Flowstone ideograms: deciphering the climate messages of Asian speleothems

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Flowstone Ideograms: Deciphering
the Climate Messages of Asian Speleothems
by
Jun Hu
Ph.D. Candidate
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)
August 2019
In loving memory of my grandmother Xianxiu Cui
ii
Acknowledgements
I wound like to extend thanks to many people, without whom this thesis would not
have been possible:
To my advisor Dr. Julien Emile-Geay, for your support, patience and guidance through-
out my whole Ph.D. I am inspired by your enthusiasm and hard working. Thank
you for leading me to paleoclimate research and training me to be a more mature
scientist. Thank you for supporting me for conferences and introducing me to great
scientists in this field. I appreciate your patience and time of polishing my every
manuscript and talk.
To my dissertation and qualifying exam committees, Dr. Sarah Feakins, Dr. Naomi
Levine, Dr. Frank Corsetti, and Dr. George Ban-Weiss, for your well-directed
comments, questions, guidance, and time.
To my other academic mentors and collaborators, Dr. Jesse Nusbaumer, Dr. Judson
Partin, Dr. Clay Tabor, Dr. Jess Adkins and Dr. David Noone: for your help,
kindness, insightful feedback, countless edits of my manuscripts. Dr. Laia Comas-
Bru for your support to joining the SISAL project.
To my labmates, Feng Zhu, Sylvia Dee, Deborah Khider, Michael Erb, and Jianghao
Wang. Thank you for being amazing labmates and friends. It has been a true joy
to work with all of you.
To my friends and every staff in the department. Thank you for your help and all the
fun we have had in the five years.
To my best friends, Qian, Stan, Feimin, Zhixiang, Chenxi, Yaoxian and particularly,
Yan. Thank you for your companionship and sharing every moment of my life.
And finally, to my parents. Thank you for your endless love, and for supporting me to
pursue a degree far away from home.
iii
Contents
Dedication ii
Acknowledgements iii
List of Tables vii
List of Figures viii
Abstract xiii
Introduction 1
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1 Correlation-basedinterpretationsofpaleoclimatedata–wherestatistics
meet past climates 18
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2 Challenge #1: serial correlation . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.2 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3 Challenge #2: Test multiplicity . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.2 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4 Challenge #3: Age uncertainties . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4.2 Compound challenges in one: the case of Crystal Cave . . . . . . . 28
1.4.3 Effect of serial correlation . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.4 Effect of test multiplicity . . . . . . . . . . . . . . . . . . . . . . . . 31
1.4.5 Effect of age uncertainties . . . . . . . . . . . . . . . . . . . . . . . 31
1.4.6 Correlations considering all three challenges . . . . . . . . . . . . . 36
1.4.7 Time-uncertain spectral analysis . . . . . . . . . . . . . . . . . . . . 39
1.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
iv
2 Impact of convective activity on precipitation δ
18
O in isotope-enabled
general circulation models 49
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.1 Correlation between stratiform precipitation fraction and δ
18
O
P
. . 55
2.3.2 Correlation between outgoing longwave radiation and δ
18
O
P
. . . . 63
2.3.3 Implications for speleothem record interpretation . . . . . . . . . . 66
2.4 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3 DecipheringChinesespeleothemswithanisotope-enabledclimatemodel 82
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Coherency analysis of Chinese speleothem δ
18
O records . . . . . . . . . . . 85
3.3 Experimental design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.1 Orbital variability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.2 Interannual variability . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4 Limited coherency of Asian speleothems over the Holocene, with impli-
cations for the Meghalayan age 108
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2.1 Data & Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2.2 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3 Spatial coherency of Holocene Asian speleothem δ
18
O . . . . . . . . . . . . 114
4.3.1 Spatial-temporal variability . . . . . . . . . . . . . . . . . . . . . . 114
4.3.2 Implications from model simulations . . . . . . . . . . . . . . . . . 117
4.4 Asian speleothem δ
18
O around 4.2 ka BP . . . . . . . . . . . . . . . . . . . 120
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Conclusion 135
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
APPENDICES 152
A Supplemental material for Chapter 2 152
v
B Supplemental material for Chapter 4 159
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
vi
List of Tables
2.1 General circulation models used in this study. * denotes a “SWING2 model” 54
4.1 Asian speleothem records used in this study. Resolution refers to the
median spacing between consecutive observations. BP means “before 1950
AD”. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
vii
List of Figures
1.1 Illustration of the influence of degrees of freedom on correlation
significance. Thep-valueandnumbersofdegreesoffreedom(DOF)ofthe
correlation (0.13) between two AR(1) time series (500 samples each) with
the changing autocorrelation φ. The green dashed line is the 5% criteria
for 5% level significance test. . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.2 Illustration of the traditional significance test procedure and the
FDR procedure on an illustrative example, with q = α = 20%.
p-values at each grid point (p
(i)
) are ranked in increasing order, plotted
against i/m, where m is the total number of grid points. The blue dashed
line is the traditional α threshold for the p-value. Green dots indicate
they are significant by both traditional and FDR procedure, and blue dots
indicate they are only significant by the traditional procedure. . . . . . . . 25
1.3 Spatial correlation of October Kirirom cellulose δ
18
O values with
the October-November-December mean of (a) CMAP precipita-
tion, and (b) NOAA interpolated OLR. Black dots indicate that the
correlation does not pass the significance test at the 90% level . . . . . . . 27
1.4 Influence of degrees of freedom and test multiplicity problem on
correlations. (a) The effective number of degrees of freedom of the corre-
lation between Crystal Caveδ
18
O and SST from 1856-2007; The correlation
between these series, considering the test multiplicity problem (False Dis-
covery Rate) (d, e), or not (b, c); considering autocorrelation (c,e) or not
(b,d). Blackdotsindicatethatthecorrelationdoesnotpassthesignificance
test at the 5% level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 The age modeling results of the Crystal cave δ
18
O record using a
Bchron age model. The gray area is the 95% confidence interval of the
age at each depth. The red lines show 10 random paths out of the 1,000 age
models generated, and the blue curve shows the StalAge-generated model
used by McCabe-Glynn et al. (2013) . . . . . . . . . . . . . . . . . . . . . 32
1.6 Crystal cave δ
18
O time series with uncertainties. Top panel: The
median, 97.5% quantile and 2.5% quantile of the ensemble of 1000 Crystal
δ
18
OtimegeneratedfromtheBchronmodel. Bottompanel: Thetimeseries
of δ
18
O record from McCabe-Glynn et al. (2013) (blue) and the median of
δ
18
O record from the Bchron model (black). . . . . . . . . . . . . . . . . . 33
viii
1.7 Correlations considering age uncertainty. The median (a), interquar-
tile range (b), the 2.5% (c) and the 97.5% (d) percentile of the correlation
between theδ
18
O age ensemble and SST. Black dots indicate that the cor-
relation does not pass the significance test at the 5% level, accounting for
serial correlation. Note that the interquartile range here is a measure of
distributional spread, and has no measure of significance attached to it. . . 35
1.8 Same as Fig. 1.7, but testing for field correlations while controlling for the
False Discovery Rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.9 Illustration of the FDR procedure on the 2.5% quantile (a) and
the 97.5% quantile (b) correlations shown in Fig. 1.8, using the
false discovery rate q = 5% (red line). p-values at each grid point (p
(i)
)
are ranked in increasing order, plotted against i/m, where m is the total
number of grid points. The blue dashed line is the traditional 5% threshold
for thep-value. Dots withp-values below this threshold are shown in blue.
Inthisexample, manydotsfallbelowthenominalthreshold(5%), butnone
fall below the red line, which means that they are not significant according
to the FDR-controlling procedure. . . . . . . . . . . . . . . . . . . . . . . . 38
1.10 The MTM-estimated spectra of the δ
18
O record from the Bchron
age model (blue line and gray shaded area) compared to the spectrum of
the published record, together with a simulated AR(1) benchmark. . . . . 40
2.1 Correlation of mean monthlyδ
18
O (GNIP station data) and strat-
iform fraction (TRMM satellite data). From Aggarwal et al. (2016),
Fig. 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2 Relationship between monthly stratiform precipitation fraction
andδ
18
O
P
at the same locations as Aggarwal et al. (2016) in (a) SPEEDY-
IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g)
iCAM5. β values are ordinary least square slopes. . . . . . . . . . . . . . . 56
2.3 Relationship between monthly stratiform precipitation fraction
and δ
18
O
P
in the tropics and mid-latitudes in (a) SPEEDY-IER; (b)
LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β
values are ordinary least square slopes. Stars behindβ values represent the
correlation passing the 95% level with an isospectral test. . . . . . . . . . . 58
2.4 Correlation between monthly stratiform precipitation fraction
and δ
18
O
P
in seven models. Dots represent correlation not passing the
95% level significance isospectral test. Yellow circles mark the locations of
GNIP stations analyzed in Aggarwal et al. (2016) . . . . . . . . . . . . . . 60
2.5 Vertical profiles of water vapor δ
18
O (a), deuterium excess (b),
and condensational heating (c) over the convective rainfall region, the
stratiform rainfall region, and no rain region in the tropics (30
◦
S-30
◦
N) in
iCAM5. The convective/stratiform rainfall region is where the proportion
of convective/stratiform rainfall to total rainfall exceeds 0.8. . . . . . . . . 62
ix
2.6 RelationshipbetweenNOAAinterpolatedOLRandGNIPδ
18
O
P
over
the stations in Aggarwal et al. (2016), excluding stations higher than 2000
m (4 stations). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.7 Correlation between monthly outgoing longwave radiation and
δ
18
O
P
in four models. Dots represent correlation not passing the 95%
level significance isospectral test. . . . . . . . . . . . . . . . . . . . . . . . 65
2.8 The variance of monthly δ
18
O
P
explained by local contributions:
stratiform rainfall fraction (a), outgoing longwave radiation (b), precipi-
tation amount (c), local water vapor convergence (d) and the total local
contributions (e) in iCAM5. Blue dots are cave sites collected in the global
speleothem database SISAL_v1, and the names of these caves are listed
besides Fig. 2.8e. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.9 Climatological stratiform fraction in the tropics from satellite obser-
vations (a) and isotope-enabled models (b,c,d,e,f,g,h). Regions where cli-
matological precipitation is less than 1.5 mm/day are masked. . . . . . . . 71
3.1 The spatial distribution of the Chinese speleothem records used
in this section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.2 Chinese speleothemδ
18
O over the last 400,000 yrs with their local
summer insolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.3 Chinese speleothem δ
18
O over Holocene with their local summer
insolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4 Tagged regions in the iCESM simulation from 1953-2012 . . . . . 90
3.5 Differencebetweenprecessionminimumandmaximum. (a)Weighted
annual mean precipitation difference; (b) JJAδ
18
O
P
and 850 hPa wind dif-
ference; (c) JJA precipitation difference . . . . . . . . . . . . . . . . . . . . 91
3.6 Climatological monthly mean precipitation in East China (20
◦
N-
35
◦
N, 105
◦
E-120
◦
E) in precession minimum (a) and precession maximum
(b). Hovmoller diagram of climatological monthly mean precipitation in
East China in precession minimum (c) and precession maximum (d). . . . 94
3.7 Contributions of four factors to seasonal variability of δ
18
O
P
over
East China from tagged regions: (a) moisture source composition, (b)
Rainout changes, (c) Condensation changes, and (d) moisture location
change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.8 Interannual variability of δ
18
O
P
in East China (a) Time series of
weighted annual mean δ
18
O
P
in East China; (b) Seasonal contributions of
four factors to weighted annual δ
18
O
P
in East China . . . . . . . . . . . . 97
3.9 Interpretation of interannual variability of Chinese δ
18
O
P
(a) Con-
tribution of moisture source location changes to East China δ
18
O
P
from
tagged regions; (b) Difference of 850 hPa wind and precipitation in ASO
between high and low δ
18
O
P
years of East China. The red rectangular is
the defined East China region (20
◦
N-35
◦
N, 105
◦
E-120
◦
E). . . . . . . . . . . 98
x
4.1 ThelocationsofeightAsianspeleothemrecordsovertheHolocene
analyzed in this study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2 MC-PCA of Asian speleothems over the Holocene. (a) The spatial
distribution of the loading of the first PC of MC-PCA; (b) The time series
of the first PC of MC-PCA (λ = 18%) . . . . . . . . . . . . . . . . . . . . 116
4.3 The spatial distribution of the loading and the time series of the
first principle component of precipitation δ
18
O and soil δ
18
O from
iCAM5-iCLM4. Both the spatial loading and principle component are
normalized by dividing the square root of their eigenvalues. . . . . . . . . . 118
4.4 Comparison between the simulated type 2 stalagmiteδ
18
O at Hes-
hang cave and type 5 stalagmite δ
18
O Lianhua cave from Karsto-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.5 Asian speleothem δ
18
O around 4.2 ka BP. The distribution of the
detrended δ
18
O values of eight Asian speleothems with 150-years intervals
based on 1000 plausible age models over Holocene. The vertical lines are
δ
18
O values around 4.2 ka BP (4,075-4,225 BP). . . . . . . . . . . . . . . . 121
A.1 Relationship between transformed monthly stratiform precipita-
tion fraction and transformedδ
18
O
P
at the same locations as Aggarwal
et al. (2016) in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e)
MIROC; (f) HadAM4; (g) iCAM5. β values are ordinary least square slopes.153
A.2 Relationship between transformed monthly stratiform precipita-
tion fraction and raw δ
18
O
P
at the same locations as Aggarwal et al.
(2016)in(a)SPEEDY-IER;(b)LMDZ;(c)CAM2; (d)isoGSM;(e)MIROC;
(f) HadAM4; (g) iCAM5. β values are ordinary least square slopes. . . . . 154
A.3 Relationship between transformed monthly stratiform precipi-
tation fraction and transformed δ
18
O
P
in the tropics and mid-
latitudes in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e)
MIROC; (f) HadAM4; (g) iCAM5. β values are ordinary least square slopes.155
A.4 Relationship between transformed monthly stratiform precipita-
tion fraction and raw δ
18
O
P
in the tropics and mid-latitudes in
(a) SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f)
HadAM4; (g) iCAM5. β values are ordinary least square slopes. . . . . . . 156
A.5 Vertical profiles of water vapor δ
18
O over the convective rainfall
region, and in the stratiform rainfall region, and no rain region
in the tropics (30
◦
S-30
◦
N) in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2;
(d) isoGSM; (e) MIROC; (f) HadAM4. The convective/stratiform rainfall
region is where the proportion of convective/stratiform rainfall to total
rainfall exceeds 0.8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
xi
A.6 Vertical profiles of deuterium excess over the convective rainfall
region, and in the stratiform rainfall region, and no rain region
in the tropics (30
◦
S-30
◦
N) in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2;
(d) isoGSM; (e) MIROC; (f) HadAM4. The convective/stratiform rainfall
region is where the proportion of convective/stratiform rainfall to total
rainfall exceeds 0.8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
B.1 TimeseriesofeightAsianspeleothemδ
18
OovertheHolocenewith
age uncertainties. Dashed lines are local summer insolation. Errorbars
above each time series are
230
Th ages and errors. . . . . . . . . . . . . . . . 161
B.2 Detrended time series of eight Asian speleothem δ
18
O over the
Holocene with age uncertainties. . . . . . . . . . . . . . . . . . . . . . 162
B.3 The age modeling results of eight Asian speleothem δ
18
O over
the Holocene using a Bchron age model. The gray area is the 95%
highest density region of the age at each depth. The red lines show 10
random paths out of the 1000 age models generated. . . . . . . . . . . . . 163
B.4 MC-PCA eigenvalue spectrum (scree plot). The blue line shows the
mean eigenvalues of 1000 ensembles with 1-sigma standard deviation. The
orange line indicates the 95% confidence level of the red noise (AR (1)
benchmark. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
B.5 The spatial distribution of the loading and the time series of
the second principle component of precipitation δ
18
O and soil
δ
18
O from iCAM5-iCLM4. Both the spatial loading and principle com-
ponent are normalized by dividing the square root of their eigenvalues. . . 165
B.6 Comparison between simulated precipitation δ
18
O and soilδ
18
O at
Heshang and Lianhua cave from iCAM5-iCLM4. . . . . . . . . . . . 166
B.7 Comparison between five types of simulated stalagmite δ
18
O at
Heshang and Lianhua cave from Karstolution. . . . . . . . . . . . . 166
B.8 The percent of Mawmluh speleothem δ
18
O values around 4.2 ka
BP falling below 5% quantile of Holocene values . . . . . . . . . . . 167
B.9 Time series of δ
18
O in stalagmites KMA (Berkelhammer et al.,
2013), ML.1 and ML.2 (Kathayat et al., 2018) in Mawmluh cave.
Light colored envelopes encompass 95% of age ensemble members. . . . . 167
xii
Abstract
Speleothems have been widely used to reconstruct past hydroclimate variability, par-
ticularly in the Asian Monsoon region. While Asian speleothem δ
18
O is traditionally
interpreted as “monsoon intensity”, recent work has proposed alternative interpretations
such as water vapor transport and changes in atmospheric circulation that challenge or
redefine this long-held concept. Also, challenges exist when we attempt to get interpre-
tations of speleothem δ
18
O by correlating them with climate variables. This thesis aims
to better understand the climate signals preserved in Asian speleothems over various
timescales ranging from orbital to interannual.
The thesis begins with the discussion of the challenges in interpreting speleothem
δ
18
O based on correlation analysis and the methods to circumvent these statistical issues
are provided. Then a state-of-the-art isotope-enabled climate model iCESM is employed
to investigate the interpretation of Asian speleothem δ
18
O. The model is capable of
credibly simulating precipitation δ
18
O and particularly the impact of convective activity
on precipitation δ
18
O. Then this model is used to quantify contributions to precipita-
tion δ
18
O over China at both orbital and interannual time scales. Results suggest that
orbital-scale speleothem δ
18
O variations at Chinese sites mainly represent the meridional
migration of the Asian monsoon circulation, accompanied by an early northward move-
ment of the East Asian rain belt. At interannual scales, Chinese speleothem δ
18
O is
also tied to the intensity of monsoonal circulation, via a change in moisture source loca-
tions: enhanced moisture delivery from remote source regions leads to depletedδ
18
O. The
results offer a re-interpretation of the concept of “monsoon intensity” as “enhanced mon-
soonalcirculation”ratherthanprecipitationamount. Despitethiscomplexity, speleothem
δ
18
O at orbital scales is coherent across Asia. However, whether the coherency still exists
at short time scales is a question. Here we synthesize Asian speleothem δ
18
O over the
xiii
Holocene to investigate whether they show coherent variability at sub-orbital scales, and
particularly focus on the “4.2 ka event” and the newly announced Meghalayan age (4.2
ka BP to present). We find no coherent variability among Asian speleothemδ
18
O at these
time scales. This asynchrony can be explained by heterogenous soil, vegetation, and karst
processes modifying coherent climate inputs. Given the relatively small amplitude (1-3
h) of speleothems at sub-orbital scales, extracting hydroclimate variability at these time
scales from speleothem δ
18
O should be done with caution, and supplemented by proxies
such as δ
13
C and trace elements ratios.
xiv
Introduction
Speleothemsaresecondarymineraldeposits(principallyofcalciumcarbonates)formed
in caves, most commonly as stalagmites, stalactites, and flowstones (White and Culver,
2011). They form when calcium carbonates are precipitated from cave dripping water
saturated with dissolved calcium, which was originally from surface precipitation that
percolated downward through soil, fractures and conduits in karst systems (McDermott,
2004, Fairchild et al., 2006). Thus, speleothems inherit signals from precipitation and
archive it, making it possible to reconstruct some aspects of paleoclimate. The main
challenge of this large and growing field, and the central question of this thesis, is "what
aspects?"
Since the 1960s, speleothems have been used to investigate past climate variability.
Early research mainly attempted to estimate paleotemperature based on oxygen isotopes
in speleothem carbonates and their fluid inclusions (Hendy and Wilson, 1968, Duplessy
et al., 1970, Thompson et al., 1974, Schwarcz et al., 1976, Harmon et al., 1979, Gascoyne
et al., 1980). This idea followed the revolutionary understanding of paleoclimate achieved
by foraminifera oxygen isotopes and ice core water isotopes at that time (Emiliani, 1955,
Shackleton, 1967, Shackleton and Opdyke, 1973, Johnsen et al., 1972). However, the
progress was hampered by difficulties with extraction and analysis of fluid inclusions, an
over-simplified view of the relationship between meteoric water isotopes and temperature,
and the need for large samples (∼10 g) for U-series dating (McDermott, 2004, Fairchild
et al., 2006).
After the late 1980s, advances in thermal ionization mass-spectrometry (TIMS) tech-
niques and U-series dating put new life into speleothem studies (Edwards et al., 1987, Li
et al., 1989). This method reduced the required sample size, but also provided dating
results almost 10 times more precise than the traditional alpha-spectrometry methods.
1
About one decade later, multi-collector inductively coupled plasma mass spectroscopy
(MC-ICP-MS) further improved the analytical precision and sample size requirements
(Shen et al., 2002, Cheng et al., 2013), especially making dating low-uranium Holocene
speleothems possible (McDermott, 2004). In addition, on the aspect of stable isotope
measurements, micro-milling, and laser-ablation techniques have improved the temporal
resolution of speleothem isotope data to monthly (Frappier et al., 2002, Spötl and Mattey,
2006).
With these technique developments, speleothems have advantages over other paleo-
climate records because they feature high dating precision which can approach±0.1%
(Cheng et al., 2013) and do not have age calibration problems, unlike radiocarbon dating
(McDermott, 2004). Also, speleothems can be found at all latitudes over continents, and
some of them grow fast (up to 1.0 mm/year), providing potentially high temporal resolu-
tion records. Finally, speleothems provide multiple paleoclimate proxies at the same site,
such as layer thickness, the stable isotopes of oxygen and carbon, clumped isotopes, and
trace elements.
These advances have led to the widespread use of speleothems as paleoclimate prox-
ies at timescales ranging from sub-annual to orbital since the late 1990s. Scientists also
shifted their focus on the global climate teleconnections captured by speleothems instead
of attempting to use speleothems to reconstruct paleotemperature. Around the year
2000, some pioneering studies found speleothems in mid-latitudes aligned with climate
variability revealed in Greenland ice-core records and significantly improved their dat-
ing (Bar-Matthews et al., 1999, Wang et al., 2001, Spötl and Mangini, 2002). Since
then, speleothem studies have entered the mainstream of paleoclimate research. Among
these studies, Wang et al. (2001) (which is so far the highest cited paper on the topic
of “speleothem” on the ISI’s Web of Science) not only revealed a well-dated speleothem
record in the monsoon region for the first time, but also refined the chronology of Green-
land ice core records. Since then Asian speleothem records have been in the spotlight
of the paleoclimate community (Yuan et al., 2004, Dykoski et al., 2005, Cosford et al.,
2008, Wang et al., 2008, Hu et al., 2008, Cheng et al., 2009, 2012, Carolin et al., 2013,
Berkelhammer et al., 2013, Cheng et al., 2016, Kathayat et al., 2016, Carolin et al., 2016,
2
Griffiths et al., 2016, Chen et al., 2016). Most of these studies are on speleothem δ
18
O
1
,
and I will also focus on Asian speleothem δ
18
O in the thesis.
The most remarkable feature of Asian speleothems is that their oxygen isotopes coin-
cide well with summer local insolation at orbital scales and their variability is dominated
by precessional cycles (23 kyrs) (Wang et al., 2008, Cheng et al., 2009, 2012, 2016).
Traditionally, this speleothem δ
18
O variability in the monsoon region has been inter-
preted as “monsoon intensity” because seasonal precipitation dominates annual precipi-
tation in these regions: based on the “amount effect” (Dansgaard, 1964), the precipitation
δ
18
O there should represent monsoon precipitation (Wang et al., 2001, Yuan et al., 2004,
Hu et al., 2008, Cheng et al., 2009, 2012). This interpretation assumes that the amount
effect is dominant in determining the variability of precipitation δ
18
O and that other
effects are secondary.
However, recent studies raise questions about the validity of the amount effect to inter-
pret speleothem δ
18
O. Local precipitation does not always correlate well with precipita-
tionδ
18
O over interannual or longer time scales (Dayem et al., 2010, Eastoe and Dettman,
2016). The variability of precipitation δ
18
O can be determined by upstream water vapor
δ
18
O (Pausata et al., 2011) or changes in moisture source, influenced by changes in atmo-
spheric circulation (Maher and Thompson, 2012, Tan, 2014). Also, many studies show
that convective activity could have a large effect on precipitation δ
18
O (Kurita, 2013,
Lekshmy et al., 2014, Aggarwal et al., 2016, Cai and Tian, 2016). On orbital time scales,
the change in seasonality also contributes to the variability of speleothem δ
18
O (Clemens
et al., 2010). Thus, it is necessary to compare the relative contributions of different factors
impacting precipitation δ
18
O, which will help constrain the interpretation of speleothem
δ
18
O. In addition, precipitationδ
18
O is not the same as speleothemδ
18
O. Climate signals
inprecipitationδ
18
Ocanbedistortedbysoil(TangandFeng,2001,GazisandFeng,2004),
vegetation (Moreira et al., 1997), karst processes (Jones et al., 2000, Baker and Bradley,
2010, Moerman et al., 2014) and kinetic effects in the deposition process (Hendy, 1971,
1
The ratio of the stable isotope oxygen
18
O and
16
O is commonly noted byδ
18
O, which is defined as:
δ
18
O=
18
O/
16
O
sample
−
18
O/
16
O
standard
18
O/
16
O
standard
×1000 (0.1)
The international reference standards are Vienna Pee Dee Belemnite (VPDB) for carbonates, and
Vienna Standard Mean Ocean Water (VSMOW) for water samples
3
Mickler et al., 2006). This is especially necessary to consider when studying speleothem
δ
18
O variability at sub-orbital scales, at which the magnitude of these effects (up to 3h)
is comparable with the amplitude of speleothem δ
18
O (Baker et al., 2013). For example,
in arid climate conditions, the evaporation could be more intense and get speleothem
δ
18
O enriched (Lachniet, 2009), and the rapid degassing of CO
2
due to the large gra-
dient of CO
2
between drip water and the cave can lead to the increase of speleothem
δ
18
O (Lachniet, 2009, Baker and Bradley, 2010), which can either amplify or mask the
climate signals in speleothems (McDermott, 2004).
Also, the term “monsoon intensity” is vague since the strength of the monsoon-related
wind and precipitation are decoupled in East Asia: a strong monsoon wind does not imply
intense precipitation, or vice versa. Modern observations reveal a “triple-mode” of the
East Asian precipitation at interannual scales (Zhou and Yu, 2005, Day et al., 2015). In
this mode, strong monsoon winds are associated with low precipitation in central China
and high precipitation in northern and southern China. In this way, spatially coherent
speleothem δ
18
O variability may correspond to asynchronous hydroclimate variability in
East Asia (Zhang et al., 2018).
Apart from interpreting speleothemδ
18
O as monsoon intensity, many studies attempt
to get other interpretations by correlating speleothem δ
18
O with climate variables and
forcings such as solar irradiance (Asmerom et al., 2007, Duan et al., 2014), the North
Atlantic Oscillation (Proctor et al., 2000, Scholz et al., 2012), the El Niño-Southern Oscil-
lation (Moerman et al., 2013, Chen et al., 2016) and sea surface temperature (McCabe-
Glynn et al., 2013). The correlation analysis in these studies often do not consider the
effects of age uncertainties, which have a leading-order effect on instrumental calibrations
(Hu et al., 2017). In addition, the loss of degrees of freedom due to the autocorrelation of
the time series (Yule, 1926, Dawdy and Matalas, 1964) and the “test multiplicity" prob-
lem associated with testing for significant correlations with an entire climate field (Holm,
1979, Benjamini and Hochberg, 1995, Storey, 2002) also affect correlation analysis and
the conclusions that stem from it.
Furthermore, traditional speleothem δ
18
O studies focus on a single site, or coherent
variability of speleothem δ
18
O records at a few sites to interpret large-scale climate vari-
ability since the variability of speleothem δ
18
O is coherent at orbital scales (Wang et al.,
4
2008, Cheng et al., 2012). However, recent studies show that Asian speleothemδ
18
O is not
synchronous at least over millennial scales (Wan et al., 2011, Chu et al., 2012, Li et al.,
2014). Thus it is necessary to synthesize Asian speleothemδ
18
O records and analyze their
spatial-temporal variability at sub-orbital scales and understand the mechanisms behind
this variability.
Therefore, the research goal of this thesis is to understand the interpretation of Asian
speleothemδ
18
O at timescales ranging from orbital to interannual. Given the large range
of timescales and the limits in models and speleothem data, I reduce this broad ques-
tion to a set of manageable problems. First, I will discuss the challenges of interpret-
ing speleothem δ
18
O based on correlation analysis and show how to circumvent these
statistical issues. In the thesis, I will employ an isotope-enabled model to disentangle
different factors affecting speleothem δ
18
O, so I will first evaluate how well the model
simulates precipitationδ
18
O, and especially focus on its capability in simulating the rela-
tionship between convective activity and precipitation δ
18
O. Then this model will be
used to quantitatively decompose precipitation δ
18
O variability in China at both orbital
and interannual time scales. Finally, I will synthesize Asian speleothem δ
18
O over the
Holocene to investigate whether they show coherent variability at sub-orbital timescales,
and particularly focus on the “4.2 ka event” and the newly announced Meghalayan age.
The thesis is structured the following four chapters:
1. Chapter 1: Correlation-based interpretations of paleoclimate data – where statis-
tics meet past climates
2. Chapter 2: Impact of convective activity on precipitation δ
18
O in isotope-enabled
general circulation models
3. Chapter 3: Deciphering Chinese speleothems with an isotope-enabled climate
model
4. Chapter 4: Limited coherency of Asian speleothems over the Holocene, with impli-
cations for the Meghalayan age
In Chapter 1, I will first review three statistical challenges in correlation analysis for
paleoclimate studies: the loss of degrees of freedom due to serial correlation (Yule, 1926,
5
Dawdy and Matalas, 1964), the test multiplicity problem in connection with a climate
field (Holm, 1979, Benjamini and Hochberg, 1995, Storey, 2002), and the presence of
age uncertainties (Crowley, 1999, Wunsch, 2003) – particularly for speleothem records.
While these issues have long been known to statisticians, they are not widely appreciated
by the wider paleoclimate community. Then I show how these challenges apply to, and
affect the published interpretations of paleoclimate proxies in the recent paleoclimate
literature (Proctor et al., 2000, Zhu et al., 2012, McCabe-Glynn et al., 2013) and suggest
that future studies should address these issues to strengthen their conclusions. Also, I will
provide methods and open-source code to circumvent these issues. For the presence of age
uncertainties, the Bayesian age model Bchron (Haslett and Parnell, 2008, Parnell et al.,
2011) will be used to translate age uncertainties to proxy uncertainties, and then the
resultant correlations will have uncertainties. This will lay a foundation for the analysis
in Chapter 4.
In the following chapters, I will employ the isotope-enabled climate model iCESM (its
atmospheric component is iCAM5 (Nusbaumer et al., 2017)) to investigate the interpreta-
tion of Asian speleothem δ
18
O and precipitationδ
18
O. Isotope-enabled models are useful
tools to study the variability of precipitation δ
18
O by separating different factors influ-
encing precipitation δ
18
O (Joussaume et al., 1984, Jouzel et al., 1987, Hoffmann et al.,
1998, Noone and Simmonds, 2002, Schmidt et al., 2007, Yoshimura et al., 2008), and a
recent version of iCESM (Nusbaumer et al., 2017) is capable of tracking moisture and
water isotopes by tagging water vapor evaporating from a specific region, which is key to
quantifying the contributions of different factors to precipitation δ
18
O.
In Chapter 2, I will first evaluate how this model simulates precipitation δ
18
O and
establish that the model is adequate to the task. As the convective activity plays an
important role in influencing precipitation δ
18
O in the Asian monsoon region, and new
observations indicate that precipitation δ
18
O is negatively correlated with the fraction of
stratiform precipitation (Aggarwal et al., 2016), I will also focus on evaluating how the
model simulates the impact of convective activity on precipitation δ
18
O. I will compare
the simulation results of iCAM5 with observations and other isotope-enabled models,
and the result will show that iCAM5 can simulate the impact of convective activity on
precipitation δ
18
O well. In addition, I will estimate the relative contributions of local
6
rainfall amount and other processes to precipitationδ
18
O variability in the Asian monsoon
regions from iCAM5. The result will suggest the local processes explain a small amount
of precipitation δ
18
O variability, highlighting the importance of contribution from water
vapor transport and large-scale circulation, which I will discuss in Chapter 3.
Then in Chapter 3, I will use the model iCESM to investigate the interpretation of
Chinese speleothem δ
18
O on both orbital and interannual time scales. To understand
speleothem δ
18
O, we need to understand precipitation δ
18
O at first, and as mentioned
above, iCESM is a good tool to separate the contributions of different factoring affecting
precipitation δ
18
O by tagging and tracking water vapors and their oxygen isotopes. The
numerical experimental results will quantify the factors affecting precipitation δ
18
O, and
the result will show that the moisture source location dominates the variability of precip-
itation δ
18
O at both time scales. For orbital scales, the simulation result will also show
that precipitationδ
18
O change corresponds to the migration of monsoon winds instead of
rainfall amount. Also, the subseasonal feature of rainbelt in China is worth investigating.
When the rainbelt stays in central China longer and “jumps” northwards later, the rainy
season there will be longer and yield more precipitation, which may affect precipitation
δ
18
O. Thus, I will also discuss the change of rainfall belt seasonality, and how it affects
the responses of precipitation and precipitation δ
18
O to precessional forcing.
Although speleothem δ
18
O at orbital scales is extremely coherent across Asia (Cheng
et al., 2012, Battisti et al., 2014, Cheng et al., 2016), this coherency vanishes at scales
shorter than millennial (Wan et al., 2011, Chu et al., 2012, Li et al., 2014). In Chapter 4,
I will investigate whether Asian speleothemδ
18
O show coherent variability at sub-orbital
scales over Holocene, and then try to employ iCESM and a karst forward model (Treble
et al., 2019) to physically explain the potential asynchrony. Also, in 2018, the Meghalayan
age (4.2 ka BP to present) of the Holocene was formally ratified by the International
Commission on Stratigraphy (Walker et al., 2018), and its global stratotype section and
point is a stalagmite record in Mawmluh cave in northeast India (Berkelhammer et al.,
2013). This chapter will investigate whether Asian speleothemδ
18
O records show the “4.2
ka event” marking the beginning of the late Holocene (Walker et al., 2012, 2018). The
extent, magnitude, and timing of the “4.2 ka event” will also be discussed in this chapter.
7
The analysis will take age uncertainties into account, which will turn out to be critical
for this discussion of abrupt climate change.
Bibliography
Aggarwal, P. K., Romatschke, U., Araguas-Araguas, L., Belachew, D., Longstaffe, F. J.,
Berg, P., Schumacher, C., andFunk, A.(2016). Proportionsofconvectiveandstratiform
precipitation revealed in water isotope ratios. Nature Geoscience, 9(8):624.
Asmerom, Y., Polyak, V., Burns, S., and Rassmussen, J. (2007). Solar forcing of Holocene
climate: New insights from a speleothem record, southwestern United States. Geology,
35(1):1–4.
Baker, A. and Bradley, C. (2010). Modern stalagmiteδ
18
O: Instrumental calibration and
forward modelling. Global and Planetary Change, 71(3):201–206.
Baker, A., Bradley, C., and Phipps, S. J. (2013). Hydrological modeling of stalag-
mite δ
18
O response to glacial-interglacial transitions. Geophysical Research Letters,
40(12):3207–3212.
Bar-Matthews, M., Ayalon, A., Kaufman, A., and Wasserburg, G. J. (1999). The Eastern
Mediterranean paleoclimate as a reflection of regional events: Soreq cave, Israel. Earth
and Planetary Science Letters, 166(1-2):85–95.
Battisti, D., Ding, Q., and Roe, G. (2014). Coherent pan-Asian climatic and isotopic
response to orbital forcing of tropical insolation. Journal of Geophysical Research:
Atmospheres, 119(21):11–997.
Benjamini, Y.andHochberg, Y.(1995). ControllingtheFalseDiscoveryRate: APractical
and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society.
Series B (Methodological), 57(1):289–300.
Berkelhammer, M., Sinha, A., Stott, L., Cheng, H., Pausata, F., and Yoshimura, K.
(2013). An abrupt shift in the Indian monsoon 4000 years ago. Climates, Landscapes,
and Civilizations, pages 75–88.
8
Cai, Z. and Tian, L. (2016). Atmospheric Controls on Seasonal and Interannual Variations
in the Precipitation Isotope in the East Asian Monsoon Region. Journal of Climate,
29(4):1339–1352.
Carolin, S. A., Cobb, K. M., Adkins, J. F., Clark, B., Conroy, J. L., Lejau, S., Malang,
J., and Tuen, A. A. (2013). Varied response of western Pacific hydrology to climate
forcings over the last glacial period. Science, page 1233797.
Carolin, S. A., Cobb, K. M., Lynch-Stieglitz, J., Moerman, J. W., Partin, J. W., Lejau,
S., Malang, J., Clark, B., Tuen, A. A., and Adkins, J. F. (2016). Northern Borneo
stalagmite records reveal West Pacific hydroclimate across MIS 5 and 6. Earth and
Planetary Science Letters, 439:182–193.
Chen, S., Hoffmann, S. S., Lund, D. C., Cobb, K. M., Emile-Geay, J., and Adkins,
J. F. (2016). A high-resolution speleothem record of western equatorial Pacific rain-
fall: Implications for Holocene ENSO evolution. Earth and Planetary Science Letters,
442:61–71.
Cheng, H., Edwards, R. L., Broecker, W. S., Denton, G. H., Kong, X., Wang, Y., Zhang,
R., and Wang, X. (2009). Ice Age Terminations. Science, 326(5950):248–252.
Cheng, H., Edwards, R. L., Shen, C.-C., Polyak, V. J., Asmerom, Y., Woodhead, J.,
Hellstrom, J., Wang, Y., Kong, X., Spötl, C., Wang, X., and Alexander, E. C. (2013).
Improvements in
230
Th dating,
230
Th and
234
U half-life values, and U–Th isotopic mea-
surements by multi-collector inductively coupled plasma mass spectrometry. Earth and
Planetary Science Letters, 371-372:82 – 91.
Cheng, H., Edwards, R. L., Sinha, A., Spötl, C., Yi, L., Chen, S., Kelly, M., Kathayat,
G., Wang, X., Li, X., et al. (2016). The Asian monsoon over the past 640,000 years
and ice age terminations. Nature, 534(7609):640–646.
Cheng, H., Sinha, A., Wang, X., Cruz, F. W., and Edwards, R. L. (2012). The Global
PaleomonsoonasseenthroughspeleothemrecordsfromAsiaandtheAmericas. Climate
Dynamics, 39(5):1045–1062.
9
Chu, P.C., Li, H.-C., Fan, C., andChen, Y.-H.(2012). Speleothemevidencefortemporal–
spatial variation in the East Asian Summer Monsoon since the Medieval Warm Period.
Journal of Quaternary Science, 27(9):901–910.
Clemens, S. C., Prell, W. L., and Sun, Y. (2010). Orbital-scale timing and mechanisms
drivingLatePleistoceneIndo-Asiansummermonsoons: Reinterpretingcavespeleothem
δ
18
O. Paleoceanography, 25(4).
Cosford, J., Qing, H., Yuan, D., Zhang, M., Holmden, C., Patterson, W., and Hai, C.
(2008). Millennial-scalevariabilityintheAsianmonsoon: Evidencefromoxygenisotope
records from stalagmites in southeastern China. Palaeogeography, Palaeoclimatology,
Palaeoecology, 266(1):3–12.
Crowley, T. J. (1999). Correlating high-frequency climate variations. Paleoceanography,
14(3):271–272.
Dansgaard, W. (1964). Stable isotopes in precipitation. Tellus, 16(4):436–468.
Dawdy, D. and Matalas, N. (1964). Statistical and probability analysis of hydrologic data,
part III: Analysis of variance, covariance and time series. McGraw-Hill.
Day, J. A., Fung, I., and Risi, C. (2015). Coupling of South and East Asian Monsoon
Precipitation in July–August. Journal of Climate, 28(11):4330–4356.
Dayem, K. E., Molnar, P., Battisti, D. S., and Roe, G. H. (2010). Lessons learned from
oxygen isotopes in modern precipitation applied to interpretation of speleothem records
of paleoclimate from Eastern Asia. Earth and Planetary Science Letters, 295(1):219–
230.
Duan, F., Wang, Y., Shen, C.-C., Wang, Y., Cheng, H., Wu, C.-C., Hu, H.-M., Kong, X.,
Liu, D., and Zhao, K. (2014). Evidence for solar cycles in a late Holocene speleothem
record from Dongge Cave, China. Scientific Reports, 4:5159.
Duplessy, J. C., Labeyrie, J., Lalou, C., and Nguyen, H. V. (1970). Continental climatic
variations between 130,000 and 90,000 years BP. Nature, 226(5246):631–633.
10
Dykoski, C. A., Edwards, R. L., Cheng, H., Yuan, D., Cai, Y., Zhang, M., Lin, Y., Qing,
J., An, Z., and Revenaugh, J. (2005). A high-resolution, absolute-dated Holocene and
deglacialAsianmonsoonrecordfromDonggeCave, China. Earth and Planetary Science
Letters, 233(1):71–86.
Eastoe, C. and Dettman, D. (2016). Isotope amount effects in hydrologic and climate
reconstructions of monsoon climates: Implications of some long-term data sets for pre-
cipitation. Chemical Geology, 430:78–89.
Edwards, R. L., Chen, J., and Wasserburg, G. (1987).
238
U-
234
U-
230
Th-
232
Th systematics
and the precise measurement of time over the past 500,000 years. Earth and Planetary
Science Letters, 81(2-3):175–192.
Emiliani, C. (1955). Pleistocene temperatures. The Journal of Geology, 63(6):538–578.
