift and
negative values net subsidence.

We then set:

seismic post seismic co seismic post
V V V
  
  
,
seismic co
ls
V
V

 
,
ls
isostacy
V
V
k  , and
seismic co
seismic post
V
V
r




and rewrite Eq. 4.1 as:

E = (1 - k)×(1 - Ω) + (r + k - 1) (4.2)

Eq. 4.2 decomposes E into three independent factors: the ratio between the volume of landslides
versus co-seismic deformation ( Ω), a factor of inter-seismic relaxation (r) and a factor of
flexural-isostatic compensation (k). Eq. 4.2 allows separate exploration of the roles of co-seismic
and inter-seismic processes in topographic growth over earthquake cycles (Figure 4.1).

4.4. Model setup
Here we briefly summarize how we simulated deformations caused by different processes.
Details are provided in the Supplementary Materials.
4.4..1. Fault implementation and co-seismic deformation
We simplified the lithosphere-asthenosphere system as an elastic layer of thickness T
e
overlying a
viscoelastic half space (Figure 4. S1, King et al., 1988). A thrust fault with specified geometry
was defined within the elastic layer. The fault’s length was determined using an empirical scaling
relation with earthquake magnitude M
w
(Leonard, 2010). Deformation was induced by imposing
a dislocation along the fault. The co-seismic deformation field (the unrelaxed asthenosphere
response to the dislocation) was calculated using an analytical solution to a 2-D dip-slip
dislocation model (Cohen, 1996).

96

4.4.2. Post-seismic relaxation
The complete time-dependent relaxed-asthenosphere solution is presented in Thatcher and
Rundle (1984) using a Thomson-Haskell propagator matrix method. For computational simplicity,
we adopted an analytical approximation of this solution at long-enough relaxation times
(>10
2
-10
3
yrs) using a plate flexure approach (Savage and Gu, 1985).

4.4.3. Seismic landslide erosion
We predicted the volume of earthquake-triggered landslides (V
ls
) using a model accounting for
earthquake magnitude, landscape sensitivity to landsliding, asperity depth, landscape steepness,
and saturation effects of seismic ground motion, after Marc et al. (2016b; see complete
expression in Supplementary Materials). For simplicity, we assume constant slope angles but
recognize that real landscapes evolve with time. For reference, we also considered an empirical
linear regression between V
ls
and M
w
(Malamud et al., 2004; see Supplementary Materials).

To model the 2-D landslide pattern, we adopted an empirical relation (Meunier et al., 2007) and
assumed a linear seismic energy source:

) exp(
0
0 0
0
 R
R d
d
R
P P
Vls

 
(4.3)

where P
Vls
is landslide volumetric density (volume of landslides in unit area, m
3
km
-2
), d is the
distance to the energy source, R
0
is the depth of the energy source, and P
0
and β are scaling
factors. β is the spatial decay factor, with higher values meaning more widely spread landsliding.
Eq. 4.3 is analogous to the law of seismic wave attenuation accounting for both geometric
spreading and quality decay, and has successfully reproduced patterns of landslides caused by the
Chi-Chi, Northridge, Finisterre (Meunier et al., 2007) and Wenchuan (Figure 4. S2) earthquakes.
The boundaries of the landsliding-impacted region were determined from an empirical relation
based on M
w
(Keefer, 1994). We assume landslide erosion results in complete removal of material
within an earthquake cycle and for simplicity do not consider the effects of sedimentation in
adjacent basins, which could be added in future work.
4.4.4. Flexural-isostatic responses
Flexural-isostatic responses were calculated as the flexure due to erosional unloading using a
flexural-isostacy model (King et al., 1988; Watts, 2001). Landslide-induced erosion was
approximated as a series of linear unloads, and the flexure caused by each segment of unloading
was calculated using the numerical approach of King et al. (1988) and the references therein.

97

4.4.5. Earthquake return time statistics
To describe earthquake occurrence over multiple earthquake cycles, we adopted a
Gutenberg-Richter M
w
-frequency relation:

Log
10
N = a - bM
w
(4.4)

where N is the number of earthquakes with magnitude ≥ M
w
within a defined time period. Setting
b = 0.9 (the global average; Malamud et al., 2004), we modeled earthquake sequences with
varying seismic product a.

4.5. Patterns of seismically induced deformations
We find that over one full earthquake cycle, different processes contribute to producing distinct
topographic structures. Co-seismic deformation creates focused uplift in a narrow zone above the
fault plane, with far field subsidence on the hanging wall, and a combination of near field
subsidence and far field bulging on the footwall (Figure 4.2a). Inter-seismic relaxation distributes
the localized, co-seismic deformation to far field areas, reducing the near field uplift and
enhancing the hanging wall’s far field uplift and the footwall’s subsidence (Figure 4.2a).
Earthquake-triggered landslide erosion mainly focuses in the same narrow zone as co-seismic
deformation and rapidly decays in the far field (Figure 4.2b). Flexural-isostatic compensation to
erosional unloading is much more widely distributed as compared to landsliding, featuring a
bulge in the near field and depressions in the far field (Figure 4.2b, c). Since seismically induced
deformations vary spatially, the total volume balance and E depend on the spatial window
analyzed. We explored a wide range of geologically reasonable values for the width of this
window, from <50 km to >500 km, and focus on two representative cases: (1) a “near field
window” on the hanging wall where most co-seismic uplift and earthquake-triggered landslide
erosion occur (e.g., Densmore et al., 2012; Li et al., 2014), with the width of this window
determined from an empirical scaling relation between earthquake magnitude, rupture length and
the area impacted by landslides (varying from ~20-200 km across M
w
5-9; Keefer, 1994; Leonard,
2010); and (2) a “far field window” centered at the fault rupture with a width of 200 km, which
covers near field deformations (both foot wall subsidence and hanging wall uplift) and a major
part of far field deformation (e.g., Marc et al., 2016a). For reference, the widths of modern-day
tectonically active mountain belts (e.g., Taiwan) are normally around 50-100 km (Hovius, 1996).

4.6. The ratio between co-seismic uplift and landslide erosion
Prior studies have explored how Ω (the ratio of co-seismic uplift to seismic landslide erosion)
varies across earthquake magnitudes (Li et al., 2014; Marc et al., 2016a) – the so-called
“co-seismic” balance. Precise quantification of Ω is complicated by several factors. First, scaling
relations between fault geometry, fault displacement, and M
w
are only weakly defined and differ
whether based on displacement models (Leonard, 2010) or purely empirical scaling relations
(Wells and Coppersmith, 1994), introducing large uncertainties to the estimate of seismic uplift
and thus Ω (Figure 4.S3). Secondly, the scaling relation between seismic landslide volume and
98

M
w
is complex. For example, predicted V
ls
depends on the extent of saturation of ground motion
at high M
w
, which in turn depends on the sensitivity of landsliding to specific frequencies of
seismic waves. If landslides are triggered by ~1 Hz seismic waves, the significant saturation
effect around a hinge magnitude of ~6.75 leads to strong non-linearity in the V
ls
-M
w
relation,
resulting in low Ω for moderate-magnitude earthquakes (Marc et al., 2016a). If instead most
landslide triggering is attributable to a different frequency of spectral acceleration, or to peak
ground velocities (PGV) or peak ground acceleration (PGA), the M
w
-dependence of V
ls
and
consequently Ω will differ (Meunier et al., 2007; Boore and Atkinson, 2008; Dreyfus et al., 2013;
Athanasopoulos-Zekkos et al., 2016; Figure 4.3). Finally, estimates of Ω depend significantly on
the scale of the spatial window considered (Figures. 4.S4, 4.S5).

Altogether, these uncertainties obscure conclusive interpretation of the interplay between
landslide erosion and co-seismic uplift, though one consistent observation is that the volume of
landslides can approach or even exceed the uplift volume ( Ω < 1) for some but not all earthquake
magnitudes (Figure 4.1). By separating Ω analytically from the inter-seismic relaxation factor r
and flexural-isostacy factor k (Eq. 4.2), we can explore how each component modulates E,
independently of the uncertainties on Ω.

4.7. Inter-seismic controls on topographic growth
In a specified setting, the normalized pattern of the structure produced by co-seismic deformation
plus inter-seismic relaxation is mainly controlled by flexural rigidity, as represented by effective
elastic thickness, T
e
(e.g., King et al., 1988). Both hanging wall uplift and footwall subsidence
decay more slowly towards far field areas with greater T
e
values (Figure 4.2). The inter-seismic
relaxation factor r, or the ratio of the relaxed volume versus the co-seismic volume, depends on
the study window. For the 200 km-wide far field window, as T
e
increases, r increases first and
then decreases, as a result of the competition between subsidence and uplift. Within the near field
window, as T
e
increases, r increases accordingly, because the near field window only includes the
uplifted part on the hanging wall and this volume gets larger as T
e
increases.

The structure of the topography produced by flexural-isostatic restoration is determined by both
the pattern of landslide-driven erosional unloading and the flexural rigidity. Thus the isostatic
factor, k, is controlled by the landslide spatial decay factor β and the effective elastic thickness T
e
.
Within the far field window, this ratio varies little around 0.8, which is the upper bound
equivalent to a situation with no flexure (i.e., the Airy isostacy). For the near field window, this
ratio varies significantly from 0.8 to 0.2, as 1/ β varies across 0-5 (e.g., for the Chi-Chi earthquake,
1/ β ~0; for the Wenchuan earthquake, 1/ β ~3; Supplementary Materials) and T
e
varies from 2-40
km, typically observed values (Densmore et al., 2012). Figure 4.3 summarizes the variation of E
versus 1/ β and T
e
over a single earthquake cycle. In the near field-window, the efficiency is quite
sensitive to changes in T
e
and β. In the far field, E seems to be less sensitive to changes in T
e
and
β, though with a relatively higher dependence on T
e
(Figure 4.3).

99

One provocative question is whether some earthquakes act to destroy rather than build
topography (Parker et al., 2011). Medium- to high-magnitude earthquakes have been proposed as
being net destructive (Li et al., 2014; Marc et al., 2016a). We find net topographic destruction (E
< 0) does occur over near field windows when Ω < 1, T
e
is high, and 1/ β is low, i.e., with more
focused landsliding (Figure 4.3). However, when considering the far field window, the effect of
inter-seismic processes means that few if any earthquakes are in fact net destructive. And
integrating across events of varying magnitude throughout earthquake cycles using
Gutenberg-Richter earthquake return time statistics yields positive long-term uplift rates.
Considering a wide range of seismicity parameters yields geologically consistent rates of ~0.1 –
10 km Myr
-1
, with high magnitude events contributing most to the total uplift (e.g., Figure 4.S1).

4.8. Wavelength of seismic topography and implications for plateau construction
Landslides are concentrated in the near field but one of their effects is to broaden the region of
deformation relative to co-seismic effects alone, because the flexural response to volume loss is
distributed as a function of lithospheric rigidity (Fig. 5). Thus we expect erosion to be focused
near range-bounding faults (Li et al., 2017), while uplift extends over wide areas. Indeed, the
wavelength of predicted topography scales with earthquake magnitude and lithospheric rigidity
(Figure 4.5). These observations point to the possibility that low-relief interior plateau of
collisional mountain ranges may be an expected consequence of aging, mature faults, generating
fewer, higher magnitude, and deeper earthquakes within thickened lithosphere. In contrast, steep,
mountainous margins may form as deformation propagates outwards (Poblet and Lisle, 2010) and
younger, less mature faults produce shorter wavelength, higher-relief topography (Figure 4.5c).
The orographic rainfall caused by the steep topography could in turn facilitate removal of the
landslide-derived sediment, though untangling cause and effect remains difficult (cf. Molnar and
England, 1990). Our analysis does not consider the evolution of relief with time, as in a
landscape evolution model, but our results suggest that the evolving characteristics of faults
themselves should be considered in such models and that the cumulative effect of multiple faults
of varying maturity may shape first-order features of large-scale topography.

4.9. Supplementary materials to Chapter 4
4.9.1. Fault settings and mechanical properties
In our fault setting, we simulate the lithosphere-asthenosphere system as a viscoelastic half space
with density ρ
A
(3300 kg m
-3
), overlaid by an elastic layer characterized by thickness T
e
and
density ρ
L
(2700 kg m
-3
). The dimensions of the seismogenic fault are determined from empirical
relations between earthquake magnitude (M
w
) and fault dimensions (Leonard, 2010). We also
conducted modeling using another group of the scaling relations reported by Wells and
Coppersmith (1994). The two studies reported similar scaling relations between surface rupture
length, fault width and M
w
, but different M
w
-fault displacement relations. Specifically, Wells and
Coppersmith (1994) used empirical regressions to describe the M
w
-displacement relation,
whereas Leonard (2010) derived a new M
w
-displacement relation based on a self-consistent
displacement model. The dip angle is set as 45°.
100

To describe the rheological properties of the system, the flexural rigidity D of the lithosphere can
be calculated as:

) 1 ( 12
2
3
 

e
ET
D (4.S1)

where E is the Young’s modulus (70×10
10
N m
-2
), and ν is the Poisson’s ratio (0.25). D varies as
a function of T
e
.

The critical wavenumber of the deformation is:

k = ( Δ ρg/D)
1/4
(4.S2)

where Δ ρ is the difference between the density of the asthenosphere versus the atmosphere and is
approximated as the density of the asthenosphere (3300 kg m
-3
), and g is the gravitational
acceleration (9.8 m s
-2
).

For computational simplicity, we assumed that the footwall and the hanging wall were two
separated blocks. The physical meaning is that over many earthquake cycles, the largest fault has
penetrated through the lithosphere with depth greater than T
e
. This assumption is reasonable as
most collisional orogens are characterized by T
e
~10-40 km, whereas a surface rupturing M
w
9
earthquake

4.9.2. Co-seismic deformation
The co-seismic deformation represents the immediate response caused by a fault dislocation,
which is the same as the deformation in an elastic half space (King et al., 1988). We calculated
the co-seismic displacement using the analytical solution reported by Cohen (1996):

)]} ( 1 [
2 ) (
) arctan(
2
{
sin
) (
2 2
x sign
D x x
xD
D
x x b
x u
D
D
co
 
 


 
 


(4.S3)

where u
co
(x) is the displacement caused by co-seismic deformation at coordinate x, b is the
imposed fault displacement (burger vector), θ is the dipping angle, and D is the fault depth, and
x
D
= Dcot θ.

4.9.3. Post-seismic relaxation
Rundle et al. (1980) first fully resolve the time-dependent post-seismic relaxation following
co-seismic faulting using the Thomson-Haskell (propagator) matrix method. In this work, we
used the solutions reported in Savage and Gu (1985) to calculate the ultimate relaxation after
101

visco-elastic readjustment to co-seismic dislocation is finished. Based on plate flexure theory, the
solutions reported in Savage and Gu (1985) are approximations for a special case-the completely
relaxed lithosphere-asthenosphere system-of the full time-dependent solutions of post-seismic
visco-elastic relaxations by Rundle et al. (1980), Rundle et al. (1982) and Rundle and Thatcher
and Rundle (1984). We adopted the solutions by Savage and Gu (1985) not only because of the
computational efficiency and simplicity, but also because the relevant timescale (Maxwell
relaxation time, ~100s-1000s yrs) in the context of tectonic mountain building. The relaxed
post-seismic deformation is calculated as:
)] ( ) cot ( ) 1 [( sin
2
1
) (
post
x C T x C b x u
e
L
A
A
L
co
   







, (x > T
e
cot θ)
(4.S4)

)] ( ) cot ( ) 1 [( sin
2
1
) (
post
x C T x C b x u
e
L
A
A
L
co
    







, (x < 0)
(4.S5)
and

)] ( ) cot ( ) 1 ( 2 [ sin
2
1
sin ) (
post
x C T x C b b x u
e
L
A
A
L
co
     







 , (0 ≤ x ≤ T
e
cot θ)
(4.S6)

where u
co+post
(x) is the displacement caused by co-seismic deformation and post-seismic
relaxation at coordinate x (Figure 4.S1), ρ
L
is the density of the lithosphere, ρ
A
is the density of
the asthenosphere, and T
e
is the effective elastic thickness.

