University of Southern California Dissertations and Theses

Questioning the cause of calamity: using remotely sensed data to assess successive fire events

(USC Thesis Other)

Questioning the cause of calamity: using remotely sensed data to assess successive fire events

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Questioning the Cause of Calamity:
Using Remotely Sensed Data to Assess Successive Fire Events

by

Joel Kerbrat

A Thesis Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)

August 2018

Copyright © 2018 by Joel Kerbrat

To my friends, Jeff Lauder and Mackenzie Kilpatrick

iv
Table of Contents
List of Figures ............................................................................................................................... vii
List of Tables ................................................................................................................................. ix
Acknowledgements ......................................................................................................................... x
List of Abbreviations ..................................................................................................................... xi
Abstract ......................................................................................................................................... xii
Chapter 1 Introduction .................................................................................................................... 1
1.1. Research Questions Investigated by this Study ..................................................................1
1.2. Project Background and Impact ..........................................................................................2
1.2.1. Historical Background ...............................................................................................3
1.2.2. Economic Impact .......................................................................................................3
1.2.3. Ecological Impact ......................................................................................................4
1.2.4. Policy Background .....................................................................................................4
1.3. Study Area and Key Terms .................................................................................................5
1.3.1. Study Area: BLM Grazing Allotments ......................................................................6
1.3.2. Rest Periods ...............................................................................................................7
1.3.3. Grazing Closure Areas ...............................................................................................8
1.3.4. High Vegetation Accumulation .................................................................................8
1.4. Structure of this Document .................................................................................................9
Chapter 2 Related Works .............................................................................................................. 10
2.1. Measuring Vegetation Remotely ......................................................................................10
2.1.1. Vegetation Indices as Estimators of Biomass and Regrowth ..................................10
2.1.2. Scaled NDVI ............................................................................................................13
2.2. The Great Basin Fire Regime ...........................................................................................15
2.3. Bromus tectorum (Cheatgrass) .........................................................................................17
v
2.4. The Role of Humans and Livestock..................................................................................18
2.4.1. Human-caused Fires.................................................................................................18
2.4.2. Grazing and Fire Recovery ......................................................................................19
2.5. Review Summary ..............................................................................................................20
Chapter 3 Data and Methodology ................................................................................................. 21
3.1. Data description ................................................................................................................21
3.1.1. US County Boundaries ............................................................................................21
3.1.2. BLM Grazing Allotments ........................................................................................21
3.1.3. BLM Nevada Wildfire Fire Perimeters....................................................................22
3.1.4. LANDSAT 5 TM Raster Images .............................................................................23
3.1.5. National Land Cover Database ................................................................................25
3.1.6. PRISM Precipitation Data........................................................................................26
3.2. Methods.............................................................................................................................27
3.2.1. Secondary Fire Identification ...................................................................................27
3.2.2. Evolution of Project Methodology Using LANDSAT5 TM Data ...........................34
3.2.3. Selected Methodology for Scaled NDVI Time Series .............................................35
3.2.4. Precipitation Data.....................................................................................................45
3.2.5. Additional Procedures ..............................................................................................46
Chapter 4 Results .......................................................................................................................... 48
4.1. Secondary Fire Search and Case Study Selection.............................................................48
4.2. Case Study Fire Events .....................................................................................................50
4.2.1. Squawvalle Fire Zones .............................................................................................52
4.2.2. Little One Fire and Green Monster Fire Zones ........................................................53
4.2.3. Rock Creek Fire Zones ............................................................................................55
4.3. Scaled NDVI Time Series and Four-Month Precipitation Totals .....................................56
vi
4.3.1. Squawvalle Fire .......................................................................................................56
4.3.2. Green Monster and Little One Fires ........................................................................58
4.3.3. Rock Creek Fire .......................................................................................................61
4.4. Other Observations ...........................................................................................................63
Chapter 5 Discussion and Conclusion .......................................................................................... 64
5.1. Discussion .........................................................................................................................64
5.2. Other Considerations ........................................................................................................66
5.3. A Better Tomorrow ...........................................................................................................71
5.4. Conclusions .......................................................................................................................73
References ..................................................................................................................................... 75

vii
List of Figures
Figure 1. BLM Grazing Allotments in Study Area......................................................................... 6
Figure 2. Recorded Fire Frequency by Year and by Month ........................................................... 8
Figure 3. Recorded Fire Perimeters in Study Area, 2000-2016 .................................................... 23
Figure 4. Footprint of LANDSAT scenes (green) from Earth Explorer ....................................... 24
Figure 5. Workflow for Secondary Fire Identification and Case Study Selection ....................... 31
Figure 6. Workflow for Vegetation Index Processing .................................................................. 36
Figure 7. Model Builder Layout for NDVI Calculations .............................................................. 40
Figure 8. Example of Center-Pivot Irrigation Systems in Nevada ............................................... 41
Figure 9. Workflow for Zonal Statistics and Time Series Creation ............................................. 43
Figure 10. Map of Fire Events from Case Studies ........................................................................ 51
Figure 11. Squawvalle Zones ........................................................................................................ 52
Figure 12. Little One and Green Monster Zones .......................................................................... 54
Figure 13. Rock Creek Zones ....................................................................................................... 55
Figure 14. Scaled NDVI over time for the Squawvalle Zones ..................................................... 57
Figure 15. Squawvalle Four-Month Precipitation Totals ............................................................. 58
Figure 16. Scaled NDVI over time for the Green Monster Zones ................................................ 59
Figure 17. Scaled NDVI over time for the Little One Zones........................................................ 59
Figure 18. Green Monster Four-Month Precipitation Totals ........................................................ 60
Figure 19. Little One Four-Month Precipitation Totals ................................................................ 61
Figure 20. Scaled NDVI over time for the Rock Creek Zones ..................................................... 62
Figure 21. Rock Creek Four-Month Precipitation Totals ............................................................. 63
Figure 22. Fire Frequency per 10x10 km Grid ............................................................................. 68
viii
Figure 23. Fire Frequency by Allotment, 2000-2016 ................................................................... 69
Figure 24. Mining Sites and Highways near Frequently Burned Allotments ............................... 71

ix
List of Tables
Table 1. Selected Three-Year Overlapping Fire Perimeters (16 out of 58 total) .......................... 49
Table 2. Squawvalle Land Cover Cell Count ............................................................................... 53
Table 3. Little One and Green Monster Land Cover Cell Count .................................................. 54
Table 4. Rock Creek Land Cover Cell Count ............................................................................... 56

x
Acknowledgements
I am grateful to my advisor, Professor Karen Kemp for her guidance and encouragement during
this project. I thank Professor Jill Heaton at the University of Nevada, Reno, for introducing me
to the topic of this project. I thank Mackenzie Kilpatrick for answering questions and providing
insight on remote sensing and fire ecology.

xi
List of Abbreviations
AVHRR Advanced Very High-Resolution Radiometer
BLM Bureau of Land Management
EVI Enhanced Vegetation Index
FWS Fish and Wildlife Service
GIS Geographic information system
MODIS Moderate Resolution Imaging Spectroradiometer
NBR Normalized Burn Ratio
NDA Nevada Department of Agriculture
NDVI Normalized Difference Vegetation Index
NOAA National Oceanic and Atmospheric Administration
NRCS Natural Resources Conservation Service
PRISM Parameter elevation Regression on Independent Slopes Model
USDA United States Department of Agriculture
USGS United States Geological Service
WFMI Wildland Fire Management Information

xii
Abstract
Bureau of Land Management policy regarding wildfire events on public rangelands dictates that
burned areas are closed to livestock grazing until the vegetation in the burned area has
reestablished itself. Ranchers and their supporters contend that extended duration of such grazing
closures increases the likelihood of subsequent fire events during the grazing rest period. The
ranchers attribute this effect to an over-accumulation of vegetation during the grazing rest period.
With the goal of testing the claim made by ranchers, this project utilized fire history records,
grazing allotment data, and remote sensing vegetation indices to identify and analyze potential
rest period fires between 2000 and 2016 in and around the Nevada counties of Humboldt and
Elko. GIS proximity tools were used to identify initial and subsequent fires on BLM grazing
allotments which met the spatial and temporal requirements of a rest period fire. The four most
likely candidates for rest period fires were selected for further examination as case studies.
Scaled NDVI was used as an estimator of vegetation cover and change between selected initial
and subsequent fires. Precipitation and land cover data were incorporated to provide further
context. Three of the four fire perimeters showed increased vegetation cover when compared to
similar nearby unburned sites during the second spring after the initial fires. This pattern
suggests that increased fuel loads before the secondary fire may have been present. Evidence of
cheatgrass and anthropogenic fire activity in the case study area suggest more complex
explanations. Ways to improve monitoring and post-fire recovery through better record keeping,
more complex sensors for satellite imagery, and targeted grazing research are discussed.

1

Chapter 1 Introduction
Land management agencies such as the Department of the Interior’s Bureau of Land
Management (BLM) often create and enforce livestock grazing closures on public grazing
allotments after wildfire events. The purpose of these closures is to allow the vegetation in a
burned area to recover or recolonize (BLM 2007). Ranchers argue that the durations of post-fire
grazing bans are longer than they need to be. Their concern is that grazing bans remain in place
long enough that plant litter accumulates more than it might otherwise, leading to a subsequent
fire reburning the area soon after the original fire. Although not specifically dealing with post-
fire bans, news articles by Halladay (2015) and Valla (2015) both feature ranchers citing grazing
bans as major factors in recent fire events. Despite ongoing disputes between policy makers,
ranching advocates, and environmentalists, there have been no major studies specifically
examining the arguments opposing such closures. The studies which do exist focus on how post-
fire grazing affects vegetation recovery, not rest period fire rates (e.g., Bruce et al. 2007).
1.1. Research Questions Investigated by this Study
The purpose of this project was to test the claim that closures and rest periods on grazing
allotments managed by the BLM lead to an increase of fuel load in the form of vegetation and
consequently to subsequent fire events during the rest period. There were two questions this
project intended to answer:
1. Which, if any, fires within BLM-managed grazing allotments burned areas previously
burned during an earlier fire season within three years?
2. If there are fires which fulfill the spatial and temporal requirements of question one, is
there evidence that these fires were preceded by greater vegetation growth/recovery than

2

similar nearby unburned areas as measured by fuel load, biomass accumulation, or a
similar indicator of vegetation health or abundance?
To answer these questions, this project utilized available spatial data to identify any
potential fires that possessed the attributes described in the first question. Upon finding any such
fires, this project acquired and processed historical satellite imagery data to construct vegetation
index time series. The time series were used to estimate vegetation recovery and/or biomass
accumulation. Because this project took the form of historical case studies, data that could
directly answer the research questions were not always available or trustworthy. In such a case:
1. Alternative methodologies were considered to answer the questions and when appropriate
data were available, some were implemented.
2. When necessary, modifications to the original questions that were answerable with
available data were made.
1.2. Project Background and Impact
The background knowledge used in this project pulls from topics such as fire science,
botany, agriculture, and the political interaction between local and federal stakeholders.
Likewise, the policy issues considered by this project are typically approached by assessing the
effect of grazing on burned rangeland. This project instead examined the fire risk of grazing
closures.
Underlining all the issues discussed in this chapter is the fact that the Bureau of Land
Management has jurisdiction over almost 47 million acres in Nevada, approximately two-thirds
of Nevada’s land (BLM 2017). Any action or policy change taken by the BLM has a
considerable impact on the state.

3

1.2.1. Historical Background
Two key milestones in the history of public lands in the American West were the passage
of the Taylor Grazing Act of 1934 and the establishment of the BLM in 1946 (Knapp 1996). The
Taylor Grazing Act regulated grazing on public lands to prevent rangeland deterioration from
overgrazing. This regulation created a system of grazing allotments that could be leased by
private operators. The private operators would have the sole rights to graze livestock on their
allotments, but also sole responsibility for degradation of the rangeland. The BLM was created
over a decade later to act as a manager for these grazing allotments and other public lands. This
aspect of federal land management has been a persistent source of contention between federal
policymakers and land users, leading to noteworthy events such as the Sagebrush Rebellion of
the late 1970s and the 2016 occupation of the Malheur Wildlife Refuge in Oregon. Given the
economic, environmental, and political impact of land use policies, it is important that such
policies are rigorously evaluated and based on both physical evidence and community needs.
1.2.2. Economic Impact
Livestock production is an important component of the Nevadan economy. In Elko
County, livestock production generated 85% of all agricultural receipts in 2012 (NDA 2015).
Cattle ranching specifically was the 12th largest industry in Elko County, generating about $83
million for the county. In Nevada, the ranching industry depends on access to public lands for
leased grazing allotments. Preventing, managing, and recovering from wildfires in the rangelands
is thus vital to the economies of rural Nevada counties such as Elko. Improved understanding of
wildfires and more effective land management could benefit these economies.