Fairchild, I. J., Smith, C. L., Baker, A., Fuller, L., Spötl, C., Mattey, D., McDermott, F.,
et al. (2006). Modification and preservation of environmental signals in speleothems.
Earth-Science Reviews, 75(1):105–153.
Frappier, A., Sahagian, D., González, L. A., and Carpenter, S. J. (2002). El Nino events
recorded by stalagmite carbon isotopes. Science, 298(5593):565–565.
Gascoyne, M., Schwarcz, H. P., and Ford, D. C. (1980). A palaeotemperature record for
the mid-Wisconsin in Vancouver Island. Nature, 285(5765):474–476.
Gazis, C. and Feng, X. (2004). A stable isotope study of soil water: evidence for mixing
and preferential flow paths. Geoderma, 119(1-2):97–111.
Griffiths, M. L., Kimbrough, A. K., Gagan, M. K., Drysdale, R. N., Cole, J. E., Johnson,
K. R., Zhao, J.-X., Cook, B. I., Hellstrom, J. C., and Hantoro, W. S. (2016). Western
Pacific hydroclimate linked to global climate variability over the past two millennia.
Nature Communications, 7:11719.
Harmon, R. S., Schwarcz, H. P., and O’Neil, J. R. (1979). D/H ratios in speleothem fluid
inclusions: a guide to variations in the isotopic composition of meteoric precipitation?
Earth and Planetary Science Letters, 42(2):254–266.
11
Haslett, J. and Parnell, A. (2008). A simple monotone process with application to
radiocarbon-dated depth chronologies. J. R. Stat. Soc. Ser. C-Appl. Stat., 57:399–418.
Hendy, C. and Wilson, A. (1968). Palaeoclimatic data from speleothems. Nature,
219(5149):48.
Hendy, C. H. (1971). The isotopic geochemistry of speleothems—i. the calculation of
the effects of different modes of formation on the isotopic composition of speleothems
and their applicability as palaeoclimatic indicators. Geochimica et cosmochimica Acta,
35(8):801–824.
Hoffmann, G., Werner, M., and Heimann, M. (1998). Water isotope module of the
ECHAM atmospheric general circulation model: A study on timescales from days to
several years. Journal of Geophysical Research: Atmospheres, 103(D14):16871–16896.
Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian
Journal of Statistics, pages 65–70.
Hu, C., Henderson, G. M., Huang, J., Xie, S., Sun, Y., and Johnson, K. R. (2008). Quan-
tification of Holocene Asian monsoon rainfall from spatially separated cave records.
Earth and Planetary Science Letters, 266(3):221–232.
Hu, J., Emile-Geay, J., and Partin, J. (2017). Correlation-based interpretations of paleo-
climate data–where statistics meet past climates. Earth and Planetary Science Letters,
459:362–371.
Johnsen, S., Dansgaard, W., Clausen, H., et al. (1972). Oxygen isotope profiles through
the Antarctic and Greenland ice sheets. Nature, 235(5339):429.
Jones, I. C., Banner, J. L., and Humphrey, J. D. (2000). Estimating recharge in a tropical
karst aquifer. Water Resources Research, 36(5):1289–1299.
Joussaume, S., Sadourny, R., and Jouzel, J. (1984). A general circulation model of water
isotope cycles in the atmosphere. Nature, 311(5981):24.
12
Jouzel, J., Russell, G., Suozzo, R., Koster, R., White, J., and Broecker, W. (1987). Sim-
ulations of the HDO and H
18
2
O atmospheric cycles using the NASA GISS General Cir-
culation Model: The seasonal cycle for present-day conditions. Journal of Geophysical
Research: Atmospheres, 92(D12):14739–14760.
Kathayat, G., Cheng, H., Sinha, A., Spötl, C., Edwards, R. L., Zhang, H., Li, X., Yi,
L., Ning, Y., Cai, Y., et al. (2016). Indian monsoon variability on millennial-orbital
timescales. Scientific Reports, 6.
Kurita, N. (2013). Water isotopic variability in response to mesoscale convective system
overthetropicalocean. Journal of Geophysical Research: Atmospheres, 118(18):10,376–
10,390.
Lachniet,M.S.(2009). Climaticandenvironmentalcontrolsonspeleothemoxygen-isotope
values. Quaternary Science Reviews, 28(5):412–432.
Lekshmy, P., Midhun, M., Ramesh, R., andJani, R.(2014).
18
Odepletioninmonsoonrain
relates to large scale organized convection rather than the amount of rainfall. Scientific
Reports, 4:5661.
Li, W.-X., Lundberg, J., Dickin, A., Ford, D., Schwarcz, H., McNutt, R., and Williams,
D. (1989). High-precision mass-spectrometric uranium-series dating of cave deposits
and implications for palaeoclimate studies. Nature, 339(6225):534.
Li,Y.,Wang,N.,Zhou,X.,Zhang,C.,andWang,Y.(2014). Synchronousorasynchronous
Holocene Indian and East Asian summer monsoon evolution: a synthesis on Holocene
Asian summer monsoon simulations, records and modern monsoon indices. Global and
Planetary Change, 116:30–40.
Maher, B. A. and Thompson, R. (2012). Oxygen isotopes from Chinese caves: records
not of monsoon rainfall but of circulation regime. Journal of Quaternary Science,
27(6):615–624.
McCabe-Glynn, S., Johnson, K. R., Strong, C., Berkelhammer, M., Sinha, A., Cheng, H.,
and Edwards, R. L. (2013). Variable North Pacific influence on drought in southwestern
North America since AD 854. Nat. Geosci., 6(8):617–621.
13
McDermott, F. (2004). Palaeo-climate reconstruction from stable isotope variations in
speleothems: a review. Quaternary Science Reviews, 23(7-8):901–918.
Mickler, P. J., Stern, L. A., and Banner, J. L. (2006). Large kinetic isotope effects in
modern speleothems. GSA Bulletin, 118(1-2):65–81.
Moerman, J. W., Cobb, K. M., Adkins, J. F., Sodemann, H., Clark, B., and Tuen, A. A.
(2013). Diurnal to interannual rainfall δ
18
O variations in northern Borneo driven by
regional hydrology. Earth and Planetary Science Letters, 369:108–119.
Moerman, J.W., Cobb, K.M., Partin, J.W., Meckler, A.N., Carolin, S.A., Adkins, J.F.,
Lejau, S., Malang, J., Clark, B., and Tuen, A. A. (2014). Transformation of ENSO-
related rainwater to dripwater δ
18
O variability by vadose water mixing. Geophysical
Research Letters, 41(22):7907–7915.
Moreira, M., Sternberg, L., Martinelli, L., Victoria, R., Barbosa, E., Bonates, L., and
Nepstad, D. (1997). Contribution of transpiration to forest ambient vapour based on
isotopic measurements. Global Change Biology, 3(5):439–450.
Noone, D. and Simmonds, I. (2002). Associations between δ
18
O of water and climate
parameters in a simulation of atmospheric circulation for 1979–95. Journal of Climate,
15(22):3150–3169.
Nusbaumer, J., Wong, T. E., Bardeen, C., and Noone, D. (2017). Evaluating hydrological
processes in the Community Atmosphere Model Version 5 (CAM5) using stable isotope
ratios of water. Journal of Advances in Modeling Earth Systems, 9(2):949–977.
Parnell, A. C., Buck, C. E., and Doan, T. K. (2011). A review of statistical chronology
models for high-resolution, proxy-based Holocene palaeoenvironmental reconstruction.
Quaternary Science Reviews, 30(21):2948–2960.
Pausata, F. S., Battisti, D. S., Nisancioglu, K. H., and Bitz, C. M. (2011). Chinese sta-
lagmiteδ
18
O controlled by changes in the Indian monsoon during a simulated Heinrich
event. Nature Geoscience, 4(7):474–480.
14
Proctor, C., Baker, A., Barnes, W., and Gilmour, M. (2000). A thousand year speleothem
proxy record of North Atlantic climate from Scotland. Climate Dynamics, 16(10-
11):815–820.
Schmidt, G. A., LeGrande, A. N., and Hoffmann, G. (2007). Water isotope expressions
of intrinsic and forced variability in a coupled ocean-atmosphere model. Journal of
Geophysical Research: Atmospheres, 112:D10103.
Scholz, D., Frisia, S., Borsato, A., Spötl, C., Fohlmeister, J., Mudelsee, M., Miorandi, R.,
and Mangini, A. (2012). Holocene climate variability in north-eastern Italy: potential
influence of the NAO and solar activity recorded by speleothem data. Climate of the
Past, 8(4):1367–1383.
Schwarcz, H. P., Harmon, R. S., Thompson, P., and Ford, D. C. (1976). Stable isotope
studies of fluid inclusions in speleothems and their paleoclimatic significance. Geochim-
ica et Cosmochimica Acta, 40(6):657–665.
Shackleton, N. (1967). Oxygen isotope analyses and Pleistocene temperatures re-assessed.
Nature, 215(5096):15–17.
Shackleton, N. J. and Opdyke, N. D. (1973). Oxygen isotope and palaeomagnetic stratig-
raphyofEquatorialPacificcoreV28-238: Oxygenisotopetemperaturesandicevolumes
on a 10
5
year and 10
6
year scale. Quaternary research, 3(1):39–55.
Shen, C.-C., Edwards, R. L., Cheng, H., Dorale, J. A., Thomas, R. B., Moran, S. B.,
Weinstein, S. E., and Edmonds, H. N. (2002). Uranium and thorium isotopic and
concentration measurements by magnetic sector inductively coupled plasma mass spec-
trometry. Chemical Geology, 185(3-4):165–178.
Spötl, C. and Mangini, A. (2002). Stalagmite from the Austrian Alps reveals Dansgaard–
Oeschger events during isotope stage 3:: Implications for the absolute chronology of
Greenland ice cores. Earth and Planetary Science Letters, 203(1):507–518.
Spötl, C. and Mattey, D. (2006). Stable isotope microsampling of speleothems for
palaeoenvironmental studies: a comparison of microdrill, micromill and laser ablation
techniques. Chemical Geology, 235(1-2):48–58.
15
Storey, J. D. (2002). A direct approach to false discovery rates. Journal of the Royal
Statistical Society: Series B (Statistical Methodology), 64(3):479–498.
Tan, M. (2014). Circulation effect: response of precipitation δ
18
O to the ENSO cycle in
monsoon regions of China. Climate Dynamics, 42(3-4):1067–1077.
Tang, K. and Feng, X. (2001). The effect of soil hydrology on the oxygen and hydrogen
isotopic compositions of plants’ source water. Earth and Planetary Science Letters,
185(3-4):355–367.
Thompson, P., Schwarcz, H. P., and Ford, D. C. (1974). Continental Pleistocene climatic
variations from speleothem age and isotopic data. Science, 184(4139):893–895.
Treble,P.,Mah,M.,Griffiths,A.,Baker,A.,Deininger,M.,Kelly,B.,Scholz,D.,andHan-
kin, S. (2019). Separating isotopic impacts of karst and in-cave processes from climate
variability using an integrated speleothem isotope-enabled forward model. EarthArXiv.
Walker, M., Head, M. H., Berklehammer, M., Bjorck, S., Cheng, H., Cwynar, L., Fisher,
D., Gkinis, V., Long, A., Lowe, J., et al. (2018). Formal ratification of the subdivision
of the Holocene Series/Epoch (Quaternary System/Period): two new Global Boundary
Stratotype Sections and Points (GSSPs) and three new stages/subseries. Episodes.
Walker, M. J., Berkelhammer, M., Björck, S., Cwynar, L. C., Fisher, D. A., Long, A. J.,
Lowe, J. J., Newnham, R. M., Rasmussen, S. O., and Weiss, H. (2012). Formal subdi-
vision of the Holocene Series/Epoch: a Discussion Paper by a Working Group of INTI-
MATE (Integration of ice-core, marine and terrestrial records) and the Subcommission
on Quaternary Stratigraphy (International Commission on Stratigraphy). Journal of
Quaternary Science, 27(7):649–659.
Wan, N.-J., Li, H.-C., Liu, Z.-Q., Yang, H.-Y., Yuan, D.-X., and Chen, Y.-H. (2011).
Spatial variations of monsoonal rain in eastern China: Instrumental, historic and
speleothem records. Journal of Asian Earth Sciences, 40(6):1139–1150.
Wang, Y., Cheng, H., Edwards, R. L., Kong, X., Shao, X., Chen, S., Wu, J., Jiang, X.,
Wang, X., and An, Z. (2008). Millennial-and orbital-scale changes in the East Asian
monsoon over the past 224,000 years. Nature, 451(7182):1090–1093.
16
Wang, Y.-J., Cheng, H., Edwards, R. L., An, Z., Wu, J., Shen, C.-C., and Dorale, J. A.
(2001). A high-resolution absolute-dated late Pleistocene monsoon record from Hulu
Cave, China. Science, 294(5550):2345–2348.
White, W. B. and Culver, D. C. (2011). Encyclopedia of caves. Academic Press.
Wunsch, C. (2003). Greenland—Antarctic phase relations and millennial time-scale cli-
matefluctuationsintheGreenlandice-cores. Quaternary Science Reviews, 22(15):1631–
1646.
Yoshimura, K., Kanamitsu, M., Noone, D., and Oki, T. (2008). Historical isotope simula-
tionusingReanalysisatmosphericdata. Journal of Geophysical Research: Atmospheres,
113:D19108.
Yuan, D., Cheng, H., Edwards, R. L., Dykoski, C. A., Kelly, M. J., Zhang, M., Qing, J.,
Lin, Y., Wang, Y., Wu, J., et al. (2004). Timing, duration, and transitions of the last
interglacial Asian monsoon. Science, 304(5670):575–578.
Yule, G. U. (1926). Why do we sometimes get nonsense-correlations between time-series
– a study in sampling and the nature of time-series. Journal of the Royal Statistical
Society, 89(1):1–63.
Zhang, H., Griffiths, M. L., Chiang, J. C., Kong, W., Wu, S., Atwood, A., Huang, J.,
Cheng, H., Ning, Y., and Xie, S. (2018). East Asian hydroclimate modulated by the
position of the westerlies during Termination I. Science, 362(6414):580–583.
Zhou, T.-J. and Yu, R.-C. (2005). Atmospheric water vapor transport associated with
typical anomalous summer rainfall patterns in China. Journal of Geophysical Research:
Atmospheres, 110(D8).
Zhu, M., Stott, L., Buckley, B., Yoshimura, K., and Ra, K. (2012). Indo-Pacific Warm
Pool convection and ENSO since 1867 derived from Cambodian pine tree cellulose
oxygen isotopes. Journal of Geophysical Research: Atmospheres, 117(D11).
17
Chapter 1
Correlation-based interpretations of paleo-
climate data – where statistics meet past cli-
mates
Abstract
Correlationanalysisisomnipresentinpaleoclimatology,andoftenservestosupportthe
proposed climatic interpretation of a given proxy record. However, this analysis presents
several statistical challenges, each of which is sufficient to nullify the interpretation: the
loss of degrees of freedom due to serial correlation, the test multiplicity problem in con-
nection with a climate field, and the presence of age uncertainties. While these issues have
long been known to statisticians, they are not widely appreciated by the wider paleocli-
mate community; yet they can have a first-order impact on scientific conclusions. Here
we use three examples from the recent paleoclimate literature to highlight how spurious
correlations affect the published interpretations of paleoclimate proxies, and suggest that
future studies should address these issues to strengthen their conclusions. In some cases,
correlations that were previously claimed to be significant are found insignificant, thereby
challenging published interpretations. In other cases, minor adjustments can be made to
Publication Details: Hu, J., Emile-Geay, J., and Partin, J. (2017). Correlation-based interpretations
of paleoclimate data–where statistics meet past climates. Earth and Planetary Science Letters, 459:362-
371, doi: 10.1016/j.epsl.2016.11.048.
18
safeguard against these concerns. Because such problems arise so commonly with paleo-
climate data, we provide open-source code to address them. Ultimately, we conclude that
statistics alone cannot ground-truth a proxy, and recommend establishing a mechanistic
understanding of a proxy signal as a sounder basis for interpretation.
1.1 Introduction
Inferring past climate conditions from proxy archives is a central tenet of paleocli-
matology. The calibration of paleoclimate proxies is accomplished in two main ways:
space-based calibrations and time-based calibrations (defined below). In space-based cal-
ibrations, the values of a proxy at different locations are calibrated to measured climate
indicators at the same locations, as exemplified by the calibration of paleothermometers
in the core-top of marine sediments (e.g. Tierney and Tingley, 2014, Khider et al., 2015).
This approach is relatively forgiving of time uncertainties, as long as core-top values are
broadly contemporaneous, in relation to the question being asked of the cores. In time-
based calibrations, on the other hand, proxy timeseries overlapping with the instrumental
era are calibrated against an instrumental target (e.g. Jones et al., 2009, Tingley et al.,
2012), via correlation analysis or the closely-related linear regression.
Thus “ground-truthing” a proxy record often involves establishing that its correlation
to an instrumental climate variable (whether local, regional, or global) is significant in
some way. Significance of correlations is most commonly assessed via a t-test, which
assumes that samples are independent, identically-distributed, and Gaussian. However,
these criteria may not be fulfilled in paleoclimate timeseries due to their intrinsic proper-
ties (Ghil et al., 2002).
Indeed, the loss of degrees of freedom due to autocorrelation has long been known to
challenge the assumption of independence (Yule, 1926), though workarounds are known
(e.g. Dawdy and Matalas, 1964). Non-Gaussianity may also prove an issue, especially for
precipitation timeseries, though relatively simple transformations may alleviate it (Emile-
Geay and Tingley, 2016).
19
Additionally, correlating proxies with instrumental climate fields is a common way of
establishing the ability of a proxy to capture large-scale climate information. Unfortu-
nately when implemented as a mining exercise using a large, spatially gridded dataset,
test multiplicity becomes a problem. We will review how this problem may be successfully
circumvented using simple statistical approaches (Benjamini and Hochberg, 1995, Storey,
2002).
Finally, the presence of age uncertainties may bring substantial uncertainties to time-
based correlations between records (e.g. Crowley, 1999, Wunsch, 2003, Black et al., 2016).
We will show a robust approach to quantifying age uncertainties and how they propagate
to correlation and other analyses.
The article is structured as follows. In Section 1.2 we show the importance of consid-
ering autocorrelation in cross-correlation analyses. In Section 1.3, we briefly introduce the
"multiplehypothesistest"problemandthefalsediscoveryrateandshowhowitaffectscor-
relations with a climate field. In Section 1.4, we introduce the effects of age uncertainties,
how they influence the interpretation of a speleothem record, and how this compounds
with the other two challenges. We finish with a discussion of the significance of these
results, and propose strategies to mitigate these statistical issues going forward.
1.2 Challenge #1: serial correlation
1.2.1 Theory
The most common way to determine the significance of Pearson’s product-moment
correlation involves at-test. Student’st distribution is fully determined by the number of
degreesoffreedomavailableinthesample(ν). ForN independentsamples,ν =N−2, but
it may be considerably lower when this assumption is violated, leading to overconfident
assessments of significance.
As an example, consider correlations between two timeseries x(t) and y(t) generated
by autoregressive processes of order 1 (a common timeseries model for serially correlated
data (e.g. Emile-Geay, 2016, Chapter 8)). Each process is evenly sampled 500 times
and their correlation coefficient is 0.13, which is significant at the 5% level assuming
independence (hence, ν = 498). However, the lag-1 autocorrelation of each time series
20
(φ) is 0.8, which is common for climate variables like temperature, as well as for many
paleoclimate records, which tend to have a red spectrum (Ghil et al., 2002). This means
that neighboring samples are highly dependent, so the effective numbers of degrees of
freedom, ν
eff
, is much lower. This number may be estimated via the following relation
(Dawdy and Matalas, 1964):
ν
eff
=N
1−φ
x
·φ
y
1 +φ
x
·φ
y
(1.1)
whereφ
x
,φ
y
are the lag-1 autocorrelation coefficients of two time series x, y respectively.
Based on equation 1.1, when either lag-1 autocorrelation coefficient increases, the
effective number of degrees of freedom decreases, and the p-value of the test increases.
In this case, the effective number of degrees of freedom decreases from 498 to 99 after
considering the autocorrelation, and the p-value rises to 0.19, suggesting the correlation
is no longer significant at the 5% level. Fig. 1.1 shows how the p-value and the degrees
of freedom change for a time series of 500 samples and a fixed correlation of 0.13 just by
changing the autocorrelation coefficients (φ
x
= φ
y
= φ for simplicity). As the autocor-
relation increases, the p-values increases, and the degrees of freedom decrease. When all
samples are independent (φ
x
= φ
y
= 0), the p-value is far smaller than 5%. When the
autocorrelation increases to about 0.65, the p-value becomes larger than 5%, making the
correlation insignificant at this level. The problem only worsens asφ increases, and as we
shall see in this article, values above 0.8 are quite typical of paleoclimate timeseries.
Autocorrelation is thus a very serious challenge, which alone can substantially raise
the bar of a significance test; if ignored, it may lead to overconfident assessments of
significance.
1.2.2 Application
To see this effect at work in the real world, consider the example of Proctor et al.
(2000), who used the band width in a stalagmite (SU-96-7) from Uamh an Tartair (north-
west Scotland) to reconstruct the North Atlantic Oscillation (NAO). The record was
dated by counting annual bands, with only 17 bands as double annual bands, implying a
counting error less than 20 years. When compared to the whole length of the entire 1087
21
0.0 0.2 0.4 0.6 0.8 1.0
Autocorrelation φ
0
100
200
300
400
500
DOF
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
p-value
Figure 1.1: Illustration of the influence of degrees of freedom on correlation sig-
nificance. Thep-valueandnumbersofdegreesoffreedom(DOF)ofthecorrelation(0.13)
between two AR(1) time series (500 samples each) with the changing autocorrelation φ.
The green dashed line is the 5% criteria for 5% level significance test.
year-long record, this amounts to only 2%. Therefore, the influence of age uncertainties
can be neglected to first order.
Theclimaticinterpretationofthestalagmitewasbasedonthehighcorrelationbetween
the band width and the temperature/precipitation ratio (r = 0.80) as well as the corre-
lation between band width and the winter NAO index (r =−0.70) by using decadally-
smoothed data. Here we apply the effective degrees of freedom in testing the significance
of correlation, since the correlation significance may be biased by autocorrelation due to
the effect of smoothing. Also, inherent aspects of these records leads to complications
using statistics based on normally distributed populations, as the band width distribu-
tion of the stalagmite record is bimodal instead of normal. The t-test for correlation
significance assumes that both time series are normally distributed, negating its use as a
statistical tool unless appropriate transformations are made.
22
Consideringtheautocorrelationofthesmootheddata, thehighcorrelationbetweenthe
band width of stalagmites and the ratio temperature/precipitation (T/P) in the instru-
mental period is not significant at 5% significance level (the adjusted p-value is 0.44).
The correlation between the band width of stalagmites and winter NAO is also not sig-
nificant, because of high autocorrelations of the smoothed time series of the band width
(φ = 0.99), T/P (φ = 0.99) and winter NAO (φ = 0.95). However, this result is based on
an assumption of normality, and as discussed above, the distribution of the band width
in this speleothem is bimodal, hence non-normal (not shown). Thus, transforming the
non-normal series to normality (Emile-Geay and Tingley, 2016) is necessary. After this
transformation, the correlations pass the significance test at the 5% level: for the corre-
lation between the band width and T/P,ν
eff
is 93 (N = 115), and thep-value is 3× 10
−3
;
for the correlation between the band width and winter NAO, ν
eff
is 95 (N = 126), and
thep-value is 4× 10
−2
, just under the 5% threshold. While we conclude that the original
interpretation is supported by our analysis, the authors reached this conclusion thanks to
error compensation, potentially undermining their point.
We note, however, that the decrease of DOF due to smoothing was considered when
this reconstruction was used for studying the long-term variability of the NAO in the
high-profile study of Trouet et al. (2009).
1.3 Challenge #2: Test multiplicity
1.3.1 Theory
Whenassessingcorrelationswithafield, multipletestsarecarriedoutatdifferentloca-
tions simultaneously. For example, if a correlation test is applied to 1000 locations with a
significance level of 5%, one would expect about 50 hypotheses to be falsely rejected (in
this case, 50 correlations would be deemed significant when in fact they are not), which
is unacceptably high. The fundamental problem is that the test level α (the probability
of false positives, i.e. the probability of falsely rejecting the null hypothesis of zero corre-
lation) applies to pairwise comparisons, but not to multiple such comparisons. The finer
the grid, the more such tests are simultaneously carried out, and the higher the risk of
23
identifying spurious correlations as “significant" when in fact they are not. This test mul-
tiplicity problem is well known in the statistical literature and solutions exist (Benjamini
and Hochberg, 1995, Storey, 2002).
In particular, the False Discovery Rate (FDR) procedure (Benjamini and Hochberg,
1995)hasbeenwidelyapplied(> 34, 000GoogleScholarcitationsatthetimeofwriting)to
controltheproportionoffalselyrejectednullhypothesesoutofallrejectednullhypotheses,
thus offering a level of scientific rigor that naive correlation testing does not afford. The
term “false discovery" here is synonymous with “falsely identified significant correlations".
Insteadofrestrictingtheoccurrenceoffalselyrejectednullhypotheses, theFDRprocedure
allows and controls the proportion of erroneously rejected null hypotheses. Setting q =
5% in the FDR procedure, guarantees that 5% or fewer of the locations where the null
hypothesis is rejected are false detections on average, so the proportion of false rejections
is controlled. The FDR procedure of Benjamini and Hochberg (1995) proceeds as follows:
1. Carry out the test (calculate p-values) at all m locations;
2. Rank the p-values p
(i)
in increasing order: p
(1)
≤p
(2)
≤...≤p
(m)
;
3. Define k as the largest i for which p
(i)
≤q
i
m
;
4. Reject hypotheses at locations i = 1, 2,...,k.
In this procedure, the FDR treats locations with low p-values as most significant,
ranking p-values from high to low. If the largest p-value is less than its corresponding
threshold, then all tests are regarded as significant. If the largest p-value is greater than
its corresponding threshold, then it is compared to the second largestp-value with a more
restricted threshold – and so on. These thresholds guarantee that the expected rate of
falsely positive hypotheses are smaller than α. Through this procedure, the number of
rejected hypotheses isk, and the expected number of falsely rejected hypotheses is smaller
than mp
(k)
, such that the fraction of falsely rejected hypotheses is smaller than mp
(k)
/k.
The third step in the FDR procedure limits the fraction of falsely rejected hypotheses to
be smaller than q, which is the threshold for the fraction of falsely rejected hypotheses.
Thus the FDR procedure ensures the fraction of erroneously detected significant locations
is smaller than a specified threshold.
24
0.0 0.2 0.4 0.6 0.8 1.0
index/m
0.0
0.2
0.4
0.6
0.8
1.0
p-value
ordered p-values
p-values rejected by the classical procedure but not by the FDR
p-values rejected by both procedures
Figure 1.2: Illustration of the traditional significance test procedure and the
FDR procedure on an illustrative example, with q = α = 20%. p-values at each
grid point (p
(i)
) are ranked in increasing order, plotted against i/m, wherem is the total
number of grid points. The blue dashed line is the traditionalα threshold for thep-value.
Green dots indicate they are significant by both traditional and FDR procedure, and blue
dots indicate they are only significant by the traditional procedure.
This procedure is graphically illustrated by Fig. 1.2, adapted from Ventura et al.
(2004). In order to clearly show the difference between the traditional and the FDR sig-
nificance test procedure, q and α were both set to 20%. At each location (p
(i)
), p-values
were ranked in increasing order and plotted against i/m, where m is the total number of
locations. The blue dashed line is the traditional significance threshold. All dots below
the blue dashed line are significant by the traditional procedure. With the FDR proce-
dure, only those p-values under the red line (green dots) are significant. Therefore, some
25
p-values are deemed significant by the traditional procedure but not by the FDR proce-
dure. Applying this methodology lowers the likelihood of identifying spurious correlations
as significant, hence making correlation tests more stringent. We should note that the
absence of significant correlations in one test does not imply that the correlation is inex-
istent – only that the data do not provide enough evidence to reject the null hypothesis
of zero correlation.
1.3.2 Application
Ventura et al. (2004) proposed a simple implementation of the FDR procedure for
climate fields, and showed that its assumptions are robust to the spatial correlation levels
typical of climate fields. Here we use the code of Benjamini and Hochberg (1995), which
is equivalent. To show how to apply this procedure to field correlations, we use the
study of Zhu et al. (2012) as an example. The authors generated a cellulose δ
18
O record
of Merkus pines for the past 140 years in Kririrom National Park, southern Cambodia
(KRPM15B, 11.29
◦
N; 104.25
◦
E; 675 m). This record was dated by ring-counting. The
authors assert that this celluloseδ
18
O record is dominantly controlled by convection over
the Indo-Pacific warm pool (IPWP), an interpretation buttressed by high correlations of
cellulose δ
18
O with instrumental precipitation and outgoing longwave radiation over the
IPWP (Fig. 7 in Zhu et al. (2012)). This explanation is reasonable because the cellulose
δ
18
O is mainly controlled by the δ
18
O in precipitation during the rainy season, which
is often depleted when the rainfall increases (the so-called “amount effect”; Dansgaard,
1964). During El Niño events, the precipitation in Southeast Asia is usually suppressed,
which would lead to higher δ
18
O values.
Here we consider the false discovery rate in this spatial correlation (Fig. 7 in Zhu
et al. (2012)). When the FDR is considered, none of the correlations are found significant,
even at the relatively permissive 10% level chosen by authors (Fig. 1.3). This calls into
question the proposed relationship between cellulose δ
18
O and interannual changes in
tropical convection at this site. However, we did find that the correlation between the
cellulose δ
18
O and (unsmoothed) NINO4 index significant at the 5% level (r = 0.45, p-
value = 2.3× 10
−6
), considering autocorrelation as above. This indicates that cellulose
δ
18
O, or at least rainfall δ
18
O at the site, may be connected with large scale changes due
26
to the El Niño-Southern Oscillation. However, the explanation that local precipitation
changes during El Niño events dominate the celluloseδ
18
O may need further examination,
for instance via forward modeling (Evans, 2007).
Figure 1.3: Spatial correlation of October Kirirom cellulose δ
18
O values with
the October-November-December mean of (a) CMAP precipitation, and (b)
NOAA interpolated OLR. Black dots indicate that the correlation does not pass the
significance test at the 90% level
27
1.4 Challenge #3: Age uncertainties
1.4.1 Theory
Age uncertainties have long been known to affect inferences made from paleoclimate
data (e.g. Clark and Thompson, 1979), including correlation analysis (see Mudelsee, 2001,
Rehfeld and Kurths, 2014, for recent examples). Indeed, Wunsch (2003) showed that even
two randomly generated, unrelated time series could appear to correlate well with each
other by adjusting the chronology within age uncertainties.
Thus, quantifying age uncertainties is critical to the analysis of paleoclimate records.
Many methods have been developed to do so (e.g. Haslett and Parnell, 2008, Blaauw and
Christen, 2011). Additionally, various methods have been devised to deal with correlation
(or covariance) under age uncertainties (Haam and Huybers, 2010).
In the next section, we illustrate how age uncertainties may influence correlation anal-
ysis, and how they may compound the effects of autocorrelation and test multiplicity by
drawing an example from a high-profile publication (McCabe-Glynn et al., 2013).
1.4.2 Compound challenges in one: the case of Crystal Cave
Recently, McCabe-Glynn et al. (2013) applied time-based calibrations to δ
18
O data
from a stalagmite from Crystal Cave, southern California. They interpreted the record
as a proxy for sea surface temperature (SST) in the Kuroshio Extension region, arguing
that warm SST in the area may generate southwesterly wind anomalies over southern
California, bringing isotopically-enriched moisture to the cave. They also found a strong
22-year periodicity, which they linked to solar cycles, and found that some southwestern
North American droughts coincided with episodes of warm Kuroshio Extension SSTs, as
reconstructed from the Crystal Cave stalagmite. Some of these conclusions are based on
a correlation analysis that encounters the three challenges of interest in this study: serial
correlation, test multiplicity and age uncertainties. Here we will show one way to properly
address these challenges.
28
1.4.3 Effect of serial correlation
Wereusetheδ
18
OdatafromstalagmiteCRC-3, collectedfromCrystalCaveinSequoia
NationalPark, California(36.59
◦
N; 118.82
◦
W; 1,386m), whichwasinterpolatedatannual
scaleandarchivedonline. Asintheoriginalstudy, therecordiscorrelatedtoSSTanomaly
data from the Kaplan SST v2 dataset (Kaplan et al., 1998) (1856-2007).
The correlation is shown in Fig. 1.4b, to be compared with Supplementary Fig. S6a in
McCabe-Glynn et al. (2013). Because both theδ
18
O series and SST field are intrinsically
autocorrelated (the autocorrelation of δ
18
O series is 0.95, and the autocorrelation of SST
is 0.11-0.76, depending on location), the effective number of degrees of freedom is much
lower than its theoretical value (N− 2 = 149). Indeedν
eff
is less than 60 in the Kuroshio
Extension region (Fig. 1.4a). Hence, when this effect is considered (Fig. 1.4c), far fewer
correlations pass the significance test (Fig. 1.4b,c). While McCabe-Glynn et al. (2013)
had considered the effect of serial correlation using the method of Macias-Fauria et al.
(2012), they did not graphically represent these results, giving little indication of where
the relationship might be reliable. Nonetheless, our result is consistent with theirs, in
that correlations over the Kuroshio Extension region pass the significance test with both
approaches.
ftp://ftp.ncdc.noaa.gov/pub/data/paleo/speleothem/northamerica/usa/california/
crystal2013.txt
29
(a) Effective number of degrees of freedom
15 30 45 60 75 90 105 120 135
(b) Correlation (no FDR, ignoring autocorrelation)
0.9 0.6 0.3 0.0 0.3 0.6 0.9
(c) Correlation (no FDR, considering autocorrelation)
0.9 0.6 0.3 0.0 0.3 0.6 0.9
(d) Correlation (FDR, ignoring autocorrelation)
0.9 0.6 0.3 0.0 0.3 0.6 0.9
(e) Correlation (FDR, considering autocorrelation)
0.9 0.6 0.3 0.0 0.3 0.6 0.9
Figure 1.4: Influence of degrees of freedom and test multiplicity problem on
correlations. (a) The effective number of degrees of freedom of the correlation between
Crystal Caveδ
18
O and SST from 1856-2007; The correlation between these series, consid-
ering the test multiplicity problem (False Discovery Rate) (d, e), or not (b, c); considering
autocorrelation (c,e) or not (b,d). Black dots indicate that the correlation does not pass
the significance test at the 5% level.
30
1.4.4 Effect of test multiplicity
Since the correlation between the δ
18
O record and SST is also a field correlation, we
need to consider test multiplicity. The result is shown in Fig. 1.4d,e. If autocorrelation
is ignored, the FDR procedure results in fewer correlations passing the significance test
(Fig. 1.4d vs Fig. 1.4b). When both autocorrelation and multiplicity are considered, no
correlationpassesthesignificancetest(Fig.1.4e). Thisresultsuggeststhatthecorrelation
between δ
18
O record and instrumental SST may not be used as a basis for the record’s
interpretation.
1.4.5 Effect of age uncertainties
Another problem compounding serial correlation stems from the fact that the age
model for the speleothem carries uncertainties of years to decades (Cheng et al., 2013),
and these uncertainties may propagate to other inferences made from the proxy. The age
uncertainties were quantified in McCabe-Glynn et al. (2013) using the StalAge algorithm
(Scholz and Hoffmann, 2011). While the StalAge code exports 95%-confidence limits for
the corresponding ages, it does not export possible age ensembles, which are essential for
propagating uncertainty to other inferences. Here we leverage the power of ensembles to
quantifying age uncertainties in correlation analysis.
Age model
We chose to model age uncertainties using Bchron (Haslett and Parnell, 2008), a
Bayesian probability model allowing for random variations in accumulation rate between
tie points. Bchron is capable of dealing with outliers and hiatuses (Parnell et al., 2011),
and naturally produces more hiatuses than StalAge. Bchron proceeds as follows: for
each dated sample, the algorithm randomly selects a calendar date consistent with the
age information (measured age and age error) and monotonicity (deeper sample with
older age). It then inserts several points (at depths we want to know ages) between dated
samplesconsistentwithmonotonicity, andthenlinearlyinterpolatesbetweenthosepoints.
Finally, it repeats this process many times until enough realizations fit the measured ages.
The main advantage of Bchron here is the ability to extract an ensemble of age models
31
all consistent with the posterior distribution of ages. For a review of different approaches
to age modeling, see Scholz et al. (2012).
0 20 40 60 80 100 120
Depth (mm)
800
1000
1200
1400
1600
1800
2000
Y ear (CE)
Crystal BChron sampling paths, 10 dates
95% CI
median
U/Th dates
sample paths
McCabe-Glynn et al., 2013
Figure 1.5: The age modeling results of the Crystal cave δ
18
O record using a
Bchron age model. The gray area is the 95% confidence interval of the age at each
depth. The red lines show 10 random paths out of the 1,000 age models generated, and
the blue curve shows the StalAge-generated model used by McCabe-Glynn et al. (2013)
The age models are compared in Fig. 1.5, and one can see that they are quite close.
The choice of age model (blue vs black curve) introduces relatively small differences in the
median age model, but the inclusion of a full ensemble of 1000 plausible age realizations
makes a great difference indeed. Each of these age realizations corresponds to a different
δ
18
O time series: Fig. 1.6 (left) shows three of them, corresponding to the lower (2.5%),
median (50%) and upper (97.5%) quantiles of the age distribution. One can see that
many of the major features of the δ
18
O timeseries can shift by about 50 years, making a
correlation to instrumental data fraught with uncertainty.
32
800 1000 1200 1400 1600 1800 2000 Year (CE)
−11 . 0 −10 . 5 −10 . 0 −9 . 5 −9 . 0 −8 . 5 −8 . 0 −7 . 5 δ
18 O Crystal δ
18 O Bchron ensemble
97.5% quantile
median
2.5% quantile
1860 1880 1900 1920 1940 1960 1980 2000 Year(CE)
−11 . 0 −10 . 5 −10 . 0 −9 . 5 −9 . 0 −8 . 5 −8 . 0 −7 . 5 δ
18 O δ
18 O record from McCabe-Glynn et al. 2013
Median of δ
18 O record from the Bchron model
800 1000 1200 1400 1600 1800 2000 Year (CE)
−11 . 0 −10 . 5 −10 . 0 −9 . 5 −9 . 0 −8 . 5 −8 . 0 −7 . 5 δ
18 O Crystal δ
18 O Bchron ensemble
97.5% quantile
median
2.5% quantile
1860 1880 1900 1920 1940 1960 1980 2000 Year(CE)
−11 . 0 −10 . 5 −10 . 0 −9 . 5 −9 . 0 −8 . 5 −8 . 0 −7 . 5 δ
18 O δ
18 O record from McCabe-Glynn et al. 2013
Median of δ
18 O record from the Bchron model
Figure 1.6: Crystal cave δ
18
O time series with uncertainties. Top panel: The
median, 97.5% quantile and 2.5% quantile of the ensemble of 1000 Crystal δ
18
O time
generated from the Bchron model. Bottom panel: The time series of δ
18
O record from
McCabe-Glynn et al. (2013) (blue) and the median ofδ
18
O record from the Bchron model
(black).
33
This may be seen in more detail on Fig. 1.5. While the median age models from the
two methods (blue and black curves) do appear quite close, there are large offsets between
δ
18
O timeseries (Fig. 1.6, right), especially before AD 1960. For instance, the large peak
ca 1890 in McCabe-Glynn et al. (2013), is centered around 1900 in the median Bchron age
model. Such differences are especially significant if one tries to correlate them to other
climatic timeseries, as we now do.
Correlations considering age uncertainty and autocorrelation
We assess the impact of chronological uncertainties by repeating the previous analysis
on each of the 1,000 realizations of the δ
18
O time series, similarly interpolated to an
annual scale to facilitate correlations to SST. In the following analysis, we will also take
autocorrelation into account.
34
(a) median
−0.9 −0.6 −0.3 0.0 0.3 0.6 0.9
(b) IQR
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
(c) 2.5% quantile
−1.0 −0.8 −0.6 −0.4 −0.2 0.0
(d) 97.5% quantile
0.0 0.2 0.4 0.6 0.8 1.0
Figure 1.7: Correlations considering age uncertainty. The median (a), interquartile
range(b),the2.5%(c)andthe97.5%(d)percentileofthecorrelationbetweentheδ
18
Oage
ensemble and SST. Black dots indicate that the correlation does not pass the significance
test at the 5% level, accounting for serial correlation. Note that the interquartile range
here is a measure of distributional spread, and has no measure of significance attached to
it.
Using the full age ensemble, one obtains 1000 δ
18
O-SST correlations for each grid
point, from which one may infer an empirical distribution, whose median, 2.5% and 97.5%
percentiles, and interquartile range (IQR) are reported in Fig. 1.7. The median is the
aspect of the distribution that most studies use exclusively (for Gaussians, the median,
mean and mode coincide). Due to the construction of our ensemble, the 2.5% quantile
gathers some of the strongest negative correlations, and the 97.5% quantile gathers some
35
of the strongest positive correlations. The IQR measures the spread between the 25% and
75% quantiles (the width of the distribution), and is therefore an indication of the spread
of correlations due to age uncertainties alone.
The pattern of correlations for the median age model is similar to Fig. 1.4b, but the
absolute values of correlations are much smaller, and the positive center in the North
Pacific is shifted southward. Also, none of the median correlations pass the significance
test at the 5% level. Since McCabe-Glynn et al. (2013) use the StalAge model, this
suggests that the correlation between the δ
18
O record and instrumental SST may be
dependent on the age model. However, Fig. 1.5 clearly shows that the StalAge model is
within the 95% confidence bounds of the Bchron age model, which underlines that age
uncertainties are generally quite large compared to the timescale of variability in the SST
record, and should therefore be accounted for in the analysis of correlations.
The pattern of the range of the correlation (IQR, Fig. 1.7b) is quite similar to Fig.
1.4b, whichindicatesthattheregionsofhighestcorrelationinMcCabe-Glynnetal.(2013)
correspond to the regions of largest uncertainties in the age models. The 97.5% quantile
and the 2.5% quantiles (Fig. 1.7c,d) also correspond to the regions of positive/negative
correlation in Fig. 1.4b, and the correlation in some regions passes the significance test.
However, the corresponding pattern differs from age ensemble to age ensemble, and from
the published result, indicating a lack of robust relationship on which to build a reliable
interpretation.
1.4.6 Correlations considering all three challenges
Adding to these challenges, when the multiple-test problem is considered, none of the
correlations for the 97.5% or 2.5% quantiles of the age ensemble pass the significance test
(Fig. 1.8). This is because, while somep-values fall below the nominal threshold (Fig. 1.9,
blue dashed lines), none drop under the FDR threshold (Fig. 1.9, red line), implying that
these correlations are not significant at the 5% level. Thus the interpretation of Crystal
Cave δ
18
O as a proxy for SST in the Kuroshio Extension (or SST anywhere) should be
revisited.
36
(a) median
0.9 0.6 0.3 0.0 0.3 0.6 0.9
(b) IQR
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
(c) 2.5% quantile
1.0 0.8 0.6 0.4 0.2 0.0
(d) 97.5% quantile
0.0 0.2 0.4 0.6 0.8 1.0
Figure 1.8: Same as Fig. 1.7, but testing for field correlations while controlling for the
False Discovery Rate.
37
(a) (b)
Figure 1.9: Illustration of the FDR procedure on the 2.5% quantile (a) and
the 97.5% quantile (b) correlations shown in Fig. 1.8, using the false discovery
rate q = 5% (red line). p-values at each grid point (p
(i)
) are ranked in increasing order,
plotted against i/m, where m is the total number of grid points. The blue dashed line is
the traditional 5% threshold for the p-value. Dots with p-values below this threshold are