4.9.4. Landslide volume
In the main text, we adopt a seismologically-based landslide volume model by Marc et al. (2016):

) ( ), exp( ) ( ) 1 (
0
mod 2
0
2
0
R a S b
T
S
I
L
a R
S b
A R a V
c
SV asp c
topo c V ls
    
(4.S7)

where V
ls
is the volume of earthquake-triggered landslides,
V
 is the material sensitivity (global
mean = 4174±212 m
3
km
-2
), a
c
is the landslide-triggering threshold acceleration (0.15±0.02 g), R
0

is the mean asperity depth, A
topo
is the proportion of landscape area with sufficient steepness for
landsliding (>~10°), L is the surface rupture length of the seismogenic fault as determined from
M
w
-based scaling relations, I
asp
is the characteristic length of an asperity patch, S
mod
is the modal
slope angle for the studied landscape, T
sv
is a steepness normalization constant (global average =
11.6±0.6).

102

We have also considered a global regression between earthquake magnitude and total landslide
volume, as reported in Malamud et al. (2004), which is an update of the regression by Keefer
(1994).

V
ls
= 1.42M
w
– 11.26(±0.52) (4.S8)

4.9.5. Flexural-isostatic compensation to landslide erosion
To model flexural-isostatic compensation to landslide erosion, we approximate the landslide
erosion as a series of linear unloads. The deflection at site x contributed by one linear unload at
site x
1
is calculated using the equation below (cf. King et al., 1988). The total deflection at site x
is obtained by integrating all deflections from each linear unload.

) ' sin ' (cos
2
) ' (
'
isostacy
kx kx
g
Fk
e x u
kx


 


(Eq. 4.S9)

where u
isostacy
means the displacement caused by flexural-isostatic responses to landslide
erosion-induced unloading, g is the gravitational acceleration, k is the wavenumber, x’ is the
distance between x and x’ (calculated as |x
1
– x|), Δ ρ is the difference between the density of the
asthenosphere versus the atmosphere and can be approximated as the density of the
asthenosphere ( ρ
A
), and F is the unload calculated from Eq. 4.S9 below:

F = -e ρ
s
gdx (Eq. 4.S10)

where e is the displacement caused by landslide erosion and is calculated from landslide
volumetric density (P
Vls
in the main text) and ρ
s
is the density of clastic sediment (2700 kg m
-3
).

Figur

Figur

Figur
of (a)
with p
landsl

res
e 4.1
e 4.1. The eff
the landslide
post-seismic r
liding-induce
ficiency of top
e volume : co-
relaxation and
d erosional un
pographic gro
-seismic uplif
d co-seismic d
nloading (k).

103
owth over a co
ft ratio ( Ω), (b
deformation (

omplete earth
b) the ratio be
(r), and (c) th
hquake cycle
etween volum
he isostatic co

as a function
mes associated
ompensation t

d
to

Figur

Figur
unloa
inter-s
erosio
comp
inter-s
flexur
and 2

e 4.2
e 4.2. Post-se
ading. Fault d
seismic relax
on varying as
ensation to l
seismic relax
ral-isostatic fa
00 km-wide f
eismic relaxa
depth 40 km.
ation normali
s a function o
landslide eros
xation factor
actor k versus
far field (dash
ation and flex
Dip angle 4
ized by maxim
of landslide sp
sional unload
r r versus T
s T
e
. k and r a
hed lines).
104
xural isostatic
5°. (a) Vertic
mum co-seism
patial decay f
ding varying
T
e
; (e) flexu
are plotted for
c compensatio
cal field of c
mic deformati
factor β; (c) p
with effectiv
ural-isostatic
r both 60 km-
on to landslid
co-seismic de
ion; (b) patter
pattern of fle
ve elastic thi
factor k ve
-wide near fie

ding erosiona
formation an
rn of landslid
exural-isostati
ckness T
e
; (d
ersus 1/ β; (f
eld (solid line
al
nd
de
ic
d)
f)
es)

Figur

Figur
thickn
km; d
and (e

e 4.3
e 4.3. The eff
ness T
e
, and th
d-f) are both c
e), E vs. 1/ β a
ficiency over
he landslide s
considered. Fa
and Ω; (c) and
one full earth
spatial decay f
ault depth 40
d (f), E vs. 1/ β

105
hquake cycle (
factor β. Near
km. Dip angl
β and T
e
.
(E) as a funct
r field (60 km
le 45°. (a) and
tion of Ω, effe
m; a-c) and far
d (d), E vs. Ω

fective elastic
r field (200
Ω and T
e
; (b)

Figur

Figur
as the
define
et al.
inter-s
as a fu
(assum
earthq
schem
differ

e 4.4
e 4.4. (a) Wav
e horizontal ra
ed as the hori
1994) as a fun
seismic proce
unction of eff
ming T
e
equat
quake magnit
matic figure sh
ent sized faul
velength of to
ange within 2
zontal width
nction of eart
esses (the sum
fective elastic
tes the depth
ude can be de
howing platea
lts and varied
opography cau
0% of maxim
of the full lan
thquake magn
m of inter-seis
c thickness T
e
of a 45° dip-s
etermined usi
au-shaped top
d lithospheric

106
used by co-se
mum deformat
ndslide erosio
nitude; (b) Wa
smic relaxatio
and equivale
slip seismoge
ing the scaling
pography exp
rigidity.
eismic deform
tion) and land
on zone, using
avelength of t
on and flexura
ent earthquake
enic fault rupt
g relations in
lained by def
mation (wavel
dslide erosion
g the relation
topography ca
al-isostatic co
e magnitude M
turing the surf
Leonard, 201
formations ca

length defined
n (wavelength
from Keefer
aused by
ompensation)
M
w
*
face, a related
10); (c) a
aused by
d
h
d

Figur
Figur

e 4.S1
e 4.S1. A sch hematic illustr ration of the f

107
fault setting.

Figur

Figur
equati

e 4.S2
e 4.S2. Wenc
ion.
chuan landslid de data (hangi

108
ing wall only y) explained b

by seismic wa ave attenuation n

Figur

Figur
co-sei
scalin

e 4.S3.
e 4.S3. Variat
ismically crea
ng relations, la
tions in the re
ated topograp
andslide mod
elations betwe
phy and co-sei
dels, and relev

109
een earthquak
ismic landslid
vant seismic g
ke magnitude
des with diffe
ground motion
and volumes
erent M
w
-fault
n parameters.

s of
t geometry
.

Figur

Figur
Lands
Wells

e 4.S4.
e 4.S4. Variat
slide models:
s and Coppers
tions of Ω acr
Keefer, 1994
smith, 1994 a
ross earthqua
4 and Marc et
and Leonard, 2

110
ake magnitude
t al., 2016; Fa
2010.
e for near fiel
ault dimension
ld and far fiel
ns-M
w
scaling

d scenarios.
g relations:

Figur

Figur
M
w
7.9
co-sei
uplift
arrow

e 4.S5
e 4.S5. Spati
9 Wenchuan
ismic displac
and landslide
w in (a); (c) Ω
ial variations
earthquake a
cement and l
e erosion vers
(V
ls
/V
co-seismic
)
s of co-seism
at the eastern
landslides cau
sus distance t
) versus the a

111
mic uplift and
n margin of
used by the
to fault trace a
absolute value
d landslide er
the Tibetan
Wenchuan e
along the dire
es of distance
rosion caused
plateau. (a)
earthquake; (
ection indicat
to fault trace
d by the 200
map view o
(b) co-seismi
ted by the gre
e.
08
of
ic
ey

Figur

Figur
seism
contri
seism
(Mala

e 4.S6
e 4.S6. Cumu
mic intensity fa
ibutions from
mic intensity fa
amud et al., 20
ulative uplift r
actor I
4
, maxi
m earthquakes
actor I
4
, the n
004); fault dim
rates over ear
imum earthqu
with differen
number of eart
mensions-M
w

112
rthquake cycle
uake magnitud
nt magnitudes
thquakes with
w
scaling relat
es (km Myr
-1
)
de in the earth
s. Seismic pro
h magnitude ≥
tion: Leonard
) plotted as fu
hquake seque
oduct a is expr
≥4 per year in
d, 2010.

unctions of
ence, and
ressed as a
n 1°×1° area
113

This page is left blank intentionally

114

Chapter 5 Conclusions and Future Work
5.1. Conclusions
In this thesis, I examined the effects of large earthquakes on mountainous landscapes, with a
focus on earthquake-triggered landslides. Taking the Wenchuan earthquake as a case study, I
explored how the Wenchuan earthquake-triggered landslides affect erosion from decadal
timescales following earthquakes towards Myr timescales, in the steep Longmen Shan mountains
at the eastern margin of the Tibetan Plateau. Building the basis of the Wenchuan observations, I
developed a generalized modeling framework to investigate how mountainous topography
evolves under the competition between earthquake-triggered landslide erosion and seismically
induced rock uplift. The main conclusions are summarized below:

(1) During the Wenchuan earthquake, landslide occurrence varied significantly in dimensions
parallel and perpendicular to the fault strike, likely caused by the variations in seismic shaking
and fault slip modes. The Wenchuan landslides preferentially located on steeper hillslopes and in
elevation zones featured by threshold landscapes where hillslope gradients approach the angle of
repose. At fault zone scales, the spatial pattern of the Wenchuan landslides correlated well with
seismic shaking intensity, and could be explained using a classical model of seismic wave
attenuation accounting for geometric spreading and quality decay, if assuming a simplified linear
energy source. At hillslope scales, the Wenchuan landslides were clustered near ridges, likely
revealing amplifications of seismic shaking at ridges. Those observations confirm a first-order
control on the spatial pattern of the Wenchuan landslides from seismological factors. The
preferred facing directions of the Wenchuan landslides varied significantly along the fault strike,
partially explained by seismic and climatic conditions, with some unexplained patterns to be
better understood.

(2) For the Wenchuan landslides, 16% of the total landslide number, 30% of the total landslide
area, and 43+9/-7% of the total landslide volume (~1.4 km
3
) was directly connected to river
channels, prone to fluvial entrainment and transport. The remaining 57+7/-9% by volume (~1.6
km
3
) was sequestered on hillslopes, beyond the immediate extent of the fluvial channel network.
Landslide-channel connectivity was to first-order controlled by seismic shaking and landslide
area, and higher landslide-channel connectivity appears in areas with higher PGA, larger
landslides, and near fault segments dominated by thrust and oblique slip. Substrate and
topography play secondary roles modulating landslide-channel connectivity. A catchment
landslide location index ψ was defined as the relative distribution of landslides versus catchment
topography as a function of upstream contributing area, providing a complementary index
describing landslide locations in landscapes independent of channel definition. In our study area,
ψ correlates well with landslide-channel connectivity, suggesting a consistence between those
landslide location metrics. Landslide-channel connectivity had limited influence on fluvial
transport of suspended sediment, likely because fine grained sediment could be rapidly mobilized
across hillslopes.
115

(3) Over the four years following the Wenchuan earthquake, the catchment-scale denudation rates
(determined from hydrological gauging) had significant correlations with landslide volumes and
PGA, indicating a dominant seismic control on denudation. “Seismic erosion rates” determined
from modeling of landslide volumes over multiple earthquake cycles were similar to the
long-term erosion rates recorded by kyr-Myr chronometers. Besides, there was a first-order
spatial coincidence of the Wenchuan landslides and the observed long-term denudation. Thus in
the Longmen Shan mountain range, earthquake-triggered landslides likely represent a major
contributor to the long-term erosional budget and sustain orogenic-scale erosional fluxes. This
result suggests that focused denudation, as commonly observed in the frontal regions of
tectonically active mountain ranges, may be partly regulated by the location and activity of
seismogenic faulting, and specifically by the resulting earthquake-triggered landslides.

(4) The 2-D surface deformation field caused by co-seismic and inter-seismic processes over a
complete earthquake cycle was simulated combining geophysical solutions to fault-related
deformations and models predicting patterns and magnitudes of earthquake-triggered landslides.
This modeling effort took into account co-seismic deformation, post-seismic relation, landslide
erosion and flexural-isostatic responses to erosional unloading. The modeling results were
summarized in the context of a new metric: E, the efficiency of earthquakes in building
topography, defined as the ratio of the volume of created topography over one full earthquake
cycle versus the co-seismically uplifted volume. E varies across earthquake magnitudes but is
primarily controlled by the flexural rigidity of the underlying lithosphere. Landslide erosion is
mainly focused over narrow spatial windows near the seismogenic fault and could cause net
topographic destruction because isostacy is limited in this narrow window. Although landslide
erosion is concentrated within the narrow window, post-seismic relaxation and flexural isostacy
distributes uplift over a wider area. Considering far field windows where isostacy acts to
efficiently counteract erosion, most earthquakes tend to be constructive. The wavelength of
seismically constructed topography correlates with earthquake magnitude, fault dimensions, and
lithospheric rigidity, suggesting that outwardly propagating fault systems in major collisional
mountain ranges might be expected to produce orogenic plateaus featured by long-wavelength,
low-relief interiors and short-wavelength, high-relief marginal mountains where erosion is
focused.

5.2. Future plans
In addition to the obtained results and conclusions, this thesis lays the foundation for future
studies on the budget and fates of sediment and carbon following the Wenchuan earthquake. This
work also pointed out several important research directions. Overall, this thesis and the related
future studies are centered around the main theme of the interactions between earthquakes,
erosion, mountain building, and the carbon cycle. The lessons learned from modern-day
observations provide key insights and evidence to understand the co-evolution of orogeny,
surface environments, biogeochemical cycles of nutrient elements, and the climate system, as
well as to predict future changes in surface environments in response to variations in climatic and
116

seismotectonic forcing. With this goal in mind, I am interested in pursuing the following research
directions in my future career:

5.3.1. Post-earthquake sediment transport
Post-earthquake sediment transport causes mass loss from mountains, thus regulating mountain
belt evolution. Landslide sediment removal after large earthquakes is also critical for
understanding secondary hazards like aggradation and flooding, which prolong the damage to
infrastructures and threatening reconstruction. Hydrological gauging of suspended sediment only
considers the fine grained sediment without accounting for coarser material (sand-sized and
larger). In the Wenchuan case, coarser material represents >90% of the total landslide volume and
is transported as river bedload. Thus understanding post-earthquake sediment transport requires
resolving bedload sediment transport. I plan to work on the post-earthquake landslide sediment
transport problem combining empirical datasets and hydraulic modeling. Whereas the
landslide-channel connectivity study constrained the locations of landslides relative to the fluvial
network, further work is needed on the grain size composition of landslide and bedload sediments,
another key parameter in hydraulic modeling.

5.3.2. Earthquakes, mountain building, the carbon cycle, and the climate system
Erosional processes are known to affect both the source and the sink terms in the carbon cycle,
including silicate chemical weathering (CO
2
sink, West et al., 2005), organic carbon burial (CO
2

sink, Galy et al., 2007) and oxidation (CO
2
source, Horan et al., 2017), and sulfide-carbonate
weathering (CO
2
source, Torres et al., 2014; Torres et al., 2017), but the net effects of erosion
remain unclear, especially in scenarios of higher erosion rates. Large earthquakes like Wenchuan
and associated landsliding represent extreme erosion events, providing a good opportunity to
examine the tectonic and erosional control on the carbon cycle and to address whether enhanced
erosion leads to more CO
2
emission or more CO
2
burial. Resolving this problem also has
profound implications for understanding how mountain uplift affects the carbon cycle and the
global climate system over the geological past. A relevant research theme is the interactions
between glacial-interglacial climatic cycles, earthquake cycles, and mountain uplift. If
seismically-driven mountain uplift perturbs the carbon cycle and leads to declines of atmospheric
CO
2
, would earthquake cycles and associated mountain uplift precondition the climate system to
enter the glacier-interglacial cycling mode, but not too cold to get into a “snowball Earth”
(Raymo, 1988; Raymo and Ruddiman, 1992; Hartmann et al., 2017)? While glacial time periods
have distinct erosion mechanisms and climatic patterns as compared to interglacial time periods,
how would the effects of earthquakes on mountain building and erosion differ in different
climatic regimes?