4

1.2.3. Ecological Impact
There are multiple ecological concerns in the region involving wildfire management. The
greater sage-grouse (Centrocercus urophasianus), a native bird species in the region, is
sagebrush obligate, and thus highly sensitive to habitat damage in the sagebrush steppes (FWS
2015). Wide-spread or uncontained wildfires can adversely affect the sage-grouse population.
This can be exacerbated by post-fire invasion by non-native grasses, most notably cheatgrass
(Bromus tectorum). Cheatgrass is less palatable than native plants to both wild and domestic
grazers while being more flammable than native plants (NRCS 2015). The post-fire spread of
cheatgrass thus induces a positive feedback loop increasing the likelihood of future fires. Nevada
experienced prolonged periods of drought during the 2000s and 2010s, with severe droughts in
2006 and 2011. The drier and warmer climate further increases the likelihood of wildfires and
the need to properly understand and manage them.
1.2.4. Policy Background
At the center of the ranchers’ claim and the research questions posed by this project is the
BLM’s “Burned Area Emergency Stabilization and Rehabilitation Handbook” (the “Handbook”)
(BLM 2007). This document is the official enunciation of BLM fire recovery policies, including
section III.B.10, which describes how fencing and other barriers should be used to protect a
burned area during recovery, and section III.B.18, which is concerned with post-fire livestock
grazing. Section III.B.18 states “Livestock are to be excluded from burned areas until monitoring
results… show emergency stabilization and rehabilitation objectives have been met… In the case
of treatment failure, other factors may need to be considered” (BLM 2007, 35).
Section III.B.10 of the handbook notes that “It often takes two years or longer to
successfully establish a new seeding” (BLM 2007, 31) which is the policy justification for two-

5

year grazing closures. The policy recommends reseeding burned areas to enable native
vegetation regrowth and to prevent post-fire colonization by invasive plants. The policy states
that shorter or longer rest periods might be used depending on climatic, meteorological, and
other environmental factors. Drought conditions may justify longer periods, while wetter climate
might require shorter periods. The handbook also states that closures lasting more than three
years are turned over to the jurisdiction of the local BLM office in charge of the allotment or
pasture. Because the federal jurisdiction over the recovery effort ends after three years, any
subsequent fire events occurring more than three years after the initial fire are excluded from the
list of candidate rest period fires examined in this study.
Section III.B.10 also guides BLM managers to limit closure areas to the minimum needed
to protect reseeding efforts from grazers (both domesticated and wild). Section III.B.18 however
suggests that it can be more cost effective in some cases to close entire allotments if the damage
is wide-spread and the cost of new fencing is not feasible. The policy gives an example of “75
percent or more of an allotment or pasture” as a situation where the entire allotment might be
placed in a grazing closure. The competing constraints of “minimum closed area” and “least
costly enclosure” lead to situations where larger burn perimeters might be enclosed using
existing fencing as a cost-saving measure. This can result in grazing closure areas that are larger
than the fire perimeter and include adjacent unburned land.
1.3. Study Area and Key Terms
This section describes the study area for the project and some of the key terms used in
this document.

6

1.3.1. Study Area: BLM Grazing Allotments
BLM grazing allotments are parcels of federally managed public rangeland which are
leased to private operators for economic use. The study area for this project consists of 339 BLM
Grazing Allotments located wholly or partly within the Nevada counties of Humboldt and Elko,
which are part of the interstate Great Basin region (Figure 1). Allotments range in size from 62.5
acres to over 1.3 million acres, with the majority being between 1,000 and 100,000 acres.
In this project, grazing allotments are the actionable spatial unit. All other events
(closures, wildfires, and vegetation growth) occur within the allotments. Events which occur
outside of allotments are not bound by BLM fire response policy and are therefore outside the
scope of this project. Understanding wildfire events at the allotment level of perception is the
primary purpose of the first component of this study.

Figure 1. BLM Grazing Allotments in Study Area

7

1.3.2. Rest Periods
The rest period of a grazing closure is the time from when the fire is contained to when
grazing resumes on the closed area. BLM post-fire recovery policy establishes a set of goals or
conditions to be met by the closure before the rest period can end (BLM 2007). These goals are
determined per-fire and can be different across fires or between two allotments affected by the
same fire. As described in the policy background section above, rest periods which extend
beyond the 3 years are outside the scope of this project.
While the true lower bound of a rest period is immediately after the fire is contained,
such a constraint makes little sense in the objectives of this project. It is not reasonable to argue
that excessive vegetation accumulation can happen within a day or two after a fire has ended. In
fact, most wildfires in the study area occur between the months of June and September (Figure
2), suggesting that rest period fires need at least one or two Spring seasons for fuel accumulation.
It is however necessary to identify all fires within the study which occurred within three
years of a previous fires. Two fires with an interval of two or three years could potentially have a
third fire occurring between them. An intervening fire would complicate the vegetation index
time series and the interpretation of the results. The fires for the case study were selected to have
two or three Spring seasons between fires and to have no recorded fire events in between.

8

Figure 2. Recorded Fire Frequency by Year and by Month
1.3.3. Grazing Closure Areas
The grazing closure area is the area which is closed to livestock grazing after a wildfire
has been contained. The purpose of closures is to allow the vegetation in a burned area to recover
before allowing livestock to graze. BLM policy mandates the use of temporary fencing to secure
the closure area (BLM 2007). In some cases, the burned area is large enough that temporary
fencing is too costly or time consuming to put up. BLM policy can instead mandate the use of
pre-existing permanent fencing used to separate individual pastures within an allotment. When
permanent fencing is authorized, the unburned areas within the pasture fencing are also closed to
grazing. In the worst scenario, a large burned area may result in the BLM closing entire
allotments.
1.3.4. High Vegetation Accumulation
High vegetation accumulation during the rest period is, according to the claims made by
ranchers, the cause of rest period fire events. Testing these claims thus requires observing
vegetation changes between the previous and subsequent fires. Chapter 2 discusses the scientific
literature about the LANDSAT sensors and their use in vegetation monitoring. This project used
data from the LANDSAT 5 Thematic Mapper sensor to record a time series for fractional

9

vegetation cover before, during and after the rest periods. The second part of Chapter 3 discusses
the methods used in this project in greater detail.
1.4. Structure of this Document
The remainder of this document details the research, methodology, and outcomes of this
project. Chapter 2 is a discussion of relevant background research, primarily on the topics of
vegetation indices, fire regimes, and Great Basin vegetation and ecology. Chapter 3 discusses the
case study selection process and the construction of vegetation index time series for each site.
Chapter 4 describes the direct outcomes of the processes described in Chapter 3, specifically
comparing the observed time series graphs to the expected time series if the ranchers’ claim is
valid. The final chapter provides qualitative context to the outcomes from Chapter 4 while also
discussing alternative explanations, additional research questions raised during this project, and
opportunities to improve rangeland policy and data collection.

10

Chapter 2 Related Works
To answer the research questions, this project requires a method to assess the historical condition
of rangeland vegetation during rest period events. To place the answers into context, a
background understanding of the Great Basin fire regime is needed. The first part of this chapter
discusses literature related to the remotely sensed vegetation indices, particularly as they related
to semi-arid shrub and grasslands. The remaining parts of this chapter review literature on
rangeland fire regimes, invasive plant species, and the human factor in wildfires.
2.1. Measuring Vegetation Remotely
The case study selection method used in this project is described in detail in the next
chapter. Suffice to say, the method in its simplest form involved finding intersecting fire
perimeters and then using attribute and areal data to find the fires which met the spatial and
temporal definitions of a rest period fire. The second part of this project dealt with accessing
vegetation growth between the initial and secondary fires. Developing the procedure for this
assessment required the following literature review.
2.1.1. Vegetation Indices as Estimators of Biomass and Regrowth
Box et al. (1989) took note of several contemporary studies which had used the NDVI
product based NOAA’s AVHRR sensor as a proxy for accessing a variety of biological
properties of vegetation. The authors devised a study which would compare AVHRR NDVI
values to field measurements of properties such as biomass and net productivity. They found that
while NDVI could be useful as proxy to measure net productivity and evapotranspiration, it was
an inconsistent tool for assessing plant biomass.

11

Santin-Janin et al. (2009) also took a critical view of their contemporaries. The authors
were concerned that other studies were using NDVI as a proxy for biomass without accounting
for the tendency for NDVI to become saturated when observing areas with high biomass. The
oversaturation is a consequence of using two-dimensional data to measure a three-dimensional
property such as biomass. The authors devised a non-linear model to fit field measurements of
vegetation on the Kerguelen Islands in the Indian Ocean to NDVI observations from AVHRR.
Of note was this paper’s use of a vegetation index time series as a visualization and analysis tool.
The analysis and visualization of changing vegetation can also be a problem for
vegetation indices. One typical method to measure vegetation change is to find the difference
between a pre-event image and a post-event image. The resulting dataset is called the “delta” or
“differenced” version of the vegetation index. One potential pitfall of this method is that the
original data (the pre-event and post-event data) are not retained with the results of the
subtraction operation (the differenced data). This can be important when trying to evaluate burn
severity, as noted in Miller and Thode (2007). In this paper, the authors were specifically
concerned with the Normalized Burn Ratio (NBR) rather than NDVI. However, the observations
they had regarding the delta NBR (dNBR) can also apply to delta NDVI and other differenced
calculations. They noted that a smaller fire can be more devastating to a lightly-vegetated area
than it would be to a densely-vegetated area. The dNBR can describe the intensity of the fire, but
the pre- and post-fire are not retained. With only the delta indices, it is not possible to accurately
describe the conditions at the site. One can only describe the absolute magnitude of the change.
Taking inspiration from Santin-Janin et al. (2009) and heeding the warnings of Miller and Thode
(2007), this project created a vegetation times series as an analysis and visualization tool.

12

Another consideration when using vegetation indices like NDVI is that they are sensitive
to photosynthetically-active green vegetation but not to photosynthetically-inactive dry
vegetation. This is important as dry vegetation is a significant factor in fire frequency and
intensity (Nagler et al. 2000, Guerschman et al. 2009). Nagler et al. (2000) developed one
potential solution to this problem through the Cellulose Absorption Index (CAI). The authors
based their work off earlier studies which noted that cellulose and lignin absorb radiation at
wavelengths of 2.1 μm. The authors took direct reflectance measurements of plant litter and soil,
both when wet and dry, and found that soils did not absorb radiation at 2.1 μm. The CAI has high
values when 2.1 μm reflectance is lower when compared to reflectance at 2.0 μm and 2.2 μm,
indicating that cellulose may be present. In material without significant cellulose, the reflectance
of the three wavelength is roughly the same, yielding a low CAI value.
Guerschman et al. (2009) developed a framework for estimating the relative surface
cover of green vegetation, dry vegetation, and bare soil by comparing NDVI and CAI values.
Study sites in the Australian savannah were measured for the relative surface cover and plants
and soils were measured for reflectance values. Data for calculating CAI for the study site were
requested from the Hyperion sensor aboard the USGS’s EO-1 satellite. NDVI values were
derived from MODIS data. The framework used NDVI to distinguish green vegetation from the
other two surface classes. CAI was then used to distinguish dry vegetation from bare soils. Areas
covered in bare soil would have low values in both indices, while dry vegetation would have
high CAI but low NDVI. The relationship between CAI and NDVI thus depends on the relative
surface cover of the three categories.
While the NDVI-CAI framework could be useful in this project, the components for
calculating CAI were not available. The EO-1 sensors were not continuously collecting data like

13

the LANDSAT and MODIS sensors. Instead, EO-1 data had to be requested by the users and
EO-1 sensor data are thus only available for locations and times that were requested.
Additionally, the sensors aboard LANDSAT and MODIS are unable to distinguish between the
three wavebands used to calculate CAI, due to the sensors grouping the three wavebands within
the larger shortwave infrared (SWIR) band.
2.1.2. Scaled NDVI
Realizing the potential weaknesses of common vegetation indices like NDVI and NBR,
Baugh and Groeneveld (2006) sought to quantitatively analyze the relative performances of 14
vegetation indices for performance in low vegetation environments. The study site they chose
was the San Luis Valley in New Mexico and they were specifically looking at how well these
indices estimated the known vegetation response to over-winter precipitation in the region. The
San Luis Valley, much like the Great Basin, is an arid and semi-arid habitat dominated by
various grass and shrub species.
The authors then acquired 14 mid-summer LANDSAT TM scenes spanning from 1986 to
2002. The scenes were processed into imagery for the vegetation indices being tested and
compared to the historical precipitation records. The authors chose sampling sites from areas that
were known to have stable groundwater levels and were unaffected by fire events, thus
precipitation would be the primary influence on the vegetation response.
The index with the best fit to the precipitation data was Scaled NDVI (NDVI*), which is
the result of taking the raw NDVI values less the NDVI value of bare soil and then dividing the
difference by the range between a saturated vegetation NDVI value and the bare soil value. Thus
NDVI* is proportional to the saturated value with respect to the bare soil value. In their testing,
NDVI* yielded an r
2
value of 0.7749 when a linear model was created to relate antecedent

14

precipitation to the value of the index. By comparison, NDVI only yielded an r
2
of 0.3686.
NDVI* differs from NDVI in that the NDVI pixels values are rescaled so that a bare soil NDVI
value (NDVI0) is set equal to 0 and an NDVI value for a cell saturated with green vegetation
(NDVIS) is set equal to 1. Baugh and Groeneveld (2006) also used center-pivot farming sites to
calculate NDVIS, as they were easily identifiable in LANDSAT imagery and were likely to have
the highest green vegetation densities in the region. Because center-pivot agriculture is also
present in northern Nevada, this method was also used for this project.
Scaled NDVI was selected as the vegetation index for this project not only because of the
favorable results it had in the Baugh and Groeneveld (2006) paper, but because it also linked the
index values to tangible environmental conditions (bare soil and saturated vegetation). Other
studies (Carlson et al. 1994, Carlson and Ripley 1997, Scanlon et al. 2002) have shown a relation
between Scaled NDVI and fractional green vegetation cover. This supports the use of Scaled
NDVI as an estimator of vegetation regrowth when values are compared over time.
One concern noted by Montandon and Small (2008) is that Scaled NDVI is sensitive to
changes in the bare soil NDVI value. Within the rescaling process, NDVI0 is used both remove
the bare soil component from cell NDVI values and to determine the rescaling factor by
removing the bare soil component from the saturated NDVI value. In dry grasslands and
shrublands, such as the Great Basin, a lower bare soil value can give the impression of
significantly higher green vegetation coverage. A higher bare soil value results in the opposite
impression that the area has much less green vegetation. Consequently, if the bare soil NDVI
cannot be directly measured, the method for determining NDVI0 must be logical and consistent.