shown in blue. In this example, many dots fall below the nominal threshold (5%), but
none fall below the red line, which means that they are not significant according to the
FDR-controlling procedure.
38
1.4.7 Time-uncertain spectral analysis
Finally, we assess the influence of chronological uncertainties on inferences made in
the spectral domain. We use the multi-taper method (Thomson, 1982), which achieves an
optimal tradeoff between leakage and resolution. We use a half-bandwidth parameter of 4,
which favors statistical significance over resolution, and is therefore the most conservative
setting. Themulti-taperspectraoftheδ
18
Orecord(855-2007AD)fromthetwoagemodels
are presented in Fig. 1.10. It shows that the δ
18
O record derived from the median of the
Bchron ensemble has a dominant period ca 18 years while the published data exhibit a
dominant period of 21 years. This is clearly within age uncertainties, despite McCabe-
Glynn et al. (2013) having used REDFIT, a variant of the LombâĂŞScargle periodogram
(Schulz and Mudelsee, 2002). Therefore, the multi-taper spectral method is suitable for
the comparison. Considering that much of the uncertainty lies above the red line and gray
area in Fig. 1.10, it is hard to distinguish significant periodicities from red noise. This
suggests that there may not be any notable harmonic cycles in the δ
18
O record at any
scale less than 1000 years. Thus more evidence would be needed to draw a connection to
the 22-year Hale solar cycle, as done in McCabe-Glynn et al. (2013).
39
200 100 50 20 10 8 6 4 2
Frequency (Years)
10
-4
10
-3
10
-2
10
-1
10
0
10
1
10
2
PSD
δ
18
O, median from Bchron
AR(1), 97.5% quantile
δ
18
O from McCabe et al. (2013)
Figure 1.10: The MTM-estimated spectra of the δ
18
O record from the Bchron
age model (blue line and gray shaded area) compared to the spectrum of the published
record, together with a simulated AR(1) benchmark.
We showed how to address the three main challenges of correlation analysis and how
these challenges weaken the conclusions of the original study. The example illustrates how
each challenge in isolation would be enough to question the main conclusions, and their
combination even more so. In closing, we note that although we do not find significant
correlations between the Crystal Cave record and sea-surface temperature, we cannot rule
out the existence of such correlations. The short calibration period and relatively large
age uncertainties simply preclude the establishment of significant correlations with the
current datasets.
Finally, we stress that the interpretation of the oxygen isotope composition of
speleothem calcite is complex (LeGrande and Schmidt, 2009), and that other factors
may influence the δ
18
O record in Crystal Cave. The variation of δ
18
O of precipitation in
California can be affected by changes in condensation height (Buenning et al., 2013), and
δ
18
O in precipitation is known to be influenced by source-water composition due to shifts
40
in storm-track location (Oster et al., 2015). Thus, the published interpretation of Crystal
Cave δ
18
O needs to be further investigated.
1.5 Discussion
In a provocative paper, Ioannidis (2005) concluded that “most published research find-
ings are false”. Central to this problem is a widespread tendency to hunt for statistical
effects (often, correlations) that pass a significance threshold of 95% (i.e. a 5% probabil-
ity of false positive), a practice often dubbed “p-hacking". While the paper focused on
biomedical research, some of its conclusions transfer to paleoclimatology, and the Earth
sciences at large. In particular, the existence of important challenges in correlation anal-
ysis (autocorrelation, false-discovery rate and age uncertainties) should be recognized by
all practicing paleoclimatologists. In this article we illustrate these challenges using three
examples, showing how published interpretations may be changed by one challenge alone,
or by a combination of them.
We beganwithshowinghowautocorrelationandnon-normalityaffectcorrelationanal-
ysis by using the example from Proctor et al. (2000), who reconstructed the NAO index
using layer thickness in a stalagmite from Scotland. We found that properly taking those
into account did not change the interpretation, though the latter would have been more
robust had the authors taken these challenges into consideration. Autocorrelation is com-
monplace in paleoclimate proxies because many processes (e.g. bioturbation in sediment
cores, groundwater mixing in speleothems, and firn diffusion in ice cores) smooth out cli-
mate signals. Autocorrelation should always be a concern unless it can be demonstrated
otherwise.
The test multiplicity problem is another serious challenge of correlation analysis, as
we showed in the example of Zhu et al. (2012). The spatial correlations between their
cellulose δ
18
O record and instrumental precipitation/outgoing longwave radiation over
the Indo-Pacific Warm Pool in this paper are not significant after considering the false
discovery rate. However, we find correlations with the NINO4 index significant, with the
index explaining about 25% of the record’s variance.
41
Age uncertainties also challenge the robustness of correlation analysis and we show
that they may also combine with other challenges in the example of McCabe-Glynn et al.
(2013), who claimed, on the basis of correlations to the SST field, to identify a relation-
ship between the δ
18
O record of a speleothem from southern California and the SST in
the Kuroshio Extension region. We show, by considering all three challenges, that no
correlation survives the test.
These three examples lead us to draw attention to the importance of:
• using established statistical procedures to guard against the misleading impacts of
spurious correlations. Several books (Wilks, 2011, Mudelsee, 2013, Emile-Geay,
2016) address these challenges in more detail.
• using the rich output of age modeling software (not just the median age) to appraise
the effect of age uncertainties on a study’s conclusions.
• establishing a mechanistic understanding for proxy signals, and only relying on
statistical approaches when there are sufficient numbers of degrees of freedom to
unequivocally reject chance correlations.
While we do not dispute that the records scrutinized here contain potentially valuable
climatic information, our main message is that it is often impossible to establish so by a
purely statistical approach. Instead, we encourage detailed process studies to elucidate
the climatic and/or hydrological controls on δ
18
O, or other proxies, in various archives.
For speleothems, possible strategies involve the forward modeling of cave processes
(such as Baker et al., 2012, Partin et al., 2013, and others), and/or cave instrumentation
(such as Spötl et al., 2005, Partin et al., 2012, and others) to better ascertain the pro-
cesses that control the recorded oxygen isotope signal. For many speleothem studies, age
modeling considerations will be of secondary importance: in studies of glacial-interglacial
cycles, for instance, age offsets of a few decades are immaterial. However, the interpreta-
tion ofδ
18
O in climate proxies (whether in terms of rainfall, temperature, or other factors)
is usually complex, and therefore should be backed by isotope-enabled models (such as
LeGrande and Schmidt, 2009, Pausata et al., 2011, and others). We note that proxy
system modeling is a burgeoning field with applications to all paleoclimate archives, not
just speleothems (e.g. Schmidt, 1999, Evans, 2007, Dee et al., 2015).
42
Forthedatacollectingandmeasuringstage, highresolutionsamplingissuggested(von
Gunten et al. (2012) suggested collecting 80-100 data points over the calibration period)
for time-based calibrations to achieve a sufficient effective sample size. If smoothing scale
is known, the record with smaller smoothing scale should be used in the time-based cali-
bration. Alsoweshouldnotethatageuncertaintiessetthelimitoftheusageofproxydata.
For example, proxy data with decadal age uncertainties cannot be used in interannual-
scale research questions, but may be used in centennial-scale research questions (Birks
et al., 2012).
Our goal in presenting these results is not to indict a particular set of authors, as
the unsophisticated data-analytical practices of the original studies are unfortunately
rather common in the paleoclimate literature. Instead, we wish to draw attention to
under-appreciated statistical issues, with the hope of lessening the occurrences of proxy
interpretations based on spurious correlations, and to improve the robustness of future
paleoclimate studies. In this spirit, we are making the Python code associated with
this study freely available at https://github.com/ClimateTools/Correlation_EPSL
in order to disseminate best practices.
Bibliography
Baker, A., Bradley, C., Phipps, S., Fischer, M., Fairchild, I., Fuller, L., Spötl, C., and
Azcurra, C. (2012). Millennial-length forward models and pseudoproxies of stalagmite
δ
18
O: an example from NW Scotland. Climate of the Past, 8(4):1153–1167.
Benjamini, Y.andHochberg, Y.(1995). ControllingtheFalseDiscoveryRate: APractical
and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society.
Series B (Methodological), 57(1):289–300.
Birks, J. B., Lotter, A. F., Juggins, S., and Smol, J. P. (2012). Tracking environmen-
tal change using lake sediments: data handling and numerical techniques, volume 5.
Springer Netherlands.
Blaauw, M. and Christen, J. A. (2011). Flexible paleoclimate age-depth models using an
autoregressive gamma process. Bayesian Analysis, 6(3):457–474.
43
Black, B. A., Griffin, D., van der Sleen, P., Wanamaker, A. D., Speer, J. H., Frank, D. C.,
Stahle, D. W., Pederson, N., Copenheaver, C. A., Trouet, V., et al. (2016). The value
of crossdating to retain high-frequency variability, climate signals, and extreme events
in environmental proxies. Global Change Biology.
Buenning,N.H.,Stott,L.,Kanner,L.,andYoshimura,K.(2013). Diagnosingatmospheric
influences on the interannual
18
O/
16
O variations in Western US precipitation. Water,
5(3):1116–1140.
Cheng, H., Edwards, R. L., Shen, C.-C., Polyak, V. J., Asmerom, Y., Woodhead, J.,
Hellstrom, J., Wang, Y., Kong, X., Spötl, C., et al. (2013). Improvements in
230
Th
dating,
230
Th and
234
U half-life values, and U–Th isotopic measurements by multi-
collector inductively coupled plasma mass spectrometry. Earth and Planetary Science
Letters, 371:82–91.
Clark, R. and Thompson, R. (1979). A new statistical approach to the alignment of time
series. Geophysical Journal International, 58(3):593–607.
Crowley, T. J. (1999). Correlating high-frequency climate variations. Paleoceanography,
14(3):271–272.
Dansgaard, W. (1964). Stable isotopes in precipitation. Tellus, 16(4):436–468.
Dawdy, D. and Matalas, N. (1964). Statistical and probability analysis of hydrologic data,
part III: Analysis of variance, covariance and time series. McGraw-Hill.
Dee, S., Emile-Geay, J., Evans, M. N., Allam, A., Steig, E. J., and Thompson, D. M.
(2015). PRYSM: An open-source framework for PRoxY System Modeling, with appli-
cations to oxygen-isotope systems. Journal of Advances in Modeling Earth Systems,
7(3):1220–1247.
Emile-Geay, J. (2016). Data Analysis in the Earth & Environmental Sciences. FigShare,
second edition.
Emile-Geay,J.andTingley,M.(2016). Inferringclimatevariabilityfromnonlinearproxies:
application to palaeo-ENSO studies. Climate of the Past, 12(1):31–50.
44
Evans, M. N. (2007). Toward forward modeling for paleoclimatic proxy signal calibration:
A case study with oxygen isotopic composition of tropical woods. Geochem. Geophys.
Geosyst, 8:Q07008.
Ghil, M., Allen, M., Dettinger, M., Ide, K., Kondrashov, D., Mann, M., Robertson, A. W.,
Saunders, A., Tian, Y., Varadi, F., etal.(2002). Advancedspectralmethodsforclimatic
time series. Reviews of Geophysics, 40(1).
Haam, E. and Huybers, P. (2010). A test for the presence of covariance between time-
uncertain series of data with application to the Dongge Cave speleothem and atmo-
spheric radiocarbon records. Paleoceanography, 25(2).
Haslett, J. and Parnell, A. (2008). A simple monotone process with application to
radiocarbon-dated depth chronologies. J. R. Stat. Soc. Ser. C-Appl. Stat., 57:399–418.
Ioannidis, J.(2005). Whymostpublishedresearchfindingsarefalse. PLoS Med, 2(8):e124.
Jones, P., Briffa, K., Osborn, T., Lough, J., van Ommen, T., Vinther, B., Luterbacher,
J., Wahl, E., Zwiers, F., Mann, M., Schmidt, G., Ammann, C., Buckley, B., Cobb,
K., Esper, J., Goosse, H., Graham, N., Jansen, E., Kiefer, T., Kull, C., Kuttel, M.,
Mosley-Thompson, E., Overpeck, J., Riedwyl, N., Schulz, M., Tudhope, A., Villalba,
R., Wanner, H., Wolff, E., and Xoplaki, E. (2009). High-resolution palaeoclimatology
of the last millennium: a review of current status and future prospects. The Holocene,
19(1):3–49.
Kaplan, A., Cane, M. A., Kushnir, Y., Clement, A. C., Blumenthal, M. B., and
Rajagopalan, B. (1998). Analyses of global sea surface temperature 1856-1991. J.
Geophys. Res.-Oceans, 103(C9):18567–18589.
Khider, D., Huerta, G., Jackson, C., Stott, L., and Emile-Geay, J. (2015). A Bayesian,
multivariate calibration for Globigerinoides ruber Mg/Ca. Geochemistry, Geophysics,
Geosystems, 16(9):2916–2932.
LeGrande, A. and Schmidt, G. (2009). Sources of Holocene variability of oxygen isotopes
in paleoclimate archives. Clim. Past, 5(3):441–455.
45
Macias-Fauria, M., Grinsted, A., Helama, S., and Holopainen, J. (2012). Persistence mat-
ters: estimation of the statistical significance of paleoclimatic reconstruction statistics
from autocorrelated time series. Dendrochronologia, 30(2):179–187.
McCabe-Glynn, S., Johnson, K. R., Strong, C., Berkelhammer, M., Sinha, A., Cheng, H.,
and Edwards, R. L. (2013). Variable North Pacific influence on drought in southwestern
North America since AD 854. Nat. Geosci., 6(8):617–621.
Mudelsee, M. (2001). The phase relations among atmospheric CO
2
content, temperature
and global ice volume over the past 420 ka. Quaternary Science Reviews, 20(4):583–589.
Mudelsee, M. (2013). Climate time series analysis. Springer.
Oster, J. L., Montañez, I. P., Santare, L. R., Sharp, W. D., Wong, C., and Cooper, K. M.
(2015). Stalagmite records of hydroclimate in central California during termination 1.
Quaternary Science Reviews, 127:199 – 214.
Parnell, A. C., Buck, C. E., and Doan, T. K. (2011). A review of statistical chronology
models for high-resolution, proxy-based holocene palaeoenvironmental reconstruction.
Quaternary Science Reviews, 30(21):2948–2960.
Partin, J., Quinn, T., Shen, C.-C., Emile-Geay, J., Taylor, F., Maupin, C., Lin, K.,
Jackson, C., Banner, J., Sinclair, D., and Huh, C.-A. (2013). Multidecadal rainfall
variability in South Pacific Convergence Zone as revealed by stalagmite geochemistry.
Geology.
Partin, J. W., Jenson, J. W., Banner, J. L., Quinn, T. M., Taylor, F. W., Sinclair, D.,
Hardt, B., Lander, M. A., Bell, T., Miklavič, B., et al. (2012). Relationship between
modernrainfallvariability, cavedripwater, andstalagmitegeochemistryinGuam, USA.
Geochemistry, Geophysics, Geosystems, 13(3).
Pausata, F. S., Battisti, D. S., Nisancioglu, K. H., and Bitz, C. M. (2011). Chinese sta-
lagmiteδ
18
O controlled by changes in the Indian monsoon during a simulated Heinrich
event. Nature Geoscience, 4(7):474–480.
46
Proctor, C., Baker, A., Barnes, W., and Gilmour, M. (2000). A thousand year speleothem
proxy record of North Atlantic climate from Scotland. Climate Dynamics, 16(10-
11):815–820.
Rehfeld, K. and Kurths, J. (2014). Similarity estimators for irregular and age-uncertain
time series. Climate of the Past, 10(1):107–122.
Schmidt, G. A. (1999). Forward modeling of carbonate proxy data from planktonic
foraminifera using oxygen isotope tracers in a global ocean model. Paleoceanography,
14(4):482–497.
Scholz, D. and Hoffmann, D. L. (2011). StalAge–An algorithm designed for construction
of speleothem age models. Quaternary Geochronology, 6(3):369–382.
Scholz, D., Hoffmann, D. L., Hellstrom, J., and Ramsey, C. B. (2012). A comparison of
different methods for speleothem age modelling. Quaternary Geochronology, 14:94–104.
Schulz, M. and Mudelsee, M. (2002). REDFIT: estimating red-noise spectra directly from
unevenly spaced paleoclimatic time series. Computers & Geosciences, 28(3):421–426.
Spötl, C., Fairchild, I. J., and Tooth, A. F. (2005). Cave air control on dripwater geo-
chemistry, Obir Caves (Austria): Implications for speleothem deposition in dynamically
ventilated caves. Geochimica et Cosmochimica Acta, 69(10):2451–2468.
Storey, J. D. (2002). A direct approach to false discovery rates. Journal of the Royal
Statistical Society: Series B (Statistical Methodology), 64(3):479–498.
Thomson, D. J. (1982). Spectrum estimation and harmonic-analysis. Proc. IEEE,
70(9):1055–1096.
Tierney, J. E. and Tingley, M. P. (2014). A Bayesian, spatially-varying calibration model
for the TEX86 proxy. Geochimica et Cosmochimica Acta, 127:83–106.
Tingley, M. P., Craigmile, P. F., Haran, M., Li, B., Mannshardt, E., and Rajaratnam, B.
(2012). Piecing together the past: statistical insights into paleoclimatic reconstructions.
Quaternary Science Reviews, 35(0):1 – 22.
47
Trouet, V., Esper, J., Graham, N. E., Baker, A., Scourse, J. D., and Frank, D. C. (2009).
Persistent positive North Atlantic Oscillation mode dominated the medieval climate
anomaly. Science, 324(5923):78–80.
Ventura, V., Paciorek, C. J., and Risbey, J. S. (2004). Controlling the Proportion of
FalselyRejectedHypotheseswhenConductingMultipleTestswithClimatologicalData.
Journal of Climate, 17(22):4343–4356.
von Gunten, L., Grosjean, M., Kamenik, C., Fujak, M., and Urrutia, R. (2012). Calibrat-
ing biogeochemical and physical climate proxies from non-varved lake sediments with
meteorological data: methods and case studies. Journal of Paleolimnology, 47(4):583–
600.
Wilks, D. S. (2011). Statistical Methods in the Atmospheric Sciences: an Introduction.
Academic Press, San Diego.
Wunsch, C. (2003). Greenland—Antarctic phase relations and millennial time-scale cli-
matefluctuationsintheGreenlandice-cores. Quaternary Science Reviews, 22(15):1631–
1646.
Yule, G. U. (1926). Why do we sometimes get nonsense-correlations between time-series
– a study in sampling and the nature of time-series. Journal of the Royal Statistical
Society, 89(1):1–63.
Zhu, M., Stott, L., Buckley, B., Yoshimura, K., and Ra, K. (2012). Indo-Pacific Warm
Pool convection and ENSO since 1867 derived from Cambodian pine tree cellulose
oxygen isotopes. Journal of Geophysical Research: Atmospheres, 117(D11).
48
Chapter 2
Impact of convective activity on precipita-
tion δ
18
O in isotope-enabled general circula-
tion models
Abstract
Theδ
18
O signal preserved in paleo-archives is widely used to reconstruct past climate
conditions. In many speleothems, this signal is classically interpreted via the “amount
effect”. However, recent work has shown that precipitation δ
18
O (δ
18
O
P
) is greatly influ-
enced by convective processes distinct from precipitation amount, and new observations
indicate that δ
18
O
P
is negatively correlated with the fraction of stratiform precipitation.
Isotope-enabled climate models have emerged as a key interpretive tool in water isotope
systematics, and it is thus important to determine to what extent they can reproduce
these relationships. Here, seven isotope-enabled models, including the state-of-the-art
model iCAM5, are evaluated to see whether they can simulate the impact of convective
activity on δ
18
O
P
in observations. The results show that, of these models, only iCAM5
can simulate the observed anticorrelation between stratiform fraction andδ
18
O
P
. Further-
more, while all models can simulate the observed relationship between outgoing longwave
Publication Details: Hu, J., Emile-Geay, J., Nusbaumer, J., and Noone, D. (2018). Impact of convec-
tive activity on precipitation δ
18
O in isotope-enabled general circulation models. Journal of Geophysical
Research: Atmospheres, 123:13595-13610, doi: 10.1029/2018JD029187.
49
radiation and δ
18
O
P
, different models achieve this via different mechanisms – some get-
ting the right answer for the wrong reasons. Because iCAM5 appears in various metrics
to correctly simulate δ
18
O
P
variability, we use it to examine longstanding interpretations
of δ
18
O
P
over Asia. We find that the contribution of convective processes is very site-
dependent, with local processes accounting for a very small amount of variance at the
sites of most Chinese cave records (speleothems). The residual is attributed to source
and transport effects. Our results imply that state-of-the-art models like iCAM5 can and
should be used to guide the interpretation of δ
18
O
P
-based proxies.
2.1 Introduction
Theδ
18
Osignalpreservedinpaleo-archives(e.g. corals, speleothem, treeringcellulose,
ice cores) is widely used to reconstruct past climate conditions. In the tropics, the inverse
relationship between precipitation δ
18
O and precipitation amount, namely the “amount
effect” (Dansgaard, 1964), is often invoked to interpret δ
18
O as a proxy for precipitation
amount (Yuan et al., 2004, Yadava et al., 2004, Cheng et al., 2006). However, recent
studies have shown that precipitation δ
18
O (δ
18
O
P
) is controlled by a wider range of
processes.
Observational studies of δ
18
O
P
reveal that convective storms, especially organized
convective systems, generate lower δ
18
O
P
than non-convective storms or disorganized
convection (Lawrence et al., 2004, Risi et al., 2008, Kurita et al., 2011, Kurita, 2013,
Moerman et al., 2013, Lekshmy et al., 2014), and suggest thatδ
18
O
P
can reflect intrasea-
sonalvariabilityliketheMadden-JulianOscillation(MJO)(Kuritaetal.,2011)ortropical
cyclone activity (Frappier et al., 2007). The mechanisms are still debated. The recycling
of low δ
18
O water vapor below the cloud base in convective systems may contribute to
the decrease of δ
18
O
P
(Risi et al., 2008), and raindrop re-evaporation depletes the sur-
rounding water vapor (Lee and Fung, 2008). Convection depth and condensation height
are other important factors. Lacour et al. (2018) find that deep convection is associated
with isotopically depleted water vapor and precipitation while Cai and Tian (2016) show
that the cloud-top height correlates well with δ
18
O
P
.
50
A recent observational study (Aggarwal et al., 2016) revealed a negative correlation
between stratiform fraction (ratio of stratiform precipitation to total precipitation) and
δ
18
O
P
, providing a new yardstick for model evaluation (R
2
=0.59, p-value<0.0001). They
usedTRMM2A25/2A23 satellite data(2.5
◦
×2.5
◦
)andδ
18
O
P
atGNIPstations from1998
to 2014, and compared the stratiform rainfall fraction to theδ
18
O
P
data (Fig. 2.1). They
proposed that convective precipitation was generated from strong updrafts, which brings
isotopically enriched water vapor up, making δ
18
O
P
relatively higher. For the stratiform
precipitation,raindropsformedfromthe
18
O-depletedwatervaporinthemid-troposphere,
resulting in more negative δ
18
O
P
.
Stratiform fraction (%)
Addis Ababa Antananarivo Ascension Is. Bangkok Bellavista
Belo Horizonte Bogota Bukit Cotonou Darwin
Dhaka Douala El Jaral El Tesoro Entebbe
Hong Kong Izobamba Jakarta Jambi Manila
Mulu
Ndola New Delhi Panama Puerto Almendras
Quito St Denis
y = −0.20x + 1.14
R
2
= 0.59, P < 0.0001
Sylhet Krakow Vienna
Stratiform fraction (%)
Monthly precipitation amount (mm)
Monthly precipitation amount (mm)
0 10 20 30 40 50 60 70 80
−20
−15
−10
−5
δ
18
O ( , VSMOW)
0
a
b
c
−20
−15
−10
−5
0
0
10
20
30
40
50
60
70
80
0 200 100 300 400 500 600 700 800 900
0 200 100 300 400 500 600 700 800 900
% % δ
18
O ( , VSMOW) % %
Figure 2.1: Correlation of mean monthly δ
18
O (GNIP station data) and strati-
form fraction (TRMM satellite data). From Aggarwal et al. (2016), Fig. 1.
Aggarwal et al. (2016)’s conclusion that more stratiform rainfall fraction is associated
with lower δ
18
O
P
does not conflict with previous findings that organized convection cor-
responds to low δ
18
O
P
. In the tropics, organized convection is associated with a higher
stratiform fraction. Stratiform precipitation is often thought to occur only in fronts and
cyclones in the mid-latitudes, but it can also occur in the tropics, and even account for a
large portion of the tropical rainfall, especially in mesoscale convective systems (MCSs).
In the tropics, precipitation is often the product of young/vigorous convection (with
strong vertical air motion), which generates convective precipitation, or old/less active
convection (with weaker vertical air motion), which generates stratiform precipitation
and shares similar characteristics to mid-latitude stratiform precipitation (Houze, 1997).
51
For example, MCSs constitute organized convection. They are very common in the trop-
ics, accounting for about 50% of tropical precipitation (Nesbitt et al., 2006), and feature
both convective and large stratiform regions. Thus, when stratiform precipitation comes
to dominate in MCSs, these organized convective systems may generate δ
18
O-depleted
precipitation.
Isotope-enabled models are useful tools to study the variability ofδ
18
O
P
by separating
different factors influencing δ
18
O
P
(Joussaume et al., 1984, Jouzel et al., 1987, Hoffmann
et al., 1998, Noone and Simmonds, 2002, Schmidt et al., 2007, Yoshimura et al., 2008),
and since they can fill in the gaps between climate variables and paleoclimate records
with the aid of proxy system models (Evans et al., 2013, Dee et al., 2015a), they can
be directly exploited to investigate the variability of paleoclimate proxies (Baker et al.,
2012, Jex et al., 2013, Dee et al., 2017). Since convective activity plays an important
role, it is important to determine to what extent isotope-enabled climate models can
reproduce the relationships between convection and δ
18
O
P
. This evaluation will give
insights into possible improvements in current isotope-enabled models. If models can
grasp the convection-δ
18
O
P
relationship, it will help justify their use in investigating the
variability of δ
18
O
P
and interpret paleoclimate records based on it.
Another reason for probing this relationship is that stable water isotopes provide
unique constraints on General Circulation Model (GCM) performance. They offer oppor-
tunities to constrain physical processes such as cloud and convection schemes in GCMs
by comparing with traditional observations (including instrumental observations from
ground stations, satellites and aircraft). For instance, previous studies have probed the
sensitivity to parameters in convection schemes, such as the timescale for consumption of
convective available potential energy (CAPE) (Lee et al., 2009, Tharammal et al., 2017),
CAPE thresholds (Nusbaumer et al., 2017), entrainment rate (Field et al., 2014), and oth-
ers (Bony et al., 2008, Risi et al., 2012). In this paper, we use the observed relationship
between convection types/depth and δ
18
O
P
as another yardstick to constrain convective
processes in GCMs.
As an application, we explore how the quantification of these relationships affects the
interpretation of paleo-hydrological records based onδ
18
O
P
, particularlyδ
18
O from Asian
speleothems. Traditionally, the δ
18
O of Asian speleothem calcite has been interpreted as
52
an indicator of (a) regional precipitation; (b) the ratio of summer to winter precipitation,
or (c) monsoon intensity (Wang et al., 2001, Dykoski et al., 2005, Wang et al., 2008,
Cheng et al., 2009). This assumes that the amount effect is dominant, though recent
studies have shown that Asian speleothem δ
18
O can also be determined by upstream
water vapor δ
18
O (Pausata et al., 2011), the variability of moisture sources (Tian et al.,
2007), changes in atmospheric circulation (Maher and Thompson, 2012, Tan, 2014), and
convective activity (Kurita, 2013, Lekshmy et al., 2014, Cai and Tian, 2016). Thus it is
necessarytocomparetherelativecontributionsimpactingδ
18
O
P
, whichwillhelpconstrain
the interpretation of these records. This is of great importance because Asian monsoon
systems ultimately provide water supporting over 4 billion people, and speleothem records
provide a unique window into the natural variability of these systems.
The paper is structured as follows: we introduce data and methods in section 2, and
evaluatehowisotope-enabledmodelssimulatetherelationshipbetweenconvectiveactivity
andδ
18
O
P
in section 3. Section 4 discusses the implications of the evaluation results and
provides our conclusions.
2.2 Data and Methods
The model outputs analyzed here are from LMDZ, CAM2, isoGSM, MIROC, and
HadAM4 as part of the Stable Water Isotope Intercomparison Group, Phase 2 (SWING2)
project (Risi et al., 2012) (https://data.giss.nasa.gov/swing2/), forced by observed
sea surface temperature and sea ice following the AMIP protocal (Hurrell et al., 2008).
The results of SPEEDY-IER and iCAM5 are also from an AMIP-style experiment (see
Table 4.1 for references). Here we call the models participating in the SWING2 project
“SWING2 models”. iCAM5 is a state-of-the-art isotope-enabled models, with finer resolu-
tion, complex convection and stratiform cloud physics schemes, including the conversions
of cloud water species (liquid, ice, vapor, snow) and subgrid scale processes in clouds
(Nusbaumer et al., 2017). In the rest of this paper, convective rainfall will be identified
with the models’s “convective precipitation” variable (CONV) and stratiform precipita-
tion with the “large-scale” (LS) precipitation variable. The impacts of this approximation
are discussed below.
53
Table 2.1: General circulation models used in this study. * denotes a “SWING2 model”
Model Resolutions Nudging Time Periods Convection Schemes Reference
LMDZ
∗
2.5
◦
× 3.75
◦
Yes 1979-2007 Emanuel and Živković-Rothman (1999) Risi et al. (2010)
CAM2
∗
2.81
◦
× 2.81
◦
No 1974-2003 Zhang and McFarlane (1995) Lee et al. (2007)
isoGSM
∗
1.9
◦
× 1.875
◦
Yes 1979-2009 Moorthi and Suarez (1992) Yoshimura et al. (2008)
MIROC
∗
2.8
◦
× 2.8
◦
No 1979-2007 Arakawa and Schubert (1974) Kurita et al. (2011)
HadAM4
∗
2.5
◦
× 3.75
◦
No 1972-2001 Gregory and Rowntree (1990) Tindall et al. (2009)
SPEEDY-IER 3.75
◦
× 3.75
◦
No 1966-2000 simplified Tiedtke (1989) Dee et al. (2015b)
iCAM5 0.9
◦
× 1.25
◦
No 1971-2005
deep convection: Zhang and McFarlane (1995)
shallow convection: Park and Bretherton (2009)
Nusbaumer et al. (2017)
TRMM 3A25 (monthly data with the resolution of 1
◦
× 1
◦
, 1998-2014) is used to cal-
culate the observed climatological stratiform rainfall fraction in the tropics. In TRMM
3A25, the rainfall type in one pixel is identified by comparing its reflectivity to the aver-
aged nearby reflectivity. If the reflectivity of a pixel exceeds the surrounding background
by a factor f, the pixel is considered to be convective. f is a function of the back-
ground reflectivity intensity, and is calibrated to match a manual separation of convec-
tive/stratiform regions where a bright band is identified in radar echoes. A bright band is
a sufficient condition for a region to be stratiform. A detailed description of this algorithm
can be found in http://www.eorc.jaxa.jp/TRMM/documents/PR_algorithm_product_
information/pr_manual/PR_Instruction_Manual_V7_L1.pdf.
GCMs commonly generate convective precipitation within their convection schemes
and produce large-scale precipitation within their cloud/microphysics schemes. In GCMs,
the convection process consumes water vapor, forming convective precipitation, with
vertical air motion and adjustments of temperature and humidity profiles. Then,
cloud/microphysics schemes produce large-scale precipitation from the remaining water
vapor if a saturation condition is reached. By definition, all precipitation formed as
part of this convection process is “convective” (CONV) precipitation. In nature, convec-
tive and stratiform precipitation occur simultaneously, and stratiform precipitation may
account for a large fraction of precipitation in convection. TRMM observational analyses
use satellite-based radar reflectivity to distinguish convective and stratiform precipita-
tion because the difference in radar reflectivity characteristics can ensure the classified
precipitation has the characteristics described in Houze (1997). Therefore, although the
separation scheme of convective and stratiform precipitation in TRMM is different from
that in models, they both intend to partition convective and stratiform precipitation as
in Houze (1997): young/vigorous convection with strong vertical motion is categorized
54
as convective precipitation, while old/inactive convection with weak vertical motion is
categorized as stratiform precipitation. Also, TRMM satellite data were used to evaluate
convective and stratiform precipitation in climate models in previous studies (Dai, 2006,
Song and Yu, 2004). Thus in this paper the simulation of convective/stratiform pre-
cipitation in GCMs is compared with TRMM observations and the results of Aggarwal
et al. (2016). The “stratiform fraction” in model simulations is calculated as the ratio of
large-scale precipitation (LS) to total precipitation (CONV+LS).
All correlations calculated in the paper are based on monthly data. When establishing
significance, we use an isospectral test (Ebisuzaki, 1997) to control for autocorrelation,
and we use the false discovery rate method (Benjamini and Hochberg, 1995) to control
for the multiple hypothesis test problem. See Hu et al. (2017) for why it is essential to
control for both effects.
2.3 Results
2.3.1 Correlation between stratiform precipitation fraction and
δ
18
O
P
Hereweevaluatewhetherisotope-enabledmodelscansimulatethenegativecorrelation
between stratiform rainfall fraction and δ
18
O
P
. Fig. 2.2 shows the relationship between
stratiform fraction and δ
18
O
P
in our 7 models, over the grid boxes colocated with the
GNIP stations analyzed in Aggarwal et al. (2016). SPEEDY-IER, CAM2, isoGSM, and
iCAM5 appear to simulate the observed anticorrelation, albeit with relatively low R
2
values; iCAM5 has the largest such value, but still underestimates the slope. Since the
distribution of stratiform fraction is non-normal, we also transform it to normality (Emile-
Geay and Tingley, 2016) (Fig. A.1 and A.2) and the results are the same.
55
Figure 2.2: Relationship between monthly stratiform precipitation fraction and
δ
18
O
P
at the same locations as Aggarwal et al. (2016) in (a) SPEEDY-IER; (b) LMDZ;
(c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β values are ordinary
least square slopes.
56
To see how models simulate this relationship at other locations, Fig. 2.3 collects grid
boxes in the tropics (red dots, 30
◦
S-30
◦
N) and mid-latitudes (blue dots, 50
◦
S-30
◦
S and
30
◦
N-50
◦
N)anddisplaystherelationshipbetweenstratiformfractionandδ
18
O
P
(resultsof
transformedstratiformfractionisshowninFig. A.3andA.4). AllmodelsexceptHadAM4
simulate negative correlations in the mid-latitudes, but in the tropics, all models except
iCAM5 simulate positive correlations, which is inconsistent with observations (Aggarwal
et al., 2016). Only iCAM5 simulates the negative correlation, though with an R
2
value
smaller than observed. Considering the uncertainties from both the observations and
models, we focus mainly on a qualitative comparison, looking at the sign of correlation,
not the magnitude. Even by this permissive criterion, only iCAM5 successfully simulates
the observed negative correlation between stratiform fraction and δ
18
O
P
in the tropics.
57
Figure 2.3: Relationship between monthly stratiform precipitation fraction and
δ
18
O
P
in the tropics and mid-latitudes in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2;
(d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β values are ordinary least square
slopes. Stars behind β values represent the correlation passing the 95% level with an
isospectral test.
58
To see this another way, Fig. 2.4 shows the spatial distribution of the correlation
between stratiform fraction andδ
18
O
P
over the globe (correlations between monthly time
series of stratiform fraction and δ
18
O
P
at each grid cell). All models simulate negative
correlations in the continental mid-latitudes, but all models except iCAM5 simulate posi-
tive correlations in the tropics. iCAM5 is the only model to simulate negative correlations
over most areas in both tropics and mid-latitudes, as in observations. LMDZ simulates
the negative correlations only over the Indo-Pacific warm pool in the tropics. In the
continental mid-latitudes, cold seasons tend to see larger stratiform fractions (smaller
convective fractions) and cold temperatures, leading to lower δ
18
O
P
. All models except
iCAM5 simulate positive correlations over the mid-latitude oceans. This cannot be con-
strained by GNIP observations since stations are all on land. One possible explanation
for this is that mid-latitude cyclones bring in warmer and more isotopically enriched air,
while at the same time generating large stratiform precipitation fractions, producing the
positive correlations. All models can simulate both relationships, so all models gener-
ate negative correlations between stratiform fraction and δ
18
O
P
there. In the tropics,
seasonal variability is small, so the correlation between stratiform fraction and δ
18
O
P
is
more dependent on convection and microphysics schemes (it is a more sensitive indicator
of model verisimilitude for these processes, and it appears that no particular type of con-
vection scheme improves simulating this relationship). Some oceanic regions immediately
west of continents show positive correlations in iCAM5. These are arid regions where cli-
matological monthly precipitation is less than 1.5 mm/day (Masked regions in Fig. 2.9),
so the uncertainty of stratiform precipitation fraction is high. Many of these regions are
also places where stratiform clouds always exists, so the separation between convective
and stratiform precipitation is somewhat arbitrary since stratiform clouds always present.
Whether the results of Aggarwal et al. (2016) apply for these regions needs future inves-
tigation.
59
(g) iCAM5
ρ
Figure 2.4: Correlation between monthly stratiform precipitation fraction and
δ
18
O
P
insevenmodels. Dotsrepresentcorrelationnotpassingthe95%levelsignificance
isospectral test. Yellow circles mark the locations of GNIP stations analyzed in Aggarwal
et al. (2016)
60
It should be noted that the observations themselves are affected by their own uncer-
tainties. For example, the stratiform fraction in Aggarwal et al. (2016) is retrieved from
satellite-based reflectivity. There are uncertainties in the satellite observation itself and
the conversion process. Also, the estimated stratiform fraction hinges on the criteria
for the classification of stratiform and convective precipitation (for example, including
shallow non-isolated precipitation in stratiform precipitation or not) (Funk et al., 2013).
Finally, the relationship revealed in Aggarwal et al. (2016) is based on a relatively short
timeseries (monthly data for 16 years vs. 29 to 35 years for GCMs).
To better understand why iCAM5 can simulate the negative relationship between
stratiform fraction and δ
18
O
P
, its cloud microphysical processes over tropical convective
and stratiform regions are diagnosed in Fig. 2.5. The water vapor δ
18
O vertical profile
showsmoredepletedvaporoverstratiformthanconvectiveregions, andmore
18
O-depleted
water vapor will form more negativeδ
18
O
P
. The deuterium excess of water vapor (d) can
be an indicator of kinetic effects, which occur during vapor deposition onto ice particles
and re-evaporation of rain in low humidity environments (Kurita et al., 2011). Fig. 2.5b
shows that iCAM5 simulates higherd values in the mid-upper troposphere (600-300 hPa)
in stratiform regions than convective regions. This indicates that more ice crystals form
in the upper troposphere over stratiform regions. The condensation heating profile (Fig.
2.5c) also confirms that stratiform regions have more condensation in the upper levels
than convective regions. Since more ice particles are generated from
18
O-depleted water
vapor, these particles have a more negative δ
18
O when they precipitate to the ground,
resulting in lower δ
18
O
P
. This process is consistent with Aggarwal et al. (2016) in that
stratiform precipitation mainly forms with the
18
O-depleted water vapor in the upper
atmosphere. Another possible explanation for the highd values in the upper troposphere
over stratiform regions in iCAM5 is that stratiform precipitation is fed by water vapor
which has been recycled via the re-evaporation of rain, following the moisture recycling
processes revealed in Risi et al. (2008) and Kurita et al. (2011). This is consistent with the
iCAM5 code, in which the large-scale cloud physics always triggers after the convection.
The vertical profiles of deuterium excess of water vapor in other models (Fig. A.6) show
that only LMDZ simulates higher d values in the mid-upper troposphere (600-300 hPa)
61
over stratiform regions like iCAM5, but its d values decreases with altitude from surface
to 400 hPa.
(a)
(b) (c)
Figure 2.5: Vertical profiles of water vapor δ
18
O (a), deuterium excess (b), and
condensational heating (c) over the convective rainfall region, the stratiform rainfall
region, andnorainregioninthetropics(30
◦
S-30
◦
N)iniCAM5. Theconvective/stratiform
rainfall region is where the proportion of convective/stratiform rainfall to total rainfall
exceeds 0.8.
62
Therefore, the reason iCAM5 can simulate the observed negative correlation between
stratiform ratio and δ
18
O
P
can be directly tied to its more faithful representation of the
vertical distribution of cloud condensate. This gives us confidence in using this model to
study the role of convection in the interpretation of δ
18
O
P
and paleohydrological records
(Section 2.3.3). Also this indicates that δ
18
O
P
is a sensitive indicator of the stratiform
cloud environment, and can therefore help inform the model development cycle.
2.3.2 Correlation between outgoing longwave radiation and
δ
18
O
P
Here we use outgoing longwave radiation (OLR) to track large-scale convective activ-
ity. Vigorous convection is usually deep, so the cloud top temperature is cold, emitting
lower outgoing longwave radiation than weak/shallow convection. Since strong convection
corresponds to lowδ
18
O
P
, OLR displays a positive correlation withδ
18
O
P
in observations
(Moerman et al., 2013, Lekshmy et al., 2014). NOAA interpolated OLR (Liebmann,
1996) and GNIP δ
18
O
P
over the stations in Aggarwal et al. (2016) also displays this
positive relationship (Fig. 2.6). Here we exclude four stations above 2000 m since the
OLR dataset cannot capture the topography of the tropical Andes due to its low spa-
tial resolution, thus may not be representative of the local climate conditions observed
by those stations. OLR is also a direct indicator of convection depth, so it can be used
to examine whether models can simulate the relationship between convection depth and
δ
18
O
P
. Deep convection (low OLR) is associated with low δ
18
O
P
in the observations
(Lacour et al., 2018), so OLR also should bear a positive correlation to δ
18
O
P
. In the
mid-latitudes, where convection is much less than the tropics, OLR is more dependent on
surface temperature. In summer, surface temperature is higher, leading to higher OLR,
and δ
18
O
P
also usually reaches its peak due to the temperature effect (Dansgaard, 1964,
Galewsky et al., 2016), and vice versa for winter. This makes the variation of OLR and
δ
18
O
P
in phase in the mid-latitudes.
63
180 200 220 240 260 280
outgoing longwave radiation (W/m
2
)
12
10
8
6
4
2
0
precipitation
18
O ( )
p-value<1e-6, =0.53
GNIP vs. OLR
Figure 2.6: Relationship between NOAA interpolated OLR and GNIP
δ
18
O
P
over the stations in Aggarwal et al. (2016), excluding stations higher than 2000 m
(4 stations).
Since only 4 out of 7 models (SPEEDY-IER, LMDZ, isoGSM and iCAM5) provide the
OLRvariable, werestricttheanalysistothesefourmodels. Allfourmodels(Fig. 2.7)sim-
ulate the positive correlations between OLR andδ
18
O
P
in both tropics and mid-latitudes,
consistent with the observations. Nonetheless, different models achieve this relationship
through different mechanisms. iCAM5, which successfully reproduces the negative corre-
lation between δ
18
O
P
and stratiform fraction, can simulate this relationship because its
strong large-scale convective regions (low OLR) are associated with more stratiform pre-
cipitation, generating lowerδ
18
O
P
. On the other hand, models like SPEEDY-IER, LMDZ
and isoGSM, which do not reproduce the anti-correlation between δ
18
O
P
and stratiform