5.3.3. Seismicity, bedrock fracturing, and landsliding
Earthquakes not only cause immediate hillslope failure but may weaken hillslope, promote
bedrock fracture, thus reducing the strength of hillslope substrates (e.g., Vanmaercke et al., 2017)
and preconditioning hillslopes to fail in future triggering events (e.g., Parker et al., 2015) or to be
117

more easily eroded. How earthquakes affect bedrock fracturing systems and how bedrock
fractures influence landsliding remain to be better understood. I am interested in exploring the
linkage between seismicity, bedrock fracturing and landsliding.

118

This page is left blank intentionally

119

Appendix A
Distribution of Earthquake-Triggered Landslides across Landscapes:
Towards Understanding Erosional Agency and Cascading Hazards

A1. Preamble
This appendix was a comprehensive analysis of the spatial pattern and the controlling factors of
the Wenchuan earthquake-triggered landslides. The presented results laid the foundation for the
three main chapters looking at how earthquake-triggered landslides affect erosion and mountain
building. This appendix reported the preferred locations of the landslides with respect to
background landscapes and fluvial networks, and identified controls from seismic processes from
fault zone scales to hillslope scales.

I was the main author, conducted the analysis of the Wenchuan landslide inventory, the digital
topography, and seismological and geological data, and wrote the manuscript. Josh West, Alex
Densmore, Robert Hilton, Zhangdong Jin, Fei Zhang and Jin Wang contributed to manuscript
editing and revision. This work was supported by U.S. National Science Foundation
(NSF-EAR/GLD grant 1053504 to Joshua West), the Chinese Academy of Sciences (YIS
fellowship grant 2011Y2ZA04 to Joshua West), a USC Dornsife college merit fellowship (to Gen
Li), and a GSA graduate research grant (to Gen Li). We thank Niels Hovius, Patrick Meunier, and
Odin Marc for helpful discussions. We thank Chong Xu for proofreading and insightful
comments. This work benefited from the workshop “Geomorphic workshops of large earthquakes”
at Durham University in May 2014. Yong-Gang Li is thanked for editorial handling.

This appendix material was published as a book chapter:
Li, G., A. J. West, A. L. Densmore, Z. D. Jin, Fei. Zhang, J. Wang, and R. G. Hilton, Distribution
of Earthquake-Triggered Landslides across Landscapes: Towards Understanding Erosional
Agency and Cascading Hazards, edited by Y. G. Li, Fault Waveguide, Strong Ground Motion,
Earthquake Hazards and Forecast, Beijing, Higher Education Press, in press.

120

A2. Introduction
A2.1. Earthquake-triggered landslides as hazards and erosional agents
In tectonically active mountain ranges, shallow, large earthquakes can induce widespread
landsliding due to strong ground motion and seismic wave propagation (Keefer, 1984; Jibson et
al., 1993; Keefer, 1994; Harp and Jibson, 1996; Guzzetti et al., 2009). Earthquake-triggered
landslides are a critical seismic hazard, not only causing severe loss of life and immediate
damage to infrastructure (Swiss Re, 2000; Petley, 2012), but also leading to prolonged and
widespread secondary impacts like channel aggradation and flooding – known as “cascading
hazards” (Korup et al., 2004; Glade and Crozier, 2005; Wang et al., 2015). Earthquake-triggered
landslides also represent an important erosional agent, converting rocks to clastic sediment and
supplying large volumes of erodible materials to rivers (Pearce and Watson, 1986; Keefer et al.,
1994; Hovius et al., 1997; Montgomery and Brandon, 2002; Larsen et al., 2010; Howarth et al.,
2012; Egholm et al., 2013). Riverine export of landslide materials means mass loss from
mountains, which could counteract coseismic rock uplift (Hovius et al., 2011; Parker et al., 2011;
Li et al., 2014; Marc et al., 2016a). Over longer timescales of multiple earthquake cycles (10s of
kyr to Myr), earthquake-triggered landslides contribute significantly to the erosional budget at
orogenic scales, and are capable of producing erosional fluxes comparable to those measured
from cosmogenic nuclide and low-temperature chronometers (Li et al., 2017). Landslides provide
the sediment that acts as effective tools shaping stream bedform and landforms as well (Sklar and
Dietrich, 2004; Yanites et al., 2010; Egholm et al., 2013). Landslides also work as a powerful
driver of the carbon cycle, harvesting organic carbon from vegetation, soil and bedrock
(Garwood et al., 1979; Hilton et al., 2011; Wang et al., 2016), as well as creating fresh mineral
surface for chemical weathering (Gabet, 2007; Jin et al., 2016; Emberson et al., 2016).

To evaluate the role of earthquake-triggered landslides in hazard generation, erosional dynamics,
mountain belt evolution, and the carbon cycle, it is critical to understand the spatial pattern of
landslides. Many studies have focused on empirical observations using landslide inventories
mapped from remotely-sensed images, and have explored how landslides are distributed across
gradients of topography and seismic shaking and among different climatic and geologic
conditions (Harp and Jibson, 1996; Meunier et al., 2007; Meunier et al., 2008; Gorum et al.,
2011; Xu et al., 2014; Li et al., 2016; Huang et al., 2017; Roback et al., 2017). Other studies
have focused on modeling the magnitude and pattern of earthquake-triggered landslides either
using an empirical approach or a forward approach based on engineering mechanics (e.g.,
Newmark analysis) or earthquake physics (Keefer, 1994; Jibson et al., 2000; Malamud et al.,
2004; Jibson et al., 2006; Gallen et al., 2015; Gallen et al., 2016; Marc et al., 2016b). Most
studies have identified a linkage between landslide spatial pattern, seismic shaking, and seismic
wave propagation (Harp and Jibson, 1996; Dadson et al., 2004; Meunier et al., 2007; Li et al.,
2016). This linkage provides the possibility for using landslide pattern to invert for fault rupture
properties and information about energy propagation during earthquakes (e.g., Meunier et al.,
2013). Many studies have also shown that hillslope gradient and properties of hillslope material
modulate landslide occurrence during earthquakes (Jibson et al., 2006; Gallen et al., 2015; Marc
121

et al., 2016b; Roback et al., 2017; Gallen et al., in press). Thus, combining landslide pattern,
topography, and seismic conditions can help to constrain properties of hillslope materials, such as
rock strength (Gallen et al., 2015). Conversely, known seismic conditions, topography, and rock
strength can be combined to quickly predict landslide occurrence and produce rapid response
landslide hazard maps following large earthquakes (e.g., Kritikos et al., 2015; Gallen et al., in
press; Robinson et al., in press).

Empirical studies have examined patterns of earthquake-triggered landslides from scales of fault
zones to individual hillslopes. At fault zone scales, for most earthquake-triggered landslide
inventories, landslide areal density (P
Als
%, the percentage of area impacted by landsliding per
unit landscape area) scales with peak ground accelerations (PGA) and decays away from the
epicenter. At hillslope scales, topographic amplifications of seismic shaking modulate how
landslides are distributed from hillslope ridges to valley bottoms as well as their preferred facing
directions (Meunier et al., 2008). These hillslope-scale landslide patterns are indicative of the
dominance of seismic triggers versus climatic triggers (Densmore and Hovius, 2000; Meunier et
al., 2008). How landslides are distributed across hillslopes also contributes to the extent to which
landslides are connected to fluvial networks, a key factor regulating post-earthquake sediment
transport (Dadson et al., 2004; Huang and Montgomery, 2014; Li et al., 2016).

A2.2. Landslides triggered by the 2008 Wenchuan earthquake
On May 12
th
, 2008, the M
w
7.9 Wenchuan earthquake occurred in the Longmen Shan mountain
range at the eastern margin of the Tibetan Plateau (Burchfiel et al., 2008; Hubbard and Shaw,
2009). Characterized by an unexpectedly high magnitude and long return time, the Wenchuan
earthquake ruptured the Yingxiu-Beichuan and Pengguan faults with a combination of thrust and
dextral strike-slip deformation. The earthquake also induced massive landsliding in the steep
Longmen Shan mountains, over dramatic gradients of topography, seismic shaking, and climate,
and among a variety of lithologic units. Many studies have produced Wenchuan landslide
inventory maps and examined patterns of the Wenchuan landslides in the context of regional
environmental conditions. Huang and Li (2009) reported preliminary observations of the
Wenchuan earthquake-triggered landslides, finding that most landslides were close to the
ruptured fault and on the hanging wall. Dai et al. (2011) mapped over 56,000 landslides using
aerial photos and satellite images, identified major landslide types by combining field
observations and interpretation of images, and discussed landslide areal and point density in the
context of distance to the fault rupture, elevation, slope, PGA, seismic intensity, and lithology.
Gorum et al. (2011) mapped ~60,000 landslides using both pre-earthquake and post-earthquake
images, reported landslide areal and point density variations versus distance to fault rupture,
co-seismic slip, and lithology, and derived empirical relations to describe the landslide pattern
with distance from the fault rupture. Parker et al. (2011) mapped >70,000 landslides by
combining automated algorithms and a series of topographic filters and visual screening. They
estimated the Wenchuan landslide volume using empirical landslide area-volume scaling
relations and found that, if landslide debris is evacuated by rivers over the earthquake cycle, the
122

landslide-induced erosion could counteract the rock uplift caused by the Wenchuan earthquake,
thus raising an intriguing question of “earthquake mass balance”, i.e., whether earthquakes build
or destroy mountainous topography. Xu et al. (2011) examined the effects of locked fault
segments where most co-seismic moment was likely released and the locations of landslides
relative to the seismic energy source. They identified a “back slope effect”, whereby hillslopes
facing the energy source that were less susceptible to landsliding, in several valleys. Fan et al.
(2012) systematically studied 828 Wenchuan earthquake-triggered landslide dams, estimated
their size distribution and storage capacity, and discussed the implications for the post-earthquake
sediment budget. Xu et al. (2014) updated the landslide inventory of Dai et al. (2011), reported a
landslide inventory map with 197,481 landslide polygons, and identified primary controls on the
Wenchuan landslides from seismic-related parameters (e.g., PGA and seismic intensity) by
adopting a bivariate statistical analysis. Yuan et al. (2013) focused on the relation between the
Wenchuan landslide pattern and seismic parameters (PGA and distance to the hypocenter) and
fitted the Wenchuan data to a landslide pattern model by assuming a point energy source
following the functional form of seismic wave attenuation (Meunier et al., 2007). Li et al. (2014)
mapped landslides using semi-automated algorithms and manual segregation of amalgamated
landslide features, and examined the Wenchuan landslide inventory in the context of a
generalized model of earthquake mass balance; following this work, Li et al. (2016) combined
the landslide inventory map and analysis of digital topography to quantify the connectivity
between earthquake-triggered landslides and the regional fluvial network. Gorum and Carranza
(2015) explored controls on the patterns of landslides from fault-types and suggested that
landslides associated with thrust slip occur over a wider zone, further away from the fault trace,
compared to strike-slip deformation. While most studies have focused on the co-seismic
landsliding, several studies have also looked into post-seismic landsliding caused by rainfall and
reactivation of the co-seismic landslide debris in the epicentral region (e.g., Tang et al., 2011;
Zhang et al., 2014).

In this contribution, we review previous studies of earthquake-triggered landslides associated
with Wenchuan and other earthquakes, and conduct new analysis to explore how the Wenchuan
landslides are distributed across the Longmen Shan range. Specifically, previous Wenchuan
studies have not fully considered the effect of the background landscape on the observed
characteristics of landslides (Meunier et al., 2008; Huang et al., 2017). For example, if landslides
randomly sample a landscape, the topographic characteristics of landslides should reproduce the
characteristics of the landscape rather than reveal information about other controlling
mechanisms. To correct for such an effect, we normalize the statistical distribution of any metric
of interest (e.g., hillslope gradient) of the landslide inventory over the distribution of this metric
of the background landscape. Combining the Wenchuan landslide inventory map reported by Li
et al. (2014) with analysis of digital topography, we identify the preferred elevation, slope and
lithologic units where the coseismic landslides were located relative to the background landscape.
We investigate the relations between landslides and seismic shaking, and model the Wenchuan
landslide pattern at orogenic scales by adopting a model of seismic energy release. We estimate
123

the average geometric profiles of hillslopes in the Longmen Shan region, quantify how
Wenchuan landslides were located relative to hillslope ridges and river channels, and discuss the
implications for landslide triggers and landslide-channel connectivity. We also track the preferred
facing directions of landslides along the Yingxiu-Beichuan fault rupture to examine the linkage
between landslide aspects and the characteristics of earthquake source, ground motion, and
energy propagation.

A3. Setting
A3.1. Topography, hydrology and climate
With > 5 km of relief over a 50 km horizontal distance from the Sichuan Basin, the Longmen
Shan mountain range marks the steepest margin of the Tibetan Plateau (Figure A.1, Densmore et
al., 2007; Burchfiel et al., 2008). Landscape relief and steepness peak at the mountain front and
gradually decrease westward toward the Tibetan Plateau. Several major tributaries of the Yangtze
River, including the Min Jiang, Fu Jiang and Tuo Jiang, drain the Longmen Shan mountains to
the Sichuan Basin (Liu-Zeng et al., 2011). The regional climate is dominated by the East Asian
monsoon, with over 70% of rainfall occurring during the wet season from June to September.
Average annual rainfall is around 1100 mm yr
-1
at the margin and decreases to ~600 mm yr
-1

toward the plateau. The regional climatic pattern is largely controlled by the high, steep
mountainous topography which works as a barrier to moisture or causes heating of the
atmosphere (Molnar et al., 2010). The orographic rainfall and steep topography lead to intensive
denudation of the mountain front, ~0.3-0.5 mm yr
-1
determined before the Wenchuan earthquake
(Liu-Zeng et al., 2011). After the Wenchuan earthquake, the seismically induced landslides have
supplied large volumes of clastic sediment to the rivers and greatly enhanced local denudation
rates to ~1-1.2 mm yr
-1
integrated over four years following the earthquake (Wang et al., 2015; Li
et al., 2017).

A3.2. Geology and tectonics
The Longmen Shan is underlain by Proterozoic basement granitoid and high-grade metamorphic
rocks, a Paleozoic passive margin sediment sequence of metamorphic sediment and granitic
intrusion, and a Mesozoic foreland basin consisting of marine and clastic sediment with limited
exposure of Cenozoic sediment (Figure A.1; Li et al., 2003; Densmore et al., 2007; Burchfiel et
al., 2008). Cenozoic deformation in the Longmen Shan is superimposed on a Mesozoic orogeny
(Burchfiel et al., 2008). River profile analysis and low temperature thermochronometry suggest
high uplift and exhumation rates (~0.5-1 km Myr
-1
), respectively, during the late Cenozoic (Kirby
et al., 2003; Kirby et al., 2008; Tian et al., 2013). The high, steep topography of the Longmen
Shan range has stimulated heated debates on different geodynamic models of plateau marginal
mountain building and on the relative importance of ductile lower crustal flow versus brittle
faulting in the upper crust (Royden et al., 1997; Clark and Royden, 2000; Tapponnier et al., 2001;
Hubbard and Shaw, 2009). The Longmen Shan is bounded by a fault system composed of three
major faults: the Wenchuan-Maowen fault, the Yingxiu-Beichuan fault and the Pengguan fault
(Chen and Wilson, 1996; Densmore et al., 2007; Liu-Zeng et al., 2009; Xu et al., 2009). The
124

Wenchuan earthquake nucleated in the southern part of the Longmen Shan, rupturing the
Yingxiu-Beichuan and Pengguan faults and propagating ~270 km to the northeast. The
co-seismic moment release varied significantly along the fault rupture but was concentrated in
the Yingxiu and Beichuan regions (Xu et al., 2009; Shen et al., 2009). Fault displacement was
mainly in the form of dextral-thrust slip in the southwestern part of the rupture and changed to
dextral strike slip towards the northeast (Liu-Zeng et al., 2009). Paleoseismology and geodetic
studies suggest a relatively long return time (~500-4000 yrs) for Wenchuan-like events in the
Longmen Shan region (Densmore et al., 2007; Liu-Zeng et al., 2009; Shen et al., 2009;
Thompson et al., 2015; Lin et al., 2016).