15

2.2. The Great Basin Fire Regime
At the core of the ranchers’ claim is the concept of areas affected by multiple fire events.
The pattern and frequency of fire events in a region are that region’s fire regime. Information
about the current, historical, and prehistorical fire regimes in the Great Basin can be used to
determine if the patterns described by the ranchers are typical or divergent with the greater
regional fire regime.
The ranchers’ claim is partially supported by Westerling et al. (2003), a report on fire
patterns in the western United States. The authors compared the spatial and temporal fire history
of the region to the Palmer Drought Severity Index (PDSI), which is based on temperature and
precipitation data. Fire history data were derived from over 410,000 reports from multiple
government land agencies, included the BLM. The data covered approximately 21 years, though
the authors noted that earlier but less reliable data were available. They found that in the Great
Basin and Mojave Desert greater fire occurrence and larger fires were correlated with
anomalously wet conditions during the previous year’s Spring and Summer. Their explanation
for this pattern is that wetter years can lead to greater vegetation growth and increased fuel load.
The vegetation dries out during the following year when normal conditions return and provides
fuel for more and larger fires.
This is further supported by Mensing et al. (2006). In this study, pollen and charcoal
sediments were recovered from dry lake bed cores from central Nevada. Lakebed core
reconstructions work on the sample principles are ice core reconstructions. Aerosolized
particulates, such as dirt, pollen, and charcoal, land on the water’s surface and settle down to the
bed. Over time, newer sediments become layered on top of older sediments. The age of the core
samples gets older as depth increases. Fire events can be inferred from peaks in the relative

16

accumulation of charcoal particulates. One of the observed charcoal peaks was correlated with a
1986 fire. The pollen can be divided into plant species favoring wetter climates and plant species
favoring drier climates. Changes to the relative abundance of the two pollen groups can indicate
changes to the climate in the area. For this paper, the authors calibrated the upper core data
against recent fire and climate records from other sources and then reconstructed the fire and
climate histories for the area. They found that fire events in sagebrush-dominated environments
were more frequent in wetter climates, which supports the ranchers’ view that increased fuel load
lead to more fires. The authors also noted that their findings could not directly establish reburn
rates, but the findings did support other models which point to multi-decadal fire intervals before
European American settlement in the region.
The multi-decadal reburn rates of the past and the sub-decadal rates of more recent times
suggest that European American settlement changed the fire regime significantly. A 1990 report
by Stephen G. Whisenant summarized previous investigations into this topic and discussed the
results of his own study. Whisenant identified several sites in the Snake River Plain of southern
Idaho and compared the fire frequency and vegetation compositions at those sites. Much like the
Great Basin, the Snake River Plain has a mix of shrublands and grasslands as the main habitat
features. Whisenant found that sites with greater species diversity were dominated by sagebrush
varieties and had reburn rates comparable to the pre-settlement rates. Sites with more frequent
fires were dominated by invasive annual grass species, most notably Bromus tectorum, known
locally as “cheatgrass”. Whisenant found a positive correlation between cheatgrass abundance at
a site and the frequency of fires at that site. He concluded that the introduction of cheatgrass
when European Americans began settling the region has been the most significant factor in the
changing fire regime. Whisenant’s report was later supported by Balch et al. (2013) which

17

conducted a similar analysis over an area including the Great Basin, the Snake River Plain, and
Eastern Oregon. The authors of this study concluded that cheatgrass dominated regions had more
frequent fires and larger fires than regions dominated by other vegetation types.
The presence of cheatgrass throughout the semi-arid shrublands and grasslands of the
American West is one of the major concerns for wildfire management. The effect cheatgrass has
on regional ecosystems and fire regimes is one of the reasons why the details of fire recovery
policy matter.
2.3. Bromus tectorum (Cheatgrass)
D’Antonio and Vitousek (1992) provides a general overview of the effects of invasive
grass species on the environments they are introduced to. The authors are especially concerned
with the ease with which grass species can alter soil nutrient cycles, regional climates, and fire
regimes. Invasive species are noteworthy because they can out-compete native plant species,
leading to the noted environmental changes. The invasive grasses may have different moisture
and nutrient requirements, which eventually results in a change in soil chemistry as the invasive
species takes over the area. Different root structures can affect the physical structure of the soil
through increased or decreased erosion. The chemical and biological structure of the invasive
species may result in a different response to fire events, when compared to native species.
In the case that an invasive species is better able to survive fire events, the invader may
expand further while the native vegetation is recovering. If that same invader has a chemical or
physical structure which increases the fuel load in the area, it can also cause or contribute to fires
which reduce the native vegetation.
A historical and environmental overview of cheatgrass can be found in an article by
Knapp (1996). According to the author, cheatgrass was introduced to the Great Basin near the

18

end of the 19th Century. The lack of other dominant annuals allowed cheatgrass to establish
itself in the region. Selective grazing of native vegetation acted as a selective pressure favoring
the spread of cheatgrass. Perhaps the most important factor is how cheatgrass altered the fire
regime in the region.
As an annual plant, cheatgrass dies off after producing seeds. The dead cheatgrass
vegetation dries out and contributes to the area’s dry fuel load. Ignition events in cheatgrass areas
use that fuel load to spread further and do more damage to native perennials. The cheatgrass
seeds are more likely to survive these fires and germinate during the winter, well before any
native perennials can recolonize the burned areas. The result of this is a positive feedback loop
where the presence of cheatgrass increases the likelihood of fire events and the fires do more
damage to biota that competes with cheatgrass for space and resources. Thus, the colonization by
cheatgrass of a burned area could be a significant contributing factor to the frequency and/or
severity of subsequent fires.
2.4. The Role of Humans and Livestock
The introduction of cheatgrass into the Great Basin was a direct but unintended result of
the introduction of European Americans and their farm animals. The influence of humans and
livestock on the fire regime of the region requires some discussion.
2.4.1. Human-caused Fires
The semi-arid climate of the Great Basin is a consequence of orographic and continental
rain shadows limiting the amount of precipitation that enters the region. Additionally, the bulk of
the precipitation occurs during the winter, with the summer months being relatively dry. One
consequence of this climate pattern is that there are far fewer lightning strikes in the Great Basin
when compared to more humid regions in the United States. Given that lightning strikes are the

19

source of most naturally occurring fire ignitions, this suggests that human activity is a significant
factor in fire ignitions in the region. The various fire monitoring agencies that operate within the
Great Basin will often identify fires as “human-caused” when the ignitions cannot be linked to
lightning events.
One study, Martínez et al. (2009) compared a 13-year fire history in Spain to multiple
factors thought to be linked to human-caused ignitions. The comparison used GIS to determine
the magnitude of the factors throughout the study area. A binary model based on the ignitions in
the fire history was used to determine whether or not a location had an ignition. A linear-
regression model was then applied to evaluate the relative influence of each factor on causing or
not causing an ignition. The authors found the most influential factors were related to
mechanized agriculture and increase incursion of urban space into wilderness areas.
This is significant because the Great Basin has some similar features. Center-pivot
agriculture and other forms of farming are prevalent in the region. Mining, which involves
extensive machinery, landscape disturbance, changes to the water table, and motorized
transportation, is common activity in northern Nevada. Many of the wilderness area, including
the grazing allotments, are accessible by automobiles and can be used for recreational activities
such as off-roading, camping, and hunting. While fuel load and fuel types are major components
to wildfires, they often require human activity, whether deliberate or accidental, to get started.
2.4.2. Grazing and Fire Recovery
The main rationale for excluding livestock from burned areas is the assumption that
grazing will further stress and damage surviving plants and seeds. Two studies, Davies et al.
(2009) and Diamond et al. (2009), however support the idea that limited and target grazing can
reduce fire danger and impede cheatgrass invasion. Both studies created plots of land to which

20

different grazing treatments could be applied. The Davies et al. study compared grazed and
historically ungrazed lands for their response to fire events. They concluded that the grazed areas
had greater volumes of plant litter removed, leading to less severe fires and more resistance to
cheatgrass invasion. The Diamond et al. study was similar in construction but focused more on
targeted grazing of cheatgrass during the Spring as method to reduce cheatgrass biomass. After
cheatgrass dies off during the early summer, the dried litter becomes less palatable to livestock
and thus reducing their willingness to eat it. Targeted Spring grazing resulted in livestock
removing 80 to 90% of the cheatgrass biomass. The reduced cheatgrass and litter volumes led to
less severe fires in subsequent years. While neither study directly tested post-fire grazing, they
do present an opportunity for targeted grazing to be a potential fire recovery tool.
2.5. Review Summary
Previous research suggests that directly assessing biomass and fuel load using only
satellite-based vegetation indices would be problematic. Scaled NDVI was identified as a way to
estimate vegetation cover as a proxy for vegetation change and recovery. The literature also
showed that the sub-decadal fire frequencies that are attested to in the ranchers’ claim are a
recent phenomenon, caused primarily by the introduction of cheatgrass by European-American
settlers during the 19
th
Century. Humans also increase fire frequency through ignitions caused by
agricultural activities, machine operation, and increased access to wilderness areas. Finally, there
is literature citing the possibility that targeted or seasonal cattle grazing could help reduce fuel
loads and fire frequency.

21

Chapter 3 Data and Methodology
This chapter describes the data and the methods used in this project. The first section covers the
data sources and gives a brief description of how the data were used. The second section
describes the methods used and any alternative methods that were considered.
3.1. Data description
Spatial data for this project are all available for free from United States government
agency websites. Once downloaded, spatial data were reprojected to UTM 11N (using the
NAD83 datum) if necessary.
3.1.1. US County Boundaries
The US County Boundaries are polygon features taken from the 2017 US Census
TIGER/Line “Counties (and equivalent)” shapefile product. This dataset is available from the US
Census website. The metadata for this dataset did not specify a spatial accuracy as the data are
created and updated from a variety of sources, including local updates from Census Bureau staff,
older datasets, and other maps. The metadata for the TIGER/Line products state that the product
is not appropriate for high-precision projects, such as property transfers and engineering projects.
However, it should be noted that the TIGER/Line products are actively maintained and go back
at least a decade. As this project only needed polygons for study area selection, the lack of “high-
precision” accuracy was not a significant concern. The polygons for Humboldt and Elko
Counties in Nevada were used to select the grazing allotments that make up the study area.
3.1.2. BLM Grazing Allotments
The grazing allotments are polygon features created by the BLM which describe the
boundaries and attributes of BLM managed grazing allotments in Nevada. This dataset was

22

recorded through GPS records of the boundaries or vertices and should have a positional
accuracy of 12 m according to Michael Schade of the Nevada BLM’s Geographic Sciences
Branch (email message to author, May 23, 2018). The general significance of the allotments to
this project was discussed in the Study Area section of Chapter 1. Allotment data were used to
construct fire histories and to select or extract data from other features.
3.1.3. BLM Nevada Wildfire Fire Perimeters
The wildfire fire perimeters are polygon features created by the BLM to depict the
boundaries and attributes of wildfire events. Between 2000 and 2016, the BLM recorded 594
Fire perimeters in the project’s study area (Figure 3). The recorded burned areas are between 10
and 10,000 acres in size. A contact at Nevada BLM stated that more recent fires perimeters are
recorded with GPS at 12 m accuracy (Stephen Levitt, email message to author, May 16, 2018).
The older fire perimeters were derived from previous maps and orthophotography of the fires
and may be less accurate. The dataset is actively maintained by the BLM and corrections can be
made, though many updates may be for recent unrecorded fires. Other fire records (which are
described later) indicated that there may be many fires smaller than 10 acres. The BLM fire
perimeter data however do not record these fires. Fire perimeter data were used to construct fire
histories, to identify rest period candidates, and to create zones for zonal statistics.

23

Figure 3. Recorded Fire Perimeters in Study Area, 2000-2016.
3.1.4. LANDSAT 5 TM Raster Images
As will be shown later in this document, the fires selected for the case study all occurred
between 2001 and 2008 and were confined to the area north-northwest of Battle Mountain, NV.
Specifically, the fires were all within two World Reference System 2 (WRS2) path/row scenes:
path 41 and rows 31 and 32 (Figure 4). The LANDSAT 5 remote sensing platform was in
operation from March of 1984 to June of 2013. The primary sensor used on LANDSAT 5 was
the Thematic Mapper (TM), which took 185 km wide swathes in 7 spectral bands. Except for
Band 6, the TM bands had 30 m by 30 m pixel resolution. The satellite could image the entire
world in 16 days.