fraction, simulate this relationship because stronger convection generates more precipita-
tion, so rainout processes produce lower δ
18
O
P
. This is not related to the discrimination
64
of isotope modules in convection and non-convection processes, so the δ
18
O
P
-OLR link is
a less sensitive metric of model performance.
From the viewpoint of convection depth, the fact that all models simulate the positive
correlations between OLR and δ
18
O
P
indirectly shows that models can correctly simu-
late the relationship between convection depth and δ
18
O
P
(Lacour et al., 2018). The
direct examination of how models simulate this relationship is currently limited by the
accessibility of the condensation heating variable in SWING2 models.
(c) iCAM5
Figure 2.7: Correlation between monthly outgoing longwave radiation and
δ
18
O
P
in four models. Dots represent correlation not passing the 95% level signifi-
cance isospectral test.
In summary, all models simulate the observed positive correlation between OLR and
δ
18
O
P
in the tropics and mid-latitudes. Since OLR is an indicator of the intensity of large-
scale convection and convection depth, this result suggests that current isotope-enabled
65
modelscanbeusedtostudytheroleofconvectiondepthorlarge-scaleconvectionintensity
in δ
18
O
P
and paleoclimate proxies.
2.3.3 Implications for speleothem record interpretation
As mentioned before, theδ
18
O of speleothem calcite is frequently used as an indicator
of past hydrological conditions (Cheng et al., 2016, Sinha et al., 2011). Since cave δ
18
O
is driven by variations in δ
18
O
P
, understanding the latter is critical to interpreting these
records, and δ
18
O
P
is controlled by multiple factors including convective activity, rather
than an indicator of regional precipitation or monsoon intensity. Thus it is necessary to
compare the relative contributions of different factors impacting δ
18
O
P
, which will help
constrain the interpretation of speleothem δ
18
O. While the spatiotemporal distribution
of available instrumental observations is very limited, isotope-enabled models provide a
perfectly-observed,physically-consistentframeworktoexploretheinterpretationofδ
18
O
P
.
Since iCAM5 is a state-of-the-art isotope-enabled model which simulates the variability
of δ
18
O
P
with high fidelity (Nusbaumer et al., 2017), and our previous result also shows
that it successfully simulates the relationship between convective activity and δ
18
O
P
, we
now use it to diagnose the causes of δ
18
O
P
variability.
Here we estimate the contribution of stratiform fraction, OLR, precipitation amount
and local water vapor advection to the percentage of variance of δ
18
O
P
at monthly scales
in iCAM5 by calculating the R
2
-value between these variables. Water vapor advection is
estimated by
Z
ptop
ps
−V·∇δ
18
O
v
dp (2.1)
which is based on Eq. (4) derived in Okazaki et al. (2015), where p
s
is surface pressure,
p
top
is pressure at the top of atmosphere,V is wind, andδ
18
O
v
is water vaporδ
18
O. Since
the advection term is calculated at each grid cell here, it only represents local water vapor
advection, and cannot describe the contribution of remote water vapor transport. Fig. 2.8
shows the distribution of R
2
-values over the Asian monsoon region. Since we are taking
the interpretation of speleothem δ
18
O as an example, many well-known Asian cave sites
are also plotted on this map. These sites are included in the recent global compilation
SISAL_v1(Atsawawaranuntetal.,2018), producedbythePAGES(PastGlobalChanges,
66
http://www.pages-igbp.org/) working group SISAL (Speleothem Isotopes Synthesis
and AnaLysis, http://www.pages-igbp.org/ini/wg/sisal/intro).
Fig. 2.8 shows that stratiform fraction contributes less than 10% of δ
18
O
P
variability
in Chinese and Indian caves, while it contributes about 20% over the Maritime continent
(Borneo). OLR can explain as much as 50% over the Indochina peninsula and 20% over
the Maritime continent (consistent with the observational result of Moerman et al. (2013)
in Borneo), and still does not contribute more than 10% variability of δ
18
O
P
over China
and India. The contribution of precipitation amount has a similar spatial distribution to
OLR over land, showing that precipitation amount accounts for less than 10% of the vari-
ability ofδ
18
O
P
in China and India (except the southeast corner of the Indian peninsula),
including the celebrated caves of Sanbao (Cheng et al., 2009, 2016) and Hulu (Wang et al.,
2001, Cheng et al., 2006, Zhang et al., 2014). This result suggests that the contribution
of convection is as important as precipitation amount over the Indochina peninsula and
the Maritime continent, and neither convection nor precipitation amount explain much of
the variability of precipitation over Chinese and Indian caves. This is in sharp contrast
to the classic view that Chinese speleothem records represent local precipitation amount
or the ratio of summer to winter precipitation (Wang et al., 2001, Dykoski et al., 2005,
Wang et al., 2008, Cheng et al., 2009). Local water vapor advection generally contributes
little to δ
18
O
P
(Fig. 2.8d). The places where local water vapor advection contributes
above 20% of δ
18
O
P
are scattered over Central Asia and Northeast Asia. To estimate
the total local contributions toδ
18
O
P
, we perform multi-linear regressions with these four
factors and obtain the R
2
statistics. The result shows that local effects can explain much
of variability of theδ
18
O
P
over the Indochina peninsula, the Maritime continent and caves
north of 40
◦
N (sum of R
2
> 90%). However, local effects explain less than 20% of the
variability ofδ
18
O
P
over eastern China, northern India and the Arabian peninsula, where
many caves exist. This suggests over these locations, remote effects like the upstream
effect raised by Pausata et al. (2011) (e.g. Chinese speleothem δ
18
O is influenced by pre-
cipitation over India) should contribute much to δ
18
O
P
variability. These contributions
will be quantified in a follow-up study. It should be noted that this result is based on
iCAM5, and iCAM5 simulates a smaller correlation of stratiform ratio and δ
18
O
P
com-
pared with the observations, as shown in section 3.1. Also, these results are based on
67
monthly data, so they mainly reflect seasonal/interannual variability of δ
18
O
P
. Over mil-
lennia/orbital scales, there is strong coherence among speleothem δ
18
O at different sites
(Battisti et al., 2014, Cheng et al., 2012) even thoughδ
18
O
P
is controlled by many factors
as we show here. This is partly because the climate signals over millennia/orbital scales
usually feature relatively large amplitude and spatial scales. Another possible reason is
δ
18
O
P
is more controlled by non-local processes, which represent more large-scale circu-
lation features, if our results hold true for millennia/orbital scales. Tabor et al. (2018)
employed iCESM/iCAM5 with the water-tagging technique to constrain the contribution
of large-scale moisture source effects at orbital scales and found that the moisture source
contributions are the dominant factor over the South Asian monsoon region.
68
(a) stratiform fraction
R
2
R
2
(c) rainfall amount
(d) water vapor convergence
(b) outgoing longwave radiation
R
2
R
2
(e) total local contributions
Figure 2.8: The variance of monthly δ
18
O
P
explained by local contributions:
stratiform rainfall fraction (a), outgoing longwave radiation (b), precipitation amount
(c), local water vapor convergence (d) and the total local contributions (e) in iCAM5.
Blue dots are cave sites collected in the global speleothem database SISAL_v1, and the
names of these caves are listed besides Fig. 2.8e.
69
2.4 Discussion and Conclusions
We evaluated the ability of seven isotope-enabled models to simulate the impact of
convective activity on δ
18
O
P
. The results show that only one (iCAM5) can simulate the
negative correlation between stratiform fraction and δ
18
O
P
discovered in observations.
The iCAM5 results are also consistent with Lacour et al. (2018) in that deeper convection
corresponds to more negative δ
18
O
P
. Models that do not simulate the anti-correlation
between stratiform fraction andδ
18
O
P
may not be suitable to study the role of precipita-
tion types onδ
18
O
P
, but they can still be useful to investigate the role of other aspects of
convection, such as convection depth, because all models investigated herein are found to
simulate the observed relationship between outgoing longwave radiation and δ
18
O
P
. But
we should also note that different models may have different mechanisms to generate this
relationship – some getting the right answer for the wrong reasons.
Although iCAM5 successfully simulates the negative correlation between stratiform
ratio and δ
18
O
P
, it should be noted that iCAM5, like other models, largely underesti-
mates stratiform fraction in the tropics (less than 10% while TRMM has over 40%) (Fig.
2.9). We also note that LMDZ simulates stratiform fraction fairly well in the tropics, even
though it does not successfully reproduce the anti-correlation between stratiform rainfall
fraction and δ
18
O
P
. The underestimation of the stratiform ratio in the tropics in climate
models is a common problem (Song and Yu, 2004), and some studies (Song and Zhang,
2011, Yang et al., 2013) proposed methods to improve its simulation, including modify-
ing microphysics parameterization schemes. In addition, the resolution of GCMs is too
coarse to represent organized convection and mesoscale convective systems (MCSs), and
MCSs have high stratiform ratios, so adding a suitable and feasible parameterization of
organized tropical convection for GCMs like that proposed by Moncrieff et al. (2017) may
improve the simulation of stratiform fraction. Finally, we note that shallow convection
precipitation in models like iCAM5 is categorized as “convective” precipitation though it
accounts for a small (< 5%) fraction of the total precipitation, but it shares some features
(e.g., relatively stable atmospheric structure) with the defined stratiform precipitation
in TRMM, and this may also underestimate the fraction of stratiform precipitation in
models.
70
(b) iCAM5
(c) (d)
(e) (f)
(g) (h)
Figure 2.9: Climatological stratiform fraction in the tropics from satellite obser-
vations (a) and isotope-enabled models (b,c,d,e,f,g,h). Regions where climatological pre-
cipitation is less than 1.5 mm/day are masked.
Lacour et al. (2018) argue that the relationship between stratiform fraction and
δ
18
O
P
can be interpreted by the depth of convection instead of cloud microphysics pro-
cesses mentioned in Aggarwal et al. (2016), because stratiform convection has a higher
condensation level, which corresponds to low δ
18
O
P
. Our results partially reconcile these
arguments,inthatcloudmicrophysicsatleastcanexplainlowδ
18
Oinstratiformprecipita-
tion because more ice particles form at high altitude with low temperature in stratiform
precipitation, resulting in low raindrop δ
18
O (Fig. 2.5b). This is one plausible reason
why deep convection may be associated with depleted δ
18
O
P
, apart from downdrafts and
reevaporation described in Lacour et al. (2018). Shallow convection also occurs in iCAM5,
and it is unclear if the explanation of Aggarwal et al. (2016) for convective precipitation
applies for both deep and shallow convection.
Lastly, we investigated the quantitative contribution of convective activity to
δ
18
O
P
variability in the Asian monsoon region in iCAM5. The result shows that the
role of convection is very important in the Indochina peninsula, where the variation of
71
outgoing longwave radiation is associated with as much as 50% of the variance of δ
18
O
P
,
and OLR is an indicator of large-scale convective activity and convection depth. This
suggests that paleoclimate records there can be partly interpreted as the variability of
large-scale convection, which can be connected to intraseasonal variability like the MJO.
However, the result shows that neither convection nor precipitation amount can explain
more than 15% ofδ
18
O
P
over China and India. This result is in stark contrast to the tra-
ditional interpretation of Chinese speleothem δ
18
O, taken to represent local precipitation
amount or the ratio of summer to winter precipitation. If so, this suggests a dominant
influence of remote water vapor transport, including the origin of water vapor source
(circulation variation), fractionation in water vapor along the transport path, and frac-
tionation at the water vapor source (e.g. SST effect, Pausata et al., 2011) – a hypothesis
thatwewillinvestigateinafollow-upstudy. Thisanalysisisbasedonmonthlymeandata,
which largely reflects the seasonal variability, and recent studies (Eastoe and Dettman,
2016) show that seasonal relationships betweenδ
18
O
P
and climate variables may not hold
true for longer time scales. Whether local contributions to the variability of δ
18
O
P
are
still small for interannual-decadal or longer time scales deserves careful investigation. We
should note that this result is based on one model, and validation by observations is also
necessary.
Compared with previous isotope-enabled model evaluations (Risi et al., 2012, Conroy
et al., 2013, Midhun and Ramesh, 2016), we mainly focused on the role of convective
activity and tried to quantify the contribution of precipitation amount, convection, and
local water vapor advection. Our results, like these previous studies, show a large model
spreadinsimulatingrelationshipsbetweenwaterisotopesandclimatevariables, indicating
the unique ability of water isotope observations to discriminate between models.
Our results suggest that a state-of-art model like iCAM5 can successfully simulate
the role of convection in the variability of δ
18
O
P
, which gives us confidence in using this
model to study the interpretation ofδ
18
O
P
and hydrological paleoclimate records. It also
implies that there are no shortcuts for isotope-enabled models to simulate the role of
precipitation types in δ
18
O
P
. The necessary processes in cloud microphysics have to be
captured if we want to use models to study the impact of convection on δ
18
O
P
. Also,
isotope-enabled models, which provide water isotope ratios that standard GCMs do not
72
track, can be used to constrain convective and microphysical processes in GCMs, which
should help improve future climate projections.
Bibliography
Aggarwal, P. K., Romatschke, U., Araguas-Araguas, L., Belachew, D., Longstaffe, F. J.,
Berg, P., Schumacher, C., andFunk, A.(2016). Proportionsofconvectiveandstratiform
precipitation revealed in water isotope ratios. Nature Geoscience, 9(8):624.
Arakawa, A. and Schubert, W. H. (1974). Interaction of a cumulus cloud ensemble with
thelarge-scaleenvironment, parti. Journal of the Atmospheric Sciences, 31(3):674–701.
Atsawawaranunt, K., Comas-Bru, L., Amirnezhad Mozhdehi, S., Deininger, M., Harrison,
S. P., Baker, A., Boyd, M., Kaushal, N., Ahmad, S. M., Ait Brahim, Y., et al. (2018).
The SISAL database: a global resource to document oxygen and carbon isotope records
from speleothems. Earth System Science Data.
Baker, A., Bradley, C., Phipps, S., Fischer, M., Fairchild, I., Fuller, L., Spötl, C., and
Azcurra, C. (2012). Millennial-length forward models and pseudoproxies of stalagmite
δ
18
O: an example from NW Scotland. Climate of the Past, 8(4):1153–1167.
Battisti, D., Ding, Q., and Roe, G. (2014). Coherent pan-Asian climatic and isotopic
response to orbital forcing of tropical insolation. Journal of Geophysical Research:
Atmospheres, 119(21):11–997.
Benjamini, Y. and Hochberg, Y. (1995). Controlling the false discovery rate: a practical
and powerful approach to multiple testing. Journal of the Royal Statistical Society.
Series B (Methodological), pages 289–300.
Bony, S., Risi, C., and Vimeux, F. (2008). Influence of convective processes on the
isotopic composition (δ
18
O and δD) of precipitation and water vapor in the tropics:
1. Radiative-convective equilibrium and Tropical Ocean–Global Atmosphere–Coupled
Ocean-Atmosphere Response Experiment (TOGA-COARE) simulations. Journal of
Geophysical Research: Atmospheres, 113:D19305.
73
Cai, Z. and Tian, L. (2016). Atmospheric Controls on Seasonal and Interannual Variations
in the Precipitation Isotope in the East Asian Monsoon Region. Journal of Climate,
29(4):1339–1352.
Cheng, H., Edwards, R. L., Broecker, W. S., Denton, G. H., Kong, X., Wang, Y., Zhang,
R., and Wang, X. (2009). Ice Age Terminations. Science, 326(5950):248–252.
Cheng, H., Edwards, R. L., Sinha, A., Spötl, C., Yi, L., Chen, S., Kelly, M., Kathayat,
G., Wang, X., Li, X., et al. (2016). The Asian monsoon over the past 640,000 years
and ice age terminations. Nature, 534(7609):640–646.
Cheng, H., Edwards, R. L., Wang, Y., Kong, X., Ming, Y., Kelly, M. J., Wang, X., Gallup,
C. D., and Liu, W. (2006). A penultimate glacial monsoon record from Hulu Cave and
two-phase glacial terminations. Geology, 34(3):217–220.
Cheng, H., Sinha, A., Wang, X., Cruz, F. W., and Edwards, R. L. (2012). The Global
PaleomonsoonasseenthroughspeleothemrecordsfromAsiaandtheAmericas. Climate
Dynamics, 39(5):1045–1062.
Conroy, J. L., Cobb, K. M., and Noone, D. (2013). Comparison of precipitation isotope
variability across the tropical Pacific in observations and SWING2 model simulations.
Journal of Geophysical Research: Atmospheres, 118(11):5867–5892.
Dai, A. (2006). Precipitation characteristics in eighteen coupled climate models. Journal
of Climate, 19(18):4605–4630.
Dansgaard, W. (1964). Stable isotopes in precipitation. Tellus, 16(4):436–468.
Dee, S., Emile-Geay, J., Evans, M. N., Allam, A., Steig, E. J., and Thompson, D. M.
(2015a). PRYSM: An open-source framework for PRoxY System Modeling, with appli-
cations to oxygen-isotope systems. Journal of Advances in Modeling Earth Systems,
7(3):1220–1247.
Dee, S., Noone, D., Buenning, N., Emile-Geay, J., and Zhou, Y. (2015b). SPEEDY-IER:
A fast atmospheric GCM with water isotope physics. Journal of Geophysical Research:
Atmospheres, 120(1):73–91.
74
Dee, S. G., Parsons, L. A., Loope, G. R., Overpeck, J. T., Ault, T. R., and Emile-Geay,
J. (2017). Improved spectral comparisons of paleoclimate models and observations via
proxy system modeling: Implications for multi-decadal variability. Earth and Planetary
Science Letters, 476(Supplement C):34–46.
Dykoski, C. A., Edwards, R. L., Cheng, H., Yuan, D., Cai, Y., Zhang, M., Lin, Y., Qing,
J., An, Z., and Revenaugh, J. (2005). A high-resolution, absolute-dated Holocene and
deglacialAsianmonsoonrecordfromDonggeCave, China. Earth and Planetary Science
Letters, 233(1):71–86.
Eastoe, C. and Dettman, D. (2016). Isotope amount effects in hydrologic and climate
reconstructions of monsoon climates: Implications of some long-term data sets for pre-
cipitation. Chemical Geology, 430:78–89.
Ebisuzaki, W. (1997). A method to estimate the statistical significance of a correlation
when the data are serially correlated. Journal of Climate, 10(9):2147–2153.
Emanuel, K. A. and Živković-Rothman, M. (1999). Development and evaluation of a
convection scheme for use in climate models. Journal of the Atmospheric Sciences,
56(11):1766–1782.
Emile-Geay,J.andTingley,M.(2016). Inferringclimatevariabilityfromnonlinearproxies:
application to palaeo-ENSO studies. Climate of the Past, 12(1):31–50.
Evans, M. N., Tolwinski-Ward, S. E., Thompson, D. M., and Anchukaitis, K. J. (2013).
Applications of proxy system modeling in high resolution paleoclimatology. Quaternary
Science Reviews, 76(0):16–28.
Field, R. D., Kim, D., LeGrande, A. N., Worden, J., Kelley, M., and Schmidt, G. A.
(2014). Evaluating climate model performance in the tropics with retrievals of water
isotopic composition from Aura TES. Geophysical Research Letters, 41(16):6030–6036.
Frappier, A. B., Sahagian, D., Carpenter, S. J., González, L. A., and Frappier, B. R.
(2007). Stalagmite stable isotope record of recent tropical cyclone events. Geology,
35(2):111–114.
75
Funk, A., Schumacher, C., and Awaka, J. (2013). Analysis of rain classifications over the
tropics by version 7 of the TRMM PR 2A23 algorithm. Journal of the Meteorological
Society of Japan. Ser. II, 91(3):257–272.
Galewsky, J., Steen-Larsen, H. C., Field, R. D., Worden, J., Risi, C., and Schneider, M.
(2016). Stable isotopes in atmospheric water vapor and applications to the hydrologic
cycle. Reviews of Geophysics, 54(4):809–865.
Gregory, D. and Rowntree, P. (1990). A mass flux convection scheme with representation
of cloud ensemble characteristics and stability-dependent closure. Monthly Weather
Review, 118(7):1483–1506.
Hoffmann, G., Werner, M., and Heimann, M. (1998). Water isotope module of the
ECHAM atmospheric general circulation model: A study on timescales from days to
several years. Journal of Geophysical Research: Atmospheres, 103(D14):16871–16896.
Houze, Jr, R. A. (1997). Stratiform precipitation in regions of convection: A meteorolog-
ical paradox? Bulletin of the American Meteorological Society, 78(10):2179–2196.
Hu, J., Emile-Geay, J., and Partin, J. (2017). Correlation-based interpretations of paleo-
climate data–where statistics meet past climates. Earth and Planetary Science Letters,
459:362–371.
Hurrell, J. W., Hack, J. J., Shea, D., Caron, J. M., and Rosinski, J. (2008). A new
sea surface temperature and sea ice boundary dataset for the Community Atmosphere
Model. Journal of Climate, 21(19):5145–5153.
Jex, C., Phipps, S., Baker, A., and Bradley, C. (2013). Reducing uncertainty in the
climatic interpretations of speleothemδ
18
O. Geophysical Research Letters, 40(10):2259–
2264.
Joussaume, S., Sadourny, R., and Jouzel, J. (1984). A general circulation model of water
isotope cycles in the atmosphere. Nature, 311(5981):24.
76
Jouzel, J., Russell, G., Suozzo, R., Koster, R., White, J., and Broecker, W. (1987). Sim-
ulations of the HDO and H
18
2
O atmospheric cycles using the NASA GISS General Cir-
culation Model: The seasonal cycle for present-day conditions. Journal of Geophysical
Research: Atmospheres, 92(D12):14739–14760.
Kurita, N. (2013). Water isotopic variability in response to mesoscale convective system
overthetropicalocean. Journal of Geophysical Research: Atmospheres, 118(18):10,376–
10,390.
Kurita, N., Noone, D., Risi, C., Schmidt, G. A., Yamada, H., and Yoneyama, K. (2011).
Intraseasonal isotopic variation associated with the Madden-Julian Oscillation. Journal
of Geophysical Research: Atmospheres, 116:D24101.
Lacour, J.-L., Risi, C., Worden, J., Clerbaux, C., and Coheur, P.-F. (2018). Importance
of depth and intensity of convection on the isotopic composition of water vapor as seen
fromIASI andTESδDobservations. Earth and Planetary Science Letters, 481:387–394.
Lawrence, J. R., Gedzelman, S. D., Dexheimer, D., Cho, H.-K., Carrie, G. D., Gasparini,
R., Anderson, C. R., Bowman, K. P., and Biggerstaff, M. I. (2004). Stable isotopic com-
position of water vapor in the tropics. Journal of Geophysical Research: Atmospheres,
109(D6).
Lee, J.-E.andFung, I.(2008). “Amounteffect”ofwaterisotopesandquantitativeanalysis
of post-condensation processes. Hydrological Processes, 22(1):1–8.
Lee, J.-E., Fung, I., DePaolo, D. J., and Henning, C. C. (2007). Analysis of the global
distribution of water isotopes using the NCAR atmospheric general circulation model.
Journal of Geophysical Research: Atmospheres, 112:D16306.
Lee, J.-E., Pierrehumbert, R., Swann, A., and Lintner, B. R. (2009). Sensitivity of stable
water isotopic values to convective parameterization schemes. Geophysical Research
Letters, 36(23):L23801.
Lekshmy, P., Midhun, M., Ramesh, R., andJani, R.(2014).
18
Odepletioninmonsoonrain
relates to large scale organized convection rather than the amount of rainfall. Scientific
Reports, 4:5661.
77
Liebmann, B. (1996). Description of a complete (interpolated) outgoing longwave radia-
tion dataset. Bull. Amer. Meteor. Soc., 77:1275–1277.
Maher, B. A. and Thompson, R. (2012). Oxygen isotopes from Chinese caves: records
not of monsoon rainfall but of circulation regime. Journal of Quaternary Science,
27(6):615–624.
Midhun, M. and Ramesh, R. (2016). Validation ofδ
18
O as a proxy for past monsoon rain
by multi-GCM simulations. Climate Dynamics, 46(5-6):1371–1385.
Moerman, J. W., Cobb, K. M., Adkins, J. F., Sodemann, H., Clark, B., and Tuen, A. A.
(2013). Diurnal to interannual rainfall δ
18
O variations in northern Borneo driven by
regional hydrology. Earth and Planetary Science Letters, 369:108–119.
Moncrieff, M. W., Liu, C., and Bogenschutz, P. (2017). Simulation, modeling, anddynam-
icallybasedparameterizationoforganizedtropicalconvectionforglobalclimatemodels.
Journal of the Atmospheric Sciences, 74(5):1363–1380.
Moorthi, S. and Suarez, M. J. (1992). Relaxed arakawa-schubert. a parameterization of
moist convection for general circulation models. Monthly Weather Review, 120(6):978–
1002.
Nesbitt,S.W.,Cifelli,R.,andRutledge,S.A.(2006). Stormmorphologyandrainfallchar-
acteristics of TRMM precipitation features. Monthly Weather Review, 134(10):2702–
2721.
Noone, D. and Simmonds, I. (2002). Associations between δ
18
O of water and climate
parameters in a simulation of atmospheric circulation for 1979–95. Journal of Climate,
15(22):3150–3169.
Nusbaumer, J., Wong, T. E., Bardeen, C., and Noone, D. (2017). Evaluating hydrological
processes in the Community Atmosphere Model Version 5 (CAM5) using stable isotope
ratios of water. Journal of Advances in Modeling Earth Systems, 9(2):949–977.
Okazaki, A., Satoh, Y., Tremoy, G., Vimeux, F., Scheepmaker, R., and Yoshimura, K.
(2015). Interannual variability of isotopic composition in water vapor over western
78
Africa and its relationship to ENSO. Atmospheric Chemistry and Physics, 15(6):3193–
3204.
Park, S. and Bretherton, C. S. (2009). The university of washington shallow convec-
tion and moist turbulence schemes and their impact on climate simulations with the
community atmosphere model. Journal of Climate, 22(12):3449–3469.
Pausata, F. S., Battisti, D. S., Nisancioglu, K. H., and Bitz, C. M. (2011). Chinese sta-
lagmiteδ
18
O controlled by changes in the Indian monsoon during a simulated Heinrich
event. Nature Geoscience, 4(7):474–480.
Risi, C., Bony, S., and Vimeux, F. (2008). Influence of convective processes on the isotopic
composition (δ
18
O andδD) of precipitation and water vapor in the tropics: 2. Physical
interpretation of the amount effect. Journal of Geophysical Research: Atmospheres,
113:D19306.
Risi, C., Bony, S., Vimeux, F., and Jouzel, J. (2010). Water-stable isotopes in the LMDZ4
general circulation model: Model evaluation for present-day and past climates and
applications to climatic interpretations of tropical isotopic records. Journal of Geo-
physical Research: Atmospheres, 115:D12118.
Risi, C., Noone, D., Worden, J., Frankenberg, C., Stiller, G., Kiefer, M., Funke, B.,
Walker, K., Bernath, P., Schneider, M., et al. (2012). Process-evaluation of tropospheric
humidity simulated by general circulation models using water vapor isotopologues: 1.
Comparison between models and observations. Journal of Geophysical Research: Atmo-
spheres, 117:D05303.
Schmidt, G. A., LeGrande, A. N., and Hoffmann, G. (2007). Water isotope expressions
of intrinsic and forced variability in a coupled ocean-atmosphere model. Journal of
Geophysical Research: Atmospheres, 112:D10103.
Sinha, A., Berkelhammer, M., Stott, L., Mudelsee, M., Cheng, H., and Biswas, J. (2011).
The leading mode of indian summer monsoon precipitation variability during the last
millennium. Geophysical Research Letters, 38(15):L15703.
79
Song, X. and Yu, R. (2004). Underestimated tropical stratiform precipitation in
the National Center for Atmospheric Research (NCAR) Community Climate Model
(CCM3). Geophysical Research Letters, 31(24):L24101.
Song, X. and Zhang, G. J. (2011). Microphysics parameterization for convective clouds
in a global climate model: Description and single-column model tests. Journal of
Geophysical Research: Atmospheres, 116:D02201.
Tabor, C. R., Otto-Bliesner, B. L., Brady, E. C., Nusbaumer, J., Zhu, J., Erb, M. P.,
Wong, T. E., Liu, Z., and Noone, D. (2018). Interpreting precession driven δ
18
O vari-
ability in the South Asian monsoon region. Journal of Geophysical Research: Atmo-
spheres, 123(5927-5946).
Tan, M. (2014). Circulation effect: response of precipitation δ
18
O to the ENSO cycle in
monsoon regions of China. Climate Dynamics, 42(3-4):1067–1077.
Tharammal, T., Bala, G., andNoone, D.(2017). Impactofdeepconvectionontheisotopic
amount effect in tropical precipitation. Journal of Geophysical Research: Atmospheres,
122(3):1505–1523.
Tian, L., Yao, T., MacClune, K., White, J., Schilla, A., Vaughn, B., Vachon, R., and
Ichiyanagi, K. (2007). Stable isotopic variations in west China: A consideration of
moisture sources. Journal of Geophysical Research: Atmospheres, 112:D10112.
Tiedtke, M. (1989). A comprehensive mass flux scheme for cumulus parameterization in
large-scale models. Monthly Weather Review, 117(8):1779–1800.
Tindall, J., Valdes, P., and Sime, L. C. (2009). Stable water isotopes in HadCM3: Isotopic
signature of El Niño–Southern Oscillation and the tropical amount effect. Journal of
Geophysical Research: Atmospheres, 114:D04111.
Wang, Y., Cheng, H., Edwards, R. L., Kong, X., Shao, X., Chen, S., Wu, J., Jiang, X.,
Wang, X., and An, Z. (2008). Millennial-and orbital-scale changes in the East Asian
monsoon over the past 224,000 years. Nature, 451(7182):1090–1093.
80
Wang, Y.-J., Cheng, H., Edwards, R. L., An, Z., Wu, J., Shen, C.-C., and Dorale, J. A.
(2001). A high-resolution absolute-dated late Pleistocene monsoon record from Hulu
Cave, China. Science, 294(5550):2345–2348.
Yadava, M., Ramesh, R., and Pant, G. (2004). Past monsoon rainfall variations in penin-
sular India recorded in a 331-year-old speleothem. The Holocene, 14(4):517–524.
Yang, B., Qian, Y., Lin, G., Leung, L. R., Rasch, P. J., Zhang, G. J., McFarlane, S. A.,
Zhao, C., Zhang, Y., Wang, H., et al. (2013). Uncertainty quantification and parame-
ter tuning in the CAM5 Zhang-McFarlane convection scheme and impact of improved
convection on the global circulation and climate. Journal of Geophysical Research:
Atmospheres, 118(2):395–415.
Yoshimura, K., Kanamitsu, M., Noone, D., and Oki, T. (2008). Historical isotope simula-
tionusingReanalysisatmosphericdata. Journal of Geophysical Research: Atmospheres,
113:D19108.
Yuan, D., Cheng, H., Edwards, R. L., Dykoski, C. A., Kelly, M. J., Zhang, M., Qing, J.,
Lin, Y., Wang, Y., Wu, J., et al. (2004). Timing, duration, and transitions of the last
interglacial Asian monsoon. Science, 304(5670):575–578.
Zhang, G. J. and McFarlane, N. A. (1995). Sensitivity of climate simulations to the
parameterization of cumulus convection in the canadian climate centre general circula-
tion model. Atmosphere-ocean, 33(3):407–446.
Zhang, W., Wu, J., Wang, Y., Wang, Y., Cheng, H., Kong, X., and Duan, F. (2014). A
detailed East Asian monsoon history surrounding the ‘Mystery Interval’derived from
three Chinese speleothem records. Quaternary Research, 82(1):154–163.
81
Chapter 3
Deciphering Chinese speleothems with an
isotope-enabled climate model
Abstract
Speleothem δ
18
O is widely used to reconstruct past hydroclimate variability, partic-
ularly over Asia. While this proxy is traditionally interpreted as “monsoon intensity”,
recent work has proposed alternative interpretations that challenge or redefine this long-
held concept. To better understand the signal preserved in speleothems over various
timescales, this study employs a state-of-the-art isotope-enabled climate model to quan-
tify contributions to the oxygen isotope composition of precipitation (δ
18
O
P
) over China.
Results suggest that orbital-scale speleothem δ
18
O variations at Chinese sites mainly
record the meridional migration of the Asian monsoon circulation, accompanied by an
early northward movement of the East Asian rain belt. At interannual scales, Chinese
speleothem δ
18
O is also tied to the intensity of monsoonal circulation, via a change in
moisture source locations: enhanced moisture delivery from remote source regions leads
to depleted δ
18
O
P
, particularly in late summer and early autumn. Our results offer a re-
interpretation of the concept of “monsoon intensity” as “enhanced monsoonal circulation”
rather than precipitation amount.
Publication Details: Hu, J., Emile-Geay, J., Tabor, C., Nusbaumer, J., Partin, J., and Adkins, J. (in
preparation). DecipheringChinesespeleothems with an isotope-enabled climate model. Paleoceanography
and Paleoclimatology.
82
3.1 Introduction
Speleothemδ
18
O is widely used to reconstruct past climate due to its high dating pre-
cision and temporal resolution (McDermott, 2004). Since the pioneering study in Hulu
cave by Wang et al. (2001), which for the first time revealed a well-dated terrestrial paleo-
climate record in the monsoon region and showed its coherent variability with Greenland
ice-cores records, Chinese speleothems have been in the spotlight of the paleoclimate com-
munity(e.g. Yuanetal.,2004,Dykoskietal.,2005,Cosfordetal.,2008,Wangetal.,2008,
Hu et al., 2008, Cheng et al., 2009, 2012, 2016). Several of Chinese speleothem records
are continuous, dating back to 600 kyrs ago, coinciding well with summer local insolation
and dominated by the precessional cycles (23 kyrs) (Wang et al., 2008, Cheng et al., 2009,
2012, 2016). These records are traditionally thought to represent “monsoon intensity”
because seasonal precipitation dominates the annual precipitation in these regions. Based
on the “amount effect” (Dansgaard, 1964), the precipitation δ
18
O there should represent
the monsoon precipitation amount (Wang et al., 2001, Yuan et al., 2004, Hu et al., 2008,
Cheng et al., 2009, 2012). This assumes that the “amount effect” is dominant, yet recent
studies have cast doubt on whether the “amount effect” can governδ
18
O
P
variability over
interannual or longer time scales (Dayem et al., 2010, Eastoe and Dettman, 2016). It has
been shown that upstream water vapor δ
18
O (Pausata et al., 2011), changes in moisture
sources or atmospheric circulation pathways (Maher and Thompson, 2012, Tan, 2014,
Tabor et al., 2018), convective activity (Kurita et al., 2011, Aggarwal et al., 2016), and
karst system processes (Baker et al., 2013) also play important roles. Thus it is necessary
to quantify each contribution to better interpret Chinese speleothem δ
18
O.
Despite this complexity, many studies have noted that speleothem δ
18
O at orbital
scales is extremely coherent across Asia, closely tracking local summer insolation (Cheng
et al., 2012, Battisti et al., 2014, Cheng et al., 2016). However, this coherency vanishes
at scales shorter than millennial (Wan et al., 2011, Chu et al., 2012, Li et al., 2014).
This raises the question about what governs Asian speleothem δ
18
O over various spatial
and temporal scales. A better interpretation of this quantity should lead to a better
understanding and improved projections of hydroclimate variability in Asia, which affects
the water supply over 4 billion people.
83
A prerequisite to understanding the isotopic composition of speleothem calcite is to
understand the isotopic composition of its primary input, precipitation δ
18
O (hence-
forth,δ
18
O
P
). Numerical simulations by isotope-enabled climate models are one of major
approaches to studying the interpretation ofδ
18
O
P
, providing a physically-grounded sim-
ulation of its relation to other climatic variables. Using such a tool, Pausata et al. (2011)
simulated a Heinrich event and found that Chinese speleothem δ
18
O is actually deter-
mined by the amount of precipitation over India, thus introducting the “upstream effect”
and challenging the traditional “monsoon intensity” view. Later, Liu et al. (2014) con-
ducted a time-slice simulation of the last 21,000 years, and reconciled these two opinions,
claiming that Chinese speleothem δ
18
O does represent the “monsoon intensity”, with
“intensity” mainly referring to southerly winds over China, working synergistically with
upstream depletion to drive Chinese speleothemδ
18
O. Over interannual time scales, Yang
et al. (2016) emphasized the role of upstream depletion, monsoon intensity and El Niño-
Southern Oscillation (ENSO) by using a simulation of IsoGSM (Yoshimura et al., 2008).
However, these simulations do not disentangle different processes, such as water vapor
transport and local rainfall amount, which influence δ
18
O
P
. A recent version of iCESM
(Brady et al., tted) is capable of tracking moisture and water isotopes by tagging water
vapor evaporating from a specific region, which is key to quantifying the contributions of
different factors to δ
18
O
P
. This approach also has advantages over the back-trajectory
method since the latter requires high-frequency measurements of circulation and mois-
ture fields to constrain errors and avoid artificial mixing (Stohl et al., 2004, Singh et al.,
2016a). Tabor et al. (2018) utilized this model and found that the change of remote mois-
ture sources played a dominant role in the orbital variation of Indian speleothemδ
18
O. In
this study, we will employ iCESM to investigate the interpretation of speleothem δ
18
O in
China at both orbital and interannual scales.
Another perspective which helps us understand the relationship between Chinese
speleothem δ
18
O and the Asian monsoon is the insight from modern monsoon studies.
Chinese records reside in the East Asian monsoon region, which is affected by both trop-
ical and mid-latitude circulations (Lau et al., 1988, Ding and Chan, 2005), unlike other
major monsoon systems such as the Indian monsoon and the west African monsoon. The
concept of “monsoon intensity” is vague since the strength of the monsoon-related wind
84
and precipitation are decoupled in East Asia: a strong monsoon wind does not imply
intense precipitation, or vice versa. Modern observations reveal a “triple-mode” of the
East Asian precipitation at the interannual scale (Fig. 1 in Zhang et al. (2018)) (Day
et al., 2015). In this mode, strong monsoon winds are associated with low precipitation
in central China. Also the East Asian monsoon has a unique subseasonal feature that the
rainbelt does not gradually move northward in summer but abruptly “jumps”. The main
rainbelt from mid-June to mid-July is called “Meiyu”, extending from central-eastern
China to Japan (where Sanbao and Hulu caves are located). After July, the rainbelt
jumps to northern China, associated with the jump of the westerly jet. If the rainbelt
stays in central China longer and jumps later, the rainy season there will be longer and
yield more precipitation, which may affect δ
18
O
P
. This unique seasonal change must be
considered when interpreting Chinese speleothem δ
18
O.
The paper is structured as follows: we first review the coherency of Asian speleothem
records in section 2, and introduce the model and numerical experiments in section 3.
Section 4 investigates the interpretations of δ
18
O
P
in China at orbital and interannual
scales. Section 5 discusses the simulation biases and implications for other time scales,
and section 6 provides our conclusions.
3.2 Coherency analysis of Chinese speleothem
δ
18
O records
In this section we review the coherency of Asian speleothems over orbital to decadal
scales. The data used are from the National Centers for Environmental Informa-
tion (NCEI) (https://www.ncdc.noaa.gov/data-access/paleoclimatology-data/
datasets/speleothem) with four speleothem records covering the last 400,000 yrs and
seven records cover the last 11,000 years. Fig. 4.1 shows the locations of these speleothem
records.
Fig. 3.2 shows the orbital variability of four speleothem δ
18
O records in China. Their
variability is very coherent, and dominated by precession (Battisti et al., 2014, Cheng
et al., 2016, Kathayat et al., 2016, Carolin et al., 2016), and correlate well with local
summer insolation. This coherency implies that Chinese speleothemδ
18
O at orbital scales
85
20°N
40°N
90°E 120°E
Dongge
Heshang
Sanbao
Jiuxian
Lianhua
Tianmen
Hulu
Xiaobailong
Figure 3.1: The spatial distribution of the Chinese speleothem records used in
this section
should represent a climate phenomenon at large spatial scales, and the prevailing inference
is the “monsoon intensity” (Cheng et al., 2016). We will examine this interpreation using
an isotope-enabled model in the next sections.
Year (kyBP)
10
5
18
O ( )
Sanbao/Hulu/Dongge
0 50 100 150 200 250 300 350 400
Year (kyBP)
15
10
18
O ( )
Xiaobailong
450
500
450
500
Local Insolation
(W m
2
)
Figure 3.2: Chinese speleothem δ
18
O over the last 400,000 yrs with their local
summer insolation
Recent studies have used speleothem δ
18
O to study hydroclimate variability at
interannual-centennial scales (Sinha et al., 2011, 2015, Tan et al., 2011, 2015, Chen et al.,
2016, Yang et al., 2016) exploiting the high temporal resolution of speleothem data. Fig.
3.3 shows Asian speleothem δ
18
O over the Holocene. All records show a trend towards
86
isotopic enrichment from 8,000 BP onwards, following the trend of their local summer
insolation. But if we zoom in time scales shorter than millennial, the coherency van-
ishes. This can be expected as the amplitude of speleothemδ
18
O variations at these time
scales is relatively small (2-3h, compared with 5-6h at orbital scales), and the regional
climate variability, karst system processes (Baker and Bradley, 2010), and age uncertain-
ties may bring in changes of speleothem δ
18
O with this amplitude. We will employ an
isotope-enabled model to explain the interannual variability of speleothem δ
18
O in Asia.
Year (BP)
9
8
7
18
O ( )
Dongge
Year (BP)
10
8
18
O ( )
Heshang
Year (BP)
10
8
18
O ( )
Jiuxian
Year (BP)
10
9
18
O ( )
Sanbao
Year (BP)
6
4
2
18
O ( )
Lianhua
0 2000 4000 6000 8000 10000
Year (BP)
20
15
18
O ( )
Tianmen
460
480
460
480
460
480
Local Insolation
(W m
2
)
460
480
460
480
460
480
Figure 3.3: Chinese speleothem δ
18
O over Holocene with their local summer
insolation
3.3 Experimental design
We employ the isotope-enabled version of the National Center for Atmospheric
Research’s Community Earth System Model Version 1 (iCESM1) (Brady et al., tted).
87
It is a state-of-the-art coupled climate model implementing water isotopes, which is com-
posed of the atmospheric model iCAM5 (Nusbaumer et al., 2017), the ocean model iPOP2
(Zhang et al., 2017), the land model iCLM4 (Wong et al., 2017), the sea ice model iCICE4,
the river transport model iRTM and the CESM Coupler. The atmosphere model we use
has a horizontal resolution of 1.875
◦
×2.5
◦
and 30 vertical levels. iCESM can simulate
the variability of δ
18
O
P
and precipitation in the Asian monsoon region with high fidelity
(Nusbaumer et al., 2017, Tabor et al., 2018, Brady et al., tted). It is the current best
isotope-enabled model in simulating the relationship betweenδ
18
O
P
and convective activ-
ity (Hu et al., 2018). Also, this version of iCESM has been modified to track moisture
and water isotopes for specific regions by tagging water vapor once it evaporates from
the surface. The tagged water vapor can be advected and goes through phase changes
and it leaves the atmosphere as precipitation. There are already several studies using this
water-tagging technique to track moisture and water isotopes (Singh et al., 2016b, Dyer
et al., 2017, Nusbaumer and Noone, 2018, Tabor et al., 2018).
For the study of the seasonal and interannual variability of speleothem δ
18
O, we con-
ducted an experiment with prescribed sea surface temperature and sea ice observations
from 1953-2012, and tag the water vapor originating from 32 regions shown in Fig. 3.4.
These regions are chosen because their total contribution of moisture contributes to 95%
of precipitation in Asia.
δ
18
O
P
at one grid cell is the sum ofδ
18
O
P
originating from all tagged regions, weighted
by their precipitation contribution:
δ
18
O
P
=
32
X
i=1
δ
18
O
P
sink
i
×
p
i
p
total
(3.1)
where i indexes a tagged region, δ
18
O
P
sink
i
and p
i
are δ
18
O
P
and precipitation falling
at this grid cell (we call this cell “sink”) whose water vapor originates from the i
th
tagged
region, respectively. p
total
is the total precipitation at this grid cell. Tabor et al. (2018)
developed a framework to decompose the change of δ
18
O
p
since δ
18
O
P
sink
i
can be seen as
the sum of water vapor δ
18
O at source location, change of water vapor δ
18
O between the
sink and source, and the difference between δ
18
O
P
and water vapor δ
18
O at the sink:
88
(δ
18
O
P
)
i
= (δ
18
O
P
sink
−δ
18
O
wv
sink
)
i
+ (δ
18
O
wv
sink
−δ
18
O
wvsource
)
i
+ (δ
18
O
wvsource
)
i
(3.2)
Thus the change of precipitation between two climate states can be decomposed into
four terms:
Δ(δ
18
O
P
)
i
= Δ