A4. Materials and methodology
We analyzed the Wenchuan earthquake-triggered landslide inventory reported in Li et al. (2014)
in the context of the digital topography of the Longmen Shan mountains. We used the void-filled
SRTM 90 digital elevation model (DEM, resolution ~87 m) from the Consultative Group for
International Agricultural Research (CGIAR, Jarvis et al., 2008). Seismological data, including
aftershock sequences and peak ground accelerations, was taken from the USGS Earthquake
Hazard program (2008). Analysis of the digital topography and the landslide inventory was
conducted in the ArcGIS platform.

Based on our analyses, we consider how the Wenchuan landslides are distributed in directions
perpendicular to and parallel to the fault trend, in different topographic and lithologic conditions,
and relative to the fluvial network. We evaluate controls from seismic processes on the pattern of
the Wenchuan landslides. We examine the relations between landslide areal density and ground
motion, whether the Wenchuan landslides can be explained using a simplified model based on
seismic energy propagation, and the imprint of a seismic trigger on landsliding at hillslope scales.

A5. Distribution of the Wenchuan landslides across the Longmen Shan
In this section, we explore the spatial patterns of the Wenchuan earthquake-triggered landslides in
directions perpendicular and parallel to the fault trend. We explore the statistical distributions of
the topographic and lithologic attributes of the landslides, accounting for the variations of the
background topography and non-uniform distribution of lithologic units across the landscape.
Finally, we compare different methods for quantifying the locations of the landslides relative to
the fluvial network to better constrain how landslides might be expected to influence
post-earthquake sediment dynamics.

A5.1. Variations of landslide pattern perpendicular to the fault trend
We plot total landslide areas in 5 km-width increments paralle to and perpendicular to the
Longmen Shan fault trend (Figure A.2). Because both seismic shaking and topography show
greatest gradients perpendicular to the fault trend (Figure A.1a and Figure A.1b), we would
expect that plotting landslide data along this trend would capture the most significant variations
of landslide occurrence. Perpendicular to the fault trend, we observe the highest landslide area
125

at the Longmen Shan mountain front, on the hangingwall and coinciding with the location of the
epicenter. Landslide area decays sharply towards the Sichuan Basin and the Tibetan Plateau
(Figure A.2a). This variation normal to the fault trend is quite similar to other cases, including the
1993 M
w
6.9/6.7 Finisterre, the 1994 M
w
6.7 Northridge, and the 1999 M
w
7.6 Chi-Chi earthquakes,
with landslides clustering at mountain fronts and decaying towards both the foreland and
hinterland (Meunier et al., 2007). This is expected because fault-bounded mountain fronts
generally have steeper topographic gradients and experience most intensive shaking, whereas the
foreland and hinterland are characterized by gentler gradients and lower fault activity (Meunier et
al., 2007; Gallen et al., 2015; Roback et al., 2017). Orographically-enhanced precipitation may
also promote landsliding by modulating rock strength (Gallen et al., 2015).

A5.2. Variations of landslide pattern along the fault trend
The complex fault slip pattern makes the Wenchuan case a good opportunity to examine fault
type controls on landsliding. Several studies have shown that, for the same initial stress
conditions, thrust faults produce much stronger ground motion compared to normal and
strike-slip faults (Oglesby and Dai, 2002; Gubachian et al., 2014). Thus, we would expect to see
some relationship between slip orientation in the earthquake and landslide occurrence. We show
that there are quite significant variations in Wenchuan earthquake-triggered landslide area
parallel to the fault trend (Figure A.2b). The maximum landslide area is located near the
southwestern end of the rupture, although offset by ~40 km from the epicenter (Figure A.2b),
whereas landslide area decreases toward the northeast. Whereas there are no dramatic gradients
in topography along the fault trend (Figure A.1a and Figure A.1b), the along-trend variations in
landsliding seem to coincide with changes in the fault slip orientation from oblique dextral-thrust
slip domains in the southwest towards strike-slip in the northeast. Gorum and Carranza (2015)
conducted a more comprehensive analysis of the along-strike variations of the Wenchuan
landslides, showing that more and larger landslides occurred on the oblique thrust-slip segment
and that, when accounting for fault types, the landslide spatial pattern could be better predicted as
a function of distance to fault than without considering of fault types. The effects of slip
orientation on landsliding thus provide additional constraints in modeling landsliding occurrence
in multi-fault type settings like Wenchuan.

A5.3. Distribution of landslides over topographic metrics and lithology
We plot the distribution of Wenchuan landslides by number as a function of topographic metrics
(gradient and elevation) and lithology (Figure A.3). Because the studied landscape does not
necessarily have a uniform distribution of topographic gradient, elevation and lithology, we also
derived the distribution of the studied metrics for all landscape cells as the background condition.
We then normalized the distribution of the metrics for the Wenchuan landslides over the
background landscape distribution, and obtained a background-corrected distribution of
landsliding susceptibility over the landscape (e.g., Meunier et al., 2008; Clark et al., 2016; Huang
et al., 2017).

126

The topographic gradients of all landslides and the studied landscape show unimodal and slightly
right-skewed distributions, with modal gradients of ~30° and ~38° for the landscape and all
landslides, respectively (Figure A.3a). Note that 30° is thought to be a threshold topographic
gradient in the study area (Ouimet et al., 2009). When corrected for the background landscape,
the Wenchuan landslides are found to preferably occur at much steeper hillslopes, which is
consistent with the Newmark sliding model: with other environmental factors set, steeper
gradients would have lower factor of safety and are more prone to fail (Jibson, 2007; Gallen et al.,
2015). The Wenchuan landsliding susceptibility also increases monotonically with increasing
gradient, underlining the importance of hillslope steepness in setting landslide occurrence. For
elevation, the Wenchuan landslides show a unimodal left-skewed distribution, quite distinct from
the underlying landscape (Figure A.3b).

The modal elevation of the landslides is around 1.5 km, roughly corresponding to the elevation of
the Longmen Shan mountain front (Figure A.3c). When corrected for the elevation distribution of
the background landscape, the Wenchuan landslides preferably occur at elevations of around 2-3
km (Figure A.3b). Notably, this elevation range corresponds to the region featured as a
ready-to-fail, threshold landscape where mean hillslope gradients approach the angle of repose
(Figure A.3c) and are insensitive to increasing erosion rates (Ouimet et al., 2009). Thus our
analyses suggest that, during the Wenchuan earthquake, although the mountain front experienced
the most intensive seismic shaking and landsliding, the region of highest landslide susceptibility
was located at slightly higher elevations of 2-3 km. This may indicate where a threshold
topographic gradient is reached, with steeper hillslopes that were more prone to fail (Figure A.3a).
Across different rock types, we find that most landslides occurred in sandstone, limestone,
mylonite, and marble (Figure A.3d). When accounting for the distribution of lithologic units
across the landscape, marble and mylonite units were most prone to landsliding (Figure A.3d).
Better understanding of the roles of lithology requires further exploration of the coupling
between lithology, rock strength, and other landscape properties, as well as the spatial
coincidence between lithologic units, seismic shaking, topography, and climate (e.g., Luo et al.,
2008; Chen et al., 2011; Gallen et al., 2015).

A5.4. Landslide locations relative to the fluvial network
Fluvial landscapes are typically composed of two geomorphic domains: streams and hillslopes.
Landslides directly connected to fluvial stream networks may more easily deliver sediment to
streams, whereas landslides located on hillslopes may be less effective at transferring sediment.
Thus, the locations of landslides relative to fluvial stream networks should provide a key
constraint on the potential effect of landslide debris on sediment transport after earthquakes.
Several studies have explored the extent to which landslides connect to fluvial streams (Dadson
et al., 2004; Meunier et al., 2008; Hilton et al., 2011; Hovius et al., 2011; Li et al., 2016; Roback
et al., 2017), while Li et al. (2016) evaluated the fluvial connectivity of Wenchuan landslides
using a raster-based approach. Here, we compare this prior approach with a vector-based
calculation to evaluate the landslide-stream connectivity for the Wenchuan earthquake (Figure
127

A.4). In both cases, to delineate streams, we determined a cutoff upstream contributing area (A
0
)
using the scaling relation between topographic gradient (S) and upstream contributing area (A)
(Figure A.4b; e.g., Montgomery, 2001). For the raster-based approach, we compared the
maximum upstream contributing area for a given landslide (A
max
) to A
0
. If A
max
was greater than
A
0
, then the landslide was assigned as “stream-connected” (Dadson et al., 2004, Li et al., 2016).
We estimate that, for the Wenchuan case, ~16% of the total landslide population is connected to
the fluvial streams (Figure A.4a), but those landslides are larger than average in both area and
volume and represent ~43% of the total landslide volume (Figure A.4c). For the vector-based
approach, we delineated the ridge lines as the boundaries of each small watershed defined by the
fluvial streams and calculated the normalized distances to ridges and streams for all landslides
(Figure A.4f). For a given landslide, the normalized distance to stream (D
nstream
) is defined as the
ratio of the distance of the lowest point within the landslide polygon to the hillslope base (stream)
over the total length of the hillslope (L), and similarly the normalized distance to ridge (D
nridge
) is
based on the highest point within the landslide polygon and the located hillslope’s ridge
(Densmore and Hovius, 2000; Meunier et al., 2008; Huang and Montgomery, 2014). For
computational simplicity, we calculated the distance as the Euclidean distance. Choosing D
nstream

= 0.2 as a cutoff for “stream-connected” (Meunier et al., 2008; Hovius et al., 2011), this second
approach reveals that ~19% of the landslide population and ~46% of the total landslide volume
are connected to the fluvial streams, consistent with the first approach and validating our estimate
of the Wenchuan landslide-stream connectivity. A vector-based approach was also adopted by
Roback et al. (2017) to evaluate landslide-stream connectivity following the 2015 Gorkha M
w
7.8
earthquake.

Building on the first approach (i.e., comparing A
0
and A
max
), we have conducted a detailed
analysis of the Wenchuan landslide-stream connectivity concerning controlling factors and
influence on sediment transport (see analysis and discussion in Li et al., 2016). We identified
primary controls on connectivity from seismic shaking and fault type, and secondary modulations
from topography and lithology. More intensive shaking triggered larger-sized landslides, which
are more likely to reach hillslope bases and connect to streams. Regions with higher drainage
density also have higher landslide-stream connectivity as more streams are available. Lithologic
units likely affect connectivity due to both different drainage densities and different capacities to
generate larger-sized landslides. Another interesting finding is that the Wenchuan
landslide-stream connectivity seems not to influence post-earthquake transport of fine landslide
sediment, perhaps because this material can easily mobilize over hillslopes via diffusive transport
processes. As a result, even those landslides not immediately connected to fluvial channels can
supply fine-grained material to rivers. In contrast, we expect that connectivity may have a more
profound influence on the export of coarser materials (grain size > 0.25 mm), which compose
~90% of the total landslide material (Wang et al., 2015; Li et al., 2016). These effects could be
modulated by mobilization of coarse-grained material in rainfall-induced debris flows in the
months to years following the earthquake. Data to evaluate these effects and test the hypothesized
roles for connectivity in transport of coarse-grained landslide debris remain lacking but will be
128

important to predicting the cascading hazards associated with sedimentation in the wake of large
earthquakes.

A6. Seismic controls on the pattern of the Wenchuan landslides
In this section, we evaluate how the spatial pattern of Wenchuan earthquake-triggered landslides
relates to the seismic trigger. We examine potential controls on landsliding from seismic shaking
and seismic energy propagation across spatial scales from the rupture zone to hillslopes.

A6.1. PGA versus landslide areal density
The most significant gradients of seismic shaking and ground motion were perpendicular to the
fault trend (Figure A.1a and Figure A.1b). Following other studies (e.g., Meunier et al., 2007), we
plot the areal density of landslides, P
Als
, against the mean peak ground acceleration (PGA) in 5
km wide increments perpendicular to the fault trend. We find a clear separation between
landslides occurring on the hanging wall versus on the footwall, and the landslide data on the
hanging wall and the footwall can be fitted separately (Figure A.5a). For data points with
P
Als
>0.1%, the regression line for the hanging wall data has a similar slope to the regression line
for the footwall data but a much smaller intercept at the x-axis, representing lower threshold
accelerations to initiate landsliding (Figure A.5a). We suggest that, in the Wenchuan region,
under similar seismic shaking, hillslopes located on the hanging wall seem to be more prone to
landsliding as compared to the footwall, but once a threshold acceleration is reached, the
landsliding response to enhanced shaking (i.e., the slope of the regression line on Figure A5a,
also termed landscape sensitivity in Marc et al., 2016b) seems to be similar. Another interesting
observation is that, , at very high PGA values, the landslide areal density in the hanging wall
seems to saturate as PGA increases (Figure A.5a), which may reveal the saturation effect of
ground motion for high magnitude events as indicated by Boore and Atkinson (2008) and Marc et
al. (2016b).

We next compare the Wenchuan P
Als
-PGA relations to those from other earthquakes compiled by
Meunier et al. (2007) (Figure A.5b). Note that Meunier et al. (2007) derived P
Als
-PGA relations
using both horizontal and vertical components of the ground motion data. Though we do not have
the horizontal and vertical PGA data from Wenchuan, our results still provide several insights
into the P
Als
-PGA relations. First, the Wenchuan case captures a greater range of variations in P
Als

(up to near 8%) and PGA (~0.2-0.7g), helping to constrain the relations between P
Als
and PGA
under much stronger seismic shaking conditions. Second, the Wenchuan case clearly shows the
different trends between hanging wall and footwall. Third, the landsliding sensitivity to PGA,
indicated by the slope of the regression line, is quite similar for the Wenchuan and the Northridge
cases, and are substantially greater than the sensitivity shown in the Chi-Chi earthquake (Figure
A.5b). This difference may be partially caused by the characteristics of seismic wave propagation
as discussed in section 5.5.2 below.

129

A6.2. Modeling of landslide pattern using a simplified seismic wave sttenuation equation
Earthquake-triggered landsliding records the surface geomorphic response to seismic energy
propagation. Following this line of thought, Meunier et al. (2007) adopted a classical model
describing seismic wave attenuation to predict the spatial pattern of earthquake-triggered
landslides:
) exp(
0 0
0 A

R R
R
R
P P
A ls

 
(A.1)
where P
Als
is landslide areal density (%), R
0
is the mean asperity depth, R is the distance to the
source, and P
A0
and β are scaling factors. β is a spatial decay factor, with higher values meaning
more widely spread landsliding. Eq. A.1 accounts for both geometric spreading (via the R
-1
term)
and quality decay (via the exponential term, which depends upon β). Using this model, Meunier
et al. (2007) successfully reproduced patterns of landslides caused by the Chi-Chi, Northridge,
and Finisterre earthquakes. Note that for those earthquakes, the seismic energy sources were
approximated as point sources – a reasonable assumption since the ratio of length to width for
those faults was close to one and those fault planes can be approximated as squares. This point
source assumption may not be appropriate for the Wenchuan earthquake, which had a rupture
length to width ratio of ~5:1. Thus, we assume a linear energy source rather than a point source.
We then calculate the corresponding distance to the seismic energy source and fit the landslide
data using Eq. A.1, treating PA0 and β as fitting parameters (Figure A.6a). We find that the
simplified linear energy source model can be used to successfully reproduce the pattern of
Wenchuan landslides, and that the hanging wall and footwall data show distinct trends. When
compared to other cases (Figure A.6b), the trends of the Wenchuan case are similar to the trends
of the Finisterre and the Northridge events, whereas the Chi-Chi case shows a near-horizontal
trend, indicating a purely geometric spreading control on the pattern of landsliding with no
quality decay during wave propagation (Meunier et al., 2007).