24

Figure 4. Footprint of LANDSAT scenes (green) from Earth Explorer. The red polygon is based
on the Sheep Fire perimeter and is used to spatially select scenes.
The LANDSAT imagery is available from the United States Geological Survey’s Earth
Explorer website. The website makes this data available as path/row scenes, which are
compressed folders containing band imagery as well as any processing quality assurance layers.
The imagery is available as “Level 2” surface reflectance data, where the top of the atmosphere
images are processed to surface reflectance values. This is a necessary step as the intervening
atmosphere can distort the reflected surface radiation. The LANDSAT 5 data for path 41, rows
31 and 32 are projected as UTM Zone 11N, however it uses the WGS84 datum instead of the
NAD83 datum used in the vector data. Using this imagery thus requires either reprojecting the

25

raster images to use NAD83 or reprojecting the polygons to use WGS84. The exact processing
method is described in the methods section of this chapter.
3.1.5. National Land Cover Database
The National Land Cover Database (NLCD) is a dataset created by the Multi-Resolution
Land Characteristics (MRLC) consortium, a collection of federal agencies and offices which
collaborate to provide land cover data for the United States. The NLCD dataset was generated by
applying a decision tree regression model to paired LANDSAT observations. Paired scenes are
used to adjust for season variations in reflectance. The NLCD has a cellular resolution of 30
meters because it is based on LANDSAT 30 m imagery. The NLCD using the 2011 methodology
is currently available for the years of 2001, 2006, and 2011. Homer et al. (2015) describes the
process of creating the NLCD in detail.
Of interest to this project are the land cover classes for Shrublands and Grasslands. The
NLCD assigned a value of 52 to shrubland pixels, which are defined as areas where shrub
canopies cover at least 20% of the surface. Species typical of shrublands in the Great Basin
region include woody shrubs like big sagebrush (Artemisia tridentata) and shadscale (Atriplex
confertifolia). The other 80% of the surface cover could be various grasses or bare soil.
Grasslands are assigned a value of 71 and defined as areas where grasses and other
herbaceous vegetation cover at least 80% of surface. Species in Great Basin grasslands could be
native perennial grasses or invasive Bromus annual grasses like cheatgrass. The NLCD
classification system does not distinguish between invasive and native grasses. Nor does it
distinguish between annual or perennial grasses.
Shrublands and grasslands are the dominant land cover classes and vegetation groups
within the grazing allotments. The two land cover classes also have different responses to

26

precipitation and fire, especially in grasslands dominated by cheatgrass. Literature described in
Chapter 2 also notes that grasslands and shrublands have different NDVI signatures throughout
the year. In this study area, the NDVI observations showed peaks during early spring and lower
values during the summer and early autumn.
The NLCD data were used to gain a general understanding of the vegetation types present
in the study area and to select secondary fire control zones that would have approximately the
same ratio of shrubland to grassland.
3.1.6. PRISM Precipitation Data
Historical precipitation data for this project were acquired from the PRISM Climate
Group at Oregon State University. The PRISM (Parameter-elevation Regressions on Independent
Slopes Model) monthly precipitation products are raster images which record the total
precipitation for each month in millimeters at a cellular resolution of 4x4 km. This data used
observations from weather stations and elevation data to generate estimates for total
precipitation. Positional accuracy of the data is based on the DEM images used and is stated to
have a circular error of 130 m with 90% probability. The PRISM model is further described in
Daly et al. (2008).
Monthly precipitation data were used to generate charts showing four-month precipitation
totals for each zone. The months were grouped based on the seasonality of fire events, such as
June through September being the peak fire season. Westerling et al. (2003) and other sources
found that reduced drought conditions often led to accelerated vegetation growth and increased
fire activity. This analysis was used to measure the influence of precipitation on the fire events.

27

3.2. Methods
This section is divided into parts describing how the methodology of this project
developed. The first section deals with the first research question.
3.2.1. Secondary Fire Identification
The desired outcome of the first research question is to create a list of fires within the
study area which are linked spatially and temporally with a previous fire event. The specific
relationship being investigated is fires which ignited within a grazing closure during the
closure’s rest period.
3.2.1.1. Original methodology based on closure and ignition data
The original methodology devised to generate the list of secondary fires would have used
the Select by Location tool (or a similar tool) to identify the ignition points within the closure
perimeters. Attribute data would then be used to keep only the ignition points where the ignition
occurred during the rest period. The resulting table of ignition points would be the list of
secondary fires answering the first question.
Thus, an ideal situation would be to have the closure perimeters and rest period dates for
post-fire grazing bans on BLM allotments. Such data would provide a complete spatial and
temporal record of the BLM policies in action. However, Paul Peterson, the BLM Nevada State
Fire Management Officer, clarified that closure area and rest period data are not available (Paul
Peterson, Oct. 11, 2017, e-mail message to author). Without this data, an alternative
methodology would need to be developed and the answer provided for the first question would
be less exact.

28

3.2.1.2. Alternative methodology using fire perimeters and ignition points
Without the closure data, there were two components that needed to be approximated: the
area closed to grazing and the duration of the rest period. Fire perimeters provided by the BLM
would be the best approximate of the grazing closure perimeters. Because closures and fire
perimeters are not perfectly aligned, there is some uncertainty of what later ignitions are closure
ignitions.
The methodology chosen to handle this approximation was to create a Near Table
comparing ignition data to fire perimeters. The Near Table would pair each fire perimeter and
ignition point within a search radius and then list the distance in map units between them. The
assumption with this methodology is that the closer the points, the greater the likelihood that the
ignition occurred within a grazing closure.
The duration, which would have been specific to each closure, was replaced by a search
window. The lower boundary of the search window was set to the spring of the year after the
initial fire. This was based on logically extending the ranchers’ argument. Fires which occurred
during the same year are not likely to have any vegetation regrowth between them, thus the
secondary fire in such pairs were not affected by BLM policy implemented on the initial fire.
Same year fires could thus be removed from the list.
The upper bound for the search window was set to three years (1095 days). This value
was derived from the BLM policy handbook (BLM 2007). If post-fire recovery goals have not
been met after three years, BLM policy relinquishes authority over fire recovery efforts to local
or regional BLM offices. The local offices can then choose to continue, cease, or modify the
grazing closure as they deem fit. Once recovery authority has been turned over to local offices,
there is greater uncertainty regarding the rest period duration.

29

The lack of closure data has introduced spatial and temporal uncertainty into the answers
for the research questions. While reasonable approximations have been selected, the answer
derived from this is not a perfect answer. Far worse for the alternative methodology was the poor
quality of the ignition points.
The National Interagency Fire Center provides a Wildland Fire Management Information
Fire Reporting Annual Dataset. This dataset is a comma delineated file containing date and point
of origin/ignition data for wildland fires. Many of the data fields in this dataset match attributes
in the BLM fire perimeter data, which would have allowed pairing between fire perimeters and
ignition points. Initial testing of the dataset, showed that 475 fires had records in both sets. The
ignition point data would have allowed this project to identify the origins of fires and related fire
events.
The fields storing the coordinates from these records were converted into point features
following the instructions provided by the source website. Ignition point data were stored as
NAD 83 Latitude and Longitude coordinates. It was necessary to convert the CSV file into Excel
format, as converting directly to an XY Event Layer in ArcGIS caused attribute data types to not
process correctly and attribute values to be lost.
Further testing revealed that there were ignition points recorded for fires that were
outside the recorded fire perimeters. The dataset was redownloaded and reprocessed, then a more
thorough comparison of the two datasets was conducted. The comparison utilized the Near Table
tool to measure distances between the ignition points and fire perimeters. The resulting table was
filtered so only rows for matched fire identifiers were kept. Of the 475 pairs, only 200 had an
ignition point within the paired fire perimeter. Of the 275 fires showing external ignition points,
239 were within 2 km of the fire perimeter. Four of the perimeters were paired with ignition

30

points that were 10 km or more outside of the perimeter. This suggested that the accuracies of the
200 ignition points within the fire perimeters were also in question.
Visual comparison of the fire perimeter polygons with differenced Normalized Burn
Ratio (dNBR) images taken from LANDSAT sensors confirmed that the datasets identified
similar regions as burned. For this reason, this project discarded the Fire Reporting Annual
Dataset. A second alternative methodology was needed to account for the unknown ignition
points of fires.
3.2.1.3. Final methodology using self-intersecting fire perimeters
The ideal method for identifying rest period fire candidates would be to compare ignition
point locations to the closure areas of previous fires and the time of the ignition to the rest period
of the previous fires. The results of such a search would yield every ignition event which
occurred within the closure area and during the rest period of a previous fire. As previously
discussed, these datasets were unavailable or unreliable which caused this project to go down a
different direction. A general workflow diagram of this final methodology is shown in Figure 5.
The BLM fire perimeter data include the areas directly affected by a fire and the
discovery and control dates for the fire events that are recorded. It should be noted that the
smallest recorded fire in the study area between 2000 and 2016 was about 9.5 acres. Any smaller
fires were not recorded by this dataset. The fire perimeters recorded are polygon vector data and
can be intersected with other polygon data.
Self-intersecting the fire perimeter data yields an unusual dataset. Given a dataset where
there is no overlapping topology within the spatial data, self-intersect does not produce any
useful results. In data with internally overlapping topology, such as the fire perimeter data,
multiple intersections are identified.

31

Figure 5. Workflow for Secondary Fire Identification and Case Study Selection

A quick way to understand this is to imagine two overlapping circles “A” and “B” in a
single dataset. When the dataset is self-intersected, it will create polygons for four overlapping
regions: “A to A”, “A to B”, “B to A”, and “B to B”. This is because the Intersect tool will create
new polygons where borders from any input layers intersect. The polygons for “A to B” and “B
to A” are bound by the same line segments (the intersecting borders of “A” and “B”) and are thus
congruent polygons. The main difference between the two is which set of attribute data are listed
first. In the case of the fire perimeters, this enabled the project to identify the first set of attributes
as the initial fire and the second set as the secondary fire.

32

A Date Difference field was added to the intersected perimeter dataset. The value of this
field (Ddiff ) was difference between the control or ending date of the first fire (Fcont) and the
discovery or starting date of the second fire (Sdisc).
Ddiff = Sdisc - Fcont (1)
The Date Difference field had values between 1 and 1095 days for pairs of fires that
occurred within 3 years of each other. Date Difference values higher than 1095 would show fires
more than three years apart. Date Difference values of 0 or lower would signify fires which
occurred before the previous fire, a logical impossibility. The unwanted intersections generated
by the self-intersecting process had Date Difference values of 0 or less, and were filtered out by
keeping only the intersections where the Date Difference value was between 1 and 1095.
The Intersect tool can split a single intersection into multiple polygons if the intersection
is crossed by the perimeter of a temporally unrelated fire. The Dissolve tool was used to
recombine split intersections. When the Dissolve tool was used for this purpose it was necessary
to select all non-spatial and non-object ID attribute fields as dissolve fields to retain the data in
those fields after the dissolve. The perimeter length, area, and object ID fields were
automatically recalculated by the Dissolve operation.
The resulting polygons identify areas which were burned by more than one fire event
within three years. This dataset was then intersected with the grazing allotment polygons to
create a dataset of three-year repeat fires that were on BLM grazing allotments. The intersection
areas from this dataset were compared to the total area of the secondary fires. A higher ratio
indicated that a greater area of the secondary fire was within the perimeter of the initial fire and
was more likely to have its ignition point within the initial fire’s perimeter. Without exact
ignition point data, this ratio became a proxy for estimating if the secondary fire ignited within

33

the fire perimeter of the previous fire. A ratio of 1:1 would indicate that the entire secondary fire,
including the ignition point, was within the perimeter of the previous fire.
One consequence of this search procedure is that secondary fires that are smaller than the
initial fires are more likely to be identified as rest period fire candidates. While larger secondary
fires could have ignited within the initial perimeters, it is a far less certain assertion to make. If
only 5% of the secondary fire overlaps with the previous fire, that leaves another 95% of the
secondary fire’s area where it could have ignited.
This also does not imply that the ratio of intersection to total area is equal to the
probability that the ignition occurred in the intersecting area. The probability of ignition events is
not homogenous throughout the fire perimeter but is greater at locations with more lightning
strikes or more human access. The relationship between ignition probability and intersection
ratio is generally fuzzy except at the ends (the 1:1 ratio case). For the purposes of this project, the
secondary fires with the greatest intersection ratios will be selected as case studies to test the
ranchers’ claim using vegetation index time series.
Implementing this methodology resulted in a list of 54 secondary fires that reburned areas
within three years of the initial fire with at least one winter in between. Due to differences
between the closure perimeters and fire perimeters, there is some uncertainty about this list
regarding answering the first research question. Likewise, the overlapping area ratio from self-
intersecting the fire perimeters is only a perfect approximation for ignition points when the ratio
is 100%, which indicates the ignition had to be within the initial fire, or 0%, which indicates the
ignition could not have possibly been within the initial fire. Four of the 54 secondary fires had
ratios at or near 100%. The 54 secondary fires and the four fires selected for the vegetation index
case study are described in greater detail in Chapter 4.