δ
18
O
P
sink
i
×
p
i
p
total
!
= Δ
h
(δ
18
O
P
sink
−δ
18
O
wv
sink
)
i
| {z }
Condensation
+ (δ
18
O
wv
sink
−δ
18
O
wvsource
)
i
| {z }
Rainout
+ (δ
18
O
wvsource
)
i
| {z }
Source composition
i
×

p
i
p
total
!
+ δ
18
O
P
sink
i
× Δ

p
i
p
total
!
| {z }
Source location changes
(3.3)
The first three terms are related the change of δ
18
O
p
at the sink. This first term
is about the change due to phase changes at the sink, which is about the change of
condensation, so we call this term “condensation changes”. The second term is about
the change of the difference between moisture source and sink, and this is generally due
to the change in travel distance and/or upstream rainout, so we call this term “rainout
changes”. The third term tracks changes due to water vapor δ
18
O at the moisture source,
so we call this term “source composition changes”. The last term arises because of the
change of precipitation contribution from each tagged region, which describes the change
of moisture source location, so we will call this term “source location change”.
For orbital variability, we analyzed the orbital forcing experiments of Tabor et al.
(2018). There are four simulations with different precession angles with the maximum
eccentricityvalueofthepast900kyrs(0.0493). Thefouranglesare: NorthernHemisphere
perihelion at autumnal equinox, winter solstice, vernal equinox, and summer solstice. The
simulations were branched from equilibrium simulations and run for 550 years and the
analyses come from the average of the last 50 years.
89
30°E 60°E 90°E 120°E 150°E 180° 150°W 120°W 90°W
20°S
0°
20°N
40°N
Figure 3.4: Tagged regions in the iCESM simulation from 1953-2012
3.4 Results
3.4.1 Orbital variability
We analyzed the experiments in the same configurations of precession minimum
(Northern Hemisphere perihelion at summer solstice, henceforth, P
min
) and precession
maximum (Northern Hemisphere perihelion at winter solstice, henceforth, P
max
) as
Tabor et al. (2018). Fig. 3.5a shows the difference of climatological weighted annual
δ
18
O
P
between P
min
and P
max
. It indicates that iCESM can simulate more depleted
weighted annual δ
18
O
P
in P
min
in the central and southern China as we expect based
on speleothem records, but the amplitude is weaker (-1.6h, compared with∼-4–5h in
speleothem δ
18
O). The negative anomaly only extends to 30
◦
N, and we will discuss the
possible reasons for this bias in Section 5.
90
Figure 3.5: Difference between precession minimum and maximum. (a) Weighted
annual mean precipitation difference; (b) JJAδ
18
O
P
and 850 hPa wind difference; (c) JJA
precipitation difference
91
The difference in circulation andδ
18
O
P
anomaly of the monsoon season (JJA) between
the P
min
and P
max
isshowninFig. 3.5b. Itgenerallydisplaysastrengtheningofthenorth-
ern part of the Asian summer monsoon low-level circulation, which indicates a northward
movement of both the Indian and East Asian summer monsoon. It is also noticeable that
the subtropical high over the North Pacific also moves northward. Thus both the sub-
tropical and monsoon systems move northward when the Northern Hemisphere receives
more summer insolation. However, this northward migration does not translate into a
uniform increase in precipitation. Fig. 3.5c shows that JJA precipitation in central China
and Japan is less in P
min
, while precipitation in the northern and southern China, India
and west Asia is larger. This pattern is similar to the leading mode of interannual precip-
itation variability in East Asia (Day et al., 2015). Speleothem trace element ratios also
suggest that depleted speleothemδ
18
O in central China can coincide with dryness instead
of wetness in that region (Zhang et al., 2018). This, however, is not enough to reach
a conclusion regarding central China receiving less precipitation in P
min
, since similar
simulations with other models like EC-Earth and GFDL-CM2.1 do not clearly show less
precipitation in central China (Bosmans et al., 2018). These models do show a northward
movement of the summer rainfall belt in East China. In short, theδ
18
O
P
change in China
at orbitals scales represents the northward extension of the East Asian monsoon circu-
lation, not the monsoon precipitation. If we have to use the term “monsoon intensity”
to interpret δ
18
O records, the intensity should refer to the intensity of monsoon winds,
and furthermore the monsoon circulation does not strengthen at its climatological loca-
tions, but moves northward. This is consistent with the conclusion made by Liu et al.
(2014), which states that the δ
18
O records represent enhanced southerly monsoon winds,
though we also point out the northward migration of the monsoon winds together with
its strengthening.
The northward extension of the monsoon circulation is accompanied by a change in
seasonality. The annual cycle of precipitation in East China (Fig. 3.6a and 3.6b) shows
that in P
min
the precipitation maximum occurs in May rather than June. This means
that the rainy season in East China starts one month earlier in P
min
than in P
max
. The
Hovmoller diagram of East Asian precipitation (Fig. 3.6c and 3.6d) shows that in P
min
the rainbelt builds in May around 30
◦
N, and then moves northward quickly to 40
◦
N in
92
June, while in P
max
the rainbelt moves northward slower, reaching 35
◦
N in August. These
results point out that the rainy season in central China, home to many famous cave sites,
comes one month earlier and its duration is shortened, which leads to less precipitation.
This is consistent with the simulation results by Kong et al. (2017), which shows an earlier
onset and a shortening of the rainy season of central China in the early Holocene when
the northern Hemisphere received more insolation in summer, comparing with the late
Holocene. Kong et al. (2017) propose that the insolation increase in summer reduces the
pole-equator temperature gradient, which lead to the earlier northward shift of westerlies,
causing the early onset and shortening of the rainy season in central China.
93
Figure3.6: ClimatologicalmonthlymeanprecipitationinEastChina (20
◦
N-35
◦
N,
105
◦
E-120
◦
E) in precession minimum (a) and precession maximum (b). Hovmoller dia-
gram of climatological monthly mean precipitation in East China in precession minimum
(c) and precession maximum (d).
Seasonal differences inδ
18
O
P
may be used an analog for precessional variations, since
the impacts of precessional forcing can be viewed as amplified summer-winter contrasts.
Thus we analyze the difference between boreal summer and winter in our AMIP-like
water-tagging experiment. With the decomposition framework, we separate the contribu-
tions from moisture source composition, rainout processes, condensation processes, and
94
moisture source location. Fig. 3.7 shows that the change of moisture source location con-
tributes most to the depleted δ
18
O
P
of East China in summer. It contributes -5.06h to
the total seasonal difference of -2.08h. The source location factor contributes to depleted
δ
18
O because more moisture comes from the Indian Ocean which makes δ
18
O
P
in China
more negative (Fig. 3.7d). If this also applies to orbital variability, this result implies
that depleted speleothem δ
18
O in China is controlled by both condensation changes and
moisture source location changes. For central China, since the model result suggests less
precipitation in P
min
, the moisture source location changes may play a more significant
role – more moisture comes from the Indian Ocean and less moisture comes from the
Northwestern Pacific.
(a) Source composition changes (b) Rainout changes
(d) Source location changes (c) Condensation changes
Figure 3.7: Contributions of four factors to seasonal variability of δ
18
O
P
over
East China from tagged regions: (a) moisture source composition, (b) Rainout changes,
(c) Condensation changes, and (d) moisture location change .
3.4.2 Interannual variability
In addition to studies about orbital variability of Chinese speleothem δ
18
O , there
are also many studies about the interannual-decadal variability of Chinese speleothem
95
records (Tan et al., 2011, 2015, Chen et al., 2016, Yang et al., 2016). The variability
of these speleothems is interpreted as monsoon intensity, ENSO, and moisture source
changes. We now explore the interpretation of speleothem δ
18
O at interannual scales.
Here we discuss what factors control the interannual variability ofδ
18
O
P
in East China
based on the AMIP-like water-tagging experiment. We spatially average the precipitation
weighted annual δ
18
O
P
in East China (red box in Fig 3.9) to generate a time series of
East China δ
18
O
P
(Fig. 3.8a). We select high δ
18
O years if their values are above one
standard deviation from the mean value, and select low δ
18
O years if their values are
below one standard deviation from the mean. We make composites of high δ
18
O years
and lowδ
18
O years. The difference of the composites are decomposed into four factors by
the previous framework. If we sum up the contribution from different tagging regions, we
obtain the contributions shown in Fig. 3.8b.
96
Source composition change
Rainout change
Condensation change
Source location change
Figure3.8: Interannualvariabilityofδ
18
O
P
inEastChina(a)Timeseriesofweighted
annual meanδ
18
O
P
in East China; (b) Seasonal contributions of four factors to weighted
annual δ
18
O
P
in East China
The result shows that, on interannual scales, δ
18
O
P
in East China is dominated by
moisture source location changes, and its main contribution is in August to October. This
suggests that the late monsoon season is likely to contribute more than other months to
the monthly-weighted δ
18
O signal, and, therefore the speleothem δ
18
O signal. When
97
unfolding the contribution from the moisture source location changes, we can see that the
Indian Ocean contributes most to positive δ
18
O anomalies (Fig. 3.9a). This is because
more moisture comes from the nearby land and ocean such as the northwestern Pacific
and the Bay of Bengal, but less moisture comes from the remote Indian Ocean. The water
vapor δ
18
O originating from remote places is generally more depleted than from nearby
regions. This result is similar to that of Tabor et al. (2018) about the orbital variability of
the Indian Monsoon in that the moisture source changes mainly controlδ
18
O
P
variability.
30°E 60°E 90°E 120°E 150°E 180° 150°W 120°W 90°W
20°S
0°
20°N
40°N
(a) Contribution of moisture source location changes ( )
0.4 0.3 0.2 0.1 0.0 0.1 0.2 0.3 0.4
2 m/s
30°E 60°E 90°E 120°E 150°E 180° 150°W 120°W 90°W
20°S
0°
20°N
40°N
(b) ASO precipitation (mm/day) and 850 hPa wind difference (m/s)
8 6 4 2 0 2 4 6 8
Figure 3.9: Interpretation of interannual variability of Chinese δ
18
O
P
(a) Contri-
bution of moisture source location changes to East China δ
18
O
P
from tagged regions; (b)
Difference of 850 hPa wind and precipitation in ASO between high and low δ
18
O
P
years
of East China. The red rectangular is the defined East China region (20
◦
N-35
◦
N, 105
◦
E-
120
◦
E).
98
The difference of atmospheric circulation and precipitation between high and low
δ
18
O
P
years is shown in Fig. 3.9b. The low-level circulation anomaly in high δ
18
O years
is akin to a weakening of monsoon winds across South and East Asia, therefore bringing
less moisture from the Indian Ocean. We can see that there are anticyclonic anomalies in
the Bay of Bengal and East China Sea. These anomalies lead to more moisture coming
from the Bay of Bengal and the northwestern Pacific. There is indeed less precipitation
in East China in high δ
18
O years, but the amplitude is too small to explain the positive
δ
18
O
P
anomalies.
In summary, the interannual variability of δ
18
O
P
in East China is mainly controlled
by moisture source changes – in years of highδ
18
O
P
, less moisture comes from the Indian
Ocean and more moisture from the northwestern Pacific, and high δ
18
O
P
represents a
weakening of the Asian monsoon winds.
3.5 Discussion
We argue that the variation of Chinese speleothemδ
18
O at orbital scales can be viewed
as the meridional migration of the monsoon winds. The mean latitude of the convergence
maxima of low-level monsoon winds over the tropical Indian Ocean (10
◦
S-10
◦
N, 35
◦
E-
60
◦
E) moves 4
◦
northward compared to the modern climatology. And the water-tagging
experiment result suggests that the change of Chinese speleothem δ
18
O at orbital scales
is largely due to the change of moisture source location.
We notice that at orbital scales, iCESM simulates a relatively weakδ
18
O
P
response to
precession forcing in East China. Based on Fig. 3.7, the moisture source location change
dominates the variation of Chinese δ
18
O
P
at orbital scales, so the bias of δ
18
O
P
can
be partly explained by the discrepancy of the model in simulating the moisture source
location change responding to precessional forcing. Fig. 3.5 shows the north Pacific sub-
tropical high strengthens together with the Asian monsoon circulation, and more moisture
from there makes Chineseδ
18
O
P
more positive since it is closer than the Indian Ocean. It
is possible that the subtropical high response is more intense than it should be in P
min
, so
its positiveδ
18
O
P
contribution dampens theδ
18
O
P
response to precession forcing in East
99
China. This weak response of δ
18
O
P
in East Asia seems common in isotope-enabled cli-
mate models such as ECHAM4.6 (Battisti et al., 2014, Roe et al., 2016) and iLOVECLIM
(Caley et al., 2014) and needs further investigation.
Our results are based on the numerical experiments, but comparing speleothem
δ
18
O with other proxy data will also provide additional evidence for the interpretation of
speleothem δ
18
O. For example, Zhang et al. (2018) compared a speleothem δ
18
O record
with a speleothem trace metal record, which is believed to represent local precipitation
and revealed the decoupling of precipitation and speleothem δ
18
O signals.
We mainly discussed the interpretation of Asian speleothemδ
18
O at orbital and inter-
annual time scales, but did not cover the time scales in between. Asian speleothem
δ
18
O are coherent at millennial scales and follow the local insolation variations (Cheng
et al., 2012, 2016), so we believe its explanations will be similar with that at orbital
scales, which is the migration of the monsoon circulation system. However, the change
of regional precipitation patterns may not be the same as the orbital variations. For
example, Pausata et al. (2011) simulated climate during the Heinrich events and found no
obvious precipitation changes in China, while our study and other simulations (Bosmans
et al., 2018) for orbital forcing show significant precipitation changes in China. There
are already studies showing the coherency of Asian speleothem δ
18
O vanishes at scales
shorter than millennial (Wan et al., 2011, Chu et al., 2012, Li et al., 2014, Hu et al., tion),
and further investigations are needed to identify at which time scales the coherency disap-
pears. Whether the external forcing or the internal variability of atmosphere-land-ocean
system dominates also needs further investigations by coupled climate models. For the
interpretation of interannual variability of speleothem δ
18
O , we mainly focus on δ
18
O
P
,
and future work needs to be done to investigate to what extent the karst system processes
and age uncertainty influence the heterogeneity of Asian speleothem records.
3.6 Conclusion
We investigated the interpretation of Chinese speleothemδ
18
O via the isotope-enabled
model iCESM. The results show that Chinese speleothemδ
18
O variations at orbital scales
mainly represent the meridional migration of the East Asian monsoon circulation. This
100
is also accompanied by the early northward movement of the East Asian rainbelt and
late retreat of the Indian monsoon precipitation. The interannual variability of Chinese
speleothem δ
18
O is mainly controlled by the change of moisture source locations. More
moisture coming from remote regions leads to depleted speleothemδ
18
O, and its influence
is greatest in late summer and early autumn. We suggest using the term “monsoon
winds” or “monsoon circulation” rather than “monsoon intensity” to interpret speleothem
δ
18
O over those regions.
Bibliography
Aggarwal, P. K., Romatschke, U., Araguas-Araguas, L., Belachew, D., Longstaffe, F. J.,
Berg, P., Schumacher, C., andFunk, A.(2016). Proportionsofconvectiveandstratiform
precipitation revealed in water isotope ratios. Nature Geoscience, 9(8):624.
Baker, A. and Bradley, C. (2010). Modern stalagmiteδ
18
O: Instrumental calibration and
forward modelling. Global and Planetary Change, 71(3):201–206.
Baker, A., Bradley, C., and Phipps, S. J. (2013). Hydrological modeling of stalag-
mite δ
18
O response to glacial-interglacial transitions. Geophysical Research Letters,
40(12):3207–3212.
Battisti, D., Ding, Q., and Roe, G. (2014). Coherent pan-Asian climatic and isotopic
response to orbital forcing of tropical insolation. Journal of Geophysical Research:
Atmospheres, 119(21):11–997.
Bosmans, J., Erb, M., Dolan, A., Drijfhout, S., Tuenter, E., Hilgen, F., Edge, D., Pope, J.,
andLourens, L.(2018). ResponseoftheAsiansummermonsoonstoidealizedprecession
and obliquity forcing in a set of GCMs. Quaternary Science Reviews, 188:121–135.
Brady, E., Stevenson, S., Bailey, D., Liu, Z., Noone, D., Nusbaumer, J., Otto-Bliesner,
B. L., Tabor, C., Tomas, R., Wong, T., Zhang, J., and Zhu, J. (submitted). The con-
nected isotopic water cycle in the Community Earth System Model version 1. Journal
of Advances in Modeling Earth Systems.
101
Caley, T., Roche, D. M., and Renssen, H. (2014). Orbital Asian summer monsoon dynam-
ics revealed using an isotope-enabled global climate model. Nature communications, 5.
Carolin, S. A., Cobb, K. M., Lynch-Stieglitz, J., Moerman, J. W., Partin, J. W., Lejau,
S., Malang, J., Clark, B., Tuen, A. A., and Adkins, J. F. (2016). Northern Borneo
stalagmite records reveal West Pacific hydroclimate across MIS 5 and 6. Earth and
Planetary Science Letters, 439:182–193.
Chen, S., Hoffmann, S. S., Lund, D. C., Cobb, K. M., Emile-Geay, J., and Adkins,
J. F. (2016). A high-resolution speleothem record of western equatorial Pacific rain-
fall: Implications for Holocene ENSO evolution. Earth and Planetary Science Letters,
442:61–71.
Cheng, H., Edwards, R. L., Broecker, W. S., Denton, G. H., Kong, X., Wang, Y., Zhang,
R., and Wang, X. (2009). Ice Age Terminations. Science, 326(5950):248–252.
Cheng, H., Edwards, R. L., Sinha, A., Spötl, C., Yi, L., Chen, S., Kelly, M., Kathayat,
G., Wang, X., Li, X., et al. (2016). The Asian monsoon over the past 640,000 years
and ice age terminations. Nature, 534(7609):640–646.
Cheng, H., Sinha, A., Wang, X., Cruz, F. W., and Edwards, R. L. (2012). The Global
PaleomonsoonasseenthroughspeleothemrecordsfromAsiaandtheAmericas. Climate
Dynamics, 39(5):1045–1062.
Chu, P.C., Li, H.-C., Fan, C., andChen, Y.-H.(2012). Speleothemevidencefortemporal–
spatial variation in the East Asian Summer Monsoon since the Medieval Warm Period.
Journal of Quaternary Science, 27(9):901–910.
Cosford, J., Qing, H., Yuan, D., Zhang, M., Holmden, C., Patterson, W., and Hai, C.
(2008). Millennial-scalevariabilityintheAsianmonsoon: Evidencefromoxygenisotope
records from stalagmites in southeastern China. Palaeogeography, Palaeoclimatology,
Palaeoecology, 266(1):3–12.
Dansgaard, W. (1964). Stable isotopes in precipitation. Tellus, 16(4):436–468.
Day, J. A., Fung, I., and Risi, C. (2015). Coupling of South and East Asian Monsoon
Precipitation in July–August. Journal of Climate, 28(11):4330–4356.
102
Dayem, K. E., Molnar, P., Battisti, D. S., and Roe, G. H. (2010). Lessons learned from
oxygen isotopes in modern precipitation applied to interpretation of speleothem records
of paleoclimate from Eastern Asia. Earth and Planetary Science Letters, 295(1):219–
230.
Ding, Y. and Chan, J. C. (2005). The East Asian summer monsoon: an overview. Mete-
orology and Atmospheric Physics, 89(1-4):117–142.
Dyer, E. L., Jones, D. B., Nusbaumer, J., Li, H., Collins, O., Vettoretti, G., and Noone,
D. (2017). Congo Basin precipitation: Assessing seasonality, regional interactions, and
sourcesof moisture. Journal of Geophysical Research: Atmospheres, 122(13):6882–6898.
Dykoski, C. A., Edwards, R. L., Cheng, H., Yuan, D., Cai, Y., Zhang, M., Lin, Y., Qing,
J., An, Z., and Revenaugh, J. (2005). A high-resolution, absolute-dated Holocene and
deglacialAsianmonsoonrecordfromDonggeCave, China. Earth and Planetary Science
Letters, 233(1):71–86.
Eastoe, C. and Dettman, D. (2016). Isotope amount effects in hydrologic and climate
reconstructions of monsoon climates: Implications of some long-term data sets for pre-
cipitation. Chemical Geology, 430:78–89.
Hu, C., Henderson, G. M., Huang, J., Xie, S., Sun, Y., and Johnson, K. R. (2008). Quan-
tification of Holocene Asian monsoon rainfall from spatially separated cave records.
Earth and Planetary Science Letters, 266(3):221–232.
Hu, J., Emile-Geay, J., McKay, N., Kathayat, G., Brahim, Y. A., and Goldsmith, Y.
(in preparation). Limited coherency of Asian speleothems over the Holocene, with
implications for the Meghalayan age. Geophysical Research Letters.
Hu, J., Emile-Geay, J., Nusbaumer, J., and Noone, D. (2018). Impact of convective
activity on precipitationδ
18
O in isotope-enabled general circulation models. Journal of
Geophysical Research: Atmospheres, 123(23):13–595.
Kathayat, G., Cheng, H., Sinha, A., Spötl, C., Edwards, R. L., Zhang, H., Li, X., Yi,
L., Ning, Y., Cai, Y., et al. (2016). Indian monsoon variability on millennial-orbital
timescales. Scientific Reports, 6.
103
Kong, W., Swenson, L.M., andChiang, J.C.(2017). Seasonaltransitionsandthewesterly
jet in the Holocene east Asian summer monsoon. Journal of Climate, 30(9):3343–3365.
Kurita, N., Noone, D., Risi, C., Schmidt, G. A., Yamada, H., and Yoneyama, K. (2011).
Intraseasonal isotopic variation associated with the Madden-Julian Oscillation. Journal
of Geophysical Research: Atmospheres, 116:D24101.
Lau, K., Yang, G., and Shen, S. (1988). Seasonal and intraseasonal climatology of summer
monsoon rainfall over Eeat Asia. Monthly Weather Review, 116(1):18–37.
Li,Y.,Wang,N.,Zhou,X.,Zhang,C.,andWang,Y.(2014). Synchronousorasynchronous
Holocene Indian and East Asian summer monsoon evolution: a synthesis on Holocene
Asian summer monsoon simulations, records and modern monsoon indices. Global and
Planetary Change, 116:30–40.
Liu, Z., Wen, X., Brady, E., Otto-Bliesner, B., Yu, G., Lu, H., Cheng, H., Wang, Y.,
Zheng, W., Ding, Y., et al. (2014). Chinese cave records and the East Asia summer
monsoon. Quaternary Science Reviews, 83:115–128.
Maher, B. A. and Thompson, R. (2012). Oxygen isotopes from Chinese caves: records
not of monsoon rainfall but of circulation regime. Journal of Quaternary Science,
27(6):615–624.
McDermott, F. (2004). Palaeo-climate reconstruction from stable isotope variations in
speleothems: a review. Quaternary Science Reviews, 23(7-8):901–918.
Nusbaumer, J. and Noone, D. (2018). Numerical Evaluation of the Modern and Future
Origins of Atmospheric River Moisture Over the West Coast of the United States.
Journal of Geophysical Research: Atmospheres, 123(12):6423–6442.
Nusbaumer, J., Wong, T. E., Bardeen, C., and Noone, D. (2017). Evaluating hydrological
processes in the Community Atmosphere Model Version 5 (CAM5) using stable isotope
ratios of water. Journal of Advances in Modeling Earth Systems, 9(2):949–977.
Pausata, F. S., Battisti, D. S., Nisancioglu, K. H., and Bitz, C. M. (2011). Chinese sta-
lagmiteδ
18
O controlled by changes in the Indian monsoon during a simulated Heinrich
event. Nature Geoscience, 4(7):474–480.
104
Roe, G. H., Ding, Q., Battisti, D. S., Molnar, P., Clark, M. K., and Garzione, C. N.
(2016). A modeling study of the response of Asian summertime climate to the largest
geologic forcings of the past 50 Ma. Journal of Geophysical Research: Atmospheres,
121(10):5453–5470.
Singh, H., Bitz, C., Nusbaumer, J., and Noone, D. (2016a). A mathematical framework
for analysis of water tracers: Part 1: Development of theory and application to the
preindustrial mean state. Journal of Advances in Modeling Earth Systems, 8(2):991–
1013.
Singh, H. K., Donohoe, A., Bitz, C. M., Nusbaumer, J., and Noone, D. C. (2016b).
Greater aerial moisture transport distances with warming amplify interbasin salinity
contrasts. Geophysical Research Letters, 43(16):8677–8684.
Sinha, A., Berkelhammer, M., Stott, L., Mudelsee, M., Cheng, H., and Biswas, J. (2011).
The leading mode of Indian Summer Monsoon precipitation variability during the last
millennium. Geophysical Research Letters, 38(15):L15703.
Sinha, A., Kathayat, G., Cheng, H., Breitenbach, S. F., Berkelhammer, M., Mudelsee, M.,
Biswas, J., and Edwards, R. L. (2015). Trends and oscillations in the Indian summer
monsoon rainfall over the last two millennia. Nature Communications, 6:ncomms7309.
Stohl, A., Cooper, O., and James, P. (2004). A cautionary note on the use of meteoro-
logical analysis fields for quantifying atmospheric mixing. Journal of the Atmospheric
Sciences, 61(12):1446–1453.
Tabor, C. R., Otto-Bliesner, B. L., Brady, E. C., Nusbaumer, J., Zhu, J., Erb, M. P.,
Wong, T. E., Liu, Z., and Noone, D. (2018). Interpreting precession driven δ
18
O vari-
ability in the South Asian monsoon region. Journal of Geophysical Research: Atmo-
spheres, 123(5927-5946).
Tan, L., Cai, Y., An, Z., Edwards, R. L., Cheng, H., Shen, C.-C., and Zhang, H. (2011).
Centennial-to decadal-scale monsoon precipitation variability in the semi-humid region,
northern China during the last 1860 years: Records from stalagmites in Huangye Cave.
The Holocene, 21(2):287–296.
105
Tan, L., Cai, Y., Cheng, H., Edwards, R. L., Shen, C.-C., Gao, Y., and An, Z. (2015). Cli-
mate significance of speleothemδ
18
O from central China on decadal timescale. Journal
of Asian Earth Sciences, 106:150–155.
Tan, M. (2014). Circulation effect: response of precipitation δ
18
O to the ENSO cycle in
monsoon regions of China. Climate Dynamics, 42(3-4):1067–1077.
Wan, N.-J., Li, H.-C., Liu, Z.-Q., Yang, H.-Y., Yuan, D.-X., and Chen, Y.-H. (2011). Spa-
tialvariationsofmonsoonalrainineasternchina: Instrumental, historicandspeleothem
records. Journal of Asian Earth Sciences, 40(6):1139–1150.
Wang, Y., Cheng, H., Edwards, R. L., Kong, X., Shao, X., Chen, S., Wu, J., Jiang, X.,
Wang, X., and An, Z. (2008). Millennial-and orbital-scale changes in the East Asian
monsoon over the past 224,000 years. Nature, 451(7182):1090–1093.
Wang, Y.-J., Cheng, H., Edwards, R. L., An, Z., Wu, J., Shen, C.-C., and Dorale, J. A.
(2001). A high-resolution absolute-dated late Pleistocene monsoon record from Hulu
Cave, China. Science, 294(5550):2345–2348.
Wong, T. E., Nusbaumer, J., and Noone, D. C. (2017). Evaluation of modeled land-
atmosphere exchanges with a comprehensive water isotope fractionation scheme in ver-
sion 4 of the Community Land Model. Journal of Advances in Modeling Earth Systems,
9(2):978–1001.
Yang, H., Johnson, K., Griffiths, M., and Yoshimura, K. (2016). Interannual controls on
oxygen isotope variability in Asian monsoon precipitation and implications for paleo-
climate reconstructions. Journal of Geophysical Research: Atmospheres, 121(14):8410–
8428.
Yoshimura, K., Kanamitsu, M., Noone, D., and Oki, T. (2008). Historical isotope simula-
tionusingReanalysisatmosphericdata. Journal of Geophysical Research: Atmospheres,
113:D19108.
Yuan, D., Cheng, H., Edwards, R. L., Dykoski, C. A., Kelly, M. J., Zhang, M., Qing, J.,
Lin, Y., Wang, Y., Wu, J., et al. (2004). Timing, duration, and transitions of the last
interglacial Asian monsoon. Science, 304(5670):575–578.
106
Zhang, H., Griffiths, M. L., Chiang, J. C., Kong, W., Wu, S., Atwood, A., Huang, J.,
Cheng, H., Ning, Y., and Xie, S. (2018). East Asian hydroclimate modulated by the
position of the westerlies during Termination I. Science, 362(6414):580–583.
Zhang, J., Liu, Z., Brady, E. C., Oppo, D. W., Clark, P. U., Jahn, A., Marcott, S. A., and
Lindsay, K. (2017). Asynchronous warming and δ
18
O evolution of deep Atlantic water
masses during the last deglaciation. Proceedings of the National Academy of Sciences,
114(42):11075–11080.
107
Chapter 4
LimitedcoherencyofAsianspeleothemsover
the Holocene, with implications for the
Meghalayan age
Abstract
Speleothems are increasingly used to reconstruct paleohydrology, particularly in the
Asian Monsoon region. Speleothems offer uniquely precise chronologies, used to pin-
point geological events: the Global Stratotype Section and Point (GSSP) for the newly
announced Meghalayan age (4.2 ky BP to present) is a stalagmite in Mawmluh cave.
The “4.2k event” is famous for its link to several civilization collapses, though contro-
versy surrounds their causes and detailed timing. Here we use a network of 8 well-dated
series to investigate the coherency of Asian speleothemδ
18
O at sub-orbital scales over the
Holocene. We find no coherent variability among Asian speleothem δ
18
O at these time
scales. This asynchrony can be explained by heterogenous soil, vegetation, and karst pro-
cesses modifying coherent climate inputs. Given the relatively small amplitude (1-3h)
of speleothems at sub-orbital scales, we recommend that extracting hydroclimate signals
from speleothem δ
18
O at these scales should be done with caution, using supplemental
proxies (e.g. δ
13
C, trace elements ratios).
PublicationDetails: Hu,J.,Emile-Geay,J.,McKay,N.,Kathayat,G.,Brahim,Y.A.(inpreparation).
Limited coherency of Asian speleothems over the Holocene, with implications for the Meghalayan age.
Geophysical Research Letters.
108
Focusing on the Meghalayan, we find no “4.2 ka event” outside of the Mawmluh cave
record of Berkelhammer et al. (2013), suggesting neither a regional nor a global climate
signal at this time. Further, two new high-resolution series from Mawmluh cave suggest
that the “4.2 ka event” is far less abrupt than originally proposed. These two observations
disqualify the publishedδ
18
O excursion as a “golden spike”. While there is evidence for a
droughtaffectingMeghalayaaround4.2kyBP,itappearstohavebeenspatiallyrestricted,
therefore atypical of global conditions over the Late Holocene.
4.1 Introduction
Speleothems have been widely used to reconstruct paleohydrology over the past few
decades. Asian speleothems, in particular, have been in the spotlight of the paleoclimate
community since several of them are well-dated, continuous, and cover several orbital
cycles. These records display coherent variability at orbital scales (dominated by the
precessionalcycles),inphasewithsummerlocalinsolation(Wangetal.,2008,Chengetal.,
2009, 2012, Battisti et al., 2014, Cheng et al., 2016). While there is general consensus for
interpretingtheseorbital-scalefluctuationsasmeasuresofmonsoonintensity(Wangetal.,
2001, Yuan et al., 2004, Hu et al., 2008, Cheng et al., 2009, 2012, Hu et al., tion), it is an
open question whether this interpretation carries over to shorter timescales. A necessary
condition for doing so would be for Asian speleothems to show coherent variability at
these scales, for instance over the Holocene. Recent studies suggest that the coherency of
Asian speleothems may vanish at scales shorter than orbital (Wan et al., 2011, Chu et al.,
2012, Li et al., 2014), but these studies do not take age uncertainties into account, which
can greatly influence the interpretation of speleothem records (Hu et al., 2017). Here we
will analyze eight high temporal resolution Asian stalagmites from a recent compilation
(Atsawawaranunt et al., 2018) over the Holocene with coherency in mind, taking age
uncertainties into account. Also, we will utilize an isotope-enabled model and a forward
model of speleothems to mechanistically interpret their spatiotemporal patterns.
Speleothems often feature high temporal precision thanks to U-Th dating techniques
(Cheng et al., 2013, Shen et al., 2013). Accordingly, the Global Stratotype Section and
Point (GSSP) of the Meghalayan age (4,200 BP to present) is a stalagmite (KM-A) from
109
Mawmluh cave (Berkelhammer et al., 2013). The Meghalayan age was formally ratified
by the International Commission on Stratigraphy (ICS) in 2018 (Walker et al., 2018) as
the latest subdivision of the Holocene Epoch, creating controversy in the paleoclimate
community (Voosen, 2018). The Mawmluh cave record in northeast India marks the “4.2
ka” event as the GSSP of the Meghalayan age (Walker et al., 2012, 2018). This event
features abrupt climate anomalies between 4.2 and 3.9 ka, evidenced in proxy records
from marine sediments, lake sediments, speleothems, and ice cores across the Middle
East (Bar-Matthews et al., 1999, Cullen et al., 2000, Arz et al., 2006), Mediterranean
(Drysdale et al., 2006, Psomiadis et al., 2018), South and East Asia (Staubwasser et al.,
2003, Berkelhammer et al., 2013, Liu and Feng, 2012, Dixit et al., 2014, Nakamura et al.,
2016), East and North Africa (Thompson et al., 2002, Stanley et al., 2003), and North
America (Fisher et al., 2008, Fisher, 2011, Menounos et al., 2008).
The “4.2 ka” event also coincides with the collapse of several civilizations across Eura-
sia and North Africa. These civilizations include the Akkadian Empire in Mesopotamia
(collapsed at 4, 170± 150 BP; Weiss et al., 1993, Cullen et al., 2000, Weiss et al., 2012),
the Old Kingdom of Egypt (whose disintegration started around 4,200 BP; Stanley et al.,
2003, Butzer, 2012, Redford, 2005), the Harappan civilization in the Indus Valley (declin-
ing from 3,900 BP onwards; Kenoyer, 1998, Staubwasser et al., 2003, Wright, 2010),
and the Longshan culture and other Neolithic cultures in China (which collapsed around
4,200-3,900 BP; Allen and Richards, 1999, Jin and Liu, 2002, Wu and Liu, 2004, Liu and
Feng, 2012).
However, there is also evidence questioning whether the “4.2 ka” event is worldwide.
Although this event has been termed a global megadrought (Weiss, 2016), many proxy
records show wet or unchanged hydroclimate anomalies during this period (Shulmeister
and Lees, 1995, Constantin et al., 2007, Tierney et al., 2011, Roland et al., 2014, Zhang
et al., 2018a, Railsback et al., 2018, Li et al., 2018). Also the timing of the 4.2 ka
event remains uncertain due to the lack of high temporal precision proxies (Staubwasser
and Weiss, 2006, Nakamura et al., 2016, Railsback et al., 2018, Kathayat et al., 2018).
The “4.2 ka event” is a good focal point to examine the coherency of Asian speleothem
δ
18
O over the Holocene. Here we will analyze the aforementioned speleothem compilation
(Atsawawaranunt et al., 2018) to in search of an extreme climate event around 4.2 ka BP.
110
The paper is structured as follows. Section 2 presents our methods. Section 3 investi-
gates the spatial-temporal variability of Asian speleothem over the Holocene and discusses
its interpretations through model simulations. Section 4 hones in on the 4.2 ka BP event,
and discusses the detailed timing of the event. We finish with a discussion of the signifi-
cance of 4.2 ka event and its relation with civilization collapses.
4.2 Methods
4.2.1 Data & Analysis
The speleothemδ
18
O records analyzed in this study are collected from the Speleothem
Isotope Synthesis and Analysis (SISAL) database version 1 (Atsawawaranunt et al., 2018)
and original authors (see Table 4.1 for references). They all cover 4.2 ka BP and longer
than 4,000 years (adequate for the analysis of sub-orbital scale variability) and the lowest
temporal resolution is about 50 years. Fig. 4.1 shows their locations.
Table 4.1: Asian speleothem records used in this study. Resolution refers to the median
spacing between consecutive observations. BP means “before 1950 AD”.
Cave site Stalagmite name Lat. Lon. Time range (BP) Resolution (y) Reference Data from SISAL
Heshang HS4 30.45 110.42 9458, 82 3 Hu et al. (2008) Yes
Sanbao SB43 31.67 110.43 10332, 157 9 Dong et al. (2010) Yes
Jiuxian C996-1 33.57 109.1 7841, -4 4 Cai et al. (2010) Yes
Lianhua A1 29.48 109.53 6588, -39 4 Cosford et al. (2009) Yes
Mawmluh KM-A 25.26 91.82 12107, 3654 4 Berkelhammer et al. (2013) Yes
Xianglong XL26 33.00 106.33 6657, 2008 5 Tan et al. (2018) No
Dongge DA 25.28 108.08 8902, -20 3 Wang et al. (2005) No
Sahiya SAH-2 30.6 77.87 5648, 2607 1 Kathayat et al. (2017) No
Age uncertainties must be considered in the investigation of the sub-orbital scale
variability and the “4.2 ka event”. Even though the most precise speleothem chronologies
canbeasaccurateasafewyears, theageuncertaintiesofmostHoloceneAsianspeleothem
records are in the tens of years. An uncertainty of 50 years can offset twoδ
18
O time series
by 100 years, thereby influencing our judgement of how abrupt or synchronous climate
conditions were. To quantify age uncertainties, we generate an ensemble of plausible age
models (1000 members) via the Bayesian probability model Bchron (Parnell et al., 2008),
and then collect 1000 possible realizations of speleothemδ
18
O time series. In this way, we
transfer age uncertainties to proxy uncertainties. (Deininger et al., 2017). This method
was applied by Hu et al. (2017) to show how age uncertainties affect correlations.
111
0°
20°N
40°N
90°E 120°E
Heshang
Sanbao
Jiuxian
Lianhua
Mawmluh
Xianglong
Dongge
Sahiya
Figure 4.1: The locations of eight Asian speleothem records over the Holocene
analyzed in this study.
There are coherent trends for each speleothem records in this study, which follow their
local summer insolation. The timescale and amplitude of the 4.2 ka event is much smaller
than this orbital variability. To emphasize the possible 4.2 ka event and also investigate
centennial to millennial variability over the Holocene, we remove the trends of all records
using the Savitzky-Golay filter (Savitzky and Golay, 1964), implemented by a Python
package Pyleoclim (Khider et al., 2018). The window size is half of the data, and the
order of the filter is four.
To investigate the spatial coherency among Asian speleothem δ
18
O at centennial-
millennial time scales, we apply Monte Carlo Principle Component Analysis (MC-PCA)
(Anchukaitis and Tierney, 2013, Deininger et al., 2017) to the detrended speleothem
records to consider age uncertainties in spatiotemporal variability analysis. We select one
realization of speleothem δ
18
O time series from each record, and then upscale them to
the same temporal resolution (50 years, which is about the lowest temporal resolution of
speleothem records used in this study) by binning and averaging with a Gaussian kernel
proposedbyRehfeldetal.(2011), andfinallyperformthePCAanalysisonthecompilation
112
of these upscaled speleothem records. Since we have 1000 members for each record, we
thenrepeatthisprocess1000times. ThedetaileddescriptionofthisroutineisinDeininger
et al. (2017) and Anchukaitis and Tierney (2013). For the selection of dominant principal
components, we generate surrogates of each speleothem ensemble member with an AR(1)
(red noise) model and test if the eigenvalues of our MC-PCA result are significantly
different above those expected from red noise (Preisendorfer, 1988, Deininger et al., 2017).
4.2.2 Modeling
We employ an isotope-enabled climate model iCAM5-iCLM4 (Wong et al., 2017, Nus-
baumer et al., 2017) and the karst forward model Karstolution (Treble et al., 2019) to
better understand the spatial heterogeneity of Asian speleothem δ
18
O over the Holocene.
We run iCAM5-iCLM4 forced by observed sea surface temperature and sea ice following
the AMIP protocol (Hurrell et al., 2008) from 1850 to 2012. The land component iCLM4
considers the water vapor flux and isotope fractionation in vegetated land. Main processes
include water isotope exchanges among soil, spaces under and above canopy, and leaves.
The land and vegetation types and amount of canopy is modern climatological mean with
seasonal cycles (Oleson et al., 2010).
Karstolution is a lumped model simulating pseudo-stalagmites given climate inputs.
The model contains two parts, the karst processes and in-cave processes. The karst
processes includes the oxygen isotope fractionation in evaporation, mixing among matrix
flow and water stores, overflow and underflow movement. The in-cave processes includes
oxygen isotope disequilibrium and evaporative fractionation effects. The model generates
five different archetypes of stalagmites with different weights in their water sources (e.g.
water store, matrix flow and overflow). We input the climate variables (precipitation,
precipitation δ
18
O, surface temperature and evaporation) simulated from iCAM5-iCLM4
with cave parameters (e.g. cave temperature, humidity, and CO
2
concentration) given by