A6.3. Landslide “clustering” as signatures of a seismic trigger
Previous studies have proposed that the distribution of landslides from hillslope top to base could
reveal information about the dominant landsliding-triggering mechanism (Densmore and Hovius,
2000; Meunier et al., 2008). Meunier et al. (2008) conducted geophysical simulations of seismic
wave propagation across variable topographic conditions, and found that during earthquakes
geometric discontinuities such as hillslope ridges and inner gorge knick points could amplify
seismic shaking and thus should experience more landsliding (i.e., landslide “clustering”, Figure
A.7a). In contrast, rainfall-triggered landslides should be distributed more evenly across
hillslopes due to the lack of such seismic amplification effects (Figure A.7b). This theorem has
successfully explained the patterns of landslides triggered by the Chi-Chi, Finisterre, and
Northridge earthquakes (Meunier et al., 2008).

In the Wenchuan case, we find a “single clustering” pattern, i.e., most Wenchuan landslides
originate from near ridges, thus reflecting the expected seismic triggering signal (Figure A.7c and
130

A.7d). This pattern may also help to explain the morphology of the hillslopes in the study area
(Figure A.8a and A.8b). For each topographic cell, we calculate the local gradient and the
normalized distances to ridges and streams (i.e., D
nridge
and D
nstream
, respectively) (Meunier et al.,
2008). We plot the mean topographic gradient as a function of D
nstream
(Figure A.8a) and convert
the gradient-D
nstream
relation to a composite or average hillslope profile (Figure A.8b, D
nstream

versus normalized height). This hillslope profile should represent the typical hillslope shape in
the Longmen Shan area. The profile consists of near-planar hillslopes without any evidence of
widespread inner gorges, suggesting topographic amplification of seismic shaking at ridge crests
is the dominant influence on hillslope form. The uppermost and lowermost parts of the hillslopes
seem to have shallower local gradients (Figure A.8b), perhaps caused by the clustered seismic
landsliding near the ridges (as shown in Figure A.7c and d) and by landslide deposition near the
bases.

A6.4. Landslide preferred aspect variation and relevance to fault slip type
The preferred aspect, or hillslope facing directions, of earthquake-triggered landslides are thought
to be indicative of seismic conditions. Hillslopes facing away from seismic energy sources are
typically more susceptible to landsliding as compared to those directly facing the energy sources
(Figure A.9) (Meunier et al., 2008).

Considering the complex energy source characteristics of the Wenchuan earthquake, we track the
variations of landslide aspect along the fault trend by dividing the study area into six segments
(Figure A.10a-g). Interestingly, we find that most of the Wenchuan landslides were preferentially
located on hillslopes facing southeast when normalized by the distribution of aspects of all
topographic cells in the study area (Figure A.10b-f), opposite to the direction (northwest) that
would experience more landsliding in theory if triggered by seismic wave propagation (Meunier
et al., 2008). This preferred facing direction is the same as the horizontal motion direction of the
hanging wall (Liu-Zeng et al., 2009), likely implying a more important role of the directivity of
ground motion rather than the relative position of seismic energy sources for triggering
landsliding. This pattern may also be caused by the fact that hillslopes facing in the
landsliding-preferred direction have more favorable climatic conditions (e.g., precipitation) and
vegetation cover for weathering and regolith development, so they may be more prone to fail
during shaking. Another interesting observation is that the landslides in swath six (Figure A.10g)
preferably face southwest, which is hard to explain by climatic factors (favoring hillslopes with a
southeast aspect), seismic energy propagation (favoring landslides with a northwest aspect), or
ground motion (favoring hillslopes with a northeast aspect). We tentatively interpret this pattern
as influenced by the clustered aftershock sequences on the eastern side of swath six (Figure
A.10a), which could induce landslides that were preferentially located on hillslopes facing away
from those aftershock energy sources, i.e., southwest. Our observations point to the need for
further studies examining the effects of seismic processes, climatic conditions, as well as rock
strength, in setting landsliding aspects.

131

A7. Conclusions, implications and future directions
How landslides are distributed across landscapes provides valuable information for hazard
management and prediction, for understanding how tectonic fault systems control erosion and the
biogeochemical cycles of essential nutrient elements, and for evaluating orogenic erosion and
evolution over tectonic timescales. Our observations from the Wenchuan earthquake validate
several theories and hypotheses related to earthquake-triggered landsliding and shed new light on
the importance of source characteristics. Specifically, we find significant variations of landslide
occurrence both perpendicular and parallel to the seismogenic fault strike, likely due to variations
in the gradients of seismic shaking and changes in fault slip modes, respectively. The Wenchuan
earthquake-triggered landslides are preferentially located on the steepest hillslopes and in
elevation zones dominated by threshold landscapes. As in the 1999 M
w
7.6 Chi-Chi earthquake,
only ~10-20% of the total landslide population is connected to the fluvial stream network, but
those stream-connected landslides are larger and represent ~40-50% of the total landslide volume.
We use two independent approaches (raster-based and vector-based) to estimate landslide-stream
connectivity and obtain consistent results. We find good correlations between landslide areal
density and seismic shaking across the Longmen Shan range, confirming a first-order control on
landslide occurrence from seismic shaking. Assuming that the Wenchuan earthquake was a linear
energy source, we demonstrate that the co-seismic landslide pattern can be modeled using a
classical model of seismic wave propagation, characterized by a geometric spreading factor and a
quality decay factor.

At hillslope scales, the Wenchuan landslides show a strong clustering near ridges, consistent with
previous findings of amplifications of seismic shaking at ridges and the consequently enhanced
landsliding. The clustering of landslides near ridges and the deposition of landslide debris near
streams may explain the rounding of the hillslope tops and bases in the study area, respectively
(e.g., Korup et al., 2007). Most Wenchuan landslides preferentially face southeast, similar to the
direction of the horizontal ground motion of the thrust-slip part of the hanging wall but facing the
seismic energy source, pointing to a more important role of ground motion and to other possible
factors like climatic conditions rather than seismic wave propagation in setting landslide aspects.

While those patterns observed in Wenchuan appear consistent between several recent and
well-studied continental earthquakes, there are some notable exceptions. For example, the
maximum landslide occurrence triggered by the 2015 M
w
7.8 Gorkha earthquake did not overlap
with the area of peak PGA (Roback et al., 2017). The Gorkha earthquake was also characterized
by large (up to M
w
7.2) that occurred far from the mainshock epicenter. Those aftershocks may be
a major contributor to the seismic moment release to the surface landscape, and may
consequently contribute significantly to landsliding (e.g., Williams et al., 2017). Altogether, these
observations raise further questions about the roles of long-term tectonic and climatic conditions
and aftershock sequences in earthquake-triggered landsliding. Moreover, long-term tectonic and
climatic conditions largely determine another critical but less well constrained component: the
bedrock fracture system, via modulations of tectonic stress and surface processes (e.g., chemical
132

weathering). The bedrock fracture system exerts a primary control on hillslope-scale rock
strength and thus on susceptibility to landsliding. Recent advances in modeling bedrock fracture
systems using topography (St. Clair et al., 2015; Moon et al., 2017) provide a promising tool to
investigate how the subsurface structures of hillslopes affect landsliding. Overall, the spatial
pattern of earthquake-triggered landslides in landscapes provides key information to validate
hypotheses and to calibrate models. Our observations highlight the importance of earthquake
source characteristics and fault rupture patterns, which could be better described in
statistically-based models. More hillslope-scale metrics (e.g., landslide-channel connectivity,
clustering pattern, and aspect) could be applied to evaluate model performance as well. In
conjunction with other models predicting the total volume of earthquake-triggered landslides
(e.g., Keefer, 1994; Malamud et al., 2004; Marc et al., 2016b), understandings of and efforts to
predict the landsliding pattern would also allow better incorporations of landsliding processes
into landscape evolution models (e.g., Bellugi et al., 2015), as well as better hazard prediction
both during earthquake events and in their aftermath as sediment is delivered to river systems.

Figur

Figur

Figur
the Lo
panel
respec
calcul
of Ea
includ
( Є: C
Meso
seque
map (

Figur

Figur
perpe
in Fig
wide

res
e A.1
e A.1. Maps
ongmen Shan
(a), A-A’ an
ctively, and t
lated. The ins
ast Asia. Pane
ding Proteroz
Cambrian, O:
zoic (T: Trias
ences, and Ce
(China Geolo
e A.2
e A.2. Dist
ndicular to an
gure A.1. The
increments. S
of the topogr
n range where
nd B-B’ are
the rectangul
set in panel (b
el (c) shows
zoic (p Є) base
Ordovician, S
ssic, J: Jurass
enozoic (Q: Q
gical Survey,
tribution of
nd (b) paralle
e total landslid
Stars indicate
raphy (a), sei
e the Wenchua
projection tr
lar box deline
b) indicates th
major litholo
ement consist
S: Silurian, D
sic, and K: C
Quaternary)
2004).
Wenchuan
el to the fault
de area is cal
the epicenter
133
smological p
an earthquake
rends perpend
eates the stu
he geographic
ogical units a
ting of granit
D: Devonian,
Cretaceous) pa
sediment, mo
earthquake-tr
strike, along
lculated as the
r.
parameters (b)
e and associat
dicular and p
dy area wher
c location of th
and faults of
toids and met
C: Carbonife
assive margin
odified from
riggered lan
g the A-A’ and
e sum of area
) and litholog
ted landslidin
parallel to th
re topograph
he study area
the Longme
tamorphic ro
erous, and P:
n and forelan
a 1: 2,500,0
ndslides in
d B-B’ trends
as of the land

gic units (c) o
ng occurred. I
he fault strike
hic metrics ar
a in the contex
n Shan range
cks, Paleozoi
Permian) an
nd sedimentar
000 geologica

directions (a
s, respectively
dslides in 5 km
of
In
e,
re
xt
e,
ic
nd
ry
al
a)
y,
m

Figur

Figur
topog
elevat
densit
repres
studie
elevat
A-A’
(d), th
landsl
litholo

re A.3
e A.3. Distr
graphic gradie
tion and grad
ty of the stud
sents the prob
ed metrics ac
tion and grad
trend, from th
he yellow and
lides, respect
ogic units nor
ribution of W
ent, (b) eleva
dient. In pane
died metrics
bability densi
cross the lan
dient data are
he Sichuan B
d black bars sh
ively, and the
rmalized by th
Wenchuan ea
ation, and (d)
els (a) and (b
of the landsc
ity of landsli
ndscape, a me
calculated as
asin to the Lo
how the prob
e grey bars sh
he distributio

134
arthquake-trig
lithology. Pa
), the grey an
cape and land
ide occurrenc
easure of lan
s the means w
ongmen Shan
bability distrib
how the prob
n of lithologi
ggered landsl
anel (c) show
nd black curv
dslides, respe
ce normalized
ndslide susce
within 89 m-w
n range and th
butions of lith
bability distri
ic units across
lides by num
ws the relation
ves indicate t
ectively, and
d by the distr
eptibility. In
wide DEM st
he Tibetan Pla
hology of the
ibution of lan
s the landscap

mber over (a
nship betwee
the probabilit
the red curv
ribution of th
panel (c), th
trips along th
ateau. In pane
landscape an
ndslides acros
pe.
a)
en
ty
ve
he
he
he
el
nd
ss

Figur

Figur
netwo
(d, e,
line in
contri
Longm
for flu
vertic
indica
fractio
indica
distrib
distan
defini
d
1
, no
(norm
landsl
e A.4
e A.4. The co
ork was evalu
f). (a) Lands
ndicating the
ibuting area
men Shan cat
uvial streams
cal dashed lin
ates the cutof
on p(N) versu
ating the assu
bution of land
nce between
itions of D
nstr
ormalized by
malized distan
lide to a strea
onnectivity b
uated using a r
slide populati
e cutoff area
(A) and mea
tchment (Duj
(Montgomer
e. (c) Cumula
ff A between
us normalized
umed cutoff
dslide volum
hillslope and
ream
(the Eucli
the hillslope
nce to hillsop
am d
2
, normal
etween Wenc
raster-based a
on fraction p
for fluvial st
an topographi
iang); the tra
ry, 2001; Dad
ative distribut
n hillslope an
d distance to
distance betw
me F(V) versu
d fluvial strea
idean distance
length revea
pe ridge, the
ized by L).
135
chuan earthqu
approach (pan
(N) versus up
treams. (b) A
ic gradient (
ansition in the
dson et al., 20
tion of landsl
nd fluvial stre
stream or hil
ween stream
us D
nstream
; the
am domains.
e between the
aled as the pla
Euclidean d
uake-triggered
nels a, b, c) an
pstream contr
A typical relat
S) of each A
e S-A relation
004; Li et al.
lide volume F
eam domains
llslope base (
and hillslope
e grey line in
(f) A schem
e lowest poin
anar straight
distance betw
d landslides a
nd a vector-b
ributing area,
tionship betw
A bin ( δlog
10
nship marks t
, 2016) and i
F(V) versus A
s. (d) Landsli
(D
nstream
), with
e domains. (e
ndicates the a
matic diagram
nt of a landsli
line distance
ween the high

and the fluvia
based approac
, with the gre
ween upstream
A = 0.1) in
the cutoff are
is marked by
A; the grey lin
ide populatio
h the grey lin
e) Cumulativ
assumed cutof
m showing th
ide to a stream
e L) and D
nridg
hest point of
al
ch
ey
m
a
ea
a
ne
on
ne
ve
ff
he
m
dge

a

Figur

Figur
accele
PGA
panel
with P
errors
et al.

e A.5
e A.5. Rela
eration (PGA
are calculated
(a), the red a
P
Als
> 0.1%,
s). In panel (b
(2007).
ationships be
) for (a) the W
d in 5 km wid
and blue lines
respectively,
b), data for th
etween lands
Wenchuan ear
de increments
are least-squ
and the grey
e Chi-Chi and

136
lide areal d
rthquake and
along the A-
uares linear fit
y shaded area
d Northridge
density (P
Als
)
for (b) other
-A’ trend, perp
ts for the han
as indicate±1
earthquakes
and mean
earthquakes.
pendicular to
ging wall and
s.d. uncertai
were compile

peak groun
P
Als
and mea
fault strike. I
d footwall dat
inties (residua
ed by Meunie
nd
an
In
ta
al
er

Figur

Figur
R for
seism
the re
footw
errors
Wenc
perpe

e A.6
e A.6. The pr
(a) the Wen
mic wave prop
ed and blue li
wall data, resp
s). In panel (b
chuan landsli
ndicular to fa
roduct of land
nchuan earthq
pagation mod
ines are least
pectively, an
b), data for th
ide P
Als
are
ault strike.
dslide areal de
quake and (b)
del to predict
t-squares line
nd the grey s
he other earthq
calculated

137
ensity and dis
) other earthq
earthquake-t
ar fits (log
10
(
shaded areas
quakes were
in 5 km-wi
stance to seism
quakes, to va
triggered land
(R×P
Als
) versu
indicate±1
compiled by
ide incremen
mic energy so
alidate the ap
dslide pattern
us R) for han
s.d. uncertain
Meunier et a
nts along th

ource R versu
pplication of
n. In panel (a
nging wall an
nties (residua
al. (2007). Th
he A-A’ tren
us
a
),
nd
al
he
nd

Figur

Figur
trigge
2008)
landsl
2000;
hillslo
symbo
versu

e A.7
e A.7. Schem
ers across hill
) and data fr
lides cluster a
; Meunier et
ope tops to ba
ol size propo
s D
nridge
, show
matic diagram
lslopes in D
n
rom the Wen
at hillslope rid
al., 2008). (b
ases. (c) Dist
ortional to la
wing a strong
ms of the distr
nridge
-D
nstream
s
nchuan earth
dges or bases
b) Rainfall-tri
tribution of th
andslide area
clustering ne