34

3.2.2. Evolution of Project Methodology Using LANDSAT5 TM Data
The second research question is a more difficult question to answer. If the ranchers’
claim is correct, the fuel load, a combination of dead plant litter and living vegetation biomass,
within the grazing closures should be higher than the fuel load in nearby areas open to grazing
during the same period. The ideal way to measure this would be to monitor the amount of
biomass, especially dry biomass, in areas closed to grazing and open to grazing. However, the
historical nature of this project limited what data were available and adjustments to the methods
and qualifications to the research question were necessary.
3.2.2.1. Original methodology using NDVI or NBR
The original design of this project involved the use of dNDVI and/or dNBR as estimators
of biomass change. The basic steps would have been to compare annual vegetation index data
starting with dates from just before the secondary fire and going back year-by-year to the time
before the initial fire. As discussed in Chapter 2, Miller and Thode (2007) demonstrated that
dNBR and other delta indices are only capable of showing how much change has occurred and
these kinds of data obscure the initial and final index values. Box et al. (1989) showed that while
NDVI was a good estimator of primary productivity, it was a poor estimator of biomass.
Based on Miller and Thode (2007) and Santin-Janin et al. (2009), futher methods were
designed around creating time series showing mean values of a vegetation index for grazed and
ungrazed regions. Due to Box et al. (2009), alternative vegetation indices were investigated.
3.2.2.2. Alternative methodology using CAI and NDVI
Guerschman et al. (2009) found that using the Cellulose Absorption Index (CAI) in
conjunction with NDVI, one could estimate the relative surface cover between green vegetation,
dry vegetation, and bare soil. The process for this would have been to calculate the two indices

35

and determine the fractional cover in a cell based on the index values. A high NDVI value would
indicate more green vegetation. Likewise, a high CAI value would indicate more dry vegetation.
Low CAI and NDVI values would indicate more bare soil. This would have been a useful tool as
NDVI is poor at distinguishing between bare soil and dry vegetation. A better estimate of dry
vegetation in the grazed and ungrazed areas would have provided a better estimate of fuel load.
This method was set aside due to the lack of necessary historical data. As discussed in
Chapter 2, the wavebands used to calculate CAI are unified into a single band within the MODIS
and LANDSAT sensors. The combined waveband prevents the calculation of CAI and any
further use of this method.
3.2.3. Selected Methodology for Scaled NDVI Time Series
Scaled NDVI was settled on as the time series vegetation index due to its relationship
with fractional green vegetation cover. The assumption here is that changes to green vegetation
cover over time can show regrowth or recolonization, senescence, and disturbance events.
Without a direct way to measure historical biomass and fuel load, vegetation regrowth in an area
may serve as a proxy for later biomass accumulation. Senescence likewise could provide an
estimate of green vegetation turning into dry vegetation during the late spring, prior to later fires.
There are three main steps to generate the time series for each of the cases: image
acquisition and preprocessing, vegetation index calculations, and the creation of zonal statistics
tables and charts. The workflow for this process is diagrammed in Figure 6.

36

Figure 6. Workflow for Vegetation Index Processing
3.2.3.1. Remote Sensing Preprocessing
Once the candidate fires for the case studies were identified, LANDSAT surface
reflectance data covering the spatial extent and timeframe of the initial and secondary fire pairs
were acquired. This data are available as a Level 2 product from the USGS Earth Explorer
website. The website allows users to upload simple (maximum 30 vertices) polygons as
shapefiles. The uploaded polygons are used to select the spatial extent of the LANDSAT data
and the time range was selected as the year of the initial fire to the year of the subsequent fire,
inclusive. Images for November through February were not included due to snow cover, cloud
cover, and almost complete absence of fire events during those months.
The file names of LANDSAT imagery provide specific information to the end user, but
the names are long and can be tedious to work with. A bulk renaming utility was used to quickly

37

rename the images to keep only path/row location, date of acquisition, and spectral band labels.
The naming system was used with Model Builder in later steps to automate most of the
procedures described. List Iterators in Model Builder were used to select scenes for processing
and to label the output files.
Working with multiple, full-sized LANDSAT 5 images can be taxing on file storage and
processing time, making it necessary to extract the relevant parts of the image before any further
processing. The enveloping rectangles, once reprojected to UTM 11N WGS84, were used with
the Extract by Mask tool to clip out the areas needed for processing and to discard the
unnecessary regions.
LANDSAT surface reflectance products include Quality Assurance (QA) images with
each scene. PIXEL_QA raster images assigns a value to each pixel based on the surface and
atmospheric quality of the pixel. Pixels showing clouds, snow, water, and other undesirable
features needed to be excluded. PIXEL_QA values of 66 and 130 identify pixels showing
unobscured surface. The PIXEL_QA values are derived from a direct classification and from a
cloud confidence calculation. In this case, the value of 66 represents “clear skies with low chance
of clouds” and the value of 130 represents “clear skies with medium chance of clouds”. Any
pixels with a high chance of clouds would not be classified as “clear skies”. Pixel values of 66
and 130 were remapped to a value of “1” and all other values were remapped to “No Data”. The
resulting raster images had only valid pixels.
RADSAT_QA is a quality assurance image which tracks over-saturated pixels in the
various spectral bands. Each pixel value in RADSAT_QA stores one bit of data for each spectral
band. These bits of data function as true or false values to determine if the pixel is oversaturated.
The bits are combined into a single byte value for the pixel. It was necessary to remove

38

oversaturated pixels as they do not represent true values of surface reflectance. This process
involved using raster arithmetic to extract the bit flags for Bands 3 and 4 from the rest of the QA
image. Finding the modulus of the QA image over 32 would yield pixels values where Bands 5
through 7 (which have bit values of 32, 64, and 128) are ignored. The pixel values in this image
will be greater than 8 if and only if Band 3 (bit value of 8) and/or Band 4 (bit value of 16) are
oversaturated.
Equation 2 can be used to identify pixels with valid saturation. The “%” in this case
represents the modulo operation which returns the remainder after division. The pixels with valid
saturation (SATv) resulting from this were assigned values of “1” and invalid pixels were
assigned to “No Data”. As with PIXEL_QA, the resulting images had only valid pixels.
If (RADSAT_QA % 32) < 8 = true, then SATv = 1. Else SATv = NoData (2)

The surface reflectance calculations can sometimes result in negative values in the
spectral bands. Negative reflectance is not possible, so such values are invalid and must also be
removed. This only required remapping negative surface reflectance values to “No Data”. Raster
multiplication with the two processed QA images worked like a Boolean AND operation. The
resulting images had negative, oversaturated, and obscured data removed.
In cases where the fire perimeter spanned multiple LANDSAT scenes, it was necessary to
create a mosaic image from the scenes after any invalid pixels were filtered out by the quality
assurance process. The mosaic process was performed at this place in the process because any
earlier mosaic may have incorporated invalid values. Performing the mosaic later (after the
vegetation index was calculated) could have affected the index values on the borders of the
scenes.

39

One potential consideration for mosaicking LANDSAT imagery is that a burn perimeter
might cross multiple swaths. One consequence of using orbit-based imaging platforms is that
adjacent swaths will not be imaged at the same time. LANDSAT 5 was no exception to this,
having used a near-polar, sun-synchronous orbit to ensure as much daylight as possible for its
images. In northern Nevada adjacent LANDSAT 5 swaths were taken either 7 or 9 days apart
because of the offset of the orbit. A fire spanning multiple swaths would need images from at
least a week apart to create the mosaic image. Fortunately, the cases investigated in this project
were all from the same swath (WRS2 Path 41), which alleviated this concern.
3.2.3.2. Vegetation index calculations
In the LANDSAT 5 Thematic Mapper, spectral band 3 records reflected energy with
wavelengths associated with red light (0.63-0.69 µm). Spectral band 4 records the reflected
energy at wavelengths labeled as Near Infrared or NIR (0.76-0.90 µm). The Normalized
Difference Vegetation Index (NDVI) provides an estimate of green vegetation productivity by
comparing the difference in NIR and visible red reflectance to the sum of the two values.
Chlorophyll activity in green plant matter absorbs red light and emits NIR radiation during the
evapotranspiration processes. High NIR reflectance coupled with low red-light reflectance can
thus be a signature of chlorophyll activity. Equation 3 is the basic algorithm for calculating
NDVI from LANDSAT 5 TM NIR (B4) and visible red (B3) surface reflectance values. Figure 7
shows how Equation 3 can be implemented in ArcGIS Model Builder.
NDVI = (B4 - B3) / Float(B4 + B3) (3)

40

Figure 7. Model Builder Layout for NDVI Calculations.
The Float operation converted the integer format of the input values into floating point
values. The integer data type was carried over from the LANDSAT surface reflectance source
data and not relevant until this point in the process. Without the Float operation, the raster
division operation in ArcGIS would detect integer-formatted inputs and yield an integer-
formatted output. On an NDVI image, this would result in NDVI values being rounded to -1, 0,
and 1. When the denominator of the Divide operation has a floating-point data type, the resulting
image will also have a floating-point data type.
The NDVI raster images were reprojected to the NAD83 datum from the WGS84 datum
of the LANDSAT 5 source data. The reprojection facilitated subsequent operations involving
NAD83 datum datasets.
Equation 4 is used to calculated Scaled NDVI (NDVI*) from NDVI values. The
calculation requires determining a saturated NDVI (NDVIS) and a bare soil NDVI (NDVI0). The
saturated NDVI represents the maximum achievable NDVI value due to vegetation. The bare soil
NDVI is the NDVI value for unvegetated areas.
NDVI* = (NDVI - NDVI0) / (NDVIS - NDVI0) (4)

41

Agricultural areas were identified by visually searching for tell-tale signs of center-pivot
agriculture (Figure 8) and using the cells classified as “Cropland” in the NLCD 06 data.
Croplands are well-suited for estimating NDVIS because farmers will optimize the growth of
crops for use and sale. Crops, when compared to wild vegetation, will have better irrigation and
better soil due to the actions and choices of the farmers. Thus it can be expected for croplands to
out-produce wild vegetation.

Figure 8. Example of Center-Pivot Irrigation Systems in Nevada
Determining NDVI0 was not as straight-forward as determining NDVIS. The idea of
using the values from nearby mining sites was considered. However, it was determined that the
mining sites were potentially too disturbed by human and industrial activity to represent the true
bare soil value. The methods described in the literature were not available for this project, as they
required either direct measurements from training sites or the use of unusual statistical analyses.
The method chosen for this project looked at the pre- and post-fire NDVI values to find the least

42

changed cells in the fire perimeters. The assumption was made that these cells represent areas
unaffected by pre-fire vegetation and post-fire charcoal. The lowest value of the least disturbed
cells was then selected as NDVI0.
A more complicated method for calculating Scaled NDVI would involve recalculating
saturated and bare soil values scene by scene. Each scene in the time series would have a specific
and relevant Scaled NDVI formula which could better account for variation over time and
seasonal variation. This method was not implemented in this project due to time constraints and
is described here as an area of future research.
Scaled NDVI raster images had values ranging from 0 (bare soil) to 1 (saturated
vegetation) or No Data. The Scaled NDVI data were used in the following steps to generate
zonal statistics and time series graphs.
3.2.3.3. Zonal statistics
To see the effects of the initial fires, control plots were created for each case study. The
control plots would have same size and shape as the secondary fire. A general workflow of this
process is shown in Figure 9. Zones were created by using the ArcGIS Edit tools to create copies
of the secondary fires and to move the copies to the control locations. One copy of the secondary
fire was used as a burned control. The burned control was placed at a location that was within
the fire perimeter of the initial fire, but was unaffected by the secondary fire. Another copy of the
secondary fire was used as an unburned control. The unburned control was placed at a location
near the secondary fire, but otherwise unaffected by any recorded fires up until and including the
secondary fire.
It should be noted that while this method of zonal selection attempted to control for
various attributes, it is not statistically strong enough to make definitive statements. A method

43

utilizing random sampling points throughout larger regions would provide a more through and
significant estimate of the Scaled NDVI values in the regions. Additional controls could be put
into place so that the random points could have the correct ratio of land cover types. Since
statistical conclusions were not the objective of this study given the nature of the research
questions, the random sample method was not used here. It is, however, definitely a direction for
future research.

Figure 9. Workflow for Zonal Statistics and Time Series Creation

The control zones were also selected based on the ratio of shrubland to grassland.
Vegetation based land cover data were used to attempt to control for the different responses and
influences that different vegetation types can have with fire events. NLCD data from the closest
year to the fire events were used to determine the ratio of shrub-dominated cover to grass-

44

dominated cover. The burned and unburned control sites for each secondary fire were selected to
have approximately the same ratio of cover types.
Based on the ranchers’ claim, the secondary fire and the burned control zone would at
first have less green vegetation and less biomass than the unburned control due to the initial fire.
However, the areas burned by the initial fire would be protected from livestock grazing allowing
for greater fuel accumulation over time than what would be seen in the unburned (and open for
grazing) control zone.
Before calculating zonal statistics, it was necessary to filter out observations where too
many of the pixels were invalidated either by the quality assurance process or by the rescaling of
the NDVI values. For the purposes of this project, it was decided to filter out observations where
less than half of the pixels for a zone were valid. To keep as many observations as possible, the
fires and control zones were analyzed separately for valid cell counts. This avoided situations
where valid data for two of the zones would have been thrown out due to missing data for the
third zone.
Determining the valid cell percentage only required generating a new raster image for
each observation where valid cells in the Scaled NDVI image were assigned a value of 1 and
invalid cells were assigned a value of 0. In ArcGIS, the IsNull tool does the opposite of this,
assigning null cells a value of 1 in the new image. Because the quality assurance and NDVI
scaling processes assigned null values to invalid cells, the IsNull tool effectively assigned values
of 1 to invalid cells. As shown in Equation 5, subtracting the IsNull result from a raster with a
constant value of 1 resulted in a raster image where valid cells in the original image were
assigned values of 1.
IsValid = 1 - IsNull(NDVI*) (5)

45

Calculating zonal statistics using the valid cell images yielded the necessary calculations
for valid cell percentage per zone. In the valid cell image, the only possible values are 1 for valid
cells and 0 for invalid cells. The mean cell value for a given zone in the valid cell image is
therefore equal to the valid cell count divided by the total cell count for the given zone. Any
observation date where the zonal mean of the valid cell image was greater than 0.5 was listed as
a valid observation for the given zone.
Once the list of valid observation dates for each zone was determined, zonal statistics for
each zone on each of the valid observation dates was calculated. The Zonal Statistics as Table
tool was used to generate a table output that could be exported and manipulated. The observation
date and zone name were added as fields to the resulting tables. The new fields were used to
uniquely identify rows once the zonal statistics tables were combined using the Merge tool. The
merged table was exported as an Excel spreadsheet.
Charts showing the changes in mean Scaled NDVI for the secondary fires and their
respective control zones over time were created in Excel from the exported tables. The time
series charts were used to analyze the ranchers’ claim that burned areas under grazing closures
will have greater regrowth than unburned lands open to grazing. If the ranchers’ claim is
accurate, the fire perimeters and burned control zones should have higher Scaled NDVI values
when compared to unburned control zones near the time of the secondary fire.
3.2.4. Precipitation Data
As described earlier, the PRISM precipitation data have a cell resolution of 4 km. It
turned out that this cell size was larger than many of the secondary fires identified in the site
selection process. The size difference prevented the Extract by Mask tool from working as
intended. The cell resolution instead warranted the use of zonal centroids and the Sample tool.