cave monitoring data (Duan et al., 2016) to simulated stalagmite δ
18
O.
113
4.3 Spatial coherency of Holocene Asian speleothem
δ
18
O
Fig. B.1 gathers the time series of eight Asian speleothem records analyzed in this
study. They all display evident trends over the Holocene, following local summer inso-
lation. Fig. B.2 shows the same time series after removing orbital trends, to empha-
size shorter-scale variability, including the 4.2 ka event. The age uncertainties of these
records are translated to δ
18
O uncertainties based on the ensembles of possible age mod-
els generated by Bchron (Fig. B.3). Records with large age uncertainties also have large
δ
18
O uncertainties.
4.3.1 Spatial-temporal variability
Here we will investigate the spatiotemporal variability of Asian speleothem at sub-
orbital scales to examine their coherency. MC-PCA (Anchukaitis and Tierney, 2013,
Deininger et al., 2017) allows to investigate whether Asian speleothem δ
18
O has a mean-
ingful spatiotemporal variability pattern, and at the same time take age uncertainties into
account. We analyze detrended δ
18
O time series because we are interested in centennial-
millennial variability and it is well known that Asian speleothem δ
18
O share common
trends over the Holocene following local insolation (Wang et al., 2008, Cheng et al., 2009,
2012, 2016, Kathayat et al., 2016).
A comparison with the red noise benchmark shows that the first PC (explaining 18%
of total variance) is above the 95% confidence level (Fig. B.4). The first PC and its
corresponding spatial loading are shown in Fig. 4.2. The time series of PC1 shows
possible major peaks (either intense dry or wet climate anomaly). The one around 8 ka
corresponds to the 8.2 ka event, which is well-accepted as a global event, and presently
separates the Early and Middle Holocene (Walker et al., 2012). There is a peak around
4 ka, but its amplitude is not significantly different from the peaks around 8 ka, 6 ka,
3 ka and 200 BP. This implies that the magnitude of the climate anomaly around 4 ka
is not unique over the Holocene, and may be part of centennial-millennial scale climate
variability. It does not appear unique enough to divide the Middle and Late Holocene.
114
The spatial pattern (Fig. 4.2, top) illustrates that Mawmluh cave δ
18
O is anti-phased
with Sahiya cave (northern India) and three caves in central and southern China, but its
variability also aligns with three caves in central China. There is evident spatial hetero-
geneity among Asian speleothem δ
18
O at centennial-millennial scales over the Holocene,
especially considering the difference between two close caves, Sanbao and Lianhua. We
now explore the source of this heterogeneity with the aid of isotope-enabled models.
115
(a)
(b)
Figure 4.2: MC-PCA of Asian speleothems over the Holocene. (a) The spatial
distribution of the loading of the first PC of MC-PCA; (b) The time series of the first PC
of MC-PCA (λ = 18%)
116
4.3.2 Implications from model simulations
Does the spatial asynchrony of speleothem records arise from hydroclimate variations,
or from non-climatic factors? iCAM5 simulates precipitation δ
18
O variability with high
fidelity(Nusbaumeretal.,2017), soweuseittoquantifyspatialcoherencyinprecipitation
δ
18
O ina“perfectmodel”scenario. iCLM4featureswaterisotopeexchangesinvegetation
and soil (Wong et al., 2017), so it provide insights into how much the climate signal or
precipitation δ
18
O can be distorted by soil and vegetation processes.
We perform PCA on weighted annual mean precipitation δ
18
O and soil δ
18
O (at a
depth of 12 cm) of the 150-year long simulation from iCAM5-iCLM4. Fig. 4.3 and B.5
show the first two modes of precipitation and soil δ
18
O respectively. The first mode of
precipitation δ
18
O shows a coherent pattern over Asia with the variability centered over
Tibet, while the second mode features an east-west dipole, which may partly explain
the anti-phasing between Mawmluh and Sahiya caves, but cannot explain the difference
between Mawmluh and some Chinese caves. The first mode of soil δ
18
O is different from
precipitationδ
18
O, with a positive center in western India and a negative center over the
Indochina peninsula. The second mode of soil δ
18
O also shows a more complex spatial
pattern than its precipitation counterpart. We also notice that both PCs of soilδ
18
O have
morelower-frequencyvariability thanthose ofprecipitationδ
18
O, whichindicatesthat soil
and vegetation processes also redden precipitation signals. In addition, we estimate the
spatial decorrelation scale of soil and precipitation δ
18
O, which is defined as the e-folding
length of spatial autocorrelation, by fitting an exponential, R =e
−x/l
d
, at each grid cell,
where R is the correlations between the time series at one grid cell and other grid cells,
x is the distance between grid cells, and l
d
is the spatial scale. We perform this at
each grid cell and then get the mean spatial scale over the Asian monsoon region. The
mean spatial scale of soil δ
18
O is 486 km, shorter than that of precipitation δ
18
O (782
km), so the spatial pattern of soil δ
18
O is less spatially coherent than precipitation δ
18
O.
And for most speleothem δ
18
O, the variability of soil δ
18
O is the more relevant input
than precipitation δ
18
O. Thus, the change from a coherent precipitation δ
18
O pattern
to a different soil δ
18
O spatial pattern suggests that soil and vegetation processes may
cause some of the asynchrony in cave records given coherent climate signals. In other
117
words, some of sub-orbital heterogeneity in Asian speleothem records may partly be due
to differences in vegetation and land types.
We should note that this result is based on the 150-year long simulation, and may not
directly apply to centennial-millennial variability. However, this is illustrative of qualita-
tive differences between the variability of soil δ
18
O and precipitation δ
18
O , even on time
scales of decades, where one would assume they might look similar. This difference in spa-
tial patterns may be quantitatively distinct at longer timescales, so further investigations
are needed.
Figure4.3: Thespatialdistributionoftheloadingandthetimeseriesofthefirst
principle component of precipitation δ
18
O and soil δ
18
O from iCAM5-iCLM4.
Both the spatial loading and principle component are normalized by dividing the square
root of their eigenvalues.
118
The MC-PCA analysis also shows distinct anti-phasing between two close caves, Lian-
hua and Heshang in China. The difference over such a short distance cannot be explained
by differences in precipitation and soil δ
18
O variability shown in Fig. 4.3 and B.5, but
differences in karst processes are a possible reason. Fig. B.6 shows the precipitation and
soilδ
18
O of these two caves from iCAM5-iCLM4, displaying high correlations as expected.
We then input climate variables (precipitation, precipitation δ
18
O, soil δ
18
O and surface
temperature) simulated from iCAM5-iCLM4 with cave parameters (e.g., cave tempera-
ture, humidity, and CO
2
concentration based on the cave monitoring data (Duan et al.,
2016)) to the karst forward model Karstolution. The model then simulates the calcite
δ
18
O of five archetypes of stalagmites based on different possible water flow pathways
before forming the stalagmites (Treble et al., 2019). This is realized in the model by spec-
ifying different weights in water sources of stalagmites (including water stores inside the
cave, epikarst, overflow from water stores, soil water and surface precipitation). These five
archetypes of stalagmites are not the only possible stalagmites. There could be infinite
simulated stalagmites based on different proportion of water sources, and here we just
show five typical types of stalagmites for the convenience of illustration.
The water sources of stalagmites type 2 and 3 are mainly soil, precipitation, and
matrix flow, so they have shorter residence times and closely track precipitation δ
18
O. In
contrast, stalagmitetypes1, 4and5receivecontributionsfromwaterstoresintheepikarst
and karst, so they have longer residence times and generally smoothe the precipitation
δ
18
O signal. Fig. B.7 shows that the correlations of the same type of stalagmites in two
caves are still high at interannual scales, ranging from 0.68 to 0.84, with p-values smaller
than 10
−6
. However, it is possible that the stalagmite types in these two caves are not
the same. For example, if the stalagmite in Heshang cave is of type 2 and the stalagmite
in Lianhua cave is of type 5, their correlation decreases to 0.25 (Fig. 4.4).
This suggests that differences in water flow pathways in karst systems can cause
significant differences in stalagmite δ
18
O, which can partly explain the different behav-
ior between Lianhua and Heshang, or more broadly, or between other proximal caves.
Since the amplitude of the millennial scale variability of Asian speleothem δ
18
O over
the Holocene is only about 1-3h (Fig. B.2), heterogenous karst processes may lead to
asynchronous variability of speleothem δ
18
O (Baker et al., 2013).
119
Together, vegetation, soilandkarstprocessmayplausiblyexplainthelackofcoherency
at centennial to millennial scales in our network of 8 speleothems.
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
18
O ( )
r=0.25, p=0.01
HS Stal2 vs. LH Stal5
Heshang
Lianhua
Figure 4.4: Comparison between the simulated type 2 stalagmite δ
18
O at Hes-
hang cave and type 5 stalagmite δ
18
O Lianhua cave from Karstolution.
4.4 Asian speleothem δ
18
O around 4.2 ka BP
What do these 8 records show regarding the “4.2 ka event”? The speleothem δ
18
O of
Mawmluh cave (KM-A), in the state of Meghalaya (Northeastern India) dramatically
increases around 4.2 ka BP (Fig. B.2), an extreme feature chosen to pinpoint the starting
date of the Meghalayan (Walker et al., 2018). Here we investigate whether this extreme
event is coeval in other Asian speleothem records.
Fig. B.2 shows that there are dry or wet anomalies in many records around 4.2 ka
BP, but these anomalies are not extreme in the context of the Holocene. To quantify how
extreme Asian speleothemδ
18
O is around 4.2 ka BP, we plot the probability distribution
of their detrendedδ
18
O values over 150-years intervals based on 1000 plausible age models
over the Holocene (Fig. 4.5) and highlight theδ
18
O values around 4.2 ka BP (4,075-4,225
BP, vertical lines in Fig. 4.5). Please see the text in the Appendix for the details.
In the Mawmluh KM-A record, theδ
18
O values around 4.2 ka BP are indeed an outlier
overtheHolocene, as88%ofthemfallbelowthe5%quantileofHolocenevalues. However,
for other Asian speleothem records, including another Indian cave record, Sahiya cave, the
period centered on 4.2 ka BP is not an outlier, being close to average for many records.
120
The δ
18
O shift at Sahiya is in fact negative during that period, opposite from Mawmluh
cave. A sensitivity analysis (Fig. B.8), shows that the selection of the interval sizes does
not change the conclusion and that the percentage of δ
18
O values around 4.2 ka falling
below the 5% quantile of Holocene values is largest when using the 150-year interval.
δ
18
O (‰) δ
18
O (‰) δ
18
O (‰)
Density Density
Density
Figure 4.5: Asian speleothem δ
18
O around 4.2 ka BP. The distribution of the
detrended δ
18
O values of eight Asian speleothems with 150-years intervals based on 1000
plausible age models over Holocene. The vertical lines are δ
18
O values around 4.2 ka BP
(4,075-4,225 BP).
Thus, we do not find evidence for a regional drought around 4.2 ka BP. This suggests
that the 4.2 ka BP megadrought recorded in the Mawmluh KMA stalagmite was a local
121
event. Droughts may have happened in other Asia locations at that time, but they do
not stand out in the context of the Holocene, therefore do not deserve to be singled out
by a chronostratigraphic denomination. In other words, there was no “Meghalayan age”
outside of Meghalaya. Also, this suggests that at this time scale, speleothem δ
18
O in the
Mawmluh cave cannot be interpreted as the large-scale circulation “monsoon intensity”.
Thus, we should be cautious about making inference from a single record before collecting
evidence from other speleothem records.
Furthermore, though the ICS proposal (Walker et al., 2018), following Berkelhammer
et al. (2013), places the starting date of Meghalayan at 4,200 BP, there are concerns about
the timing of this new age (Railsback et al., 2018). Recently, Kathayat et al. (2018) ana-
lyzed two new high-precision stalagmites (ML.1 and ML.2) in Mawmluh cave (red curves
in Fig. B.9). Comparing with the old Mawmluh stalagmite KMA (Berkelhammer et al.,
2013), which serves as the GSSP of Meghalayan age, the new Mawmluh stalagmites do
not show an abrupt increase ofδ
18
O values around 4.2 ka, but they show gradual increases
over that period, interspersed with some short-term droughts. Taking age uncertainties
into account (Fig. B.9), we notice that the ML.2 stalagmite displays an evident peak of
δ
18
O values starting about 3.8 kaBP. This indicates that even within Mawmluh cave, the
timing of this extreme event is variable.
4.5 Discussion
We analyzed speleothems in China and India and the result shows that there is no pan-
Asia 4.2 ka event. What about other regions of the world? Mesopotamia and west Asia is
where the 4.2 ka event was first proposed to explain the collapse of the Akkadian Empire
(Weiss et al., 1993), and it is well supported by a multiproxy marine sediment from the
RedSea(Arzetal.,2006)andadustproxyintheGulfofOman(Cullenetal.,2000). This
seems at least to indicate that there was a megadrought in this region. However, only one
of nine existing speleothem records in this region shows an abrupt climate change around
4.2 ka (Carolin et al., 2019). There are multi-proxy records in Europe (Constantin et al.,
2007, Railsback et al., 2011, Roland et al., 2014), Canada (Menounos et al., 2008), East
Africa (Tierney et al., 2011, Konecky et al., 2011), and the Indian Ocean (Tierney et al.,
122
2012, Li et al., 2018) not showing abrupt climate change or even showing wet climate
conditions around 4.2 ka BP.
For the timing of the “4.2 ka event”, Railsback et al. (2018) found that the starting
time in the literature supporting this event ranged from 4.28 to 3.41 ka BP. They also
presented a stalagmite in south Africa showing two-pulse wet events around 4.2 ka instead
of a one-pulse event.
The “4.2 ka event” is often connected to the collapses of civilizations during that
period, butthereareconcernsthatsocietalreactionstoabruptclimatechangearecomplex
(McAnany and Yoffee, 2009, Jaffe et al., 2019). Civilizations may collapse partially, or
adapt. After the mature phase of the Harappan Civilization, people migrated out of
cities and settled in places more favorable for farming (Sarkar et al., 2016). Spurts of
settlement increases more in arid periods after 4 ka in Mesopotamia (Lawrence et al.,
2016). Also, there are many other reasons for civilization collapses, such as internal
political or economic crises, as well as epidemics (Butzer, 2012). Finally, the chronology
of these civilizations is still controversial (Butzer and Endfield, 2012). For example, there
is no generally accepted chronology for Mesopotamia before 4.1 ka BP and the dating of
the First Intermediate Period of Egypt (after the Old Kingdom) is not well constrained.
Thus it is controversial to directly link these civilization collapses to a climate event.
We used numerical experiments with a isotope-enabled model to explain the asyn-
chrony of Asian speleothemδ
18
O over the Holocene. Our simulations were short, however
(150y), so future studies should consider longer simulations such as the Last Millennium
Ensemble (Otto-Bliesner et al., 2016) and TraCE-21ka experiments (He et al., 2013) to
investigate decadal-centennial variability of precipitation and soil δ
18
O. The karst model
Kastolution currently assumes the capacity of water stores in the karst system is constant
through time, while in reality it may vary and the water sources of one stalagmite can
also change with time at centennial-millennial time scales. This will also contribute to the
spatial heterogeneity of speleothem δ
18
O at time scales shorter than orbital. To better
constrain the non-climate factors from speleothem records, analyzing other proxies such
as δ
13
C and trace elements with δ
18
O at the same time will be helpful (Cosford et al.,
2009, Zhang et al., 2018b).
123
4.6 Conclusion
Our analysis finds no coherent sub-orbital scale variability among Asian speleothem
δ
18
O over the Holocene. This asynchrony can be explained by heterogenous soil, vege-
tation, and karst processes. Given its relatively small amplitude, detecting and under-
standing sub-orbital scale climate variability over the Holocene is challenging. Apart from
enlarging the proxy network, analyzing other proxies such asδ
13
C and trace elements with
δ
18
O at the same time will help exclude non-climate factors in speleothem records and
better interpret speleothem δ
18
O (Cosford et al., 2009, Zhang et al., 2018b).
We also examined whether there was an extreme climate event at 4.2 ka BP, the
proposed onset of the Meghalayan age. We analyzed Asian speleothem δ
18
O during this
period by taking age uncertainties into account, and none of them shows an extreme
climate change in 4.2 ka BP except the original Mawmluh cave record. Also, a new high-
resolution stalagmite in Mawmluh cave questions the exact timing of the “4.2 ka event”.
Therefore there was no pan-Asian 4.2 ka event, so the “Meghalayan age” is a misnomer,
and Mawmluh cave does not offer a “golden spike” for the Late Holocene. This is in
contrast to the “8.2 ka event”, which is broadly expressed throughout the globe, thus
eligible to divide the Early and Middle Holocene (Walker et al., 2012, 2018).
Bibliography
Allen, R. and Richards, D. S. (1999). The Cambridge history of ancient China: From the
origins of civilization to 221 BC. Cambridge University Press.
Anchukaitis, K. J. and Tierney, J. E. (2013). Identifying coherent spatiotemporal modes
in time-uncertain proxy paleoclimate records. Climate Dynamics, 41(5-6):1291–1306.
Arz, H.W., Lamy, F., andPätzold, J.(2006). Apronounceddryeventrecordedaround4.2
ka in brine sediments from the northern Red Sea. Quaternary Research, 66(3):432–441.
Atsawawaranunt, K., Comas-Bru, L., Amirnezhad Mozhdehi, S., Deininger, M., Harrison,
S. P., Baker, A., Boyd, M., Kaushal, N., Ahmad, S. M., Ait Brahim, Y., et al. (2018).
124
The SISAL database: a global resource to document oxygen and carbon isotope records
from speleothems. Earth System Science Data, 10:1687–1713.
Baker, A., Bradley, C., and Phipps, S. J. (2013). Hydrological modeling of stalag-
mite δ
18
O response to glacial-interglacial transitions. Geophysical Research Letters,
40(12):3207–3212.
Bar-Matthews, M., Ayalon, A., Kaufman, A., and Wasserburg, G. J. (1999). The Eastern
Mediterranean paleoclimate as a reflection of regional events: Soreq cave, Israel. Earth
and Planetary Science Letters, 166(1-2):85–95.
Battisti, D., Ding, Q., and Roe, G. (2014). Coherent pan-Asian climatic and isotopic
response to orbital forcing of tropical insolation. Journal of Geophysical Research:
Atmospheres, 119(21):11–997.
Berkelhammer, M., Sinha, A., Stott, L., Cheng, H., Pausata, F., and Yoshimura, K.
(2013). An abrupt shift in the Indian monsoon 4000 years ago. Climates, Landscapes,
and Civilizations, pages 75–88.
Butzer, K. W. (2012). Collapse, environment, and society. Proceedings of the National
Academy of Sciences, 109(10):3632–3639.
Butzer, K. W. and Endfield, G. H. (2012). Critical perspectives on historical collapse.
Proceedings of the National Academy of Sciences, 109(10):3628–3631.
Cai, Y., Tan, L., Cheng, H., An, Z., Edwards, R. L., Kelly, M. J., Kong, X., and Wang,
X. (2010). The variation of summer monsoon precipitation in central China since the
last deglaciation. Earth and Planetary Science Letters, 291(1):21–31.
Carolin, S. A., Walker, R. T., Day, C. C., Ersek, V., Sloan, R. A., Dee, M. W., Talebian,
M., and Henderson, G. M. (2019). Precise timing of abrupt increase in dust activity
in the Middle East coincident with 4.2 ka social change. Proceedings of the National
Academy of Sciences, 116(1):67–72.
Cheng, H., Edwards, R. L., Broecker, W. S., Denton, G. H., Kong, X., Wang, Y., Zhang,
R., and Wang, X. (2009). Ice Age Terminations. Science, 326(5950):248–252.
125
Cheng, H., Edwards, R. L., Shen, C.-C., Polyak, V. J., Asmerom, Y., Woodhead, J.,
Hellstrom, J., Wang, Y., Kong, X., Spötl, C., Wang, X., and Alexander, E. C. (2013).
Improvements in
230
Th dating,
230
Th and
234
U half-life values, and U–Th isotopic mea-
surements by multi-collector inductively coupled plasma mass spectrometry. Earth and
Planetary Science Letters, 371-372:82 – 91.
Cheng, H., Edwards, R. L., Sinha, A., Spötl, C., Yi, L., Chen, S., Kelly, M., Kathayat,
G., Wang, X., Li, X., et al. (2016). The Asian monsoon over the past 640,000 years
and ice age terminations. Nature, 534(7609):640–646.
Cheng, H., Sinha, A., Wang, X., Cruz, F. W., and Edwards, R. L. (2012). The Global
PaleomonsoonasseenthroughspeleothemrecordsfromAsiaandtheAmericas. Climate
Dynamics, 39(5):1045–1062.
Chu, P.C., Li, H.-C., Fan, C., andChen, Y.-H.(2012). Speleothemevidencefortemporal–
spatial variation in the East Asian Summer Monsoon since the Medieval Warm Period.
Journal of Quaternary Science, 27(9):901–910.
Constantin, S., Bojar, A.-V., Lauritzen, S.-E., and Lundberg, J. (2007). Holocene
and Late Pleistocene climate in the sub-Mediterranean continental environment: a
speleothem record from Poleva Cave (Southern Carpathians, Romania). Palaeogeogra-
phy, Palaeoclimatology, Palaeoecology, 243(3-4):322–338.
Cosford, J., Qing, H., Mattey, D., Eglington, B., andZhang, M.(2009). Climaticandlocal
effects on stalagmite δ
13
C values at Lianhua Cave, China. Palaeogeography, Palaeocli-
matology, Palaeoecology, 280(1):235–244.
Cullen, H. M., deMenocal, P. B., Hemming, S., Hemming, G., Brown, F. H., Guilderson,
T., and Sirocko, F. (2000). Climate change and the collapse of the Akkadian empire:
Evidence from the deep sea. Geology, 28(4):379–382.
Deininger, M., McDermott, F., Mudelsee, M., Werner, M., Frank, N., and Mangini, A.
(2017). Coherency of late Holocene European speleothemδ
18
O records linked to North
Atlantic Ocean circulation. Climate Dynamics, 49(1-2):595–618.
126
Dixit, Y., Hodell, D. A., and Petrie, C. A. (2014). Abrupt weakening of the summer
monsoon in northwest India∼ 4100 yr ago. Geology, 42(4):339–342.
Dong, J., Wang, Y., Cheng, H., Hardt, B., Edwards, R. L., Kong, X., Wu, J., Chen, S.,
Liu, D., Jiang, X., et al. (2010). A high-resolution stalagmite record of the Holocene
EastAsianmonsoonfromMtShennongjia, centralChina. The Holocene, 20(2):257–264.
Drysdale, R., Zanchetta, G., Hellstrom, J., Maas, R., Fallick, A., Pickett, M., Cartwright,
I., and Piccini, L. (2006). Late Holocene drought responsible for the collapse of Old
World civilizations is recorded in an Italian cave flowstone. Geology, 34(2):101–104.
Duan, W., Ruan, J., Luo, W., Li, T., Tian, L., Zeng, G., Zhang, D., Bai, Y., Li, J., Tao,
T., et al. (2016). The transfer of seasonal isotopic variability between precipitation and
drip water at eight caves in the monsoon regions of China. Geochimica et Cosmochimica
Acta, 183:250–266.
Fisher, D., Osterberg, E., Dyke, A., Dahl-Jensen, D., Demuth, M., Zdanowicz, C., Bour-
geois, J., Koerner, R. M., Mayewski, P., Wake, C., et al. (2008). The Mt Logan
Holocene—late Wisconsinan isotope record: tropical Pacific—Yukon connections. The
Holocene, 18(5):667–677.
Fisher, D. A. (2011). Connecting the Atlantic-sector and the north Pacific (Mt Logan)
ice core stable isotope records during the Holocene: The role of El Niño. The Holocene,
21(7):1117–1124.
He, F., Shakun, J. D., Clark, P. U., Carlson, A. E., Liu, Z., Otto-Bliesner, B. L., and
Kutzbach, J. E. (2013). Northern Hemisphere forcing of Southern Hemisphere climate
during the last deglaciation. Nature, 494(7435):81.
Hu, C., Henderson, G. M., Huang, J., Xie, S., Sun, Y., and Johnson, K. R. (2008). Quan-
tification of Holocene Asian monsoon rainfall from spatially separated cave records.
Earth and Planetary Science Letters, 266(3):221–232.
Hu, J., Emile-Geay, J., and Partin, J. (2017). Correlation-based interpretations of paleo-
climate data–where statistics meet past climates. Earth and Planetary Science Letters,
459:362–371.
127
Hu, J., Emile-Geay, J., Tabor, C., Nusbaumer, J., Partin, J., and Adkins, J. (in prepara-
tion). Deciphering Chinese speleothems with an isotope-enabled climate model. Pale-
oceanography and Paleoclimatology.
Hurrell, J. W., Hack, J. J., Shea, D., Caron, J. M., and Rosinski, J. (2008). A new
sea surface temperature and sea ice boundary dataset for the Community Atmosphere
Model. Journal of Climate, 21(19):5145–5153.
Jaffe, Y., Bar-Oz, G., and Ellenblum, R. (2019). Improving integration in societal
consequences to climate change. Proceedings of the National Academy of Sciences,
116(11):4755–4756.
Jin, G. and Liu, D. (2002). Mid-Holocene climate change in North China, and the effect
on cultural development. Chinese Science Bulletin, 47(5):408–413.
Kathayat, G., Cheng, H., Sinha, A., Berkelhammer, M., Zhang, H., Duan, P., Li, H., Li,
X., Ning, Y., and Edwards, R. L. (2018). Evaluating the timing and structure of the 4.2
ka event in the Indian summer monsoon domain from an annually resolved speleothem
record from Northeast India. Climate of the Past, 14(12):1869–1879.
Kathayat, G., Cheng, H., Sinha, A., Spötl, C., Edwards, R. L., Zhang, H., Li, X., Yi,
L., Ning, Y., Cai, Y., et al. (2016). Indian monsoon variability on millennial-orbital
timescales. Scientific Reports, 6.
Kathayat, G., Cheng, H., Sinha, A., Yi, L., Li, X., Zhang, H., Li, H., Ning, Y., and
Edwards, R. L. (2017). The Indian monsoon variability and civilization changes in the
Indian subcontinent. Science Advances, 3(12):e1701296.
Kenoyer, J. M. (1998). Ancient cities of the Indus valley civilization. American Institute
of Pakistan Studies.
Khider, D., Zhu, F., Hu, J., and Emile-Geay, J. (2018). LinkedEarth/Pyleoclim_util:
Pyleoclim release v0.4.8.
Konecky, B. L., Russell, J. M., Johnson, T. C., Brown, E. T., Berke, M. A., Werne, J. P.,
and Huang, Y. (2011). Atmospheric circulation patterns during late Pleistocene climate
changesatLakeMalawi, Africa. Earth and Planetary Science Letters, 312(3-4):318–326.
128
Lawrence, D., Philip, G., Hunt, H., Snape-Kennedy, L., and Wilkinson, T. J. (2016). Long
term population, city size and climate trends in the fertile crescent: A first approxima-
tion. PLOS ONE, 11(3):1–16.
Li, H., Cheng, H., Sinha, A., Kathayat, G., Spötl, C., André, A. A., Meunier, A., Biswas,
J., Duan, P., Ning, Y., et al. (2018). Hydro-climatic variability in the southwestern
Indian Ocean between 6000 and 3000 years ago. Climate of the Past, 14(12):1881–1891.
Li,Y.,Wang,N.,Zhou,X.,Zhang,C.,andWang,Y.(2014). Synchronousorasynchronous
Holocene Indian and East Asian summer monsoon evolution: a synthesis on Holocene
Asian summer monsoon simulations, records and modern monsoon indices. Global and
Planetary Change, 116:30–40.
Liu, F. and Feng, Z. (2012). A dramatic climatic transition at∼ 4000 cal. yr BP and its
cultural responses in Chinese cultural domains. The Holocene, 22(10):1181–1197.
McAnany, P. A. and Yoffee, N. (2009). Questioning collapse: human resilience, ecological
vulnerability, and the aftermath of empire. Cambridge University Press.
Menounos, B., Clague, J. J., Osborn, G., Luckman, B. H., Lakeman, T. R., and Minkus,
R. (2008). Western Canadian glaciers advance in concert with climate change circa 4.2
ka. Geophysical Research Letters, 35(7).
Nakamura, A., Yokoyama, Y., Maemoku, H., Yagi, H., Okamura, M., Matsuoka, H.,
Miyake, N., Osada, T., Adhikari, D. P., Dangol, V., Ikehara, M., Miyairi, Y., and
Matsuzaki, H. (2016). Weak monsoon event at 4.2 ka recorded in sediment from lake
rara,himalayas. Quaternary International,397:349–359. JapaneseQuaternaryStudies.
Nusbaumer, J., Wong, T. E., Bardeen, C., and Noone, D. (2017). Evaluating hydrological
processes in the Community Atmosphere Model Version 5 (CAM5) using stable isotope
ratios of water. Journal of Advances in Modeling Earth Systems, 9(2):949–977.
Oleson, K. W., Lawrence, D. M., Gordon, B., Flanner, M. G., Kluzek, E., Peter, J., Levis,
S., Swenson, S. C., Thornton, E., Feddema, J., et al. (2010). Technical description of
version 4.0 of the Community Land Model (CLM). Technical report, National Center
for Atmospheric Research.
129
Otto-Bliesner, B. L., Brady, E. C., Fasullo, J., Jahn, A., Landrum, L., Stevenson, S.,
Rosenbloom, N., Mai, A., and Strand, G. (2016). Climate variability and change since
850 CE: an ensemble approach with the Community Earth System Model. Bulletin of
the American Meteorological Society, 97(5):735–754.
Parnell, A. C., Haslett, J., Allen, J. R., Buck, C. E., and Huntley, B. (2008). A flexible
approach to assessing synchroneity of past events using Bayesian reconstructions of
sedimentation history. Quaternary Science Reviews, 27(19):1872–1885.
Preisendorfer, R. (1988). Principal component analysis in meteorology and oceanography.
Elsevier Sci. Publ., 17:425.
Psomiadis, D., Dotsika, E., Albanakis, K., Ghaleb, B., and Hillaire-Marcel, C. (2018).
Speleothem record of climatic changes in the northern Aegean region (Greece) from the
Bronze Age to the collapse of the Roman Empire. Palaeogeography, Palaeoclimatology,
Palaeoecology, 489:272–283.
Railsback, L. B., Liang, F., Brook, G., Voarintsoa, N. R. G., Sletten, H. R., Marais,
E., Hardt, B., Cheng, H., and Edwards, R. L. (2018). The timing, two-pulsed nature,
and variable climatic expression of the 4.2 ka event: A review and new high-resolution
stalagmite data from Namibia. Quaternary Science Reviews, 186:78–90.
Railsback, L. B., Liang, F., Romaní, J. R. V., Grandal-d’Anglade, A., Rodríguez, M. V.,
Fidalgo, L. S., Mosquera, D. F., Cheng, H., and Edwards, R. L. (2011). Petrographic
and isotopic evidence for Holocene long-term climate change and shorter-term environ-
mental shifts from a stalagmite from the Serra do Courel of northwestern Spain, and
implications for climatic history across Europe and the Mediterranean. Palaeogeogra-
phy, Palaeoclimatology, Palaeoecology, 305(1-4):172–184.
Redford, D. B. (2005). The Oxford encyclopedia of ancient Egypt. Oxford University
Press.
Rehfeld, K., Marwan, N., Heitzig, J., and Kurths, J. (2011). Comparison of correla-
tion analysis techniques for irregularly sampled time series. Nonlinear Processes in
Geophysics, 18(3):389–404.
130
Roland, T. P., Caseldine, C., Charman, D., Turney, C., and Amesbury, M. (2014). Was
there a ‘4.2 ka event’in Great Britain and Ireland? Evidence from the peatland record.
Quaternary Science Reviews, 83:11–27.
Sarkar, A., Mukherjee, A. D., Bera, M. K., Das, B., Juyal, N., Morthekai, P., Deshpande,
R. D., Shinde, V. S., and Rao, L. S. (2016). Oxygen isotope in archaeological bioap-
atites from India: Implications to climate change and decline of Bronze Age Harappan
civilization. Scientific Reports, 6:26555 EP –.
Savitzky, A. and Golay, M. J. (1964). Smoothing and differentiation of data by simplified
least squares procedures. Analytical Chemistry, 36(8):1627–1639.
Shen, C.-C., Lin, K., Duan, W., Jiang, X., Partin, J. W., Edwards, R. L., Cheng, H., and
Tan, M. (2013). Testing the annual nature of speleothem banding. Scientific Reports,
3:2633.
Shulmeister, J. and Lees, B. G. (1995). Pollen evidence from tropical Australia for the
onset of an ENSO-dominated climate at c. 4000 BP. The Holocene, 5(1):10–18.
Stanley, J.-D., Krom, M. D., Cliff, R. A., and Woodward, J. C. (2003). Short contribu-
tion: Nile flow failure at the end of the Old Kingdom, Egypt: strontium isotopic and
petrologic evidence. Geoarchaeology, 18(3):395–402.
Staubwasser, M., Sirocko, F., Grootes, P. M., and Segl, M. (2003). Climate change at
the 4.2 ka BP termination of the Indus valley civilization and Holocene south Asian
monsoon variability. Geophysical Research Letters, 30(8).
Staubwasser, M. and Weiss, H. (2006). Holocene climate and cultural evolution in late
prehistoric–early historic West Asia. Quaternary Research, 66(3):372–387.
Tan, L., Cai, Y., Cheng, H., Edwards, L. R., Gao, Y., Xu, H., Zhang, H., and An,
Z. (2018). Centennial-to decadal-scale monsoon precipitation variations in the upper
Hanjiang River region, China over the past 6650 years. Earth and Planetary Science
Letters, 482:580–590.
131
Thompson, L. G., Mosley-Thompson, E., Davis, M. E., Henderson, K. A., Brecher, H. H.,
Zagorodnov, V. S., Mashiotta, T. A., Lin, P.-N., Mikhalenko, V. N., Hardy, D. R., et al.
(2002). Kilimanjaro ice core records: evidence of Holocene climate change in tropical
Africa. Science, 298(5593):589–593.
Tierney, J. E., Oppo, D. W., LeGrande, A. N., Huang, Y., Rosenthal, Y., and Linsley,
B. K. (2012). The influence of Indian Ocean atmospheric circulation on Warm Pool
hydroclimate during the Holocene epoch. Journal of Geophysical Research: Atmo-
spheres, 117(D19).
Tierney, J. E., Russell, J. M., Damsté, J. S. S., Huang, Y., and Verschuren, D. (2011).
Late Quaternary behavior of the East African monsoon and the importance of the
Congo Air Boundary. Quaternary Science Reviews, 30(7-8):798–807.
Treble,P.,Mah,M.,Griffiths,A.,Baker,A.,Deininger,M.,Kelly,B.,Scholz,D.,andHan-
kin, S. (2019). Separating isotopic impacts of karst and in-cave processes from climate
variability using an integrated speleothem isotope-enabled forward model. EarthArXiv.
Voosen, P. (2018). New geological age comes under fire. Science, 361(6402):537–538.
Walker, M., Head, M. H., Berklehammer, M., Bjorck, S., Cheng, H., Cwynar, L., Fisher,
D., Gkinis, V., Long, A., Lowe, J., et al. (2018). Formal ratification of the subdivision
of the Holocene Series/Epoch (Quaternary System/Period): two new Global Boundary
Stratotype Sections and Points (GSSPs) and three new stages/subseries. Episodes.
Walker, M. J., Berkelhammer, M., Björck, S., Cwynar, L. C., Fisher, D. A., Long, A. J.,
Lowe, J. J., Newnham, R. M., Rasmussen, S. O., and Weiss, H. (2012). Formal subdi-
vision of the Holocene Series/Epoch: a Discussion Paper by a Working Group of INTI-
MATE (Integration of ice-core, marine and terrestrial records) and the Subcommission
on Quaternary Stratigraphy (International Commission on Stratigraphy). Journal of
Quaternary Science, 27(7):649–659.
Wan, N.-J., Li, H.-C., Liu, Z.-Q., Yang, H.-Y., Yuan, D.-X., and Chen, Y.-H. (2011).
Spatial variations of monsoonal rain in eastern China: Instrumental, historic and
speleothem records. Journal of Asian Earth Sciences, 40(6):1139–1150.
132
Wang, Y., Cheng, H., Edwards, R. L., He, Y., Kong, X., An, Z., Wu, J., Kelly, M. J.,
Dykoski, C. A., and Li, X. (2005). The Holocene Asian monsoon: links to solar changes
and North Atlantic climate. Science, 308(5723):854–857.
Wang, Y., Cheng, H., Edwards, R. L., Kong, X., Shao, X., Chen, S., Wu, J., Jiang, X.,
Wang, X., and An, Z. (2008). Millennial-and orbital-scale changes in the East Asian
monsoon over the past 224,000 years. Nature, 451(7182):1090–1093.
Wang, Y.-J., Cheng, H., Edwards, R. L., An, Z., Wu, J., Shen, C.-C., and Dorale, J. A.
(2001). A high-resolution absolute-dated late Pleistocene monsoon record from Hulu
Cave, China. Science, 294(5550):2345–2348.
Weiss, H. (2016). Global megadrought, societal collapse and resilience at 4.2–3.9 ka BP
across the Mediterranean and west Asia. PAGES, 24:62–63.
Weiss, H., Courty, M.-A., Wetterstrom, W., Guichard, F., Senior, L., Meadow, R., and
Curnow, A. (1993). The genesis and collapse of third millennium north Mesopotamian
civilization. Science, 261(5124):995–1004.
Weiss, H., Manning, S., Ristvet, L., Mori, L., Besonen, M., McCarthy, A., Quenet, P.,
Smith, A., and Bahrani, Z. (2012). Tell Leilan Akkadian imperialization, collapse, and
short-lived reoccupation defined by high-resolution radiocarbon dating. Seven Genera-
tions Since the Fall of Akkad, ed Weiss H (Harrassowitz Verlag, Wiesbaden, Germany).
Wong, T. E., Nusbaumer, J., and Noone, D. C. (2017). Evaluation of modeled land-
atmosphere exchanges with a comprehensive water isotope fractionation scheme in ver-
sion 4 of the Community Land Model. Journal of Advances in Modeling Earth Systems,
9(2):978–1001.
Wright, R. P. (2010). The ancient Indus: urbanism, economy, and society, case studies
in early societies. Cambridge: Cambridge University Press.
Wu, W. and Liu, T. (2004). Possible role of the “Holocene Event 3” on the collapse
of Neolithic Cultures around the Central Plain of China. Quaternary International,
117(1):153–166.
133
Yuan, D., Cheng, H., Edwards, R. L., Dykoski, C. A., Kelly, M. J., Zhang, M., Qing, J.,
Lin, Y., Wang, Y., Wu, J., et al. (2004). Timing, duration, and transitions of the last
interglacial Asian monsoon. Science, 304(5670):575–578.
Zhang, H., Cheng, H., Cai, Y., Spötl, C., Kathayat, G., Sinha, A., Edwards, R. L., and
Tan, L. (2018a). Hydroclimatic variations in southeastern China during the 4.2 ka
event reflected by stalagmite records. Climate of the Past, 14(11):1805–1817.
Zhang, H., Griffiths, M. L., Chiang, J. C., Kong, W., Wu, S., Atwood, A., Huang, J.,
Cheng, H., Ning, Y., and Xie, S. (2018b). East Asian hydroclimate modulated by the
position of the westerlies during Termination I. Science, 362(6414):580–583.
134
Conclusion
Summary
The objective of this thesis was to better interpret Asian speleothem δ
18
O. I tack-
led this problem with statistical and climate model tools to disentangle different factors
controlling speleothem δ
18
O. The thesis accomplished this over four chapters, with the
first two chapters laying down the statistical and modeling foundations, and the last
two chapters applying them to the interpretation of Chinese cave records and Holocene
hydroclimate variability over Asia.
Chapter 1 discussed the challenges in interpreting speleothem δ
18
O based on correla-
tion analysis alone. These challenges are: the loss of degrees of freedom due to serial cor-
relation, the test multiplicity problem in connection with a climate field, and the presence
of age uncertainties. I showed how these challenges applied to, and affected, the pub-
lished interpretations of paleoclimate proxies by taking the recent paleoclimate literature
(Proctor et al., 2000, Zhu et al., 2012, McCabe-Glynn et al., 2013) as examples. In some
cases, correlations that were previously claimed to be significant were found insignificant,
thereby challenging published interpretations. Thus we suggested that future paleocli-
mate studies (including speleothem-related studies) should address these challenges in
correlation analysis. More importantly, we argued that correlation analysis should not be
the sole basis for interpretation. Also, the methods and open-source code to circumvent
these issues were provided. This will help future studies when they attempt to draw
conclusions based on correlation analysis.
Since I employed the isotope-enabled climate model iCESM (and its atmospheric com-
ponent iCAM5) to investigate the interpretation of Asian speleothem δ
18
O in subsequent
135
chapters, Chapter 2 evaluated whether the model was capable of credibly simulating vari-
ability in precipitationδ
18
O (δ
18
O
P
). I focused on how the model simulated the impact of
convective activity on δ
18
O
P
compared with six other isotope-enabled models, as convec-
tive activity plays an important role in influencing δ
18
O
P
in the Asian monsoon region.
This model comparison will also help the development of isotope-enabled models, espe-
cially constraining convective processes in general circulation models (Bony et al., 2008,
Lee et al., 2009, Risi et al., 2012, Tharammal et al., 2017). The result showed that only
iCAM5 can simulate the observed anticorrelation between stratiform fraction and δ
18
O
P
,
and that gave us confidence in employing this model to investigateδ
18
O
P
variability in the
following studies. It also implied that there were no shortcuts for isotope-enabled models
to simulate the role of precipitation types in δ
18
O
P
. The necessary processes in cloud
microphysics have to be captured if one wants to use general circulation models to study
the impact of convection on δ
18
O
P
. In addition, iCAM5 allowed to estimate the relative
contributions of local rainfall amount, local water vapor convergence, convective activity
to precipitation δ
18
O variability in the Asian monsoon regions. I found that the contri-
bution of convective processes was very site-dependent, with local processes accounting
for a very small amount of variance at the sites of most Asian speleothem records. This
implied a major contribution of water source and transport effects, which I discussed in
the following chapter.
In Chapter 3, iCESM was utilized to investigate the interpretation of Chinese
speleothem δ
18
O at orbital and interannual scales. Based on Chapter 2, iCESM could
simulate the variability of δ
18
O
P
well and it was able to track and tag water vapor and
its oxygen isotopes once it evaporates from specific regions (Tabor et al., 2018). With
the aid of this water vapor tagging technique, I decomposed the change of δ
18
O
P
at both
orbital and interannual time scales into four parts: changes of moisture source composi-
tion, rainout changes, condensation changes, and changes of moisture locations and the
results showed that the change of moisture source locations dominate the change ofδ
18
O
P
at both time scales. At orbital scales, the variability of Chinese speleothem δ
18
O mainly
represents the meridional migration of the East Asian monsoon circulation, instead of
precipitation over Chinese speleothem sites. In precession minima (when the Northern
Hemisphere was towards the sun in summer solstice), the East Asian rainfall belt moved
136
northward earlier than today, so the rainy season in central China was shortened. At
interannual scales, more moisture coming from remote regions (mainly the Indian Ocean)
leads to depleted speleothem δ
18
O, and its influence is greatest in late summer and early
autumn. This work offers a re-interpretation of the concept of “monsoon intensity” as
“enhanced monsoonal circulation” rather than precipitation amount.
Despite the coherency of Asian speleothem δ
18
O at orbital scales (Cheng et al., 2012,
Battisti et al., 2014, Cheng et al., 2016), whether this coherency exists at sub-orbital
scales remains a question. Chapter 4 investigated this question over the Holocene. The
result shows no coherent variability among Asian speleothem δ
18
O at these time scales.
Based on the simulation results from iCESM (Wong et al., 2017, Nusbaumer et al., 2017)
and a karst forward model (Karstolution, Treble et al., 2019), this asynchrony can be
explained by heterogeneous soil, vegetation, and karst processes modifying coherent cli-
mate inputs. This indicated that extracting hydroclimate variability at sub-orbital scales
from speleothem δ
18
O should be done with caution, given the relatively small amplitude
(1-3h) of speleothems at these time scales. In addition, the Global Stratotype Section
and Point (GSSP) for the newly announced Meghalayan age (4.2 ky BP to present) is a
stalagmite from Mawmluh cave, in the Indian state of Meghalaya. Thus I investigated
whether Asian speleothem δ
18
O records express the “4.2 ka event”. We found no “4.2
ka event” outside of the Mawmluh cave record of Berkelhammer et al. (2013), suggesting
no regional or global climate signal at this time. Further, two new high-resolution series
from Mawmluh cave suggested that the “4.2 ka event” was far less abrupt than originally
published. Although there was evidence for a drought affecting Meghalaya around 4.2
ky BP, it appeared to have been spatially restricted, questioning whether Meghalaya was
representative of global conditions over the Late Holocene.
In summary, at orbital scales, Asian speleothem δ
18
O is mainly controlled by the
large-scale atmospheric circulation–moisture source location change and it represents the
meridional migration of monsoon winds instead of rainfall amount. At sub-orbital scales,
when the amplitude of δ
18
O is small, Asian speleothem δ
18
O variability is not coherent
and greatly influenced by soil, vegetation, karst processes, and age uncertainties. Thus we
should be especially cautious about the interpretation of speleothem δ
18
O at these time
scales. Cave monitoring, analyzing multi-proxies of speleothems, comparing with other
137
paleoclimate proxies, and taking account of age uncertainties can help us better decipher
climate signals from speleothems.
Caveats and Future Work
The thesis employed iCESM to understand the interpretation of speleothemδ
18
O, and
asallmodelsareapproximaterepresentationsoftherealworld, ithaslimitations. Chapter
3 shows that precipitation south of the Tibet was overestimated in the model, which was
partly due to the low resolution used in the thesis. Gao et al. (2011) showed that zooming
to 0.5
◦
spatial resolution dramatically improved the simulation of precipitation andδ
18
O
P
in one isotope-enabled model (LMDZ). Thus increasing the spatial resolution in the model
is worth doing in the future. Also, the response ofδ
18
O
P
to precessional forcing over East
Asia is relatively weak compared with speleothem records. It can be partly explained
by the different responses in the North Pacific subtropical high and the Asian monsoon
circulation when the Northern Hemisphere receives more insolation. When the former
response is stronger than it should be, the δ
18
O
P
change in East Asia is dampened. This
weak response of δ
18
O
P
in East Asia seems common in isotope-enabled climate models
such as ECHAM4.6 (Battisti et al., 2014, Roe et al., 2016) and iLOVECLIM (Caley et al.,
2014) and needs further investigation. Finally, I only used one model in the study. It
is worth checking whether other isotope-enabled models generate similar results in the
same experimental settings. Recent studies also show that multi-model ensembles can
help constrain model uncertainties (Piani et al., 2005, Knutti and Sedláček, 2013). Thus
running isotope-enabled models under the same experiment settings such as the same
orbital configurations will give us more confidence in the interpretation of δ
18
O
P
and
speleothem δ
18
O in the future.
In Chapter 3, I mainly focused on Chinese speleothems. The variability of Asian
speleothem δ
18
O in other places such as India and the Maritime Continent is also worth
investigating (Berkelhammer et al., 2013, Carolin et al., 2013, 2016, Griffiths et al., 2016,
Chen et al., 2016, Kathayat et al., 2016). For example, the variability of speleothem
δ
18
O in Borneo, Malaysia is quite different from Chinese speleothems at both orbital and
138
interannual time scales. Its interannual variability seems strongly connected to El Niño-
Southern Oscillation (Moerman et al., 2013, Chen et al., 2016), and its orbital variability
aligns more with its local fall (September, October, and November) insolation instead of
summer insolation like speleothems in China (Carolin et al., 2016). Bischoff et al. (2017)
proposed a theoretical model to explain the phase lagging of precipitation at different
latitudes in response to insolation, but there are still no direct explanations for δ
18
O
P
or speleothem δ
18
O yet. Also, the reason why speleothem δ
18
O responds most in the
precessional band is still mysterious, and recently Clemens et al. (2018) found that marine
sediments in East Asia did not feature precession-scale variance. Further investigations
including isotope-enabled general circulation models, theoretical models for the response
of δ
18
O
P
to insolation, and new records (other proxies) at orbital scales are necessary.
There are other experiments can be done with the isotope-enabled model and the
water tagging technique to understand the interpretation of speleothemδ
18
O. In Chapter
3, water tagging experiments were conducted at orbital and interannual time scales, but
not at time scales between them. It would be interesting to know whether, at centennial-
millennial time scales, the change of moisture source location still dominates the δ
18
O
P
variability. Also, Chapter 4 showed that Asian speleothemδ
18
O were not coherent at sub-
orbital (centennial-millennial) scales, and we also would like to know whether the isotope-
enabled model can simulate similar patterns in δ
18
O
P
and soil δ
18
O. Thus conducting
experiments similar to Chapter 3 at centennial-millennial time scales is necessary in the
future. In addition, limited by computational resources, I ran the precessional forcing
experiments based on orbital configurations at one single time, while in reality, the orbital
forcing changes through time. In the future, experiments with transient orbital forcing
should be conducted by isotope-enabled models, even if by intermediate-complexity mod-
els (Severijns and Hazeleger, 2010, Roche, 2013). Finally, the water tagging techniques
not only can tag water vapor from different moisture source regions but also tag water
vapor condensed at different heights, or raindrops with different life cycles (like how many
times it has been through condensation and evaporation processes). In this way, we can
better understand δ
18
O
P
variability physically.
Chapter 4 discussed the influence of non-climate factors such as soil, vegetation and
karstprocessestospeleothemδ
18
O,andakarstmodel(Karstolution)wasusedtoillustrate
139
the role of karst processes. The model simplifies complex water paths in caves and gives
five archetypes of stalagmites, and assumes the capacity of water stores in the karst
systems to be constant through time, which may not be true in reality. Also, to better
use this model, the knowledge of water sources of the studied speleothem is necessary.
This can be achieved through detailed hydrological investigations or cave monitoring by
comparing δ
18
O
P
and drip water δ
18
O. Collecting and analyzing cave monitoring data
with speleothem proxies will help us better constrain and understand the influence of non-
climate factors on speleothem records (Baker et al., 2018). In addition, Chapter 4 showed
a strong asynchrony of Asian speleothem δ
18
O at sub-orbital scales, but the coherency of
speleothem records at other short-time scales from interannual to centennial scales were
not investigated.
Although strong asynchrony of speleothemδ
18
O is shown at sub-orbital scales, it does
not mean that it is hopeless to correctly interpret speleothem δ
18
O at these time scales.
There are several approaches to improving the robustness of the interpretation. The
reproducibility of speleothem records should be checked first if available, especially for
speleothems in the same cave. This method has been used in many speleothem studies
to ensure the published speleothem records represent the regional variability instead of
karst system processes (Wang et al., 2001, Partin et al., 2007, Wang et al., 2008, Cheng
et al., 2009, Moseley et al., 2014). The coherency of speleothem records gives confidence
that they share a common hydroclimatic driver. Secondly, the cave monitoring is helpful
to interpret speleothem δ
18
O since we can compare modern drip water δ
18
O to rainfall
amount and temperature, revealing the role of karst system processes (Moerman et al.,
2014, Duan et al., 2016, Baker et al., 2018). Thirdly, other speleothem observations
such as δ
13
C and other trace elements (such as Sr/Ca, Ba/Ca and Mg/Ca) also help us
understand the interpretation of speleothem δ
18
O (Fairchild et al., 2000, Johnson et al.,
2006, Fairchild and Treble, 2009). Trace element ratios of speleothems are believed to be
more sensitive the local hydroclimate change thanδ
18
O
P
(Liu et al., 2013, Griffiths et al.,
2016, Zhang et al., 2018). Comparing multiple data streams from the same speleothem
record will help us better understand speleothem δ
18
O. Furthermore, there are other
hydroclimate proxies in lake sediments (Ji et al., 2005, Morrill et al., 2006), ice cores
(Thompson et al., 2000, Pang et al., 2014), tree ring (Feng et al., 1999, Zhang et al.,
140
2003), loess (Zhisheng et al., 2001, Qiang et al., 2001) in the Asian monsoon region.
Thus, comparing these with speleothem records will also help us extract climate signals
from speleothems. Also, these proxies are often located in regions where speleothems do
not grow, and thus give us a more complete picture of the spatial-temporal variability of
hydroclimate variability in Asia.
Outlook
To better interpret speleothemδ
18
O and understand the past hydroclimate variability,
there is abundant room for future progress in modeling, data collection, and analysis
methods. Isotope-enabled models and karst forward models provide opportunities to
conduct numerical experiments to understand what speleothemδ
18
O represent. With the
improvementof themodels, computation resources, andthesourcetaggingtechniqueused
in this thesis, we can test hypotheses about the speleothem variability which were not
tested before. One hypothesis yet to be tested is whether the phase lag between Chinese
and Borneo speleothem δ
18
O at orbital scales needs the modulation of ENSO (Carolin
et al., 2016). Also, when one connects speleothem records with climate variables, these
models are effective tools to test the robustness of their relationship.
Cave monitoring data is important for us to better understand speleothem
δ
18
Obecauseitgivesusreliabledatasetsofdripwaterδ
18
O, surfaceprecipitationδ
18
Oand
otherclimatevariables, whichgivesusthechancetolearnaboutthehydrologicalprocesses
in the cave, but unfortunately the current monitoring data is still insufficient, especially
the long-term monitoring (Baker et al., 2018). Thus, building a consistent, long-term cave
monitoring network is necessary to ground-truth speleothem records.
For the speleothem collection, there are still evident spatial gaps among current pub-
lished Asian speleothem records. For instance, few long-term high-resolution speleothem
recordshavebeenfoundinnorthernChinaandtheregionbetweenSanbaoandHulucaves,
where the summer main rainbelt dwells. Filling these gaps will help us better understand
the spatial-temporal hydroclimate variability and test the hypothesis that the modern
triple-mode pattern of Chinese precipitation at the interannual scale also applies for the
past climate at longer time scales. In addition, since it is promising to use trace element
141
ratios of speleothems to better understand speleothem δ
18
O (Liu et al., 2013, Griffiths
et al., 2016, Zhang et al., 2018), it would be helpful for the speleothem community to
collect these records globally alongside δ
18
O and δ
13
C.
As shown in the thesis, age uncertainties are important when analyzing speleothem
data at time scales shorter than centennial. We encourage the speleothem community
to translate the age uncertainties to proxy uncertainties in the analysis by generating
age ensembles via Bayesian age models like we did in Chapter 1 and 4. In this way,
the age uncertainties are indeed incorporated the analysis, and there will be an ensemble
of chronologies, rather than a single one for each speleothem record. Publicly available
speleothem records often only provide the estimated age along with proxy values, but the
depth of samples is necessary for other scientists to generate age ensembles.
Furthermore, comparing speleothem δ
18
O with other proxies can always help better
understand the interpretation of speleothems. For instance, the recent finding of marine
sediments in East China Sea representing the monsoon runoff that did not feature the
23-kyr cycles shown in speleothem records at orbital scales (Clemens et al., 2018) makes
us rethink why speleothemδ
18
O primarily responds to insolation in the precessional band.
Another example is that comparing lake sediments with speleothem records, which can
question the interpretation of speleothem δ
18
O as rainfall amount (Liu et al., 2015, Rao
et al., 2016). Thus integrating speleothem δ
18
O with other climate sensors will give us a
more comprehensive picture of past hydroclimate variability.
Finally, the proper interpretation of Asian speleothems will give us a more accurate
picture of the past hydroclimate in Asia, which can ultimately help to better project
future hydroclimate variability in this region, where over 4 billion people live. There are
still large uncertainties of the future projection of the Asian monsoon or precipitation
patterns among models under high-CO
2
scenarios (Lee and Wang, 2014, Wang et al.,
2014). Speleothem records can be used to characterize or even reduce these uncertainties.
One part of the projection uncertainties stems from models’ capability to simulate
feedbacks internal to the climate system. More reliable interpretations of speleothem
records can offer the opportunity to characterize and understand this natural variability.
One of the relevant characteristics is the variability of the Asian monsoon or rainfall pat-
terns. As suggested from Chapter 3, the Asian speleothem δ
18
O represents the migration
142
of the Asian monsoon circulation, so we could infer the magnitude of the monsoon migra-
tion from speleothems at different time scales, its responses to various orbital forcing,
and whether this variation exhibits periodic behavior. Also, we are interested in learning
from speleothems whether the triple-pattern of East Asian summer rainfall at interan-
nual scales applies for decadal and longer scales, and how this pattern moves, extends or
changes its magnitude through time. In short, learning the magnitude, the periodicity
of the monsoon and its related circulation from speleothems should help understand the
projection uncertainties due to continental-scale climate dynamics, and help us better
understand GCM projections and attribute climate variability.
The characteristics of past hydroclimate can be also inferred by data assimilation,
which provide physically coherent climate fields rather than isolated time series. In the
assimilation framework, the climate model is coupled with proxy forward models to pro-
ducepseudoproxies, andclimatevariablesateachstepareadjustedaccordingtothediffer-
ence between observed proxies and pseudoproxies to fit the past climate, and speleothems
are one important proxy used. Once we have better modeling of speleothem δ
18
O and
incorporate more speleothem data into the framework, the assimilation result will be
improved. One example of the data assimilation project is the Last Millennium climate
Reanalysis (Hakim et al., 2016), and its result has been applied to study the multidecadal
climate variability (Singh et al., 2018). With the development of speleothem forward
models and Asian speleothem data, the assimilation method will reveal a more compre-
hensive picture of the past hydroclimate in Asia. We should note that speleothems are
not coherent at sub-orbital scales, based on Chapter 4, so they may not provide a reliable
constraint at these shorter scales, but they can be very useful at time scales longer than
millennial.
Speleothem records can also be used to constrain projection uncertainties. The proper
interpretation of speleothems can give us inferred climate variables such as precipitation
and the northern boundary of the monsoon circulation, and they can be used as metrics
to constrain projections. The uncertainties due to model physics can be constrained by
Bayesian selection of parameters from perturbed physics ensembles (Collins et al., 2012,
Schmidt et al., 2013). The idea is to run the model with different sets of input physics
143
parameters. At first, each ensemble member has the same weight. Then the probabilis-
tic weights of each ensemble member are calculated according to their performance in
reproducing metrics derived from observations. The weighted ensemble is the posterior
estimate of projections. Also, the model spread can be constrained by discriminating
the performance of models based on climate variables inferred from speleothems. For
instance, Schmidt et al. (2013) selected models simulating the rainfall pattern of South
America better than other CMIP5 models in the mid-Holocene and reduced the model
spread of rainfall change there in the future.
Lastly, as speleothems feature high dating precision and coincide with millennial-scale
global climate signals sunch as glacial-interglacial cycles and Dansgaard-Oeschger events,
they are often used to constrain or tune other chronologies (e.g., Greenland ice cores,
Bar-Matthews et al., 1999, Wang et al., 2001; South China Sea sediments, Caballero-
Gill et al., 2012). However, sometimes the tuning is subjective in selecting peaks. A
better interpretation of speleothem records can let us know why the subtropical climate
was linked to polar climate change, whether they occurred simultaneously (whether lags
exist) and what climate events are suitable for tuning. In this way, we can gain confidence
in the chronologies refined by speleothems.
Bibliography
Baker, A., Duan, W., Cuthbert, M., Treble, P., Banner, J., and Hankin, S. (2018). Cli-
maticinfluencesontheoffsetbetweenδ
18
Oofcavedripwatersandprecipitationinferred
from global monitoring data. EarthArXiv.
Bar-Matthews, M., Ayalon, A., Kaufman, A., and Wasserburg, G. J. (1999). The Eastern
Mediterranean paleoclimate as a reflection of regional events: Soreq cave, Israel. Earth
and Planetary Science Letters, 166(1-2):85–95.
Battisti, D., Ding, Q., and Roe, G. (2014). Coherent pan-Asian climatic and isotopic
response to orbital forcing of tropical insolation. Journal of Geophysical Research:
Atmospheres, 119(21):11–997.
144
Berkelhammer, M., Sinha, A., Stott, L., Cheng, H., Pausata, F., and Yoshimura, K.
(2013). An abrupt shift in the Indian monsoon 4000 years ago. Climates, Landscapes,
and Civilizations, pages 75–88.
Bischoff, T., Schneider, T., and Meckler, A. N. (2017). A conceptual model for the
response of tropical rainfall to orbital variations. Journal of Climate, 30(20):8375–8391.
Bony, S., Risi, C., and Vimeux, F. (2008). Influence of convective processes on the
isotopic composition (δ
18
O and δD) of precipitation and water vapor in the tropics:
1. Radiative-convective equilibrium and Tropical Ocean–Global Atmosphere–Coupled
Ocean-Atmosphere Response Experiment (TOGA-COARE) simulations. Journal of
Geophysical Research: Atmospheres, 113:D19305.
Caballero-Gill,R.P.,Clemens,S.C.,andPrell,W.L.(2012). DirectcorrelationofChinese
speleothem δ
18
O and South China Sea planktonic δ18O: Transferring a speleothem
chronology to the benthic marine chronology. Paleoceanography, 27(2).
Caley, T., Roche, D. M., and Renssen, H. (2014). Orbital Asian summer monsoon dynam-
ics revealed using an isotope-enabled global climate model. Nature Communications,
5.
Carolin, S. A., Cobb, K. M., Adkins, J. F., Clark, B., Conroy, J. L., Lejau, S., Malang,
J., and Tuen, A. A. (2013). Varied response of western Pacific hydrology to climate
forcings over the last glacial period. Science, page 1233797.
Carolin, S. A., Cobb, K. M., Lynch-Stieglitz, J., Moerman, J. W., Partin, J. W., Lejau,
S., Malang, J., Clark, B., Tuen, A. A., and Adkins, J. F. (2016). Northern Borneo
stalagmite records reveal West Pacific hydroclimate across MIS 5 and 6. Earth and
Planetary Science Letters, 439:182–193.
Chen, S., Hoffmann, S. S., Lund, D. C., Cobb, K. M., Emile-Geay, J., and Adkins,
J. F. (2016). A high-resolution speleothem record of western equatorial Pacific rain-
fall: Implications for Holocene ENSO evolution. Earth and Planetary Science Letters,
442:61–71.
145
Cheng, H., Edwards, R. L., Broecker, W. S., Denton, G. H., Kong, X., Wang, Y., Zhang,
R., and Wang, X. (2009). Ice Age Terminations. Science, 326(5950):248–252.
Cheng, H., Edwards, R. L., Sinha, A., Spötl, C., Yi, L., Chen, S., Kelly, M., Kathayat,
G., Wang, X., Li, X., et al. (2016). The Asian monsoon over the past 640,000 years
and ice age terminations. Nature, 534(7609):640–646.
Cheng, H., Sinha, A., Wang, X., Cruz, F. W., and Edwards, R. L. (2012). The Global
PaleomonsoonasseenthroughspeleothemrecordsfromAsiaandtheAmericas. Climate
Dynamics, 39(5):1045–1062.
Clemens, S., Holbourn, A., Kubota, Y., Lee, K., Liu, Z., Chen, G., Nelson, A., and Fox-
Kemper, B. (2018). Precession-band variance missing from East Asian monsoon runoff.
Nature Communications, 9(1):3364.
Collins, M., Chandler, R. E., Cox, P. M., Huthnance, J. M., Rougier, J., and Stephenson,
D. B. (2012). Quantifying future climate change. Nature Climate Change, 2(6):403.
Duan, W., Ruan, J., Luo, W., Li, T., Tian, L., Zeng, G., Zhang, D., Bai, Y., Li, J., Tao,
T., et al. (2016). The transfer of seasonal isotopic variability between precipitation and
drip water at eight caves in the monsoon regions of china. Geochimica et Cosmochimica
Acta, 183:250–266.
Fairchild, I. J., Borsato, A., Tooth, A. F., Frisia, S., Hawkesworth, C. J., Huang, Y.,
McDermott, F., and Spiro, B. (2000). Controls on trace element (Sr–Mg) composi-
tions of carbonate cave waters: implications for speleothem climatic records. Chemical
Geology, 166(3-4):255–269.
Fairchild, I. J. and Treble, P. C. (2009). Trace elements in speleothems as recorders of
environmental change. Quaternary Science Reviews, 28(5-6):449–468.
Feng, X., Cui, H., Tang, K., and Conkey, L. E. (1999). Tree-ring δD as an indicator of
Asian monsoon intensity. Quaternary Research, 51(3):262–266.
Gao, J., Masson-Delmotte, V., Yao, T., Tian, L., Risi, C., and Hoffmann, G. (2011).
Precipitation water stable isotopes in the south Tibetan Plateau: observations and
modeling. Journal of Climate, 24(13):3161–3178.
146
Griffiths, M. L., Kimbrough, A. K., Gagan, M. K., Drysdale, R. N., Cole, J. E., Johnson,
K. R., Zhao, J.-X., Cook, B. I., Hellstrom, J. C., and Hantoro, W. S. (2016). Western
Pacific hydroclimate linked to global climate variability over the past two millennia.
Nature Communications, 7:11719.
Hakim, G. J., Emile-Geay, J., Steig, E. J., Noone, D., Anderson, D. M., Tardif, R.,
Steiger, N., and Perkins, W. A. (2016). The Last Millennium Climate Reanalysis
Project: Framework and First Results. Journal of Geophysical Research: Atmospheres.
Ji, J., Shen, J., Balsam, W., Chen, J., Liu, L., and Liu, X. (2005). Asian monsoon
oscillations in the northeastern Qinghai–Tibet Plateau since the late glacial as inter-
preted from visible reflectance of Qinghai Lake sediments. Earth and Planetary Science
Letters, 233(1):61–70.
Johnson, K. R., Hu, C., Belshaw, N. S., and Henderson, G. M. (2006). Seasonal trace-
element and stable-isotope variations in a Chinese speleothem: The potential for high-
resolution paleomonsoon reconstruction. Earth and Planetary Science Letters, 244(1-
2):394–407.
Kathayat, G., Cheng, H., Sinha, A., Spötl, C., Edwards, R. L., Zhang, H., Li, X., Yi,
L., Ning, Y., Cai, Y., et al. (2016). Indian monsoon variability on millennial-orbital
timescales. Scientific Reports, 6.
Knutti, R. and Sedláček, J. (2013). Robustness and uncertainties in the new CMIP5
climate model projections. Nature Climate Change, 3(4):369.
Lee, J.-E., Pierrehumbert, R., Swann, A., and Lintner, B. R. (2009). Sensitivity of stable
water isotopic values to convective parameterization schemes. Geophysical Research
Letters, 36(23):L23801.
Lee, J.-Y. and Wang, B. (2014). Future change of global monsoon in the CMIP5. Climate
Dynamics, 42(1-2):101–119.
Liu, J., Chen, J., Zhang, X., Li, Y., Rao, Z., and Chen, F. (2015). Holocene East
Asian summer monsoon records in northern China and their inconsistency with Chinese
stalagmite δ
18
O records. Earth-Science Reviews, 148:194–208.
147
Liu, Y., Henderson, G., Hu, C., Mason, A., Charnley, N., Johnson, K., and Xie, S. (2013).
Links between the East Asian monsoon and North Atlantic climate during the 8,200
year event. Nature Geoscience, 6(2):117–120.
McCabe-Glynn, S., Johnson, K. R., Strong, C., Berkelhammer, M., Sinha, A., Cheng, H.,
and Edwards, R. L. (2013). Variable North Pacific influence on drought in southwestern
North America since AD 854. Nat. Geosci., 6(8):617–621.
Moerman, J. W., Cobb, K. M., Adkins, J. F., Sodemann, H., Clark, B., and Tuen, A. A.
(2013). Diurnal to interannual rainfall δ
18
O variations in northern Borneo driven by
regional hydrology. Earth and Planetary Science Letters, 369:108–119.
Moerman, J.W., Cobb, K.M., Partin, J.W., Meckler, A.N., Carolin, S.A., Adkins, J.F.,
Lejau, S., Malang, J., Clark, B., and Tuen, A. A. (2014). Transformation of ENSO-
related rainwater to dripwater δ
18
O variability by vadose water mixing. Geophysical
Research Letters, 41(22):7907–7915.
Morrill, C., Overpeck, J. T., Cole, J. E., Liu, K.-b., Shen, C., and Tang, L. (2006).
Holocene variations in the Asian monsoon inferred from the geochemistry of lake sedi-
ments in central Tibet. Quaternary Research, 65(2):232–243.
Moseley, G. E., Spötl, C., Svensson, A., Cheng, H., Brandstätter, S., and Edwards, R. L.
(2014). Multi-speleothem record reveals tightly coupled climate between central Europe
and Greenland during Marine Isotope Stage 3. Geology, 42(12):1043–1046.
Nusbaumer, J., Wong, T. E., Bardeen, C., and Noone, D. (2017). Evaluating hydrological
processes in the Community Atmosphere Model Version 5 (CAM5) using stable isotope
ratios of water. Journal of Advances in Modeling Earth Systems, 9(2):949–977.
Pang, H., Hou, S., Kaspari, S., andMayewski, P.(2014). Influenceofregionalprecipitation
patterns on stable isotopes in ice cores from the central Himalayas. The Cryosphere,
8(1):289–301.
Partin, J. W., Cobb, K. M., Adkins, J. F., Clark, B., and Fernandez, D. P. (2007).
Millennial-scale trends in west Pacific warm pool hydrology since the Last Glacial Max-
imum. Nature, 449(7161):452.
148
Piani, C., Frame, D., Stainforth, D., and Allen, M. (2005). Constraints on climate change
from a multi-thousand member ensemble of simulations. Geophysical Research Letters,
32(23).
Proctor, C., Baker, A., Barnes, W., and Gilmour, M. (2000). A thousand year speleothem
proxy record of North Atlantic climate from Scotland. Climate Dynamics, 16(10-
11):815–820.
Qiang, X., Li, Z., Powell, C. M., and Zheng, H. (2001). Magnetostratigraphic record of
the Late Miocene onset of the East Asian monsoon, and Pliocene uplift of northern
Tibet. Earth and Planetary Science Letters, 187(1):83–93.
Rao, Z., Li, Y., Zhang, J., Jia, G., and Chen, F. (2016). Investigating the long-term
palaeoclimatic controls on theδD andδ
18
O of precipitation during the Holocene in the
Indian and East Asian monsoonal regions. Earth-Science Reviews, 159:292–305.
Risi, C., Noone, D., Worden, J., Frankenberg, C., Stiller, G., Kiefer, M., Funke, B.,
Walker, K., Bernath, P., Schneider, M., et al. (2012). Process-evaluation of tropospheric
humidity simulated by general circulation models using water vapor isotopologues: 1.
Comparison between models and observations. Journal of Geophysical Research: Atmo-
spheres, 117:D05303.
Roche, D. (2013). δ
18
O water isotope in the iLOVECLIM model (version 1.0)–Part 1:
Implementation and verification. Geoscientific Model Development, 6(5):1481–1491.
Roe, G. H., Ding, Q., Battisti, D. S., Molnar, P., Clark, M. K., and Garzione, C. N.
(2016). A modeling study of the response of Asian summertime climate to the largest
geologic forcings of the past 50 Ma. Journal of Geophysical Research: Atmospheres,
121(10):5453–5470.
Schmidt, G., Annan, J., Bartlein, P., Cook, B., Guilyardi, É., Hargreaves, J., Harrison,
S., Kageyama, M., LeGrande, A., Konecky, B., et al. (2013). Using palaeo-climate
comparisons to constrain future projections in CMIP5. Climate of the Past, 10(1):221–
250.
149
Severijns, C. and Hazeleger, W. (2010). The efficient global primitive equation climate
model SPEEDO V2. 0. Geoscientific Model Development, 3(1):105.
Singh, H. K., Hakim, G. J., Tardif, R., Emile-Geay, J., and Noone, D. C. (2018). Insights
into Atlantic multidecadal variability using the Last Millennium Reanalysis framework.
Climate of the Past, 14(2):157.
Tabor, C. R., Otto-Bliesner, B. L., Brady, E. C., Nusbaumer, J., Zhu, J., Erb, M. P.,
Wong, T. E., Liu, Z., and Noone, D. (2018). Interpreting precession driven δ
18
O vari-
ability in the South Asian monsoon region. Journal of Geophysical Research: Atmo-
spheres, 123(5927-5946).
Tharammal, T., Bala, G., andNoone, D.(2017). Impactofdeepconvectionontheisotopic
amount effect in tropical precipitation. Journal of Geophysical Research: Atmospheres,
122(3):1505–1523.
Thompson, L. G., Yao, T., Mosley-Thompson, E., Davis, M., Henderson, K., and Lin,
P.-N. (2000). A high-resolution millennial record of the South Asian monsoon from
Himalayan ice cores. Science, 289(5486):1916–1919.
Treble,P.,Mah,M.,Griffiths,A.,Baker,A.,Deininger,M.,Kelly,B.,Scholz,D.,andHan-
kin, S. (2019). Separating isotopic impacts of karst and in-cave processes from climate
variability using an integrated speleothem isotope-enabled forward model. EarthArXiv.
Wang, B., Yim, S.-Y., Lee, J.-Y., Liu, J., and Ha, K.-J. (2014). Future change of Asian-
Australian monsoon under RCP 4.5 anthropogenic warming scenario. Climate Dynam-
ics, 42(1-2):83–100.
Wang, Y., Cheng, H., Edwards, R. L., Kong, X., Shao, X., Chen, S., Wu, J., Jiang, X.,
Wang, X., and An, Z. (2008). Millennial-and orbital-scale changes in the East Asian
monsoon over the past 224,000 years. Nature, 451(7182):1090–1093.
Wang, Y.-J., Cheng, H., Edwards, R. L., An, Z., Wu, J., Shen, C.-C., and Dorale, J. A.
(2001). A high-resolution absolute-dated late Pleistocene monsoon record from Hulu
Cave, China. Science, 294(5550):2345–2348.
150
Wong, T. E., Nusbaumer, J., and Noone, D. C. (2017). Evaluation of modeled land-
atmosphere exchanges with a comprehensive water isotope fractionation scheme in ver-
sion 4 of the Community Land Model. Journal of Advances in Modeling Earth Systems,
9(2):978–1001.
Zhang, H., Griffiths, M. L., Chiang, J. C., Kong, W., Wu, S., Atwood, A., Huang, J.,
Cheng, H., Ning, Y., and Xie, S. (2018). East Asian hydroclimate modulated by the
position of the westerlies during Termination I. Science, 362(6414):580–583.
Zhang, Q.-B., Cheng, G., Yao, T., Kang, X., and Huang, J. (2003). A 2,326-year tree-ring
record of climate variability on the northeastern Qinghai-Tibetan Plateau. Geophysical
Research Letters, 30(14).
Zhisheng, A., Kutzbach, J. E., Prell, W. L., and Porter, S. C. (2001). Evolution of Asian
monsoonsandphasedupliftoftheHimalaya–TibetanPlateausinceLateMiocenetimes.
Nature, 411(6833):62–66.
Zhu, M., Stott, L., Buckley, B., Yoshimura, K., and Ra, K. (2012). Indo-Pacific Warm
Pool convection and ENSO since 1867 derived from Cambodian pine tree cellulose
oxygen isotopes. Journal of Geophysical Research: Atmospheres, 117(D11).
151
Appendix A
Supplemental material for Chapter 2
152
FigureA.1: Relationshipbetweentransformedmonthlystratiformprecipitation
fraction and transformed δ
18
O
P
at the same locations as Aggarwal et al. (2016) in (a)
SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5.
β values are ordinary least square slopes.
153
FigureA.2: Relationshipbetweentransformedmonthlystratiformprecipitation
fraction and rawδ
18
O
P
at the same locations as Aggarwal et al. (2016) in (a) SPEEDY-
IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β values
are ordinary least square slopes.
154
FigureA.3: Relationshipbetweentransformedmonthlystratiformprecipitation
fraction and transformed δ
18
O
P
in the tropics and mid-latitudes in (a) SPEEDY-
IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β values
are ordinary least square slopes.
155
FigureA.4: Relationshipbetweentransformedmonthlystratiformprecipitation
fraction and raw δ
18
O
P
in the tropics and mid-latitudes in (a) SPEEDY-IER; (b)
LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f) HadAM4; (g) iCAM5. β values are
ordinary least square slopes.
156
Figure A.5: Vertical profiles of water vapor δ
18
O over the convective rainfall
region, and in the stratiform rainfall region, and no rain region in the tropics
(30
◦
S-30
◦
N) in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f)
HadAM4. The convective/stratiform rainfall region is where the proportion of convec-
tive/stratiform rainfall to total rainfall exceeds 0.8.
157
Figure A.6: Vertical profiles of deuterium excess over the convective rainfall
region, and in the stratiform rainfall region, and no rain region in the tropics
(30
◦
S-30
◦
N) in (a) SPEEDY-IER; (b) LMDZ; (c) CAM2; (d) isoGSM; (e) MIROC; (f)
HadAM4. The convective/stratiform rainfall region is where the proportion of convec-
tive/stratiform rainfall to total rainfall exceeds 0.8.
158
Appendix B
Supplemental material for Chapter 4
• The following list shows the steps to produce each subplot in Fig. 4.5.
– For each record, we can get 1000 realizations of speleothem δ
18
O time series
– Cut all these time series into 150-year long slices
– Get the median δ
18
O values of all these slices
– Plot the distribution of all median values, which is shown in histograms and
curves in Fig. 4.5
– Select the slices covering 4,200 BP and collect their median δ
18
O values
– Then plot these δ
18
O values around 4,200 BP by vertical lines as in Fig. 4.5
Bibliography
Aggarwal, P. K., Romatschke, U., Araguas-Araguas, L., Belachew, D., Longstaffe, F. J.,
Berg, P., Schumacher, C., andFunk, A.(2016). Proportionsofconvectiveandstratiform
precipitation revealed in water isotope ratios. Nature Geoscience, 9(8):624.
Berkelhammer, M., Sinha, A., Stott, L., Cheng, H., Pausata, F., and Yoshimura, K.
(2013). An abrupt shift in the indian monsoon 4000 years ago. Climates, landscapes,
and civilizations, pages 75–88.
Kathayat, G., Cheng, H., Sinha, A., Berkelhammer, M., Zhang, H., Duan, P., Li, H., Li,
X., Ning, Y., and Edwards, R. L. (2018). Evaluating the timing and structure of the 4.2
159
ka event in the Indian summer monsoon domain from an annually resolved speleothem
record from Northeast India. Climate of the Past, 14(12):1869–1879.
160
Year (BP)
10
8
Heshang
30 N
Year (BP)
11
10
9
Sanbao
31 N
Year (BP)
12.5
10.0
7.5
Jiuxian
33 N
Year (BP)
8
6
4
2
Lianhua
29 N
Year (BP)
8
6
Mawmluh
25 N
Year (BP)
12
10
8
6
Xianglong
32 N
Year (BP)
10
9
8
7
Dongge
25 N
0 2000 4000 6000 8000 10000
Year (BP)
10
9
Sahiya
30 N
460
470
480
490
460
470
480
490
460
470
480
490
460
470
480
490
Local Insolation
(W m
2
)
460
470
480
460
470
480
490
460
470
480
460
470
480
490
Figure B.1: Time series of eight Asian speleothem δ
18
O over the Holocene with
age uncertainties. Dashed lines are local summer insolation. Errorbars above each time
series are
230
Th ages and errors.
161
Year (BP)
2
0
Heshang
Year (BP)
0.5
0.0
0.5
Sanbao
Year (BP)
5.0
2.5
0.0
2.5
Jiuxian
Year (BP)
2
0
2
Lianhua
Year (BP)
1
0
1
Mawmluh
Year (BP)
4
2
0
Xianglong
Year (BP)
1
0
1
Dongge
0 2000 4000 6000 8000 10000
Year (BP)
1
0
1
Sahiya
Figure B.2: Detrended time series of eight Asian speleothem δ
18
O over the
Holocene with age uncertainties.
162
Heshang Sanbao
Jiuxian Lianhua
Mawmluh Xianglong
Dongge
Sahiya
Figure B.3: The age modeling results of eight Asian speleothem δ
18
O over the
Holocene using a Bchron age model. The gray area is the 95% highest density region
of the age at each depth. The red lines show 10 random paths out of the 1000 age models
generated.
163
1 2 3 4 5 6 7 8
Rank
8
10
12
14
16
18
20
Explained Variance (%)
95% of Monte-Carlo AR1
Eigenvalues
Figure B.4: MC-PCA eigenvalue spectrum (scree plot). The blue line shows the
mean eigenvalues of 1000 ensembles with 1-sigma standard deviation. The orange line
indicates the 95% confidence level of the red noise (AR (1) benchmark.
164
Figure B.5: The spatial distribution of the loading and the time series of the
second principle component of precipitation δ
18
O and soil δ
18
O from iCAM5-
iCLM4. Both the spatial loading and principle component are normalized by dividing
the square root of their eigenvalues.
165
1860 1880 1900 1920 1940 1960 1980 2000 2020
Time(months)
14
12
10
8
6
18
O ( )
r=0.61, p<1e-6
Precipitation
18
O
Heshang
Furong
1860 1880 1900 1920 1940 1960 1980 2000 2020
Time(months)
13
12
11
10
9
8
18
O ( )
r=0.79, p<1e-6
Soil
18
O
Heshang
Furong
Figure B.6: Comparison between simulated precipitation δ
18
O and soil δ
18
O at
Heshang and Lianhua cave from iCAM5-iCLM4.
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
11
10
18
O ( )
r=0.79, p<1e-6
Stal 1
Heshang
Lianhua
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
18
O ( )
r=0.68, p<1e-6
Stal 2
Heshang
Lianhua
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
18
O ( )
r=0.71, p<1e-6
Stal 3
Heshang
Lianhua
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
11
18
O ( )
r=0.84, p<1e-6
Stal 4
Heshang
Lianhua
1860 1880 1900 1920 1940 1960 1980 2000 2020
Year
16
15
14
13
12
11
18
O ( )
r=0.69, p<1e-6
Stal 5
Heshang
Lianhua
Figure B.7: Comparison between five types of simulated stalagmite δ
18
O at
Heshang and Lianhua cave from Karstolution.
166
50 100 150 200 250 300
Interval
78
80
82
84
86
88
Percent of passing the significance test (5% level)
Mawmluh
Figure B.8: The percent of Mawmluh speleothem δ
18
O values around 4.2 ka BP
falling below 5% quantile of Holocene values
Year (BP)
7
6
5
KMA
Year (BP)
6.5
6.0
5.5
ML.1
0 2000 4000 6000 8000 10000
Year (BP)
6.0
5.5
5.0
ML.2
Figure B.9: Time series of δ
18
O in stalagmites KMA (Berkelhammer et al.,
2013), ML.1 and ML.2 (Kathayat et al., 2018) in Mawmluh cave. Light colored
envelopes encompass 95% of age ensemble members.
167

Asset Metadata

Creator Hu, Jun (author)

Core Title Flowstone ideograms: deciphering the climate messages of Asian speleothems

Contributor Electronically uploaded by the author (provenance)

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Geological Sciences

Publication Date 07/23/2019

Defense Date 05/13/2019

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

Tag climate model,monsoon,OAI-PMH Harvest,speleothem,water isotope

Format application/pdf (imt)

Language English

Advisor Emile-Geay, Julien (committee chair), Corsetti, Frank (committee member), Feakins, Sarah (committee member), Levine, Naomi (committee member)

Creator Email hoo.jun.1990@gmail.com,hujun@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-188846

Unique identifier UC11663168

Identifier etd-HuJun-7581.pdf (filename),usctheses-c89-188846 (legacy record id)

Legacy Identifier etd-HuJun-7581.pdf

Dmrecord 188846

Document Type Dissertation

Format application/pdf (imt)

Rights Hu, Jun

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

Abstract (if available)

Abstract Speleothems have been widely used to reconstruct past hydroclimate variability, particularly in the Asian Monsoon region. While Asian speleothem δ¹⁸O is traditionally interpreted as “monsoon intensity”, recent work has proposed alternative interpretations such as water vapor transport and changes in atmospheric circulation that challenge or redefine this long-held concept. Also, challenges exist when we attempt to get interpretations of speleothem δ¹⁸O by correlating them with climate variables. This thesis aims to better understand the climate signals preserved in Asian speleothems over various timescales ranging from orbital to interannual. The thesis begins with the discussion of the challenges in interpreting speleothem δ¹⁸O based on correlation analysis and the methods to circumvent these statistical issues are provided. Then a state-of-the-art isotope-enabled climate model iCESM is employed to investigate the interpretation of Asian speleothem δ¹⁸O. The model is capable of credibly simulating precipitation δ¹⁸O and particularly the impact of the convective activity on precipitation δ¹⁸O. Then this model is used to quantify contributions to precipitation δ¹⁸O over China at both orbital and interannual time scales. Results suggest that orbital-scale speleothem δ¹⁸O variations at Chinese sites mainly represent the meridional migration of the Asian monsoon circulation, accompanied by an early northward movement of the East Asian rain belt. At interannual scales, Chinese speleothem δ¹⁸O is also tied to the intensity of monsoonal circulation, via a change in moisture source locations: enhanced moisture delivery from remote source regions leads to depleted δ¹⁸O. The results offer a re-interpretation of the concept of “monsoon intensity” as “enhanced monsoonal circulation” rather than precipitation amount. Despite this complexity, speleothem δ¹⁸O at orbital scales is coherent across Asia. However, whether the coherency still exists at short time scales is a question. Here we synthesize Asian speleothem δ¹⁸O over the Holocene to investigate whether they show coherent variability at sub-orbital scales, and particularly focus on the “4.2 ka event” and the newly announced Meghalayan age (4.2 ka BP to present). We find no coherent variability among Asian speleothem δ¹⁸O at these time scales. This asynchrony can be explained by heterogenous soil, vegetation, and karst processes modifying coherent climate inputs. Given the relatively small amplitude (1-3 ‰) of speleothems at sub-orbital scales, extracting hydroclimate variability at these time scales from speleothem δ¹⁸O should be done with caution, and supplemented by proxies such as δ¹⁸C and trace elements ratios.