138
ribution of lan
space (panels
hquake (pane
s due to ampl
iggered lands
he Wenchuan
a. (d) The W
ear ridges and
ndslides indu
a and b; ada
ls c and d).
ification effec
slides are dis
n landslides in
Wenchuan lan
d consistent w
uced by seism
apted from M
(a) Seismic
cts (Densmor
stributed mor
n D
nridge
-D
nstre
ndslide popul
with a seismic

mic and rainfa
Meunier et al
cally triggere
re and Hovius
e evenly from
eam
space, wit
lation fractio
trigger.
all
l.,
ed
s,
m
th
on

Figur

Figur
averag
D
nstrea
deviat

re A.8
e A.8. Local
ge hillslope p
am
bin ( ΔD
ns
tions.
l topographic
profile in the s
stream
=0.02),
c gradient ve
study area. In
and the sol

139
rsus (a) D
nstr
n panel (a), th
lid lines are
ream
and (b) t
he red dots are
e the bounda
the resulting
e the mean gr
aries defined
composite o
radient in eac
d by standar
or
ch
rd

Figur

Figur
and th
prefer
asymm
of lan
landsc

Figur

Figur
earthq
locati
swath
aspec
Data a

e A.9
e A.9. Schem
he correspond
rentially loca
metric amplif
ndslide aspec
cape) corresp
e A.10
e A.10. Va
quake-trigger
on (star), maj
h profiles. (b
t distribution
are grouped i
matic diagram
ding distributi
ated on hills
fication of gro
cts (normalize
ponding to the
Variability in
ed landslides
jor aftershock
) to (g) Lan
n) and landsca
n 10° bins; th
illustrating th
ion of landslid
slopes facing
ound motion,
ed by the dis
e scenario in p
n preferred
along fault s
ks (circles) an
ndslide aspect
ape aspect di
he stars indica

140
he effect of as
de aspects acr
g away from
, as shown by
stribution of
panel (a).
hillslope a
strike. (a) Ma
nd surface rup
t distribution
istributions (g
ate the relativ
spect on earth
ross landscap
m the seism
y Meunier et
aspects of a
aspect direct
ap view of co
upture (red lin
ns (red curve
grey curves)
e locations of
hquake-trigge
pes. (a) Lands
mic energy s
al. (2008). (b
all hillslopes
tions of th
o-seismic upl
ne), and locat
es, corrected
for each num
f the epicente

ered landslide
slides are ofte
source due t
b) Distributio
in the studie

he Wenchua
ift, mainshoc
ions of the si
for landscap
mbered swath
er.
es
en
to
on
ed
an
ck
ix
pe
h.
141

This page is left blank intentionally

142

Bibliography
Allen, R. B., P. J. Bellingham, and S. K. Wiser (1999), Immediate damage by an earthquake to a
temperate montane forest, Ecology, 80(2), 708-714.
Ansberque, C., V. Godard, O. Bellier, J. De Sigoyer, J. Liu-Zeng, X. Xu, Z. Ren, Y. Li, and A. S.
T. E. R. Team (2015), Denudation pattern across the Longriba fault system and implications
for the geomorphological evolution of the eastern Tibetan margin, Geomorphology, 246,
542-557.
Arne, D., B. Worley, C. Wilson, S. F. Chen, D. Foster, Z. L. Luo, S. G. Liu, and P. Dirks (1997),
Differential exhumation in response to episodic thrusting along the eastern margin of the
Tibetan Plateau, Tectonophysics, 280(3–4), 239-256.
Attal, M., and J. Lavé (2006), Changes of bedload characteristics along the Marsyandi River
(central Nepal): Implications for understanding hillslope sediment supply, sediment load
evolution along fluvial networks, and denudation in active orogenic belts, Geological Society
of America Special Papers, 398, 143-171.
Attal, M., S. M. Mudd, M. D. Hurst, B. Weinman, K. Yoo, and M. Naylor (2015), Impact of
change in erosion rate and landscape steepness on hillslope and fluvial sediments grain size in
the Feather River basin (Sierra Nevada, California), Earth Surf. Dynam., 3(1), 201-222.
Avouac, J. P. (2007), 6.09 - Dynamic Processes in Extensional and Compressional Settings –
Mountain Building: From Earthquakes to Geological Deformation, in Treatise on Geophysics,
edited by S. Editor-in-Chief: Gerald, pp. 377-439, Elsevier, Amsterdam.
Barlow, J., I. Barisin, N. Rosser, D. Petley, A. Densmore, and T. Wright (2015),
Seismically-induced mass movements and volumetric fluxes resulting from the 2010 M
w
7.2
earthquake in the Sierra Cucapah, Mexico, Geomorphology, 230(0), 138-145.
Beaumont, C., R. A. Jamieson, M. H. Nguyen, and B. Lee (2001), Himalayan tectonics explained
by extrusion of a low-viscosity crustal channel coupled to focused surface denudation, Nature,
414(6865).
Bellugi, D., D. G. Milledge, W. E. Dietrich, J. A. McKean, J. T. Perron, E. B. Sudderth, and B.
Kazian (2015), A spectral clustering search algorithm for predicting shallow landslide size and
location, Journal of Geophysical Research: Earth Surface, 120(2), 300-324.
Bennett, G. L., S. R. Miller, J. J. Roering, and D. A. Schmidt (2016), Landslides, threshold slopes,
and the survival of relict terrain in the wake of the Mendocino Triple Junction, Geology, 44 (5),
363-366.
Boore, D. M., and G. M. Atkinson (2008), Ground-Motion Prediction Equations for the Average
Horizontal Component of PGA, PGV , and 5%-Damped PSA at Spectral Periods between 0.01 s
and 10.0 s, Earthquake Spectra, 24(1), 99-138.
Brardinoni, F., and M. A. Hassan (2006), Glacial erosion, evolution of river long profiles, and the
organization of process domains in mountain drainage basins of coastal British Columbia,
Journal of Geophysical Research, 111, F01013.

143

Burbank, D. W., A. E. Blythe, J. Putkonen, B. Pratt-Sitaula, E. Gabet, M. Oskin, A. Barros, and T.
P. Ojha (2003), Decoupling of erosion and precipitation in the Himalayas, Nature, 426(6967),
652-655.
Burbank, D. W., J. Leland, E. Fielding, R. S. Anderson, N. Brozovic, M. R. Reid, and C. Duncan
(1996), Bedrock incision, rock uplift and threshold hillslopes in the northwestern Himalayas,
Nature, 379(6565), 505-510.
Burchfiel, B. C., C. Zhiliang, L. Yupinc, and L. H. Royden (1995), Tectonics of the Longmen
Shan and adjacent regions, Central China, International Geology Review, 37(8), 661-735.
Burchfiel, B. C., L. H. R., R.D. van der Hilst, B.H. Hager, Z. Chen, R.W. King, C. Li, J. Lü, H.
Yao, and E. Kirby (2008), A geological and geophysical context for the Wenchuan earthquake
of 12 May 2008, Sichuan, People's Republic of China, GSA Today, 18(7), 5.
Chen, H., G.-W. Lin, M.-H. Lu, T.-Y . Shih, M.-J. Horng, S.-J. Wu, and B. Chuang (2011). Effects
of topography, lithology, rainfall and earthquake on landslide and sediment discharge in
mountain catchments of southeastern Taiwan, Geomorphology, 133(3-4), 132-142.
Chen, J. S., F. Y. Wang, X. H. Xia, and L. T. Zhang (2002), Major element chemistry of the
Changjiang (Yangtze River), Chemical Geology, 187(3-4), 231-255.
Chen, Q., H. Cheng, Y. Yang, G. Liu, and L. Liu (2014), Quantification of mass wasting volume
associated with the giant landslide Daguangbao induced by the 2008 Wenchuan earthquake
from persistent scatter InSAR, Remote Sensing of Environment, 152(0), 125-135.
Chen, S., and Wilson, C. J. L. (1996). Emplacement of the Longmen Shan Thrust-Nappe Belt
along the eastern margin of the Tibetan Plateau, Journal of Structural Geology, 18, 413-430.
Chetelat, B., C. Q. Liu, Z. Q. Zhao, Q. L. Wang, S. L. Li, J. Li, and B. L. Wang (2008),
Geochemistry of the dissolved load of the Changjiang Basin rivers: Anthropogenic impacts
and chemical weathering, Geochimica Cosmochimica Acta, 72(17), 4254-4277.
China Geological Survey (2004). China Geological Base Map and Instructions (1:2,500,000),
edited by K. Huang, SinoMaps Press, Beijing.
Clark, K. E., A.J. West, R.G. Hilton, G.P. Asner, C.A. Quesada, M.R. Silman, S.S. Saatchi, W.
Farfan-Rios, R.E. Martin, A.B. Horwath, K. Halladay, M. New, and Y. Malhi (2016).
Storm-triggered landslides in the Peruvian Andes and implications for topography, carbon
cycles, and biodiversity, Earth Surface Dynamics, 4(1), 47-70.
Clark, M.K., and L.H. Royden (2000). Topographic ooze: Building the eastern margin of Tibet by
lower crustal flow, Geology, 28(8), 703-706.
Cohen, S. C. (1996), Convenient formulas for determining dip-slip fault parameters from
geophysical observables, Bulletin of the Seismological Society of America, 86(5), 1642-1644.
Cruden, D.M. and Varnes, D.J. (1996), Landslide Types and Processes: Chapter 3 in Turner, A.K,
and Schuster, R.L., (Editors) 1996: Landslides - Investigation and mitigation. Special Report
247, Transportation Research Board, National Research Council. National Academy Press
Washington D. C, 1996, 36-75.
CSN Catalog (2015), data collected by the China Seismograph Network stations, Copyrighted &
produced by China Earthquake Networks Center (CENC)
(http://www.csndmc.ac.cn/wdc4seis@bj/earthquakes/csn_catalog_p001.jsp).
144

Cui, J.W., J.G. Zhang, D. Gao, J.X. Duan and T. Wang (2012), The ground motion attenuation
relation for the Mountainous area in Sichuan and Yunnan, 15th World Conference on
Earthquake Engineering 2012, 1, 668-677.
Cui, Y., G. Parker, J. Pizzuto, and T. E. Lisle (2003), Sediment pulses in mountain rivers: 2.
Comparison between experiments and numerical predictions, Water Resource Research, 39(9).
Dadson, S. J., N. Hovius, H. Chen, W. B. Dade, J.-C. Lin, M.-L. Hsu, C.-W. Lin, M.-J. Horng,
T.-C. Chen, J. Milliman, C. P. Stark (2004), Earthquake-triggered increase in sediment delivery
from an active mountain belt, Geology, 32(8), 733-736.
Dai, F. C., C. Xu, X. Yao, L. Xu, X. B. Tu, and Q. M. Gong (2011). Spatial distribution of
landslides trig-gered by the 2008 Ms 8.0 Wenchuan earthquake, China, Journal of Asian Earth
Sciences, 40(4), 883-895.
Day, D. G. (1980), Lithologic controls of drainage density, Catena, 7(1), 339-351.
Densmore, A. L. (2014). Workshop on the geomorphic effects of large earthquakes, Durham
University, Durham, UK.
Densmore, A. L., and N. Hovius (2000). Topographic fingerprints of bedrock landslides, Geology,
28(4), 371-374.
Densmore, A. L., M. A. Ellis, Y. Li, R. J. Zhou, G. S. Hancock, and N. Richardson (2007a),
Active tectonics of the Beichuan and Pengguan faults at the eastern margin of the Tibetan
Plateau, Tectonics, 26(4).
Densmore, A. L., R. N. Parker, N. J. Rosser, M. de Michele, Y. Li, R. Q. Huang, S. Whadcoat,
and D. N. Petley (2012), Isostasy can't be ignored reply, Nature Geoscience, 5(2), 83-84.
Densmore, A. L., S. Gupta, P. A. Allen, and N. H. Dawers (2007b), Transient landscapes at fault
tips, Journal of Geophysical Research, 112, F03S08.
DiBiase, R. A., K. X. Whipple, A. M. Heimsath, and W. B. Ouimet (2010), Landscape form and
millennial erosion rates in the San Gabriel Mountains, CA, Earth and Planetary Science Letters,
289(1-2), 134-144.
Dingman, S. L. (1978), Drainage density and streamflow: A closer look, Water Resource
Research, 14(6), 1183-1187.
Duvall, A., E. Kirby, and D. Burbank (2004), Tectonic and lithologic controls on bedrock channel
profiles and processes in coastal California, J. Geophys. Res., 109, F03002.
Egholm, D. L., M. F. Knudsen, and M. Sandiford (2013), Lifespan of mountain ranges scaled by
feedbacks between landsliding and erosion by rivers, Nature, 498(7455), 475-478.
Egholm, D.L., M.F. Knudsen, and M. Sandiford (2013). Lifespan of mountain ranges scaled by
feed-backs between landsliding and erosion by rivers, Nature, 498(7455), 475-478.
Emberson, R., N. Hovius, A. Galy, and O. Marc (2016), Chemical weathering in active mountain
belts controlled by stochastic bedrock landsliding, Nature Geoscience, 9(1), 42-45.
England, P., and P. Molnar (1990), Surface Uplift, Uplift of rocks, and exhumation of rocks,
Geology, 18(12), 1173-1177.
Fan, X.M., C.J. van Westen, O. Korup, T. Gorum, Q. Xu, F.C. Dai, R.Q. Huang, and G.H. Wang
(2012). Transient water and sediment storage of the decaying landslide dams induced by the
2008 Wenchuan earthquake, China, Geomorphology, 171, 58-68.
145

Ferguson, R. I., M. Church, C. D. Rennie, and J. G. Venditti (2015), Reconstructing a sediment
pulse: Modeling the effect of placer mining on Fraser River, Canada. Journal of Geophysical
Research: Earth Surf., 120, 1436-1454.
Gabet, E.J. (2007). A theoretical model coupling chemical weathering and physical erosion in
landslide-dominated landscapes, Earth and Planetary Science Letters, 264(1–2), 259-265.
Gabuchian, V ., A.J. Rosakis, N. Lapusta, and D.D. Oglesby (2014). Experimental investigation of
strong ground motion due to thrust fault earthquakes, Journal of Geophysical Research: Solid
Earth, 119(2), 1316-1336.
Gallen, S. F., M. K. Clark, and J. W. Godt (2015), Coseismic landslides reveal near-surface rock
strength in a high-relief, tectonically active setting, Geology, 43(1), 11-14.
Gallen, S. F., M. K. Clark, J. W. Godt, K. Roback, and N. A. Niemi. Application and evaluation
of a rapid response earthquake-triggered landslide model to the 25 April 2015 M
w
7.8 Gorkha
earthquake, Nepal, in press, Tectonophysics.
Galy, V., C. France-Lanord, O. Beyssac, P. Faure, H. Kudrass, and F. Palhol (2007), Efficient
organic carbon burial in the Bengal fan sustained by the Himalayan erosional system, Nature,
450(7168), 407-410.
Garwood, N.C., D.P. Janos, and N. Brokaw (1979). Earthquake-caused landslides: A major
disturbance to tropical forests, Science, 205(4410), 997-999.
Glade, T., and M.J. Crozier (2005). The nature of landslide hazard impact, in Landslide Hazard
and Risk, edited, pp. 41-74, John Wiley & Sons, Ltd.
Godard, V., J. Lave, J. Carcaillet, R. Cattin, D. Bourles, and J. Zhu (2010), Spatial distribution of
denudation in Eastern Tibet and regressive erosion of plateau margins, Tectonophysics,
491(1-4), 253-274.
Godard, V., R. Pik, J. Lave, R. Cattin, B. Tibari, J. de Sigoyer, M. Pubellier, and J. Zhu (2009),
Late Cenozoic evolution of the central Longmen Shan, eastern Tibet: Insight from (U-Th)/He
thermochronometry, Tectonics, 28, TC5009.
Gorum, T., and E. J. M. Carranza (2015), Control of style-of-faulting on spatial pattern of
earthquake-triggered landslides, International Journal of Environmental Science and
Technology, 12(10), 3189-3212.
Gorum, T., X. M. Fan, C. J. van Westen, R. Q. Huang, Q. Xu, C. Tang, and G. H. Wang (2011),
Distribution pattern of earthquake-induced landslides triggered by the 12 May 2008 Wenchuan
earthquake, Geomorphology, 133(3-4), 152-167..
Guzzetti, F., F. Ardizzone, M. Cardinali, M. Rossi, and D. Valigi (2009), Landslide volumes and
landslide mobilization rates in Umbria, central Italy, Earth and Planetary Science Letters,
279(3-4), 222-229.
Harp, E.L., and R.W. Jibson (1996). Landslides triggered by the 1994 Northridge, California,
earthquake, Bulletin of the Seismological Society of America, 86(1B), S319-S332.
Hartmann, J., G. Li, and A. J. West (2017), Running out of gas: Zircon
18
O-Hf-U/Pb evidence for
Snowball Earth preconditioned by low degassing, Geochemical Perspectives Letters, 4, 41-46.