46

The Feature to Point tool created the centroids for the secondary fire perimeters and the
control zones. The Sample tool took the series of precipitation imagery and found the values of
cells marked by the centroid features. The sampled values were then added to a table containing
the list of zones and columns for each monthly precipitation estimate.
This table was exported to Excel and line charts were generated for each zone and fire
perimeter. Precipitation estimates for winter months on a year-by-year basis were also calculated.
The four-month precipitation totals were used to analyze the influence of precipitation on the
Scaled NDVI in the zones.
3.2.5. Additional Procedures
The grazing allotment data were spatially joined with burn perimeter data within ArcGIS.
As part of the spatial joining process, a new field was calculated containing the count of burn
perimeters for each allotment. The data resulting from this operation have the same geometry
and topology as the grazing allotment data and included the count of burn perimeters. This data
were used to identify fire hot spots within the grazing allotments. Intersecting the allotment and
burn perimeter data provided a list of fire events affecting each allotment. A similar analysis was
performed using a grid with 10 x 10 km cells as the spatial base. The gridded fire count was
created to control for the Modifiable Areal Unit Problem (MAUP). Grazing allotments, which
have a variety shapes and sizes, are prone to MAUP concerns.
As discussed in Chapter 2, Martinez et al. (2009) studied various factors to determine
which were more correlated with human-caused ignitions. They found that sites with highly
partitioned and mechanized agriculture were more prone to human caused fires. Other factors
were related to increased development near the sites, such as increased access to the wildlands by
humans. Imagery of the study site was overlaid with the perimeters of grazing allotments that

47

had the highest fire occurrence during the study’s timeframe to provide a qualitative estimate of
the influence of human factors.

48

Chapter 4 Results
This chapter reports the results of carrying out the selected project methodology. The first
section goes over the results of the secondary fire search/case study selection process and
provides a basic overview of the selected case study sites. The next section discusses the
vegetation index time series for each of the cases. The preliminary methods considered in
Chapter 3 were not carried out and are therefore not discussed.
4.1. Secondary Fire Search and Case Study Selection
Of the 594 fires investigated as part of this study, 58 fires had a fire perimeter which
overlapped with another fire within the previous three years. Additionally, four of the 58
secondary fires occurred on the same year as the initial fires, with intervals ranging between
seven and 38 days. The ranchers’ claim is based on vegetation regrowth, thus same-year fires are
not considered valid rest period fire candidates. These four were included in the initial results out
of a concern that other fire pairs might have been affected by a same-year reburn. The four
incidents fortunately did not affect the fires selected for the case studies. The 54 remaining
secondary fires, which reburned areas within three years but not during the same year, are the
best answers for the first research question given the available data.
Table 1 lists selected pairs of fires which had overlapping perimeters within three years.
The top four unshaded rows are the fire pairs where 99% or more of the secondary fire was
contained in the initial fire perimeter. These four secondary fires (Little One, Green Monster,
Rock Creek, and Squawvalle) and the respective initial fires (Winters, Sheep, and Hot Lake)
were selected for the case studies.
The blue-shaded rows are two fires pairs where more than half of the secondary fire was
within the initial perimeter. While there is a chance that the secondary fires ignited within the

49

respective initial perimeters, it is not a near certainty as in the case of the four pairs that were
selected. The probability that an unknown ignition point was at a specific spot is not equally
likely throughout the fire perimeter. Areas prone to lightning strikes or with more human activity
(the primary sources of ignitions) are more likely to contain the point of ignition. As such, a 67%
overlapping area does not mean a 67% chance that the ignition point was within the overlap.
Nonetheless, these two pairs were noted as possible alternative cases if needed.
Table 1. Selected Three-Year Overlapping Fire Perimeters (16 out of 58 total). Column names
starting with “I.” indicate initial fire attributes and “S.” indicate secondary fire attributes. Shaded
regions are described in the text.

The yellow-shaded rows after that are the five pairs where 10 to 50% of the secondary
fire were within the initial perimeters. These pairs are included in the table to demonstrate how
quickly the overlapping area percentage decreased in this dataset. Only 11 of the 58 Three-Year
Overlaps had areas that were 10% or more of the secondary fire’s area.
The red-shaded rows are five fire pairs that involve either the Winters Fire or the Sheep
Fire, which are initial fires for three of the case study fires. If these five pairs represented fires
which met the requirement of the ranchers’ claims, they could have implied that there are cycles

50

of wildfires and grazing closures in the region. As it stands, the five pairs have overlapping areas
that are less than 0.5% of the burnt area of the secondary fires. The low overlapping percentages
suggest that the five listed initial fires had little to no impact on the Winters and Sheep Fires.
In all four of the case study pairs the secondary fires are smaller than the initial fires. As
described in Chapter 3, this is a consequence of using the overlapping percentage value as a
filter. It does not indicate that all secondary fires are smaller than the initial fires. Rather, missing
ignition point data makes it unlikely, if not impossible, to determine if a larger secondary fire
ignited within the perimeter of a smaller initial fire.
4.2. Case Study Fire Events
The case study selection process described in Chapter 3 identified four fire events that,
given the available data and selected methodology, best met the spatial and temporal criteria of a
rest period fire. All four events were concentrated in the area north-northeast of the town of
Battle Mountain (Figure 10). Two of the events (Little One and Green Monster) shared the same
initial fire (the Winters Fire). Three of the events (Little One, Green Monster and Rock Creek)
covered the same timeframe, having the initial fires in 2006 and the secondary fires in 2008.
While the 2004 Squawvalle Fire and the three initial fires were listed as having natural causes,
the other three secondary fires were listed as human-caused. Additionally, the Squawvalle Fire of
2004 has an overlap with the Sheep Fire of 2006. The overlap percentage for this pair is less than
0.25%, so it is not likely that the Squawvalle Fire had a significant effect on the Sheep Fire.

51

Figure 10. Map of Fire Events from Case Studies

52

4.2.1. Squawvalle Fire Zones
The Squawvalle Fire was discovered 18 km south-southeast of Midas, NV, on June 23,
2004. It was preceded by the Hot Lake Fire, which was contained on September 30, 2001, 997
days earlier. Approximately 0.2% of the area of the Squawvalle Fire extends past the perimeter
of the Hot Lake Fire. The Squawvalle Fire is on the western edge of the Hot Lake Fire’s eastern
lobe (Figure 11). The Burned Control Zone for this case was placed to the immediate northeast
of the fire perimeter. Because the Squawvalle Fire was centrally located within the Hot Lake
Fire’s fire perimeter, the Unburned Control Zone was placed about 15 km to the west-southwest
of the fire perimeter.

Figure 11. Squawvalle Zones
Control Zones were selected to have approximately the same ratio of shrub-dominated
area to grass-dominated area based on the 2001 NLCD data. Table 2 shows the total counts for
both categories in all three zones and compares the counts in the control zones to the areas burnt
by the Squawvalle Fire. Error for this table is the difference in the counts of the two categories.

53

The Burned and Unburned Control Zones had classification errors of 1 cell and 8 cells,
respectively. The difference in the total cell count between zones is due to boundary errors
between vector masks and raster images when using the Extract by Mask or Zonal Statistics
tools. Adjusting the position of the vector mask or zone can cause the tools to include or exclude
cells at the vector boundary. According to the 2001 NLCD, approximately 90% of the cells in the
Squawvalle Fire and the control zones represented shrub-dominated cover.
Table 2. Squawvalle Land Cover Cell Count

4.2.2. Little One Fire and Green Monster Fire Zones
Figure 12 shows the locations of the Little One and Green Monster Fires, which are
notable for being physically adjacent events that occurred only a month and a half apart. Both
fires followed the Winters Fire that was contained on August 3, 2006. The Little One Fire was
discovered 727 days later, on July 30, 2008. The Green Monster Fire was discovered later in the
season, on September 16, 2008. The boundary between the two fires is a small creek passing
through the area. Both fires as well as the initial Winters Fire are all north of Midas, NV.
The control zones for the Green Monster Fire ended up being relatively close to the fire
perimeter. The burned control is just east of the fire, while the unburned control is about 18 km
to the west. The control zones for the Little One Fire were set further away due to land cover
balancing. The burned control is about 10 km northwest of the fire perimeter, while the unburned
control is west of Midas.

54

Figure 12. Little One and Green Monster Zones
Table 3 shows the relative count of shrub and grass land cover types based on the 2006
version of the NLCD. The burned controls for both fires were better matched to the fire
perimeter land covers, but the category errors for all four control zones were less than 1% of the
cell count. The Little One Fire had a ratio of about 10 to 3 in favor of shrubland cover. The ratio
for the Green Monster Fire was closer at 8 to 5 in favor of the shrubland.
Table 3. Little One and Green Monster Land Cover Cell Count

55

4.2.3. Rock Creek Fire Zones
The Rock Creek Fire (Figure 13) was discovered on August 10, 2008. This was 700 days
after the Sheep Fire was contained, on September 10, 2006. The Rock Creek Fire is centrally
located in the lower lobe of the Sheep Fire. The Rock Creek Fire is the largest secondary fire
selected for the case studies. Control zones were set further away (approximately 18 km to the
northwest) to find locations with comparable land cover ratios. The Unburned Control Zone is
divided by the Rock Creek Road, which appears to divide the shrubland and grassland cover
types.

Figure 13. Rock Creek Zones
Finding ideal locations for the control zones was problematic due to the abundance of
grassland dominant cells. Unlike the previous fires, The Rock Creek fire perimeter has a roughly

56

even ratio of shrublands to grass lands (Table 4). While the absolute classification error for both
zones was around 1% of the total number of cells.
Table 4. Rock Creek Land Cover Cell Count

4.3. Scaled NDVI Time Series and Four-Month Precipitation Totals
Due to the spatial and temporal proximity of the Little One, Green Monster, and Rock
Creek Fires, common values for NDVIs and NDVI0 were used for these three case studies. A
different set of values for NDVIs and NDVI0 were used for earlier Squawvalle Fire.
NDVI0 values were calculated from the least disturbed post-fire pixels in the Squawvalle
perimeter (for the 2001 to 2004 data) and the Rock Creek perimeter (for the 2006 to 2008 data).
NDVIS values for the two time periods were calculated from the maximum observed NDVI in
cropland pixels during the time periods. For the Squawvalle Fire case study, NDVIs was set to
0.9237 and NDVI0 was set to 0.1154. For the 2008 case studies, NDVIs was set to 0.9021 and
NDVI0 was set to 0.0835.
4.3.1. Squawvalle Fire
The mean Scaled NDVI values for the Squawvalle Perimeter and Burned Control Zones
during the spring after the Hot Lake Fire are lower than both the previous spring and the
subsequent spring. By comparison, the mean Scaled NDVI values for the Unburned Control
show less change during this time period. By the second spring after the fire, the mean values of

57

Scaled NDVI in burned areas are similar to mean values in the unburned area. This confirms
there was some form of vegetation regrowth in the burned area within two years of the initial
fire. This would lend support to the ranchers’ claim that vegetation in areas under a grazing
closure were recovering to levels comparable to the nearby unburned area.

Figure 14. Scaled NDVI over time for the Squawvalle Zones
The four-month precipitation series for the Squawvalle Zones (Figure 15) has a few
interesting features to discuss. The four-month periods ending May 2000, September 2003, and
September 2004 all had greater precipitation than comparable periods in other years. Oppositely,
the January 2003 precipitation totals are much lower than previous January observations. The
spike in Spring Scaled NDVI values in 2003 and 2004 do not seem to correlate well with any of
the precipitation observations.

58

Figure 15. Squawvalle Four-Month Precipitation Totals
4.3.2. Green Monster and Little One Fires
Figures 16 and 17 show the Scaled NDVI graphs for the Green Monster Fire and the
Little One Fire, respectively. Like with the Squawvalle data, the mean Scaled NDVI values show
a post-fire drop of live vegetation cover in burned areas during the first spring after the initial
fire. By the second spring, the live vegetation cover in burned areas is comparable to the cover in
the unburned area. These results also suggest that the ranchers’ claim regarding post-fire
regrowth may be valid.

59

Figure 16. Scaled NDVI over time for the Green Monster Zones

Figure 17. Scaled NDVI over time for the Little One Zones

60

The Four-Month Precipitation Totals for both fires (Figures 18 and 19) tell similar
stories. The Winter and Spring months in 2005 and 2006 have higher precipitation totals on
average than the same months in 2007 and 2008. This is consistent with Westerling et al. (2003),
as 2006 was a peak fire year while 2008 had very few recorded fire events (Figure 2).

Figure 18. Green Monster Four-Month Precipitation Totals

61

Figure 19. Little One Four-Month Precipitation Totals
4.3.3. Rock Creek Fire
The Scaled NDVI time series for the Rock Creek Zones is shown in Figure 20. Due to
cloud cover, some of the observations of the initial Sheep Fire were excluded as invalid data. The
signature of the Rock Creek Fire can be seen as the drop in the mean Scaled NDVI values of Fire
Perimeter zone at the end of the series. Perhaps the most unusual aspect of this time series is the
drop in the Unburned Control Zone values during 2008. The drop in the Unburned Control could
be the result of a region-wide disturbance. A more thorough analysis method and further research
would be needed to confirm that possibility. The Rock Creek data, with a seemingly constant
mean Scaled NDVI in the burned areas during spring observations, seem to be the least
consistent with the ranchers’ claim.

62

Figure 20. Scaled NDVI over time for the Rock Creek Zones
The Four-Month Precipitation Totals for the Rock Creek Zones (Figure 21) are consistent
with the precipitation data from the Little One and Green Monster Fires (Figures 18 and 19).
2005 and 2006 appear to be wetter years than 2007 and 2008, which is consistent with the fire
occurrence per year data (Figure 2). Since all three cases were concerned with spatially
proximate and contemporary fires, it is not surprising that the precipitation data are consistent for
all three.

63

Figure 21. Rock Creek Four-Month Precipitation Totals
4.4. Other Observations
In Chapter 1, Figure 2 included a bar chart for fire frequency per year. It is interesting to
note that all the case studies had initial fires during years with more fires (2001 and 2006) and
secondary fires during years with fewer fires (2004 and 2008). This appears to be another artifact
of the case study selection process (with the larger initial fires occurring in the more active
years). The selection filter favored larger initial fires and smaller secondary fires. As noted
Westerling et al. (2003), larger and more frequent fires are associated with wetter climates in the
previous year.

64

Chapter 5 Discussion and Conclusion
This chapter summarizes the main findings of the case studies, suggests methods that could be
used to further explore the study objectives, discusses other factors that might affect the
frequency of fire events or the perception of fire frequency, and describes potential developments
that could improve post-fire vegetation monitoring in the future.
5.1. Discussion
This study originally intended to use a method based on comparing ignition data to the
exact closure areas and rest periods. As described in Chapter 3, the fire identification
methodology was revised to use self-intersections of recorded fire perimeters due to the
unavailability of the closure dataset and unreliability of the ignition dataset. The self-intersected
perimeter data showed 54 fire events which reburned areas affected by fires one, two, or three
years before. The reburned areas for these fires covered almost 13,000 acres in total. While these
fires represent reburned areas, they do not represent the true closure areas or ignition points.
Other fires could have been added to this list if the ignition points and grazing closures were
known. Some of the 54 could also be removed from this list if the actual rest periods were
known. The search window of “less than three years but not the same year” was chosen as a
proxy for rest period durations based on the nature of the ranchers’ claim and the details of BLM
policy.
Of the 54 secondary fires, four were entirely or almost entirely within the perimeter of the
previous fire, indicating that the ignitions for the four fires were within the previous fire
perimeter. Because of this, these four fire pairs were selected for further analysis in the
vegetation case studies designed to answer the second research question.

65

To answer the second research question, a method to assess the changes to fuel load and
biomass during grazing rest periods was needed. Multiple methods to measure fuel load and/or
vegetation biomass in the case study zones were considered. After reviewing background
literature on vegetation indices and what historical data were available, options such as using
unmodified NDVI, differenced imagery, and CAI were ruled out. Finally, Scaled NDVI was
selected for use as an estimator of green vegetation cover and thus live vegetation recovery in the
case studies.
The advantages of Scaled NDVI were that it was easily calculated from base NDVI
values and that it related the observed values to bare soil and irrigated cropland. Time series
graphs were created as analysis and visualization tools for Scaled NDVI observations as an
alternative to creating differenced vegetation index imagery from pre- and post-event data.
LANDSAT 5 Thematic Mapper data were used to calculate NDVI because it was available as
high-resolution imagery of the case study locations during the periods of time covered by the
case studies.
Three of the four Scaled NDVI time series showed vegetation recovery in burned areas
within two years of the initial fire as estimated by the changes in vegetation cover. The recovery
in the Little One Fire, the Squawvalle Fire, and the Green Monster Fire suggests the possibility
that biomass and fuel load may also have recovered as would be expected under the ranchers’
claim, though further research would be needed to verify those conditions. The Rock Creek
Unburned Control zone had an unexplained drop of mean Scaled NDVI values in spring of 2008,
while the Burned Control and Fire Perimeter zones had stable values across all spring
observations.

66

Precipitation data seem to be consistent with the observations made by Westerling et al.
(2003) and other literature sources. Wetter years seem to result in more and larger fires in the
following year. The case study selection method, which preferred smaller secondary fires and
larger initial fires, also found secondary fires which occurred during low fire years, which may
be inconsistent with ranchers’ claim that the secondary fire resulted from accumulated vegetation
litter. However, if these are just artifacts of the search algorithm, it is possible there are other
examples of rest period fires that are more consistent with the ranchers’ claim.
Further investigation into these claims using a method such as sampling a sufficient
number of randomly selected cells from the secondary perimeters and from burned and unburned
regions for statistical analysis would be necessary to demonstrate statistically whether the
vegetation cover regrowth was significant enough to support the ranchers’ claim. Given the ratio
of shrubland to grassland in the burned areas, a random points method would have to be
stratified to balance that ratio for each case study fire. This could potentially be achieved by
splitting sampling regions by land cover classification. Additionally, the unburned sampling
region could be taken from the grazing allotments west of the Squawvalle and Rock Creek Fires
or similar allotments northeast of the Little One and Green Monster Fires. These allotments had
seven or fewer fires during the study period. This would require further consideration by
researchers utilizing the random points method.
5.2. Other Considerations
There is a high level of consensus in the literature that cheatgrass invasion is a primary
component of the current fire regime in the Great Basin. This project was concerned with fires
that occurred within three years of each other. This three-year frequency is itself a product of the
cheatgrass-modified fire regime. While the ranchers’ claim is only but justifiably concerned with

67

fuel accumulation during rest periods, the source of that fuel is also important. As such, ranchers
in areas with more frequent fires might consider targeted grazing in the Spring to reduce
cheatgrass biomass at their ranches.
Although fire suppression is no longer a popular land management practice, it may be
necessary when fires start near cheatgrass patches. Cheatgrass benefits too much from large and
uncontrolled fires to justify a hands-off approach in this region. Likewise, simply removing
livestock and ceasing ranching operation will not result in the restoration of the native
vegetation. Some level of direct management is necessary to remove cheatgrass and reestablish
the original fire regime.
It may be helpful to consider the spatial distribution of fire events relative to the grazing
allotments. In Figure 22, a grid with 10 km by 10 km cells was generated to cover the spatial
extent of the study area grazing allotments. The grid was spatially joined to the fires investigated
by the project so that a count of fires in each cell could be calculated. At this scale, there appears
to be a crescent of high fire frequency in the central part of the study area, with lesser hotspots at
the northwestern, northeastern, and southeastern corners. Also at this scale, the maximum fires
recorded in a cell was 11, which would be a frequency greater than one fire per year.

68

Figure 22. Fire Frequency per 10x10 km Grid
Another way to look at the spatial distribution of fire events is by allotment. A spatial
join was used to count the number of study area fires that occurred within each study area
grazing allotment. From the spatial join, a choropleth map using five classes grouped by natural
breaks and a sixth class for allotments with no fires was created (Figure 23). It should be noted
that the allotments have a wide range of sizes and a variety of shapes. As such, the Modifiable
Areal Unit Problem (MAUP) applies to this information. Although there are exceptions, the
larger allotments tend to have more fires, which is expected because the larger allotment have
much more space for fires to start or to spread.

69

Figure 23. Fire Frequency by Allotment, 2000-2016
The real message from this map is somewhat anecdotal. The people and businesses who
lease and manage these allotments are probably not going to care about the MAUP (though it
would be helpful to spatial scientists if they did). They are going to care that their allotment had
19 fires or 37 fires in 17 years. The case studies in this project were attempts to find evidence for
claims made by the lease holders and managers. The claims themselves were a consequence of
the events aggregated in this map.
A frequency of at least one fire every three years, which was the maximum interval for
this project, would result in 5 or more fires over the 17 years from 2000 to 2016. That frequency
includes many of the light green colored allotments and all yellow, orange, and red allotments.
While this study has shown that only 58 of the three-year fire pairs have overlapping perimeters,

70

the situation could appear as a continuous series of closures and fires to the ranchers or managers
present at these allotments.
Figure 24 shows all grazing allotments in the study area that had 17 or more intersecting
fire perimeters between 2000 and 2016, yielding an average of one or more recorded fire events
per year during the study period. The Twenty Five allotment, with 37 recorded fires between
2000 and 2016 (including the Rock Creek and Sheep fires) was the most frequently burned
allotment. While all four allotments are among the larger allotments, size alone does not explain
the frequency of fires in these allotments. One possible factor contributing to fire occurrence in
these allotments is human activity. All four are near towns or unincorporated inhabited places
(Winnemucca, Battle Mountain, Carlin, and Midas). There are two active mines in the area,
including the Goldstrike Mine which occupies the northwestern corner of the T Lazy S
allotment. All except the Squaw Valley allotment border Interstate 80. Farming sites with center-
pivot irrigation can easily be identified near, adjacent to, and within these allotments. The ease of
access and closeness to agricultural equipment and machinery could be signs that human activity
in the area has contributed to the increased fire count.

71

Figure 24. Mining Sites and Highways near Frequently Burned Allotments. The four allotments
pictured were affected by 18 or more fire events during the 17-year study period. The image was
created by importing the allotment shapefile into Google Earth.
5.3. A Better Tomorrow
As discussed in earlier chapters, the original concept for this project would have involved
spatial data for closure areas and ignition points, as well as the true duration of any rest periods.
Without access to accurate and complete copies of these data, this project instead identified
repeat fires by looking at intersecting fire perimeters and assuming a rest period duration of three
years, based on limits and descriptions from BLM wildland fire policies.
While the historical data are limited to what was collected at the time, attempts to track
rest period fires in the future could benefit from improved spatial data collection and better
record keeping. Accurate positional data could pinpoint spatially related events while rest period
histories could be used to create a timeline of events and observations.

72

In addition to the assumptions about the duration of rest periods, another assumption
made in this project was that the closure areas would include the entire fire perimeter. It is
possible that a low-intensity fire might only cause severe damage sporadically within the fire
perimeter. In such a scenario, the closure area might be smaller than the fire perimeter. This is
case where the missing data might hurt the ranchers’ claim. If two fires overlap and the
intersecting area was not part of the first fire’s closure area, then the second fire could not be the
result of the grazing rest period. It is not possible to be certain about any rest period fire if the
spatial and temporal information about the closure is not accurate and available.
The Cellulose Absorption Index (CAI) was also considered as a possible tool for
measuring rest period vegetation growth. CAI would have been useful in distinguishing between
dry vegetation and bare soil. However, the necessary bands to calculate CAI are all in the
shortwave infrared (SWIR) range and grouped together as a single band in all available
LANDSAT sensors. Information from the LANDSAT website suggests that the sensors on the
upcoming LANDSAT 9 will continue to group SWIR as a single band. CAI is not likely going to
be available through LANDSAT anytime soon. The remote sensing of rangeland health could
benefit if more complex sensors and new vegetation indices were developed to address the
shortcomings of current technology.
The final consideration is the lack of a means to assess the validity of the results
statistically. While a method such as the use of random sampling points could provide a
statistical result, the method used in this project was sufficient to demonstrate that there may be
some validity in the ranchers’ claim, suggesting that further research is called for. The method
described here provides a foundation for further development of appropriate statistical methods.
It also highlighted some of the main data concerns that can occur with a study such as this, such

73

as the missing closure data, the inaccurate ignition point data, and the merged waveband data on
LANDSAT sensors which prevented the use of CAI.
5.4. Conclusions
The two objectives of this project were to find fires that had the spatial and temporal
qualities described by the ranchers’ claim and then to analyze the accumulation of fuel in areas
under a grazing closure compared to areas open to grazing. Due to the data concerns discussed
throughout this document, the first objective was modified to find reburns of three-years or less
and then identify candidates for rest period fires by estimating the likelihood of ignition within
the reburned area by looking at the overlapping area. The second objective was modified to
consider live vegetation cover as proxy for vegetation recovery as direct methods to measure
historical biomass were not available.
The modified first question was successfully answered by identifying 54 recorded fire
events which affected areas previously burned up to three years earlier and by further identifying
four of those fires where the ignition point was within the previous fire perimeter.
The modified second objective was addressed by constructing Scaled NDVI time series
for the case study areas. The Scaled NDVI time series created for three of the four fires showed
vegetation cover similar to nearby unburned areas within two years of the initial fires, which
suggests live vegetation recovery contrasting with the BLM statement that two to three years
might be needed to see full recovery. Confirming this indication would require a stronger
statistical analysis, such as the random points sampling method discussed in Chapters 3 and 4.
Also, it is possible that the observed recovery is actually cheatgrass invasion, in which case
methods to remove cheatgrass should be considered for revisions to BLM policy.

74

This project has shown that it is possible that rest period fire events are a valid concern
for ranchers and land management agencies. Better data, both in the public records of grazing
closures and in more complex sensors for observation satellites, and more statistically thorough
methods will be needed to more confidently identify rest period fire events and to measure fire
danger. The candidate fire selection method was the direct result of the ideal datasets being
unavailable or unreliable. The selection method also had some unforeseen consequences, such as
only selecting fires which were smaller than the initial fires.
In the future, whether a stakeholder is a rancher seeking to profit from a ranching
operation or an environmentalist trying to protect Sage Grouse habitats, this study has
demonstrated that all parties in the grazing lands of Nevada and elsewhere in the American West
would benefit from better informed management practices developed from better data and
statistically strong analysis of that data.

75

References
Balch, Jennifer K., Bethany A. Bradley, Carla M. D'Antonio, and José Gómez-Dans. 2013.
“Introduced annual grass increases regional fire activity across the arid western USA
(1980-2009).” Global Change Biology 19, no. 1: 173-183.

Baugh, W. M. and D. P. Groeneveld. 2006. “Broadband vegetation index performance evaluated
for a low-cover environment.” International Journal of Remote Sensing 27, no. 21: 4715-
4730.

Box, Elgene O., Brent N. Holben, and Virginia Kalb. 1989. “Accuracy of the AVHRR
Vegetation Index as a Predictor of Biomass, Primary Productivity and Net CO₂ Flux.”
Vegetatio 80, no. 2: 71-89.

Bruce, L. B., Barry Perryman, Ken Conley, and Kent McAdoo. 2007. “Grazing Management on
Seeded and Unseeded Post-Fire Public Rangelands.” The Professional Animal Scientist
23, no. 3: 285-290.

Bureau of Land Management. 2007. “Burned Area Emergency Stabilization and Rehabilitation.”
BLM Handbook H-1742-1, Washington, D.C. Accessed June 18, 2018.
https://www.blm.gov/sites/blm.gov/files/uploads/Media_Library_BLM_Policy_Handboo
k_h1742-1.pdf.

____. 2017. “Public Land Statistics 2016.” Washington, D.C. Accessed June 18, 2018.
https://www.blm.gov/sites/blm.gov/files/PublicLandStatistics2016.pdf.

Carlson, Toby N., Robert R. Gillies, and Eileen M. Perry. 1994. “A method to make use of
thermal infrared temperature and NDVI measurements to infer surface soil water content
and fractional vegetation cover.” Remote Sensing Reviews 9, no. 1-2: 161-173.

Carlson, Toby N. and David A. Ripley. 1997. “On the relation between NDVI, fractional
vegetation cover, and leaf area index.” Remote Sensing of Environment 62, no. 3: 241-
252.

Daly, Christopher, Michael Halbleib, Joseph I. Smith, Wayne P. Gibson, Matthew K. Doggett,
George H. Taylor, Jan Curtis, and Phillip P. Pasteris. 2008. “Physiographically sensitive
mapping of climatological temperature and precipitation across the conterminous United
States.” International Journal of Climatology 28, no. 15: 2031-2064.

D'Antonio, Carla M. and Peter M. Vitousek. 1992. “Biological Invasions by Exotic Grasses, the
Grass Fire Cycle, and Global Change.” Annual Review of Ecology and Systematics 23,
no. 1: 63-87.

Davies, K. W., T. J. Svejcar, and J. D. Bates. 2009. “Interaction of historical and nonhistorical
disturbances maintains native plant communities.” Ecological Applications 19, no. 6:
1536-1545.

76

Diamond, Joel M., Christopher A. Call, and Nora Devoe. 2010. “Effects of targeted cattle
grazing on fire behavior of cheatgrass-dominated rangeland in the northern Great Basin,
USA.” International Journal of Wildland Fire 18, no. 8: 944-950.

Fry, Joyce, George Z. Xian, Suming Jin, Jon Dewitz, Collin G. Homer, Limin Yang, Christopher
A. Barnes, N. D. Herold, and J. D. Wickham. 2011. “Completion of the 2006 national
land cover database for the conterminous united states.” Photogrammetric Engineering
and Remote Sensing 77, no. 9: 7.

Guerschman, Juan Pablo, Michael J. Hill, Luigi J. Renzullo, Damian J. Barrett, Alan S. Marks,
and Elizabeth J. Botha. 2009. “Estimating fractional cover of photosynthetic vegetation,
non-photosynthetic vegetation and bare soil in the Australian tropical savanna region
upscaling the EO-1 Hyperion and MODIS sensors.” Remote Sensing of Environment 113,
no. 5: 928-945.

Halladay, Kerry. 2015. “Grazing resumes at Battle Mountain.” Western Livestock Journal.
Accessed Sept. 14, 2017. https://wlj.net/article-permalink-11680.html.

Homer, Collin, Jon Dewitz, Limin Yang, Suming Jin, Patrick Danielson, George Xian, John
Coulston, Nathaniel Herold, James Wickham, and Kevin Megown. 2015. “Completion of
the 2011 National Land Cover Database for the conterminous United States-representing
a decade of land cover change information.” Photogrammetric Engineering & Remote
Sensing 81, no. 5: 345-354.

Knapp, Paul A. 1996. “Cheatgrass (Bromus tectorum L) dominance in the Great Basin Desert:
history, persistence, and influences to human activities.” Global Environmental Change
6, no. 1: 37-52.

Martínez, Jesús, Cristina Vega-Garcia, and Emilio Chuvieco. 2009. “Human-caused wildfire risk
rating for prevention planning in Spain.” Journal of environmental management 90, no.
2: 1241-1252.

Mensing, Scott, Stephanie Livingston, and Pat Barker. 2006. “Long-term fire history in Great
Basin sagebrush reconstructed from macroscopic charcoal in spring sediments, Newark
Valley, Nevada.” Western North American Naturalist 66, no. 1: 64-77.

Miller, Jay D. and Andrea E. Thode. 2007. “Quantifying burn severity in a heterogeneous
landscape with a relative version of the delta Normalized Burn Ratio (dNBR).” Remote
Sensing of Environment 109, no. 1: 66-80.

Montandon, L. M. and E. E. Small. 2008. “The impact of soil reflectance on the quantification of
the green vegetation fraction from NDVI.” Remote Sensing of Environment 112, no. 4:
1835-1845.

77

Nagler, P. L., C. S. T. Daughtry, and S. N. Goward. 2000. “Plant Litter and Soil Reflectance.”
Remote Sensing of Environment 71, no. 2: 207-215.

Nevada Department of Agriculture. 2015. “An Economic Analysis of the Food and Agriculture
Sector: Nevada’s Counties 2015.” Carson City, NV.
Santin-Janin, H., M. Garel, J. -L. Chapuis, and D. Pontier. 2009. “Assessing the performance of
NDVI as a proxy for plant biomass using non-linear models: a case study on the
Kerguelen archipelago.” Polar Biology 32, no. 6: 861-871.

Scanlon, Todd M., John D. Albertson, Kelly K. Caylor, and Chris A. Williams. 2002.
“Determining land surface fractional cover from NDVI and rainfall time series for a
savanna ecosystem.” Remote Sensing of Environment 82, no. 2-3: 376-388.

USDA Natural Resources Conservation Service. 2015. “Plant Guide: Cheatgrass” The PLANTS
Database. Greensboro, NC. Accessed June 18, 2018.
http://plants.usda.gov/plantguide/pdf/pg_brte.pdf

U.S. Fish & Wildlife Service. 2015. “Greater Sage-Grouse: 2015 Not Warranted Finding Under
the Endangered Species Act.” Denver. Accessed June 18, 2018.
http://www.fws.gov/greaterSageGrouse/PDFs/GrSG_Finding_FINAL.pdf
Valla, Emily. 2015. “Ranchers: Grazing could have impacted size of Soda Fire.” KTVB.
Accessed June 18, 2018. https://www.ktvb.com/article/news/local/ranchers-grazing-
could-have-impacted-size-of-soda-fire/175295752.

Westerling, A. L., A. Gershunov, T. J. Brown, D. R. Cayan, and M. D. Dettinger. 2003. “Climate
and Wildfire in the Western United States.” Bulletin of the American Meteorological
Society 84, no. 5: 595-604.

Whisenant, Stephen G. 1990. “Changing Fire Frequencies on Idaho's Snake River Plains:
Ecological and Management Implications”. Ogden, UT: U.S. Department of Agriculture,
Forest Service, Intermountain Research Station.

Abstract (if available)

Abstract Bureau of Land Management policy regarding wildfire events on public rangelands dictates that burned areas are closed to livestock grazing until the vegetation in the burned area has reestablished itself. Ranchers and their supporters contend that extended duration of such grazing closures increases the likelihood of subsequent fire events during the grazing rest period. The ranchers attribute this effect to an over-accumulation of vegetation during the grazing rest period. With the goal of testing the claim made by ranchers, this project utilized fire history records, grazing allotment data, and remote sensing vegetation indices to identify and analyze potential rest period fires between 2000 and 2016 in and around the Nevada counties of Humboldt and Elko. GIS proximity tools were used to identify initial and subsequent fires on BLM grazing allotments which met the spatial and temporal requirements of a rest period fire. The four most likely candidates for rest period fires were selected for further examination as case studies. Scaled NDVI was used as an estimator of vegetation cover and change between selected initial and subsequent fires. Precipitation and land cover data were incorporated to provide further context. Three of the four fire perimeters showed increased vegetation cover when compared to similar nearby unburned sites during the second spring after the initial fires. This pattern suggests that increased fuel loads before the secondary fire may have been present. Evidence of cheatgrass and anthropogenic fire activity in the case study area suggest more complex explanations. Ways to improve monitoring and post-fire recovery through better record keeping, more complex sensors for satellite imagery, and targeted grazing research are discussed.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Use of remotely sensed imagery to map sudden oak death (Phytophthora ramorum) in the Santa Cruz Mountains

PDF

Effect of spatial patterns on sampling design performance in a vegetation map accuracy assessment

PDF

Increase in surface temperature and deep layer nitrate in the California Current: a spatiotemporal analysis of four-dimensional hydrographic data

PDF

Comparing Landsat7 ETM+ and NAIP imagery for precision agriculture application in small scale farming: a case study in the south eastern part of Pittsylvania County, VA

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

Building better species distribution models with machine learning: assessing the role of covariate scale and tuning in Maxent models

PDF

Using Landscape Integrity Index to evaluate the cumulative impacts of BLM resource management programs

PDF

Using Landsat and a Bayesian hard classifier to study forest change in the Salmon Creek Watershed area from 1972–2013

PDF

Exploring remote sensing and geographic information systems technologies to understand vegetation changes in response to land management practices at Finke Gorge National Park, Australia Between ...

PDF

Assessing the transferability of a species distribution model for predicting the distribution of invasive cogongrass in Alabama

PDF

Monitoring parks with inexpensive UAVs: cost benefits analysis for monitoring and maintaining parks facilities

PDF

Finding the green in greenspace: an examination of geospatial measures of greenspace for use in exposure studies

PDF

Modeling burn probability: a Maxent approach to estimating California's wildfire potential

PDF

Using geospatial technology to establish marsh bird monitoring sites for a pilot study in Maine in accordance with the North American Marsh Bird Monitoring Protocol

PDF

Precision agriculture and GIS: evaluating the use of yield maps combined with LiDAR data

PDF

Using census data, urban-land cover classification, and dasymetric mapping to measure urban growth of the lower Rio Grande Valley, Texas

PDF

A comparison of urban land cover change: a study of Pasadena and Inglewood, California, 1992‐2011

PDF

Is the likelihood of waterfowl presence greater on conserved lands?

PDF

Spatial analysis of human activities and wildfires in the Willamette National Forest

PDF

Using Maxent to model the distribution of prehistoric agricultural features in a portion of the Hōkūli‘a subdivision in Kona, Hawai‘i

Asset Metadata

Creator Kerbrat, Joel Andre, Jr. (author)

Core Title Questioning the cause of calamity: using remotely sensed data to assess successive fire events

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 07/12/2018

Defense Date 06/01/2018

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

Tag BLM,cheatgrass,Fire,grazing,landsat,NDVI,Nevada,OAI-PMH Harvest,Ranching,Vegetation

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen (committee chair), Lee, Su Jin (committee member), Longcore, Travis (committee member)

Creator Email joelakerbrat@gmail.com,kerbrat@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-16510

Unique identifier UC11668788

Identifier etd-KerbratJoe-6395.pdf (filename),usctheses-c89-16510 (legacy record id)

Legacy Identifier etd-KerbratJoe-6395.pdf

Dmrecord 16510

Document Type Thesis

Format application/pdf (imt)

Rights Kerbrat, Joel Andre, Jr.

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