146

Hilton, R. G., P. Meunier, N. Hovius, P. J. Bellingham, and A. Galy (2011), Landslide impact on
organic carbon cycling in a temperate montane forest, Earth Surface Processes and Landforms,
36(12), 1670-1679.
Horan, K., R. G. Hilton, D. Selby, C. J. Ottley, D. R. Gröcke, M. Hicks, and K. W. Burton (2017),
Mountain glaciation drives rapid oxidation of rock-bound organic carbon, Science Advances,
3(10).
Hovius, N., C.P. Stark, and P.A. Allen (1997). Sediment flux from a mountain belt derived by
landslide mapping, Geology, 25(3), 231-234.
Hovius, N., P. Meunier, L. Ching-Weei, C. Hongey, C. Yue-Gau, S. Dadson, H. Ming-Jame, and
M. Lines (2011), Prolonged seismically induced erosion and the mass balance of a large
earthquake, Earth and Planetary Science Letters, 304(3-4), 347-355.
Howarth, J. D., S. J. Fitzsimons, R. J. Norris, and G. E. Jacobsen (2012), Lake sediments record
cycles of sediment flux driven by large earthquakes on the Alpine fault, New Zealand, Geology,
40(12), 1091-1094.
Hu, M.-H., R. F. Stallard, and J. M. Edmond (1982), Major ion chemistry of some large Chinese
rivers, Nature, 298(5874).
Huang, A. Y. L., and D. R. Montgomery (2014), Topographic locations and size of earthquake
and typhoon-generated landslides, Tachia River, Taiwan, Earth Surface Processes and
Landforms, 39(3), 414-418.
Huang, J.-C., J.D. Milliman, T.-Y. Lee, Y.-C. Chen, J.-F. Lee, C.-C. Liu, J.-C. Lin, and S.-J. Kao
(2017). Terrain attributes of earthquake- and rainstorm-induced landslides in orogenic
mountain Belt, Taiwan, Earth Surface Processes and Landforms, 42(10), 1549-1559.
Huang, R. Q., and X. M. Fan (2013), The landslide story, Nature Geoscience, 6(5), 325-326.
Hubbard, J., and J. H. Shaw (2009), Uplift of the Longmen Shan and Tibetan plateau, and the
2008 Wenchuan (M=7.9) earthquake, Nature, 458(7235), 194-197.
Jarvis, A., H.I. Reuter, A. Nelson, E. Guevara, 2008, Hole-filled SRTM for the globe Version 4,
available from the CGIAR-CSI SRTM 90m Database (http://srtm.csi.cgiar.org).
Jibson, R.W., and D.K. Keefer (1993). Analysis of the seismic origin of landslides: Examples
from the New Madrid seismic zone, Geological Society of America Bulletin, 105(4), 521-536.
Jin, Z., A. J. West, F. Zhang, Z. An, R. G. Hilton, J. Yu, J. Wang, G. Li, L. Deng, and X. Wang
(2016), Seismically enhanced solute fluxes in the Yangtze River headwaters following the A.D.
2008 Wenchuan earthquake, Geology, 44(1), 47-50.
Keefer, D. K. (1984). Landslides caused by earthquakes, Geological Society of America Bulletin,
95(4), 406-421.
Keefer, D.K. (1994). The Importance of Earthquake-Induced Landslides to Long-Term Slope
Erosion and Slope-Failure Hazards in Seismically Active Regions, Geomorphology, 10(1-4),
265-284.
King, G. C. P., R. S. Stein, and J. B. Rundle (1988), The Growth of Geological Structures by
Repeated Earthquakes 1. Conceptual Framework, Journal of Geophysical Research: Solid
Earth, 93(B11), 13307-13318.

147

Kirby, E., K. Whipple, and N. Harkins (2008), Topography reveals seismic hazard, Nature
Geoscience, 1(8), 485-487.
Kirby, E., K. X. Whipple, W. Tang, and Z. Chen (2003), Distribution of active rock uplift along
the eastern margin of the Tibetan Plateau: Inferences from bedrock channel longitudinal
profiles, Journal of Geophysical Research: Solid Earth, 108, 2217.
Kirby, E., P. W. Reiners, M. A. Krol, K. X. Whipple, K. V . Hodges, K. A. Farley, W. Tang, and Z.
Chen (2002), Late Cenozoic evolution of the eastern margin of the Tibetan Plateau: Inferences
from 40Ar/39Ar and (U-Th)/He thermochronology, Tectonics, 21(1).
Korup, O., J. J. Clague, R. L. Hermanns, K. Hewitt, A. L. Strom, and J. T. Weidinger (2007),
Giant landslides, topography, and erosion, Earth and Planetary Science Letters, 261(3–4),
578-589.
Korup, O., M. J. McSaveney, and T. R. H. Davies (2004), Sediment generation and delivery from
large historic landslides in the Southern Alps, New Zealand, Geomorphology, 61(1-2),
189-207.
Kritikos, T., T. R. Robinson, and T. R. H. Davies (2015), Regional coseismic landslide hazard
assessment without historical landslide inventories: A new approach, Journal of Geophysical
Research: Earth Surface, 120(4), 711-729.
Larsen, I. J., D. R. Montgomery, and H. M. Greenberg (2014), The contribution of mountains to
global denudation, Geology, 42(6), 527-530.
Larsen, I.J., D.R. Montgomery, and O. Korup (2010). Landslide erosion controlled by hillslope
material, Nature Geoscience, 3(4), 247-251.
Lavé, J., and D. Burbank (2004), Denudation processes and rates in the Transverse Ranges,
southern California: Erosional response of a transitional landscape to external and
anthropogenic forcing, Journal of Geophysical Research: Earth Surface, 109(F1), 1-31.
Leonard, M. (2010), Earthquake fault scaling: Self-consistent relating of rupture length, width,
average displacement, and moment release, Bulletin of the Seismological Society of America,
100(5A), 1971-1988.
Li, G., A. J. West, A. L. Densmore, D. E. Hammond, Z. Jin, F. Zhang, J. Wang, and R. G. Hilton
(2016), Connectivity of earthquake-triggered landslides with the fluvial network: Implications
for landslide sediment transport after the 2008 Wenchuan earthquake, Journal of Geophysical
Research: Earth Surface, 121(4), 703-724.
Li, G., A. J. West, A. L. Densmore, Z. Jin, F. Zhang, J. Wang, M. Clark, and R. G. Hilton (2017),
Earthquakes drive focused denudation along a tectonically active mountain front, Earth and
Planetary Science Letters, 472, 253-265.
Li, G., A. J. West, A. L. Densmore, Z. Jin, R. N. Parker, and R. G. Hilton (2014), Seismic
mountain building: Landslides associated with the 2008 Wenchuan earthquake in the context
of a generalized model for earthquake volume balance, Geochemistry, Geophysics,
Geosystems, 15(4), 833-844.
Li, Y, P. A. Allen, A. L. Densmore, and X. Qiang (2003), Evolution of the Longmen Shan
foreland basin (western Sichuan, China) during the Late Triassic Indosinian orogeny, Basin
Research, 15(1), 117-138.
148

Lin, A., B. Yan, and G. Rao (2016). Millennium recurrence interval of morphogenic earthquakes
on the Qingchuan fault, northeastern segment of the Longmen Shan Thrust Belt, China,
Journal of Seismology, 20(2), 535-553.
Liu, F, B.H. Fu, S.H. Yang (2013) Quantitative estimation of the evacuation time of landslide
mass and sediment induced by the 2008 Wenchuan great earthquake along the Minjiang River,
Longmen Shan in east Tibet, Chinese Journal of Geophysics, 56(5), 1517-1525. (in Chinese
with English abstract)
Liu, F., J. Li, and S. Yang (2015), Landslide erosion associated with the Wenchuan earthquake in
the Minjiang River watershed: Implication for landscape evolution of the Longmen Shan,
eastern Tibetan Plateau, Natural Hazards, 76(3), 1911-1926.
Liu-Zeng, J., L. Wen, M. Oskin, and L. S. Zeng (2011), Focused modern denudation of the
Longmen Shan margin, eastern Tibetan Plateau, Geochemistry Geophysics Geosystems, 12,
Q1107.
Liu-Zeng, J., Z. Zhang, L. Wen, P. Tapponnier, J. Sun, X. Xing, G. Hu, Q. Xu, L. Zeng, L. Ding,
C. Ji, K.W. Hudnut and J. van der Woer (2009), Co-seismic ruptures of the 12 May 2008, M
s

8.0 Wenchuan earthquake, Sichuan: East-west crustal shortening on oblique, parallel thrusts
along the eastern edge of Tibet, Earth and Planetary Science Letters, 286(3-4), 355-370.
Luo, W., and T. Stepinski (2008), Identification of geologic contrasts from landscape dissection
pattern: An application to the Cascade Range, Oregon, USA, Geomorphology, 99(1–4), 90-98.
Ma, H., J. Heyman, X. Fu, F. Mettra, C. Ancey, and G. Parker (2014), Bed load transport over a
broad range of timescales: Determination of three regimes of fluctuations, Journal of
Geophysical Research: Earth Surface, 119(12), 2653-2673.
Malamud, B. D., D. L. Turcotte, F. Guzzetti, and P. Reichenbach (2004), Landslides, earthquakes,
and erosion, Earth and Planetary Science Letters, 229(1–2), 45-59.
Marc, O., and N. Hovius (2015), Amalgamation in landslide maps: effects and automatic
detection, Natural Hazards and Earth System Sciences, 15(4), 723-733.
Marc, O., N. Hovius, and P. Meunier (2016a). The mass balance of earthquakes and earthquake
sequences, Geophysical Research Letters, 43(8), 3708-3716.
Marc, O., N. Hovius, P. Meunier, T. Gorum, and T. Uchida (2016b), A seismologically consistent
expression for the total area and volume of earthquake-triggered landsliding, Journal of
Geophysical Research: Earth Surface, 121(4), 2015JF003732.
Marc, O., N. Hovius, P. Meunier, T. Uchida, and S. Hayashi (2015), Transient changes of
landslide rates after earthquakes, Geology, 43(10), 883-886.
McPhillips, D., P. R. Bierman, and D. H. Rood (2014), Millennial-scale record of landslides in
the Andes consistent with earthquake trigger, Nature Geoscience, 7(12), 925-930.
Meade, B. J. (2007), Present-day kinematics at the India-Asia collision zone, Geology, 35(1),
81-84.
Meng, X.-L., R. Rosenthal, and D. B. Rubin (1992), Comparing correlated correlation
coefficients, Psychological Bulletin, 111(1), 172-175.
Meunier, P., N. Hovius, and A. J. Haines (2007), Regional patterns of earthquake-triggered
landslides and their relation to ground motion, Geophysical Research Letters, 34(20), L20408.
149

Meunier, P., N. Hovius, and A.J. Haines (2008). Topographic site effects and the location of
earthquake induced landslides, Earth and Planetary Science Letters, 275(3-4), 221-232.
Moglen, G. E., E. A. B. Eltahir, and R. L. Bras (1998), On the sensitivity of drainage density to
climate change, Water Resource Research, 34(4), 855-862.
Molnar, P. (2012), Isostasy can't be ignored, Nature Geoscience, 5(2), 83-83.
Molnar, P., and P. England (1990), Late Cenozoic uplift of mountain ranges and global climate
change: chicken or egg?, Nature, 346(6279), 29-34.
Molnar, P., W.R. Boos, and D.S. Battisti (2010), Orographic controls on climate and paleoclimate
of Asia: thermal and mechanical roles for the Tibetan Plateau, Annual Review of Earth and
Planetary Sciences, 38(1), 77-102.
Montgomery, D. R. (2001), Slope distributions, threshold hillslopes, and steady-state topography,
American Journal of Science, 301(4-5), 432-454.
Montgomery, D. R., and E. Foufoula-Georgiou (1993), Channel network source representation
using digital elevation models, Water Resource Research, 29(12), 3925-3934.
Montgomery, D. R., and J. M. Buffington (1997), Channel-reach morphology in mountain
drainage basins, Geological Society of America Bulletin, 109(5), 596-611.
Montgomery, D. R., and M. T. Brandon (2002), Topographic controls on erosion rates in
tectonically active mountain ranges, Earth and Planetary Science Letters, 201(3–4), 481-489.
Moon, S., J.T. Perron, S.J. Martel, W.S. Holbrook, and J. St. Clair (2017). A model of
three-dimensional topographic stresses with implications for bedrock fractures, surface
processes, and landscape evolution, Journal of Geophysical Research: Earth Surface, 122(4),
823-846.
Oglesby, D.D., and S.M. Day (2002). Stochastic fault stress: Implications for fault dynamics and
ground motion, Bulletin of the Seismological Society of America, 92(8), 3006-3021.
Ouimet, W. B. (2010), Landslides associated with the May 12, 2008 Wenchuan earthquake:
Implications for the erosion and tectonic evolution of the Longmen Shan, Tectonophysics,
491(1-4), 244-252.
Ouimet, W. B., K. X. Whipple, and D. E. Granger (2009), Beyond threshold hillslopes: Channel
adjustment to base-level fall in tectonically active mountain ranges, Geology, 37(7), 579-582.
Parker, R.N., A.L. Densmore, N.J. Rosser, M. de Michele, Y. Li, R.Q. Huang, S. Whadcoat, and
D.N. Petley (2011), Mass wasting triggered by the 2008 Wenchuan earthquake is greater than
orogenic growth, Nature Geoscience, 4(7), 449-452.
Parker, R. N., G. T. Hancox, D. N. Petley, C. I. Massey, A. L. Densmore, and N. J. Rosser (2015),
Spatial distributions of earthquake-induced landslides and hillslope preconditioning in the
northwest South Island, New Zealand, Earth Surf. Dynam., 3(4), 501-525.
Pearce, A. J., and A. J. Watson (1986), Effects of earthquake-induced landslides on sediment
budget and transport over a 50-yr period, Geology, 14(1), 52-55.
Petley, D. (2012), Global patterns of loss of life from landslides, Geology, 40(10), 927-930.
Ramos Scharrón, C. E., E. J. Castellanos, and C. Restrepo (2012), The transfer of modern organic
carbon by landslide activity in tropical montane ecosystems, Journal of Geophysical Research:
Biogeosciences, 117(G3), G03016.
150

Raymo, M. E., and W. F. Ruddiman (1992), Tectonic forcing of Late Cenozoic climate, Nature,
359(6391), 117-122.
Raymo, M. E., W. F. Ruddiman, and P. N. Froelich (1988), Influence of Late Cenozoic mountain
building on ocean geochemical cycles, Geology, 16(7), 649-653.
Reiners, P. W., and M. T. Brandon (2006), Using thermochronology to understand orogenic
erosion, Annual Reviews of Earth and Planetary Sciences, 34, 419–466.
Richardson, N. J., A. L. Densmore, D. Seward, A. Fowler, M. Wipf, M. A. Ellis, L. Yong, and Y.
Zhang (2008), Extraordinary denudation in the Sichuan Basin: Insights from low-temperature
thermochronology adjacent to the eastern margin of the Tibetan Plateau, Journal of
Geophysical Research: Solid Earth, 113(B4).
Roback, K., M. Clark, A.J. West, D. Zekkos, G. Li, S.F. Gallen, D. Champlain, J. Godt (2017).
The size, distribution, and mobility of landslides caused by the 2015 Mw7.8 Gorkha
earthquake, Nepal, in press, Geomorphology
Robinson, T.R., N.J. Rosser, A.L. Densmore, J.G. Williams, M.E. Kincey, J. Benjamin, and H.J.A.
Bell, Rapid post-earthquake modeling of coseismic landsliding intensity and distribution for
emergency response decision support, in press, Natural Hazards and Earth System Sciences.
Royden, L.H., B.C. Burchfiel, R.W. King, E. Wang, Z. Chen, F. Shen, and Y. Liu (1997). Surface
deformation and lower crustal flow in eastern Tibet, Science, 276(5313), 788-790.
Scordilis, E. M. (2006), Empirical global relations converting M
s
and M
b
to Moment Magnitude,
Journal of Seismology, 10(2), 225-236.
Shen, Z.K., J.B. Sun, P.Z. Zhang, Y.G. Wan, M. Wang, R. Burgmann, Y.H. Zeng, W.J. Gan, H.
Liao, and Q.L. Wang (2009). Slip maxima at fault junctions and rupturing of barriers during
the 2008 Wenchuan earthquake, Nature Geoscience, 2(10), 718-724.
Simpson, G. (2015), Accumulation of permanent deformation during earthquake cycles on
reverse faults, Journal of Geophysical Research: Solid Earth, 120(3), 1958-1974.
Sklar, L. S., and W. E. Dietrich (2004), A mechanistic model for river incision into bedrock by
saltating bed load, Water Resource Research, 40(6), 1-21.
Sklar, L., and W. E. Dietrich (2013), River Longitudinal Profiles and Bedrock Incision Models:
Stream Power and the Influence of Sediment Supply, in Rivers Over Rock: Fluvial Processes
in Bedrock Channels, edited, pp. 237-260, American Geophysical Union.
St. Clair, J.,S. Moon, W.S. Holbrook, J.T. Perron, C.S. Riebe, S.J. Martel, B. Carr, C. Harman, K.
Singha, and D.D. Richter (2015). Geophysical imaging reveals topographic stress control of
bedrock weathering, Science, 350(6260), 534-538.
Steer, P., M. Simoes, R. Cattin, and J. B. H. Shyu (2014), Erosion influences the seismicity of
active thrust faults, Nature Communications, 5, 5564.
Stock, J., and W. E. Dietrich (2003), Valley incision by debris flows: Evidence of a topographic
signature, Water Resource Research, 39(4), 1089.
Summerfield, M. A., and N. J. Hulton (1994), Natural controls of fluvial denudation rates in
major world drainage basins, Journal of Geophysical Research: Solid Earth, 99(B7),
13871-13883.

151

Sutherland, D. G., M. H. Ball, S. J. Hilton, and T. E. Lisle (2002), Evolution of a
landslide-induced sediment wave in the Navarro River, California, Geological Society of
America Bulletin, 114(8), 1036-1048.
Swiss Re (2000). Random occurrence or Random Occurrence Or Predictable Disaster?: New
Models in Earthquake Probability Assessment. Swiss Re, Zurich.
Tang, C., J. Zhu, X. Qi, and J. Ding (2011), Landslides induced by the Wenchuan earthquake and
the subsequent strong rainfall event: A case study in the Beichuan area of China, Engineering
Geology, 122(1–2), 22-33.
Tapponnier, P., X. Zhiqin, F. Roger, B. Meyer, N. Arnaud, G. Wittlinger, and Y. Jingsui (2001),
Oblique stepwise rise and growth of the Tibet Plateau, Science, 294(5547), 1671-1677.
Thompson, T. B., A. Plesch, J. H. Shaw, and B. J. Meade (2015), Rapid slip-deficit rates at the
eastern margin of the Tibetan Plateau prior to the 2008 M
w
7.9 Wenchuan earthquake,
Geophysical Research Letters, 42(6), 1677-1684.
Tian, Y., B. P. Kohn, A. J. W. Gleadow, and S. Hu (2013), Constructing the Longmen Shan
eastern Tibetan Plateau margin: Insights from low-temperature thermochronology, Tectonics,
32(3), 576-592.
Torres, M. A., A. J. West, and G. Li (2014), Sulphide oxidation and carbonate dissolution as a
source of CO2 over geological timescales, Nature, 507(7492), 346-349.
Torres, M. A., N. Moosdorf, J. Hartmann, J. F. Adkins, and A. J. West (2017), Glacial weathering,
sulfide oxidation, and global carbon cycle feedbacks, Proceedings of the National Academy of
Sciences, 114(33), 8716-8721.
Tsai, Z.-X., G. J. Y. You, H.-Y. Lee, and Y.-J. Chiu (2013), Modeling the sediment yield from
landslides in the Shihmen Reservoir watershed, Taiwan, Earth Surface Processes and
Landforms, 38(7), 661-674.
Tucker, G. E., and R. L. Bras (1998), Hillslope processes, drainage density, and landscape
morphology, Water Resour Res, 34(10), 2751-2764, doi: 10.1029/98wr01474.
USGS Earthquake Hazard Program, 2008 May 12 06:28:01 UTC
(http://earthquake.usgs.gov/earthquakes).
Utsu, T. (1999), Representation and analysis of the earthquake size distribution: a historical
review and some new approaches, Pure and Applied Geophysics, 155(2), 509-535.
Vanmaercke, M., F. Ardizzone, M. Rossi, and F. Guzzetti (2017), Exploring the effects of
seismicity on landslides and catchment sediment yield: An Italian case study, Geomorphology.
Wang, E., and Q. Meng (2009), Mesozoic and Cenozoic tectonic evolution of the Longmenshan
fault belt, Science in China Series D: Earth Sciences, 52(5), 579-592.
Wang, E., E. Kirby, K. P. Furlong, M. van Soest, G. Xu, X. Shi, P. J. J. Kamp, and K. V. Hodges
(2012), Two-phase growth of high topography in eastern Tibet during the Cenozoic, Nature
Geoscience, 5(9), 640-645.
Wang, J., Z. Jin, R. G. Hilton, F. Zhang, A. L. Densmore, G. Li, and A. J. West (2015a), Controls
on fluvial evacuation of sediment from earthquake-triggered landslides, Geology, 43(2),
115-118.

152

Wang, J., Z. Jin, R. G. Hilton, F. Zhang, G. Li, A.L. Densmore, D.R. Gröcke, X. Xu, and A.J.
West (2016). Earthquake-triggered increase in biospheric carbon export from a mountain belt,
Geology, 44(9), 414-417.
Wang, M., D. Jia, J. H. Shaw, J. Hubbard, A. Plesch, Y. Li, and B. Liu (2014), The 2013 Lushan
earthquake: Implications for seismic hazards posed by the Range Front blind thrust in the
Sichuan Basin, China, Geology, 42(10), 915-918.
Wang, Y., K. Zhang, Q. Gan, W. Zhou, L. Xiong, S. Zhang, and C. Liu (2015b), Bayesian
probabilities of earthquake occurrences in Longmenshan fault system (China), Journal of
Seismology, 19(1), 69-82.
Watts, A. B. (2001), Isostacy and flexure of the lithosphere, Cambridge University Press,
Cambridge.
Wells, D. L., and K. J. Coppersmith (1994), New empirical relationships among magnitude,
rupture length, rupture width, rupture area, and surface displacement, Bulletin of the
Seismological Society of America, 84(4), 974-1002.
West, A. J., A. Galy, and M. Bickle (2005), Tectonic and climatic controls on silicate weathering,
Earth and Planetary Science Letters, 235(1-2), 211-228.
West, A. J., C. W. Lin, T. C. Lin, R. G. Hilton, S. H. Liu, C. T. Chang, K. C. Lin, A. Galy, R. B.
Sparkes, and N. Hovius (2011), Mobilization and transport of coarse woody debris to the
oceans triggered by an extreme tropical storm, Limnology and Oceanography, 56(1), 77-85.
West, A. J., R. Hetzel, G. Li, Z. Jin, F. Zhang, R. G. Hilton, and A. L. Densmore (2014), Dilution
of
10
Be in detrital quartz by earthquake-induced landslides: Implications for determining
denudation rates and potential to provide insights into landslide sediment dynamics, Earth and
Planetary Science Letters, 396(0), 143-153.
Whadcoat, S. K. (2011), Landsliding and sediment dynamics following the 2008 Wenchuan
Earthquake in the Beichuan area of China, Durham theses, Durham University (Durham
E-Theses Online: http://etheses.dur.ac.uk/1369).
Williams, J. G., N. J. Rosser, M. E. Kincey, J. Benjamin, K. J. Oven, A. L. Densmore, D. G.
Milledge, and T. R. Robinson (2017), Satellite-based emergency mapping: Landslides
triggered by the 2015 Nepal earthquake, Natural Hazards and Earth System Sciences
Discussion.
Wong, M., and G. Parker (2006), Reanalysis and correction of bed-load relation of Meyer-Peter
and Müller using their own database, Journal of Hydraulic Engineering, 132(11), 1159-1168.
Xu, C., X. Xu, and J. B. H. Shyu (2015), Database and spatial distribution of landslides triggered
by the Lushan, China Mw 6.6 earthquake of 20 April 2013, Geomorphology, 248, 77-92.
Xu, C., X. Xu, X. Yao, and F. Dai (2014), Three (nearly) complete inventories of landslides
triggered by the May 12, 2008 Wenchuan M
w
7.9 earthquake of China and their spatial
distribution statistical analysis, Landslides, 11(3), 441-461.
Xu, Q., S. Zhang, and W. Li (2011). Spatial distribution of large-scale landslides induced by the
5.12 Wenchuan Earthquake, Journal of Mountain Science, 8(2), 246.

153

Xu, X., X. Wen, G. Yu, G. Chen, Y. Klinger, J. Hubbard, and J. Shaw (2009). Coseismic reverse-
and oblique-slip surface faulting generated by the 2008 M
w
7.9 Wenchuan earthquake, China,
Geology, 37(6), 515-518.
Yang, S. L., K. H. Xu, J. D. Milliman, H. F. Yang, and C. S. Wu (2015), Decline of Yangtze River
water and sediment discharge: Impact from natural and anthropogenic changes, Scientific
reports, 5, 12581.
Yanites, B. J., G. E. Tucker, K. J. Mueller, and Y. G. Chen (2010), How rivers react to large
earthquakes: Evidence from central Taiwan, Geology, 38(7), 639-642.
Yuan, R.M., Q.H. Deng, D. Cunningham, C. Xu, X.W. Xu, and C.P. Chang (2013). Density
distribution of landslides triggered by the 2008 Wenchuan earthquake and their relationships to
peak ground acceleration, Bulletin of the Seismological Society of America, 103(4),
2344-2355.
Zhang, J., W. W. Huang, M. G. Liu, and Q. Zhou (1990), Drainage basin weathering and major
element transport of two large Chinese rivers (Huanghe and Changjiang), Journal of
Geophysical Research: Oceans, 95(C8), 13277-13288.
Zhang, P. Z., Z. Shen, M. Wang, W. Gan, R. Bürgmann, P. Molnar, Q. Wang, Z. Niu, J. Sun, J.
Wu, H. Sun, X. You, (2004), Continuous deformation of the Tibetan Plateau from global
positioning system data, Geology, 32(9), 809-812.
Zhang, S., L. M. Zhang, and T. Glade (2014). Characteristics of earthquake- and rain-induced
landslides near the epicenter of Wenchuan earthquake, Engineering Geology, 175, 58-73.
Zhang, Y., R. Wang, Y.-T. Chen, L. Xu, F. Du, M. Jin, H. Tu, and T. Dahm (2014), Kinematic
rupture model and hypocenter relocation of the 2013 M
w
6.6 Lushan earthquake constrained by
strong-motion and teleseismic data, Seismological Research Letters, 85(1), 15–22.

Abstract (if available)

Abstract In tectonically active regions, earthquakes are a key driver of landscape evolution, erosional effluxes, mountain building, and the carbon cycle. Seismic processes deform the lithosphere and create permanent topographic features, like mountain belts, at Earth’s surface. However, strong ground motion during large earthquakes also causes widespread mass wasting which collectively generates large volumes of clastic sediment and enhances fluvial erosional fluxes out of mountains. Due to the rarity of high magnitude earthquakes, limited studies have provided direct observations on how earthquakes affect landscapes and mass fluxes into and out of mountain ranges. Whereas numerous studies have looked into how earthquakes induce rock uplift, the erosive power of earthquakes remains to be better understood, especially in the context of the tectonic evolution of mountain belts. Considering those knowledge gaps, this thesis is aimed at understanding the effects of large earthquakes on mountainous landscapes, with an emphasis on earthquake-triggered landslides. Specifically, this thesis addressed the research question of how earthquakes affect landscapes from three aspects: (1) the locations of earthquake-triggered landslides in landscapes and relative to fluvial networks, as presented in Chapter 2 and the Appendix

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Landslide inventory associated with the 2008 Wenchuan Earthquake and implications for seismic mountain building

PDF

Tectonic control on landsliding revealed by the 2015 Gorkha earthquake

PDF

Extreme events as drivers of landscape evolution in active mountain belts

PDF

Germanium and silicon isotope geochemistry in terrestrial and marine low-temperature environments

PDF

Chemical weathering across spatial and temporal scales: from laboratory experiments to global models

PDF

Critical zone response to perturbation: from mountain building to wildfire

PDF

Tectonic and kinematic evolution of the Alborán domain during late orogenic extension

PDF

The impact of faulting and folding on earthquake cycles in collisional orogens

PDF

Lithium isotopes as a carbon cycle proxy: examining the effects of high-temperature weathering, understanding brachiopod archives, and validating Cenozoic records

PDF

Diagenesis of C, N, and Si in marine sediments from the Western Tropical North Atlantic and Eastern Subtropical North Pacific: pore water models and sedimentary studies

PDF

New insights into glacial-interglacial carbon cycle: multi-proxy and numerical modeling

PDF

A trace metal investigation of the radiocarbon anomaly in the eastern equatorial Pacific during the last glacial termination

PDF

Characterizing fault behavior and earthquake surface expression on timescales of single events to multiple millennia

PDF

Concentration and size partitioning of trace metals in surface waters of the global ocean and storm runoff

PDF

On protracted Laurentian rifting, continental freeboard, and cap carbonates in Neoproterozoic time

PDF

Carbonate dissolution at the seafloor: fluxes and drivers from a novel in situ porewater sampler

PDF

Ages, origins and biogeochemical role of water across a tropical mountain to floodplain transition

PDF

Identifying long-runout landslides on the surface of Mars

PDF

Towards an understanding of fault-system mechanics: from single earthquakes on isolated faults to millenial-scale collective plate-boundary fault-system behavior

PDF

Modeling deep ocean water and sediment dynamics in the eastern Pacific Ocean using actinium-227 and other naturally occurring radioisotopes

Asset Metadata

Creator Li, Gen (author)

Core Title Earthquake-driven landsliding, erosion and mountain building: from the eastern Tibetan mountains towards global models

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Geological Sciences

Publication Date 11/13/2019

Defense Date 06/21/2017

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag denudation,Earthquakes,Erosion,Landslides,mountain building,OAI-PMH Harvest,Tibetan mountains

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor West, Abraham Joshua (committee chair), Hammond, Douglas E. (committee member), Robinson, Alexandar (committee member)

Creator Email genli@usc.edu,ligenlg@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-454787

Unique identifier UC11264314

Identifier etd-LiGen-5905.pdf (filename),usctheses-c40-454787 (legacy record id)

Legacy Identifier etd-LiGen-5905.pdf

Dmrecord 454787

Document Type Dissertation

Rights Li, Gen

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA