Digital soil mapping of soil burn severity

Abstract

Fire alters soil hydrologic properties leading to increased risk of catastrophic debris flows and post‐fire flooding. As a result, US federal agencies map soil burn severity (SBS) via direct soil observation and adjustment of rasters of burned area reflectance. We developed a unique application of digital soil mapping (DSM) to map SBS in the Creek Fire which burned 154,000 ha in the Sierra Nevada. We utilized 169 ground‐based observations of SBS in combination with raster proxies of soil forming factors, pre‐fire fuel conditions, and fire effects to vegetation to build a digital soil mapping model of soil burn severity (DSMSBS) using a random forest algorithm and compared the DSMSBS map to the established SBS map. The DSMSBS model had a cross‐validation accuracy of 48%. The established technique had 46% agreement between field observations and pixels. However, since the established technique is manual, it could not be compared to the DSMSBS model via cross‐validation. We produced SBS class uncertainty maps, which showed high prediction probabilities around field observations, and low probabilities away from field observations. SBS prediction probabilities could aid post‐fire assessment teams with sample prioritization. We report 107 km² more area classified as high and moderate SBS compared to the established technique. We conclude that blending soil forming factors based mapping and vegetation burn severity mapping can improve SBS mapping. This represents a shift in SBS mapping away from validating remotely sensed reflectance imagery and toward a quantitative soil landscape model, which incorporates both fire and soils information to directly predict SBS.

Full text

Received: 22 December 2023 Accepted: 17 April 2024
DOI: 10.1002/saj2.20702
ORIGINAL ARTICLE
Special Section: North American Forest Soils Conference
Digital soil mapping of soil burn severity
Stewart G. Wilson1Samuel Prentice2
1Department of Natural Resources
Management and Environmental Sciences,
California Polytechnic State University, San
Luis Obispo, California, USA
2Sierra National Forest, United States Forest
Service, North Fork, California, USA
Correspondence
Stewart G. Wilson, Department of Natural
Resources Management and Environmental
Sciences, California Polytechnic State
University, San Luis Obispo, CA 93407,
USA.
Email: swilso49@calpoly.edu
Assigned to Associate Editor Dave Morris
Funding information
California State University Agricultural
Research Institute; USDA National Institute
of Food and Agriculture
(NIFA)-McIntire-Stennis Forest Capacity
Grant
Abstract
Fire alters soil hydrologic properties leading to increased risk of catastrophic debris
flows and post-fire flooding. As a result, US federal agencies map soil burn severity
(SBS) via direct soil observation and adjustment of rasters of burned area reflectance.
We developed a unique application of digital soil mapping (DSM) to map SBS in the
Creek Fire which burned 154,000 ha in the Sierra Nevada. We utilized 169 ground-
based observations of SBS in combination with raster proxies of soil forming factors,
pre-fire fuel conditions, and fire effects to vegetation to build a digital soil map-
ping model of soil burn severity (DSMSBS) using a random forest algorithm and
compared the DSMSBS map to the established SBS map. The DSMSBS model had
a cross-validation accuracy of 48%. The established technique had 46% agreement
between field observations and pixels. However, since the established technique is
manual, it could not be compared to the DSMSBS model via cross-validation. We
produced SBS class uncertainty maps, which showed high prediction probabilities
around field observations, and low probabilities away from field observations. SBS
prediction probabilities could aid post-fire assessment teams with sample prioritiza-
tion. We report 107 km2more area classified as high and moderate SBS compared
to the established technique. We conclude that blending soil forming factors based
mapping and vegetation burn severity mapping can improve SBS mapping. This rep-
resents a shift in SBS mapping away from validating remotely sensed reflectance
imagery and toward a quantitative soil landscape model, which incorporates both
fire and soils information to directly predict SBS.
1INTRODUCTION
Megafires (fires >10,000 ha) are increasing in size and sever-
ity and are amplifying hillslope denudation processes and
Abbreviations: BAER, burned area emergency response; BAI, burned area
index; BARC, Burned Area Reflectance Classification; dNBR, difference
normalized burn ratio; DSM, digital soil mapping; DSMSBS, digital soil
mapping of soil burn severity; EVI, enhanced vegetation index; FSSBS,
Forest Service Soil Burn Severity Mapping; GEE, Google Earth Engine;
NBRT, normalized burn ratio thermal; NDVI, normalized difference
vegetation index; NDWI, normalized difference water index; RF, random
forest; SBS, soil burn severity.
©2024 The Authors. Soil Science Society of America Journal ©2024 Soil Science Society of America.
potential soil damage in western US forests (Cole et al., 2020;
Fraser et al., 2020; Stephens et al., 2014; Stevens et al., 2017;
Westerling, 2016; Williams et al., 2023). These megafires
impact soil properties, including degradation of hydrologic
function and destruction of soil aggregates with cascading
effects to watershed function, and potentially catastrophic
effects to life and property via flooding and landslide haz-
ards (Certini, 2005; Kean et al., 2019; Rengers et al., 2020;
Stevens et al., 2017; Wagenbrenner et al., 2023). Furthermore,
catastrophic megafires and wildfire triggered landslides are
expected to increase with climate change (Cannon & DeGraff,
2009; Stevens et al., 2017; Westerling, 2016). The risk of
Soil Sci. Soc. Am. J. 2024;1–23. wileyonlinelibrary.com/journal/saj2 1

2WILSON and PRENTICE
catastrophic debris flow is directly related to the degree of
thermal damage to soils, often referred to as the soil burn
severity (SBS), which impacts soil infiltration rates and soil
structure, directly contributing to debris-flow risks and post-
fire flooding (Certini, 2005; Cheung & Giardino, 2023;Cole
et al., 2020; Kean et al., 2019; Rengers et al., 2020; Thomas
et al., 2023).
The impact to property, infrastructure, and water quality
from post-fire watershed responses tied to SBS is significant,
with individual losses from single catastrophic events upward
of $1 billion, and the cumulative annual effect of smaller
slides in the hundreds of millions of dollars (Dolan, 2018;
Fraser et al., 2022; Lukashov et al., 2019). In an extreme
example, just a few weeks after the Thomas fire was con-
tained, heavy rains triggered a catastrophic debris-flow in
Montecito, CA, that killed 23 people and closed a major US
freeway for weeks at an estimated cost exceeding $1 billion
(Dolan, 2018; Kean et al., 2019; Lukashov et al., 2019). In
events like this, SBS maps are used to identify potential flood-
ing and debris flow risks and to raise public safety awareness
(Kean et al., 2019). Aside from risks to life, risks to infrastruc-
ture and property are also significant, with the impacts from
multiple small fires leading to hundreds of millions of dollars
of losses related to post-fire flooding and landslides (Fraser
et al., 2022).
To manage these hazards, spatially explicit maps of
post-fire conditions are generated and provisioned to land
managers, first responders, hydropower authorities, and the
public. The principle among these products is the SBS map
produced by soil scientists within federal post-fire emergency
assessment teams (burned area emergency response [BAER])
and California Department of Forestry and Fire Protection’s
Watershed Emergency Response Team. SBS maps are a cor-
nerstone dataset for an interagency decision support system
tasked with identifying and, where possible, mitigating threats
to life, property, and water supplies (Parsons et al., 2010). Fur-
thermore, SBS maps feed the critical United States Geologic
Survey (USGS) post-fire debris-flow model (Staley et al.,
2016). Moreover, the SBS map informs decisions relating to
water quality, erosion, aquatic habitat, forest recovery, tim-
ber harvest, and other forest and water quality interventions
(Bladon et al., 2014; Cheung & Giardino, 2023; Cole et al.,
2020; Hohner et al., 2019; Morgan et al., 2015; Robichaud
et al., 2020, 2021; Smith et al., 2011; Uzun et al., 2020;
Wagenbrenner et al., 2023).
The federal process that birthed post-fire SBS map-
ping (BAER) was conceived independent of soil-landscape
concept models, and before geospatial methods became ubiq-
uitous (Bobbe et al., 2001). Concurrent with a major shift
toward integrating geospatial information in the early 2000s,
spatially explicit post-fire burn severity maps were generated
to test the deployment of rapidly available remotely sensed
products that supported an emerging community of diverse,
Core Ideas
∙A digital soil mapping method to map wildfire soil
burn severity (DSMSBS) was developed.
∙Soil field observations were combined with rasters
of environmental covariates and fire effects.
∙Excellent fidelity between field observations of
soil burn severity (SBS) and the final DSMSBS
map was reported.
∙Direct classification of SBS improves SBS map-
ping compared to validation of remotely sensed
burn severity.
∙Class probabilities generated for SBS may aid post-
fire assessment.
integrated post-fire first responders (Clark et al., 2003). The
primary of these products was a four-class (high, moderate,
low, and unburned) derivative ofa difference normalized burn
ratio (dNBR) image. This simple product, known as a Burned
Area Reflectance Classification (BARC) map, was blended
with a ground-based suite of soil thermal damage observa-
tions (Parsons, 2003). In concert, these form an operational
protocol for SBS mapping that is standard among soil scien-
tists supporting post-fire assessments but separate in evolution
from soil mapping conventions. As such, the SBS mapping
process as codified has untested remote sensing assumptions
and it was not equipped to keep pace with advances in digital
soil mapping (DSM) (Thompson et al., 2020).
Established SBS mapping utilizes soil field observations
to validate satellite imagery estimates of vegetation burn
severity and adjust classes on a classified vegetation burn
severity map, called a BARC map, in an attempt to cap-
ture soil effects (Parsons et al., 2010; Safford et al., 2008).
The SBS field assessment involves direct soil observation
of surface litter charring and consumption, ash color and
thickness, aggregate stability, hydrophobicity, and charring
and consumption of roots to develop the SBS site rating
(high, moderate, low, and unburned SBS) (Parsons et al.,
2010). Due to the physical disconnect between ground-based
SBS determinations and satellite-based vegetation burn sever-
ity, the classified vegetation burn severity map (e.g., the
BARC map) is manually adjusted in a geographic information
system (GIS) to account for ground observations (Fernández-
Guisuraga et al., 2023; García-Llamas et al., 2019; Saberi
et al., 2022; Soverel et al., 2010). This adjustment approach
does not maintain direct fidelity to ground-based observations
and spatial patterns. Alternatively, ground-based SBS obser-
vations can drive a spatial model of SBS based on correlation
with remotely sensed vegetation burn severity. Furthermore,
incorporation of other environmental predictor variables that
influence both fire behavior and the spatial distribution of
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 3
soil properties, such as topography, may improve SBS map-
ping. Improvements in SBS mapping have been identified as
a critical research need for more accurate post-fire debris flow
risk assessment and natural resource management (Cheung
& Giardino, 2023; Cole et al., 2020; Thomas et al., 2023;
Wagenbrenner et al., 2023).
DSM uses remotely sensed proxies for the soil form-
ing factors to map soil properties or types in unmapped
or under mapped locations (McBratney et al., 2003). DSM
derives quantitative relationships (e.g., models) between spa-
tially explicit data (i.e., rasters of vegetation indices and
terrain indices) and sparse point data (i.e., direct soil measure-
ments). This quantitative relationship, called “SCORPAN”
(an acronym for Soils data, Climate, Organisms, Relief, Parent
material, Age, and spatial locatioN), is then used to predict the
target soil attribute across the unsampled mapping area based
on the modeled relationship. The technique is widespread and
has been used to map diverse soil properties across a variety
of spatial extents ranging from fields to countries, continents,
and the globe (Chaney et al., 2016, 2019; Hengl & MacMillan,
2019; Hengl, Leenaars, et al., 2017; Hengl, Mendes de Jesus,
et al., 2017;Hengletal.,2021; Poggio et al., 2021; Ramcha-
ran et al., 2018). However, despite the widespread use of DSM
and the importance and impact of SBS maps, we report, to the
authors’ knowledge, the first study that integrates DSM into
SBS mapping (Levi & Bestelmeyer, 2018).
Soils are highly spatially variable, and burn severity is often
patchy, resulting in a mosaic of soil properties and fire effects
(Soverel et al., 2010; Stanley et al., 2023). While dNBR and
other remotely sensed attributes have successfully mapped
the vegetation centric composite burn index (CBI), a discon-
nect exists between remotely sensed surface reflectance-based
burn severity (e.g., dNBR) and SBS (Fallon et al., 2024;
Fernández-Guisuraga et al., 2023; García-Llamas et al., 2019;
Harvey et al., 2019; Saberi et al., 2022). In some instances,
vegetation burn severity may be higher than SBS, and vice
versa (Fallon et al., 2024; Fernández-Guisuraga et al., 2023).
This disconnect between remotely sensed burn severity and
SBS is due to the interaction of soil variability, fire behavior,
and the difficulty translating vegetation surface reflectance to
soil impacts (Harvey et al., 2019; Saberi & Harvey, 2023;
Saberi et al., 2022). Furthermore, antecedent soil properties
such as texture can determine the potential impact of fire to
soil (Woods & Balfour, 2010). Therefore, incorporating addi-
tional predictor variables (e.g., climate, vegetation, and terrain
attributes), which are known to influence both soil and fire
spatial variability, may improve SBS mapping (Behrens et al.,
2010; Dahlgren et al., 1997; Dillon et al., 2011). Since both
fire and soil are spatially heterogeneous, combining methods
which account for soil variability and burn severity variability
may improve assessment of the spatial distribution of SBS.
Here, we explore a digital soil mapping of soil burn severity
(DSMSBS), which combines soil-landscape predictive fac-
tors (SCORPAN), with raster proxies of fire effects to soil
and vegetation. Our objectives are to (1) implement a viable
DSMSBS approach for generating a post-fire SBS map, (2)
assess fidelity of the DSMSBS to ground-based observations,
and (3) compare and contrast the accuracy of DSMSBS and
established techniques to evaluate the use of DSMSBS as a
potential tool to aid post-fire soil assessment.
2MATERIALS AND METHODS
2.1 Site description
The Creek Fire (154,000 ha) began on September 4, 2020, in
the Big Creek subdrainage of the San Joaquin River canyon
on the Sierra National Forest, and was fully contained on
December 24, 2020. It burned in a highly mixed fire intensity
pattern, consuming over 70% of its area in the first 2 weeks
and 90% after 2 months (Stephens et al., 2022). The climate
in the subregion is Mediterranean-type with cool to cold wet
winters and warm to hot, dry summers that vary systemat-
ically with elevation from 300 to 3200 m (Dahlgren et al.,
1997). As such, the burn footprint captures extensive areas
of a modal climosequence supporting the six primary soil-
ecological zones of the Sierra Nevada west slope (Figure 1)
(D. Beaudette et al., 2023; D. E. Beaudette et al., 2013;
Dahlgren et al., 1997).
The pre-fire forest footprint mostly consisted of conifer
species but spanned thermic (California blue oak) to mesic
(ponderosa pine, white fir, and incense cedar) to cryic (lodge-
pole pine) forest climate types. This forest climosequence
is supported by an established archetype of granitic soil
development from low elevation/moisture-limited to high
elevation/temperature-limited settings, with a Goldilocks soil
development sweet spot at roughly the Holocene rain-snow
interface (∼1500 m) (Figure 1; Appendix A) (D. Beaudette
et al., 2023; D. E. Beaudette et al., 2013; Dahlgren et al.,
1997; Huntington, 1954). At this mid-elevation soil devel-
opment optimum, aboveground productivity is maximized
and strongly benefits from exceptionally thick, low-coarse
fragment epipedons overlying deeply weathered regolith that
provisions annual winter moisture through hot, dry summers
(Graham et al., 2010). Soils at lower elevations trend more
chemically weathered, while soils upslope of the optimum are
progressively coarser textured and clastic, grading to a discon-
tinuous soil mantle interrupted by glaciated bedrock outcrops
in the upper reaches of mapped soils (Giger & Schmitt, 1983).
In the published soil survey, soil organic horizon inputs
and thicknesses (i.e., combustible surface fuel) mirror for-
est productivity and litter decomposition processes across
the moisture- to temperature-limited elevational gradient
(Giger & Schmitt, 1983). Mapped O horizons are not
recorded on the most moisture-limited (Vista series) and most
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

4WILSON and PRENTICE
FIGURE 1 Generalized soil transect profiles of major soil types in the study area. Data from Dahlgren et al. (1997), illustrations from soilDB
R package (D. Beaudette et al., 2023;D.E.Beaudetteetal.,2013).
temperature-limited (Chiquito series) soil map units, but peak
at the forest soil productivity optimum (Musick/Shaver series
transition) (Figure 1). However, at local scale, O horizon and
litter layers are nearly ubiquitous across all soil types, with
thicknesses and surface continuity varying greatly depend-
ing on vegetation distribution (e.g., canopy cover) and legacy
management (e.g., plantations and fire exclusion). Pre-fire
surface fuels were also strongly modified by two recent
large fires (Aspen Fire 2011 and French Fire 2012), and an
exceptional range wide drought resulting in upward of 70%
mortality in the dry mixed conifer zone of the Sierra National
Forest (Restaino et al., 2019).
2.2 Soil sampling and observation
Ground-based sampling of soil thermal damage was led and
performed by soil scientists deployed on the USFS post-
fire assessment team (BAER) concurrent with late stages of
wildfire suppression. To guide ground-based SBS sampling
locations, the burn scar was conceptually stratified using a
classified vegetation burn severity index (BARC map) and
soil-landscape variability conventions (Hudson, 1992). Diag-
nostic indicators of loss of soil cover, aggregate collapse, root
combustion, and alteration were taken at 169 points across
the burn scar (Figure 2). Observations were made directly
at all locations to the depth of live root mortality caused by
wildfire heat pulse. Prolonged extreme heating that presents
as highly oxidized soil (roughly redder than 5YR) was noted
in some instances; however, the depth of color change never
exceeded root mortality depth. Samples were taken in tran-
sects or clusters that spanned multiple adjacent BARC map
pixels to adequately capture ground-based spatial variabil-
ity and uncertainty implicit in BARC map pixels. Unburned
locations were sampled outside the burn footprint to ver-
ify reference conditions. Each sample point was assigned
a qualitative SBS of underburned, low, moderate, and high
per federal post-fire SBS mapping protocols (Parsons et al.,
2010).
2.3 Soil burn severity modeling
2.3.1 BARC adjustment method (Forest
Service Soil Burn Severity Mapping [FSSBS])
To approximate SBS at unsampled locations across the burn
scar within a rapid assessment framework (i.e., BAER),
vegetation burn class values (BARC) were extracted at all
ground-based SBS point locations and these paired data were
checked for dis/agreement. Per standard SBS mapping proto-
cols, the BARC map color ramp was manually adjusted within
GIS software to approximate ground-based SBS data. The
burn scar was divided into three zones to better approximate
SBS patterns, with each zone adjusted separately. The merged
result was recast as the final SBS map (FSSBS).
2.3.2 Digital soil mapping of soil burn
severity
To prepare a DSMSBS, a raster stack of environmental predic-
tor variables accepted as proxies for the soil forming factors
was coupled with remotely sensed variables that account for
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 5
FIGURE 2 Post-fire image displaying surface reflectance of
three bands from Sentinel-2 (shortwave infrared [SWIR], near infrared
[NIR], and red) from September 28, 2020 (post-fire but before full
containment). Points are field observations of soil burn severity (SBS)
colored by SBS class. Inset map is California state outline.
fire effects to vegetation (e.g., dNBR) and fuels data. Terrain
attributes included topographic position index, topographic
ruggedness index, topographic wetness index, surface rough-
ness, slope, and aspect, derived from the National Elevation
Dataset 30-m digital elevation model and generated in the R
raster package (Hijmans, 2024). Additional terrain attributes
included various forms of surface curvature from algorithms
executed in ArcPro. Proxies for lithology came from the
USGS aero-radiometric dataset (Duval et al., 2005; Wilford,
2012). Given that soil variation in the region is significantly
impacted by the climate soil forming factor, we approximated
climate from 30-year normal mean rainfall and temperature
data from the PRISM climate group at 800 m resolution
(Dahlgren et al., 1997; Daly et al., 2008).
Normalized difference vegetation index (NDVI) and
enhanced vegetation index (EVI) were used as indicators
of general plant vigor, and the normalized difference water
index (NDWI) was used to represent apparent moisture in
vegetation (Gao, 1996). We extracted the Landsat 8 Collec-
tion 1 Tier 1 32-day composite images from Google Earth
Engine (GEE) for NDVI, EVI, and NDWI for months preced-
ing the fire (June, July, and August 2020). Remotely sensed
indices thought to approximate fire effects to soils and veg-
etation, namely the normalized burn ratio thermal (NBRT)
and the burned area index (BAI) were extracted from the
Landsat 8 Collection 1 Tier 1 32-day composite images in
GEE for the fire period (September, October, and Novem-
ber 2020). NBRT is a thermally enhanced spectral index for
mapping fire perimeters (Holden et al., 2005) and BAI a spec-
tral index designed to highlight the departure of a pixel from
the spectral signal of charcoal (Chuvieco et al., 2002). Pre-
fire and post-fire normalized burn ratio indices (e.g., dNBR)
were generated from Sentinel and Landsat 8 in GEE accord-
ing to standard practice, with the pre-fire image window
from July 1, 2020, to August 31, 2020, and post-fire image
window from October 31 to December 31 (Key & Benson,
2006). For Landsat 8 from GEE, we utilized the “LAND-
SAT/LC08/C01/T1_SR” image collection, and for Sentinel,
we utilized the “COPERNICUS/S2” image collection. The
least cloudy pixels were selected for the time frame, and a
cloud and cloud shadow mask were applied to the Land-
sat 8 and Sentienal-2 image collections, to obtain the least
cloudy image for the analysis window. An additional dNBR
image (Sentinel-2) of the Creek Fire was obtained from pub-
licly available incident data, with a window of pre-fire date
of September 3, 2020, and a post-fire image from Septem-
ber 28, 2020. Fuels and vegetation data were from Landfire
(https://landfire.gov/getdata.php). Datasets included canopy
fuel data (forest canopy cover, forest canopy base height, and
forest canopy bulk density), fuel vegetation data (fuel vege-
tation type and fuel vegetation height), and data from fuels
models (13 Anderson fire behavior fuel models, 40 Scott &
Burgan fire behavior fuel models, and fuel characteristics
classification system). Descriptions of Landfire variables can
be found at https://www.landfire.gov/data_overviews.php and
in Ryan and Opperman (2013). All raster data were resampled
to 30 m resolution to match the Landsat 30 m resolution with
the resample feature of the R raster package. Due to small
amounts of smoke or cloud obscured pixels in images taken
from GEE and the USFS incident data, the final predicted map
had some missing pixels in the fire area. Therefore, to accu-
rately compare the DSMSBS map with the USFS SBS map,
the USFS SBS was cropped by the DSMSBS map.
2.3.3 Model fitting, testing, validation, and
uncertainty
Ground-based SBS observations were used as training data
for a random forest (RF) model (DSMSBS) with SBS data
as the response variable and collocated raster data of terrain,
vegetation, climate, lithology, fire effects, and fuels data as
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

6WILSON and PRENTICE
predictor variables (Breiman, 2001; James et al., 2013). RF
uses many (500+) random decision trees based on boot-
strapped training samples (Grimm et al., 2008; James et al.,
2013). Each split of each tree assesses a random subset of the
predictors at the split (the random part of RF). All trees are
bagged into a “forest” to make a single prediction. In classifi-
cation, majority vote defines the correct classification of the
many trees in the forest (Malone et al., 2017). For SBS clas-
sification, the RF model was trained in r caret,with3×10
fold cross-validation, wherein the data are split into 10 ran-
dom groups of 90% training data and 10% testing data, and
these are iterated three times (Kuhn, 2008). The final model
was then fit in the randomForest package to all available SBS
observations, and a confusion matrix was generated from the
internal cross-validation in the package, with two-thirds of
the data as training data and one-third as test data (Hengl &
MacMillan, 2019).
We compared producer’s and user’s accuracy for each class,
and overall model accuracy from cross-validation. User’s
accuracy can be explained as, of all predictions of a par-
ticular class, how many ground-based observations matched
that predicted class (errors of commission). Similarly, pro-
ducer’s accuracy can be explained as how many ground-based
observations were correctly classified (errors of omission).
Overall accuracy is the number of field observations correctly
classified for the entire model set. We utilized the Cohen’s
κstatistic to indicate the reliability of the model compared
to chance agreement. A κthat is <0 indicates model results
are indistinguishable from chance classification, 0.01–0.20
indicates that a limited number of classifications cannot be
explained by random chance, 0.21–0.40 indicates a fair num-
ber of classifications that cannot be explained by random
chance, 0.61–0.80 as substantial number of classifications that
cannot be explained by random chance, and a κof 0.8–1.0
indicates near perfect model distinctness compared to random
chance classification (McHugh, 2012).
After validation, the model was fit to all SBS ground-based
observations, and predicted to the entire Creek Fire extent,
creating the final DSMSBS map. Since we could not compare
cross-validation scores (like accuracy) between the DSMSBS
map and the FSSBS map, we assessed how many ground-
based observations were sitting in pixels of the same class
between the two maps. From these final maps, we gener-
ated a pseudo-confusion matrix which compared agreement
between ground-based observations and the pixel classifica-
tions of SBS for each map (Parks et al., 2018; Soverel et al.,
2010). These are pseudo-confusion matrices because they are
not from cross-validation, but represent the relative fidelity of
two mapping techniques, and are not statistical comparisons.
These pseudo-confusion matrices were used for comparison
of user’s and producer’s accuracy between the DSMSBS map
and the FSSBS map. For reference, accuracy of the DSMSBS
model from cross-validation is referred to as model accuracy
in the text, and agreement between ground-based observa-
tions and the DSMSBS map or FSSBS map, is referred to as
map accuracy. However, it is important to note that we fit the
DSMSBS model to all available observations to generate the
DSMSBS map, as using all observations will generate the best
predictions across the entire map extent (>5 million pixels)
at the expense of overfitting at the 169 pixels that contained
ground-based observations.
To assess overall map reliability, we generated two uncer-
tainty estimates for the map extent. First, we utilized the class
probabilities generated by RF to generate prediction proba-
bilities maps for each SBS class (e.g., four maps, one each for
high, moderate, low, and unburned SBS). Second, we utilized
the maps of prediction probabilities to generate a Shannon’s
entropy (log base 2) map to highlight the relative purity of
the predictions at each pixel (Kempen et al., 2009). If the
predictions are more certain, there will be a high prediction
probability of one class and a low prediction probability of
the other three classes, which corresponds to a lower Shan-
non’s entropy (log base 2) value. To provide some interpreted
value for Shannon’s entropy, we generated different Shannon
entropy estimates based on what we subjectively considered
“good” versus “bad” combinations of class probabilities. For
example, with four classes, a Shannon’s entropy of 0.75 cor-
responds to a prediction probability of 80% for the first class,
10% for the second, and 2.5% for the third and fourth classes.
Therefore, Shannon’s entropy of 0.75 and below could be con-
sidered exceptional. Similarly, a Shannon’s entropy of 1.25
corresponds to a prediction probability in the first class of
70%, the second class probability of 20%, the third class
probability of 7.5%, and the fourth class of 2.5%, and so we
considered a range from 0.75 to 1.25 to be excellent. Follow-
ing this same framing, a 65% probability of the first class,
a 12% probability for the second and third classes, and a
11% probability of the fourth class have Shannon’s entropy of
∼1.49. So, we considered Shannon entropy values from 1.25
to 1.49 to be good. Shannon’s entropy of 1.84 corresponds
to a prediction probability of 45% for the first class, 25% for
the second, and 15% for the third and fourth classes. There-
fore, we arbitrarily considered Shannon’s entropy from 1.49
to 1.84 tolerable. Finally, Shannon’s entropy of 2 is an equal
probability of all four classes. We intersected our ground-
based SBS observations with the prediction probability maps,
and the Shannon’s entropy map to understand the integrated
relative certainty of the SBS field observations. Finally,
we took 30,000 random samples from the entire Shannon’s
entropy uncertainty map to provide an estimate of average
uncertainty for the entire Creek Fire extent. Unfortunately,
the FSSBS technique is strictly cartographic, and we can-
not compare uncertainty between the FSSBS and DSMSBS
maps.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 7
TABLE 1 Confusion matrix for random forest generated digital soil mapping of soil burn severity (DSMSBS) map (digital soil map of soil
burn severity by random forest).
Reference SBS
DSMSBS predictions
High Moderate Low Unburned Total PA (%)
High 29 14 7 0 50 58
Moderate 15 20 15 252 38
Low 2 15 21 8 46 46
Unburned 1 3 9 7 20 35
Total 47 52 52 17 168
UA (%) 62 38 40 41
Abbreviations: PA, producer accuracy; reference SBS, ground-based SBS observations; UA, user accuracy.
3 RESULTS
3.1 Model performance
Overall DSMSBS model performance from cross-validation
had an accuracy of 0.48 and a Cohen’s κof 0.25. We uti-
lized a confusion matrix to assess how well the DSMSBS
model performed at identifying SBS classes in held out obser-
vations (Table 1). The confusion matrix suggests that the
DSMSBS model is most sensitive to high SBS, with a pro-
ducer’s accuracy of 58%, meaning that 58% of high SBS
ground observations were correctly classified as high SBS by
the model in cross-validation. It was least sensitive tothe mod-
erate SBS class, where only 38% of moderate SBS ground
observations were classified accurately in cross-validation.
Users’ accuracy was similar to producer’s accuracy. Among
high DSMSBS predictions, 62% of ground observations were
in high SBS pixels, while only 38% of ground-based obser-
vations were sitting in pixels predicted as moderate SBS.
Errors in predicting moderate SBS ground observations were
perfectly split, with ∼30% of moderate field observations mis-
classified as high and ∼30% of moderate field observations
misclassified as low. Overall, cross-validation results suggest
that the DSMSBS model was better at predicting high SBS
field observation than at predicting low SBS field observa-
tions, and better at predicting low observations than moderate
SBS field observations.
Environmental covariates ranked in order of variable
importance from RF highlight dNBR as a dominant pre-
dictor in the DSMSBS model (Figure 3). This suggests a
relatively strong relationship between remotely sensed vegeta-
tion burn severity and SBS. Other important variables include
the BAI which approximates the spectral signature of char-
coal, and the Landsat 8 6/7 ratio which has been correlated
to free hydroxyls in exposed geologic materials (shortwave
infrared 1/shortwave infrared 2) (Boettinger et al., 2008).
These remotely sensed proxies (dNBR, BAI, and band 6/band
7 ratio) were coupled with two curvature terrain attributes as
the top predictors of SBS. Notably, remotely sensed variables
that approximate fire severity, BAI, NBRT, and dNBR, were
all in the top predictors. Of the top 20 predictors, typical DSM
covariates were confined to terrain attributes related to curva-
ture, vegetation data from Landfire (fuel vegetation type and
canopy height) and mean annual precipitation (Figure 3). Data
from fuels models were also included in the top 20 predictors,
including the Landfire 40 Scott & Burgan Fire Behavior Fuel
Models (Scott and Burgan Fuel Model), Landfire 13 Ander-
son Fire Behavior Fuel Models (Anderson Fuel Model), and
the Landfire Fuel Characteristics Classification System (Fuel
Characteristics). Overall, remotely sensed indicators of veg-
etation burn severity drove the SBS map predictions, while
data from Landfire fuels and vegetation models, as well as
curvature-based terrain attributes, also contributed to the SBS
predictions, but to a lesser extent than the remotely sensed
indicators of vegetation burn severity.
3.2 Comparisons between established
technique map (FSSBS map) and digital soil
mapping technique map (DSMSBS map)
The FSSBS and DSMSBS maps show the spatial distribution
of SBS based on BARC adjustment method and DSM method,
respectively (Figure 4). Ground observations of SBS are over-
laid onto both maps as a general assessment of dis/agreement
between field observed and modeled SBS. The FSSBS model
predicts extensive areas of moderate SBS, with large patchy
areas of high SBS intermingled with moderate SBS. The
edges of the fire scar are classified as low and unburned SBS
(Figure 4a). The DSMSBS map shows more homogeneous
patches of each severity class (Figure 4b). High DSMSBS
patches are larger than high FSSBS patches and appear to
occur adjacent to low and unburned SBS with more regu-
larity. Severity classes appear to generally be adjusted up a
class relative to the FSSBS map, meaning that large areas that
were classified as moderate in the FSSBS map were classified
as high in the DSMSBS map, areas of low in the USFSSBS
map were classified as moderate in the DSMSBS map, and
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

8WILSON and PRENTICE
FIGURE 3 Random forest variable importance. dNBR USFS,
difference normalized burn ratio from Landsat 8; dNBR Sentinel,
difference normalized burn ratio from Sentinel-2, Landsat 8 spectral
ratios (band 6/band 7); September burned area index (Sept. Burned
Area Index); November burned area index (Nov. Burned Area Index);
December burned area index (Dec. Burned Area Index); Gaussian
Curvature, Casorati Curvature, Landsat post-fire reflectance (USFS
Surface Reflectance); November normalized burn ratio thermal (Nov.
(Continues)
FIGURE 3 (Continued)
Normalized Burn Ratio Thermal); Landfire Classified Fuel Vegetation
Type (Fuel Vegetation Type); Landfire 40 Scott & Burgan Fire
Behavior Fuel Models (Scott and Burgan Fuel Model); Landfire 13
Anderson Fire Behavior Fuel Models (Anderson Fuel Model); Landfire
Fuel Characteristics Classification System (Fuel Characteristics); Land
Fire Forest Canopy Height (Canopy Height 200); PRISM 30-year mean
annual precipitation (Mean Annual Precipitation). Variables ranking is
based on relative increase in node purity (Gini Index).
so on. Additionally, high fidelity between field observations
of SBS and pixel classification is maintained in the DSMSBS
map, such that more field observations are sitting in patches
of pixels with the same classification in the DSMSBS map.
To visually demonstrate differences in fidelity between
SBS ground observations and pixel classifications of SBS
among both maps, and to investigate uncertainty in the
DSMSBS map, we focused the map area of interest on a
region with 12 SBS ground observations encompassing all
four SBS classes (Upper Chiquito Creek; Figure 5). In the
FSSBS map (Figure 5a), large patches of pixels are classified
as unburned (green) and moderate (yellow), with six of the
12 ground observations incorrectly classified. This disagree-
ment is highlighted by a transect of four high SBS ground
observations sitting in pixels classified as moderate SBS.
This contrasts with the DSMSBS map (Figure 5b) where the
map area is generally adjusted up a class compared to the
FSSBS map, with large areas of low SBS and high SBS as
opposed to unburned SBS and moderate SBS in the FSSBS
map. Importantly, in the DSMSBS map, there was excellent
agreement between ground observations of SBS and pixel
classifications of SBS, with all field observations of SBS
sitting in pixels with the same classification. However, this
can be misleading as RF generates final predictions based
on majority vote across all trees in the forest, even if predic-
tion probabilities for that class were relatively low. Ultimately,
agreement between ground-based observations and pixel clas-
sifications only provides information on the 169 pixels where
field observations were taken, whereas the map extent is >5
million pixels. In this respect, prediction probability maps
provide more information about uncertainty outside of the
pixels where ground-based observations were located, and
therefore, a better assessment of overall map utility (Figure 6).
For example, around the transect of four high SBS field obser-
vations, prediction probabilities are high in the pixels where
the ground-based SBS observations are located (>80%), while
the probabilities become lower as one moves away from the
ground-based observations of high SBS (Figure 6a). Over-
all, the large patch of high severity in the final DSMSBS
map (e.g., Figure 5b) generally has a probability greater than
40% in the class probability map (Figure 6a). However, as
one moves away from the high SBS ground-based observa-
tions, and toward the low SBS ground-based observations,
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 9
FIGURE 4 (a) US Forest Service soil burn severity (SBS) map, Creek Fire (2020). Map shows pixels classified in each SBS class using the
established, Burned Area Reflectance Classification (BARC) adjustment technique, and field observation in each soil burn severity class as recorded
by soil scientists on the ground immediately after fire. (b) Random forest soil burn severity map, Creek Fire (2020). Map shows pixels classified in
each soil burn severity class by random forest digital soil mapping technique and field observation in each soil burn severity class as recorded by soil
scientists on the ground immediately after fire.
FIGURE 5 (a) Comparison between field observations of soil burn severity (points) and USFS soil burn severity classifications in the Upper
Chiquito Creek area and (b) comparison between field observations of soil burn severity (points) and random forest generated (digital soil mapping
technique) soil burn severity classifications in the Upper Chiquito Creek area; Creek Fire (2020).
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

10 WILSON and PRENTICE
FIGURE 6 (a) High soil burn severity (SBS) class probabilities from random forest generated (digital soil mapping technique) soil burn
severity classifications, Upper Chiquito Creek area Creek Fire (2020), and SBS ground-based (soil) observations; (b) Moderate soil burn severity
(SBS) class probabilities from random forest generated (digital soil mapping technique) soil burn severity classifications, Upper Chiquito Creek area
Creek Fire (2020), and SBS ground-based (soil) observations; (c) Low soil burn severity (SBS) class probabilities from random forest generated
(digital soil mapping technique) soil burn severity classifications, Upper Chiquito Creek area Creek Fire (2020), and SBS ground-based (soil)
observations; and (d) unburned soil burn severity (SBS) class probabilities from random forest generated (digital soil mapping technique) soil burn
severity classifications, Upper Chiquito Creek area Creek Fire (2020), and SBS ground-based (soil) observations.
the probability of high SBS classification declines, and the
probability of low SBS classification increases (Figure 6c).
It is important to note that the transitional areas between
patches of high and low SBS in the DSMSBS map have low
certainty of any class (Figure 6a–d). The model is unable
to provide crisp boundaries between relatively homogeneous
predictions of each class, leading to diffuse prediction bound-
aries between higher confidence areas, with low confidence
of any class in these boundary areas and therefore a high
Shannon’s entropy (Shannon’s entropy >1.84; Appendix C).
The overall Shannon’s entropy from 30,000 random samples
was 1.71, suggesting tolerable average uncertainty across the
Creek Fire area (Appendix C). For class probability maps for
the entire Creek Fire area, see Appendix B. Since the FSSBS
method was a cartographic method, we have no uncertainty
estimates to compare the DSMSBS map predictions against.
We cannot cross-validate the FSSBS map, as it is man-
ual technique and does not have cross-validation accuracy
scores (e.g., κand percent accuracy). As such, we cannot
compare the FSSBS against DSMSBS via model diagnostics.
Therefore, to compare the relative accuracy of the FSSBS
map and the DSMSBS map, we used the full DSMSBS
model to predict SBS across the Creek Fire extent (e.g.,
we generated the DSMSBS map from the fit RF model)
and compared the mapped predictions to the ground-based
observations. Similarly, we compared FSSBS pixel classi-
fications to ground observed SBS. We used these data to
generate two pseudo-confusion matrices, one for each map-
ping technique (see Section 2). The pseudo-confusion matrix
for the FSSBS technique shows low producer’s accuracy for
high SBS field observations, with only 29% of high SBS
field observations classified as high (Table 2). Conversely,
the FSSBS method classified moderate SBS well, with 65%
of moderate SBS field observations classified as moderate.
The producer’s accuracy of FSSBS classification of low and
unburned was 38% and 53%, respectively. While acknowledg-
ing overfitting at ground-based observations used to build the
model, the DSMSBS map showed nearly 100% producer’s and
user’s accuracy on training data (Table 3).
To provide a clearer picture of the uncertainty at ground-
based observations, uncertainty estimates of predictions of
the training data were included in parentheses in Table 3.For
example, for the average class probabilities of all high SBS
field observations, 78% of trees in RF classified that obser-
vation as high, 13% classified it as moderate, 7% as low,
and 2% trees classified it as unburned, with majority vote
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 11
TABLE 2 Pseudo-confusion matrix for the USFS soil burn severity (SBS) map (FSSBS).
Reference SBS
FSSBS predictions
High Moderate Low Unburned Total PA
High 15 32 4 0 51 29%
Moderate 10 34 7 1 52 65%
Low 1 17 18 11 47 38%
Unburned 0 2 7 10 19 53%
Total 26 85 36 22 169
UA 58% 40% 50% 45%
Note: The table shows the agreement between the FSSBS map and the field observations of SBS. The mapped predictions run horizontally. Therefore, a total of 26 field
observations were in pixels with the classification of high SBS. The user’s accuracy (UA) is how many of those predicted high SBS predictions were true to the field
observations. For example, for high SBS, there were 26 field observations that were classified as high and 15 of those observations were actually high,sotheUAwas
58%. The producer’s accuracy (PA) is how many of the field observations were correctly classified. In this case, there were 51 field observations that were recorded as
high SBS. Of those 51 field observations of high SBS, only 15 were correctly classified as high SBS. Therefore, the PA was 29%.
Abbreviation: FSSBS, Forest Service soil burn severity mapping; reference SBS, ground-based SBS observations.
TABLE 3 Pseudo-confusion matrix for the digital soil mapping of soil burn severity (DSMSBS) map (soil burn severity map generated by
random forest fit to all the data using a digital soil mapping method).
DSMSBS predictions
Reference SBS High Moderate Low Unburned Total PA Shannon’s entropya
High 50 (0.78) 0(0.13) 0(0.07) 0(0.02) 50 100% 0.96
Moderate 0(0.12) 52 (0.77) 0(0.09) 0(0.03) 52 100% 1.02
Low 0 (0.07) 0(0.11) 46 (0.76) 1(0.07) 47 98% 1.02
Unburned 0(0.05) 0(0.07) 0(0.17) 18 (0.71) 18 100% 1.13
Total505246 19 167
UA 100% 100% 100% 95%
Note: The table shows the agreement between the DSMSBS map and the field observations of soil burn severity (SBS). Subscripts are the class probabilities based on the
proportion of classes in the forest ensemble, and report full model uncertainties for predictions based on model fit to all available data. For example, for moderate SBS of
the 500 trees, 0.77 of the trees predicted moderate SBS, while 0.12 predicted high SBS. The user’s accuracy (UA) is how many of those predicted high SBS predictions
were true to the field observations. For example, for high SBS, there were 50 field observations that were classified as high and 50 of those observations were actually
high, so the UA was 100%. The producer’s accuracy (PA) is how many of the field observations were correctly classified. In this case, there were 50 field observations
that were recorded as high SBS. Of those 50 field observations of high SBS, all were correctly classified as high SBS. Therefore, the PA was 100%. Shannon’s entropy
log base 2 is a measure of uncertainty. It was calculated from class prediction probabilities for each class from model fit to all available data.
Abbreviation: reference SBS, ground-based SBS observations.
aMean of Shannon’s entropy log base 2 for predictions fit to all available data.
resulting in 100% high SBS predictions at those field obser-
vations. For high SBS, the mean Shannon’s entropy (log base
2) of all high SBS field observations used to train the model
was 0.96, indicating little uncertainty around SBS predictions
at pixels that contained ground-based SBS training data. In
final predictions based on all available data, only one SBS
field observation was misclassified with one low SBS ground
observation classified as unburned. However, as previously
noted, uncertainty appeared to increase as distance from field
observations increased (Figure 6a–d).
The FSSBS had more moderate severity pixels compared
to the DSMSBS map, whereas the DSMSBS map had many
more high SBS pixels (Table 4). Overall, the DSMSBS map
predicted nearly double the area of high severity with the
DSMSBS map predicting about 132 km2more area of high
severity relative to the FSSBS map (48% more high SBS in
DSMSBS). Overall, the DSMSBS map had notably less area
classified as unburned, somewhat less area classified as low
and moderate SBS, and markedly more area classified as high
SBS. Overall, there were ∼91 km2fewer areas classified as
low and unburned SBS and ∼103 km2more areas classified
as high and moderate SBS in the DSMSBS map. This equates
to 13% more high and moderate SBS across the Creek Fire
scar using the DSMSBS method.
4 DISCUSSION
Here we report a spatially explicit SBS model (DSMSBS)
which combines remotely sensed indicators of vegetation
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

12 WILSON and PRENTICE
TABLE 4 Number of square kilometer in each soil burn severity class for two mapping methods, the Forest Service Soil Burn Severity
Mapping (FSSBS) map (Burned Area Reflectance Classification adjustment, final USFS map) and the digital soil mapping of soil burn severity
(DSMSBS) map (digital soil mapping using random forest) (Creek Fire, Sierra National Forest, 2020).
Mapping method Unburned (km2)Low(km
2) Moderate (km2)High(km
2)
FSSBS 348.8 407.2 614.1 145.1
DSMSBS 281.0 384.1 585.0 277.5
Abbreviations: BARC, Burned Area Reflectance Classification; FSSBS, Forest Service soil burn severity mapping; DSMSBS, digital soil mapping of soil burn severity.
burn severity (e.g., dNBR) with a DSM-based mapping
methodology and compares it against the conventional
approach that qualitatively approximates SBS from vegeta-
tion burn severity (FSSBS). The DSMSBS model had an
overall accuracy of 48% across all pixels (cross-validation).
In contrast, FSSBS prediction accuracy is unobtainable out-
side of ground sample points (i.e., no cross-validation). The
DSMSBS model predictions relied heavily on reflectance-
based burn severity indicators (dNBR, BAI, and NBRT),
terrain curvature, and data from fuels models (fuel vegetation
type, Scott and Burgan fuels model, Anderson fuel model, and
fuel characteristics).
The DSMSBS map had 100% producer’s and user’s accu-
racy at ground observations, which reflects the majority vote
of all trees in the RF for class membership of a pixel that con-
tained a training point. In contrast, FSSBS map accuracy at
ground-based observations was 45%. Perfect DSMSBS accu-
racy is overly optimistic at ground-based observations (e.g.,
overfit), as the actual map accuracy was 48%. High prediction
accuracy at ground-based observations reflects high predic-
tion probabilities at locations used to train the model, while
the uncertainty away from these ground-based observations
is a more informative estimate of map reliability. However,
since the objective was to generate the best map across the
5 million pixels, we traded overfitting at the 169 pixels with
ground sample locations, with using all ground-based obser-
vations to generate a more reliable map outside of where the
ground-based observations were located. This is reflected in
the overall uncertainty of the DSMSBS map, which is a mix
of high and moderate uncertainty (mean Shannon’s entropy
log base 2 =1.71; Appendix C). However, this is contrasted
against the FSSBS map, which has no accuracy estimates
and no uncertainty estimates, giving us no way to assess
confidence outside of where the ground-based observations
were located (45% accuracy for FSSBS and ∼100% accu-
racy for the DSMSBS). Since ground-based observations are
used to manually adjust the BARC map to generate the final
FSSBS map, we can assume some bias toward field obser-
vations in the FSSBS map. However, without accuracy and
uncertainty estimates, the reliability of the FSSBS map out-
side of where the ground-based observations are located is
unknown.
4.1 Remote sensing of fire effects on soil
Historically, CBI explored correlations between ground-
based data and remote sensing to spatially predict fire effects.
CBI is an index of fire impacts to surface soil, litter, and
understory and overstory vegetation measurements taken at
30-m plot scale to align with Landsat sensor resolution. Map-
ping CBI has typically relied on correlations with dNBR
(Fernández-Guisuraga et al., 2023; García-Llamas et al.,
2019; Kasischke et al., 2008; Key & Benson, 2006;Parks
et al., 2018; Soverel et al., 2010). Results from CBI map-
ping are not directly transferable to soils impacts, as CBI
is biased toward vegetation, and classification accuracies of
thermal impacts to soil are very often hampered by the dis-
connect between vegetation severity and substrate or soil
impacts (Fernandez-Garcia et al., 2018; Fernández-Guisuraga
et al., 2023; Morresi et al., 2022; Sobrino et al., 2019). For
example, poor classification of moderate burn severity has
been attributed to mixed fire effects between forest under-
story and overstory (Morresi et al., 2022; Parks et al., 2018;
Soverel et al., 2010). A direct relationship between surface
burn severities and remote sensing indices such as dNBR
is inconsistent, especially in Mediterranean type ecosystems
(Fernández-Guisuraga et al., 2023).
There are few investigations that have built predictive maps
of SBS using correlations between ground-based measure-
ments and remotely sensed fire effects data to provide context
to our results. For example, in a small scale unmanned air-
craft system study, classification accuracy in cross-validation
was reportedly high using NDWI as a singular predictor (over-
all accuracy =83%, κ=0.74) (Beltrán-Marcos et al., 2021).
While these results appear promising, they were built on high
ground density ground data (one sample per hectare) which is
unrealistic for large areas or any rapid hazard assessment sce-
nario. By comparison, Creek Fire field SBS data density was
roughly one observation per 900 ha. Other forays into mod-
eling soil thermal impacts from remotely sensed data were
not cross-validated, and so their producer’s and user’s accu-
racy are more comparable to our DSMSBS map fit to all the
training data. For example, a study on a moderately sized
fire in Spain reported good producer’s accuracy, but results
were not comparable to the Creek Fire due to a smaller area
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 13
(15,000 ha), higher sample density, and lack of any cross-
validation accuracy assessment (Sobrino et al., 2019). Other
studies have reported that remotely sensed indices poorly
describe SBS variance (Hudak et al., 2007; Morgan et al.,
2014; Robichaud et al., 2007). For example, Hudak et al.
(2007) concluded that the dNBR maps better reflected post-
fire vegetation conditions than soil conditions, with a low
correlation between the water drop test (r<0.1), exposed sur-
face soil (r<0.35) and dNBR. We suggest that the DSMSBS
method is an improvement compared to these few other exam-
ples in several ways. First, we mapped a large area (∼154,000
ha) with a very low point density. Second, DSMSBS com-
bines multiple remotely sensed indices and terrain attributes,
not just dNBR, potentially capturing some previously unas-
signed variance in which studies that only utilized dNBR or
other spectral indices may have missed. Finally, DSMSBS
includes assessments of uncertainty, improving map utility for
users and practitioners, by providing some indication of map
reliability in areas that lacked ground-based observations.
4.2 The disconnect between remotely
sensed burn severity and soil burn severity
In terms of ecosystem, hydrologic, and soil processes, the
distinction between fire effects on soil and fire effects on veg-
etation is well known and comprise the justification for a risk
assessment process that relies on soil conditions and measure-
ments (Safford et al., 2008). The BAER operational model
(FSSBS) implicitly acknowledges that loss of canopy cover,
forest litter combustion, and compromised soil hydrologic
properties set the boundary conditions for hillslope erosion
and amplified watershed response. Less recognized is the
disconnect between fire effects on vegetation canopy as mea-
sured by spectral indices (i.e., dNBR) compared to fire effects
on near-soil surface fuels (Busse et al., 2014).Thelackof
a spatial model driven by soils information and strong pixel
level covariates becomes problematic, as does the absence of
SBS class accuracy and uncertainty estimates in the final map
(Morales-Barquero et al., 2019). For example, the accepted
SBS mapping schema uses a classified thematic map of veg-
etation burn severity (BARC map) to guide stratified field
sampling of SBS. BARC map classes are initially grouped into
four bins—unburned, low, moderate, and high—using custom
breaks of normalized intensity values (0–256) in GIS. The
BARC map is subsequently converted into a SBS map by iter-
atively adjusting vegetation burn severity class break points
until the classes visually align with field SBS data points
and the map approximates the producer’s expert judgement
of ground-based SBS patterns. Notably, break point adjust-
ments are domain wide (i.e., they affect all pixels regardless of
their relative BARC severity) and one directional (less or more
severe). Typically, BARC burn classes are down-adjusted
across the entire SBS feature space reflecting a generalized
assumption of lower severity soil thermal damage compared
to the canopy overstory (Safford et al., 2008). Accordingly,
high BARC is often reclassed to moderate SBS in the final
FSSBS map, and so on for each class (Parsons et al., 2010;
Safford et al., 2008). In this approach, the influence of ground-
based soil conditions and measurements is diminished, and
the assigned SBS classes often may contradict direct SBS
observations in the final SBS map.
The disconnect between remotely sensed vegetation
burn severity and ground-based SBS can be significant
(Fernández-Guisuraga et al., 2023). The causes of this dis-
connect are case dependent but may be due to disease and
infestation effects, differences between crown fires and sur-
face fires in landscapes with mature fire-resistant trees, or a
general disconnect between remote sensing of fire effects to
canopy vegetation relative to fire effects to the soil or soil
surface (Chen et al., 2015; Fernández-Guisuraga et al., 2023;
Saberi & Harvey, 2023; Saberi et al., 2022). In forests infected
by pests or disease, a disconnect exists between canopy burn
severity and substrate burn severity, and this disconnect is
greatest in areas with more infected (dead and dying) trees or
longer time since infestation (Chen et al., 2015; Hicke et al.,
2012; Morresi et al., 2022). For example, in the Big Sur area of
California, there was little to no reported correlation between
surface reflectance and substrate burn severity in diseased
forests (R2from 0.04 to 0.15), but a very strong relationship in
healthy forests (R2from 0.52 to 0.77) (Chen et al., 2015). Sim-
ilarly, it has been suggested that pine bark beetle infestations
in forests may lead to higher substrate burn severities relative
to the apparent vegetation damage due to higher surface fuel
loads (Hicke et al., 2012). In the area of the Creek Fire (Sierra
Nevada of California), a striking 49% of trees have died from
a combination of drought and beetle infestation, which led
to increased fuels from dead trees and over thickened sur-
face fuels from needle cast and downed wood, which may
have increased the disconnect between vegetation burn sever-
ity and SBS (Fettig et al., 2019, 2021). Moreover, Saberi et al.
(2022) note that vegetation burn severity may be decoupled
from surface burn severity in areas like the Creek Fire with
large fire-adapted trees. In these areas, substrate effects may
be greater than vegetation effects, especially in sites with over-
thickened O horizons, contributing to the disconnect between
remotely sensed vegetation burn severity indicators and SBS.
This disconnect necessitates building models from ground-
based SBS observations, as opposed to relying on remotely
sensed vegetation burn severity to approximate soil impacts.
In this respect, the DSMSBS model may better account for
the disconnect between vegetation burn severity and SBS,
since the SBS ground-based observations drive the spatial
model from the “bottom up,” as opposed to using the ground-
based observations to validate a remotely sensed indicator of
vegetation burn severity.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

14 WILSON and PRENTICE
Comparing the nature of the producer’s and user’s
errors between the cross-validated DSMSBS model and the
FSSBS map can illuminate the tendency of each mapping
system to shift the final SBS map up or down relative to
ground-based observations. Generally, the FSSBS map was
adjusted down to a greater degree than the cross-validated
DSMSBS model. For example, in the FSSBS map, 61% of
high SBS observations were misclassified lower, whereas in
the cross-validated DSMSBS model, 42% of high observa-
tions were adjusted down. For moderate observations, the
cross-validated DSMSBS model had a low producer’s accu-
racy (38%), with half of the classification errors adjusted up,
and the other half adjusted down, while for the FSSBS map,
10 errors were due to an adjustment up and eight were due
to an adjustment down. The FSSBS map favored moderate
SBS over all classes, with 85 of the 169 field observations
classified as moderate in the FSSBS map. This leads to a gen-
eral bias toward moderate SBS classes in the FSSBS map,
despite the relatively even distribution of SBS classes in
ground-based observation.
4.3 Digital soil mapping of soil burn
severity: A novel tool to map soil burn severity
The DSMSBS map requires no up or down adjustment of
classes, as it is not a validation of a remotely sensed vege-
tation burn severity, but instead a spatial model trained on
ground-based observations of SBS which then directly pre-
dicts SBS. This represents a valuable improvement in SBS
mapping, away from validation of remote sensing, and toward
a quantitative soil landscape model based on remotely sensed
proxies for soil variability and fire impacts. The DSMSBS
model adjusts classification based on multiple environmen-
tal variables and therefore adjusts the final SBS map pixel
by pixel, without the challenges associated with the BARC
manual adjustment technique (FSSBS), wherein a single
adjustment of the BARC classes affects all pixels in the extent.
In contrast, the DSMSBS technique allows for local scale
machine adjustment of pixel level SBS, improving flexibil-
ity and reproducibility. Furthermore, providing uncertainty
estimates, not available in the conventional method, informs
map confidence across the extent. This can assist post-fire
hazard mapping teams by identifying areas with high uncer-
tainty to target limited ground resources. For example, areas
of high uncertainty around high-risk watersheds or values at
risk would be potential targets for additional sampling.
A good model of SBS should account for both the variabil-
ity inherent to soils and the variability attributable to the fire
associated disturbance. The SBS ground-based observation is
meant to capture the change to soil physical and morpholog-
ical characteristics following thermal damage and the SBS
map is meant to capture the spatial extent of the physical and
morphological change to soils caused by fire. As expected,
rasters of fire effects to vegetation (e.g., dNBR) had the high-
est variable importance in the DSMSBS model. While dNBR
does not always correlate with SBS, it is still the most impor-
tant variable for predicting thermal impacts to soils in this
study, corroborating other reports (Hudak et al., 2007;Morgan
et al., 2014; Morresi et al., 2022; Parks et al., 2018; Robichaud
et al., 2007; Soverel et al., 2010). Anecdotally, we expected
terrain to figure more prominently in the DSMSBS model,
given the known relationship between terrain and soil proper-
ties in the region, as well as established correlations between
terrain and vegetation burn severity (Dahlgren et al., 1997;
Dillon et al., 2011; Holden et al., 2009; Viedma et al., 2015).
Topography influences the distribution of vegetation, as well
as fuels and fire behavior, impacting burn severity (Dillon
et al., 2011; Holden et al., 2009). However, terrain attributes
were secondary to rasters of fire effects in mapping SBS,
contrasting with reports of terrain as an important driver of
CBI or dNBR-based severity indicators (Dillon et al., 2011;
Holden et al., 2009). However, studies that cited topography
as a driver of severity were attempting to predict potential
severity (e.g., potential dNBR), and did not include dNBR
as a predictor variable. Nonetheless, terrain (curvature) was
the most important variable not directly associated with fire
effects, in agreement with our understanding of both soil and
fire variability (Behrens et al., 2010; Dillon et al., 2011).
Vegetation is a key driver of burn severity (Odion et al.,
2004), and remotely sensed indicators of vegetation have
been important predictors in DSM (Boettinger et al., 2008).
However, remotely sensed indicators of vegetation (e.g.,
NDVI, NDWI, and EVI) were not important predictors in the
DSMSBS model. Speculatively, the Landsat derived Landfire
variables such as fuel vegetation type and canopy height, may
have explained the variance that could have been attributable
to the Landsat spectral ratios used in this study (e.g., NDVI).
Additionally, the Creek Fire spanned many different vege-
tation types and a very large area, which may have led to
excessive variability in the spectral signature of pre-fire veg-
etation conditions, reducing the predictive power of pre-fire
vegetation reflectance. In this respect, spectral indicators that
calculate change from pre-fire to post-fire conditions, such as
dNBR, may better normalize the effect of fire on the diverse
vegetation types in the study, versus spectral ratios of pre-fire
conditions alone. Therefore, a vegetation greenness differ-
ence metric (i.e., pre-fire NDVI minus post-fire NDVI) may
better approximate fire impacts to vegetation and improve
SBS predictions, as has been reported for vegetation burn
severity (Chen et al., 2015; Fernandez-Garcia et al., 2018;
García-Llamas et al., 2019).
Central to any fire effects model performance is an accurate
accounting of pre-fire fuels (i.e., all combustible vegetation)
that are differentially consumed and result in variable fire
effects. Although canopy fuel combustion may contribute
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 15
via radiative thermal damage and ground fire spotting, fire
effects on soils are most strongly driven by combustion of
O horizon and forest detritus at or near the mineral soil
surface (Busse et al., 2014). In this mechanistic soil process,
the volume, size, and horizontal arrangement of combustible
materials are highly variable in composition (Van Wagten-
donk et al., 1998) and spatial range (Loudermilk et al., 2012)
with highest spatial variation in the fine litter components
(Keane, 2016; Vakili et al., 2016). Along the vertical dimen-
sion, soil fuel bed thickness controls moisture sorption and
retention by strongly mediating O horizon porosity (i.e., aera-
tion) which in turn attenuates SBS (Busse et al., 2005;Kreye
et al., 2014). Birch et al. (2023) have further shown that O
horizon mass (duff +litter) and downed coarse wood are
the primary fuels driving wildfire energy and fire burn sever-
ity. Accordingly, three-dimensional pre-fire soil surface fuel
bed volume and architecture present as a critical variable for
explaining wildfire SBS and SBS patterns. This conclusion is
consistent with the Creek Fire study area, where exceptionally
high pre-fire loadings of dead and cured forest fuels caused
unusually intense internal heating following initial convection
(Lee et al., 2023; Stephens et al., 2022). Such energy release
patterns of protracted fuel bed consumption and residence
times coincide with the most extensive high SBS patches in
both FSSBS and DSMSBS maps (Figure 3).
While an overthickened surface fuel bed caused by pre-fire
drought, disease, or fire exclusion may drive energy release
and SBS patterns, our results demonstrate that a 30-m fuel bed
does not summon the predictive power inherent in mechanistic
fire and SBS processes. In the DSMSBS model, pre-fire fuel
type and distribution are represented by Landfire fuels data.
These data figured prominently but were not a primary pre-
dictor of SBS (Figure 3). We suggest that improved remotely
sensed covariates representing soil O horizon thickness and
surface fuel bed components would likely yield relatively
substantial improvements in our ability to model SBS. Such
covariates may be in the form of finer spatial resolution (e.g.,
low altitude unmade aerial aircraft), active sensor technolo-
gies such as radar (Saatchi et al., 2007), or rasterized fire
spread rate maps as proxies for fire residence time.
Our study represents a unique application of DSM to soil
disturbance mapping. No other studies, to the authors’ knowl-
edge, have incorporated fire effects rasters into a soil forming
factors-based model (e.g., SCORPAN +Fire), to predict the
impact of fire to soils. In this unique application of DSM,
the goal of the mapping is not to map a specific soil prop-
erty per se, but instead to map the magnitude of a disturbance
to the soil. Therefore, it is reasonable that the proxy for dis-
turbance in this study (e.g., dNBR) was the best predictor
of the impact of that disturbance to soils, as opposed to the
SCOPRAN factors which are meant to explain the distribu-
tion of other, non-fire affected, soil properties. We hope that
this study will provide a framework for additional DSM-based
investigations into fire effects to soils, both for prediction of
fire effects to specific soil properties and for inference into
potential landscape scale drivers of these effects.
4.4 Limitations and improvements
Improvements in model fitting and validation, and considera-
tion of additional variables that describe fire behavior and fire
conditions, could improve the DSMSBS model. For exam-
ple, to compare the DSMSBS map against the FSSBS map,
we fit the final DSMSBS model to all available data, lead-
ing to overfitting around ground-based observations. A larger
training dataset would allow for more latitude in leaving field
observations out for validation. For example, to better esti-
mate the reliability and utility of the DSMSBS approach,
models could be built on many available fires, with the model
accuracy evaluated on fires the model has not seen. This
may also improve accuracy, as many more SBS field obser-
vations could be included across many fires with different
severities, increasing sample number, and presumable reli-
ability. This would give practitioners a better sense of the
reliability of the model and accuracy of predictions. More-
over, with a larger DSMSBS model, additional model types,
such as deep learning, which require larger numbers of train-
ing data, could be tested. Finally, additional raster layers
that capture fire radiative heat, fire residence time, or better
estimates of vegetation condition and disease impacts could
also improve predictions. For example, tree mortality from
drought and bark beetles impacts vegetation burn severity,
and the Sierra Nevada had been in a multi-year drought with
millions of dead trees leading up to the Creek Fire (Fet-
tig et al., 2019, 2021; Littell et al., 2016). Longer droughts
lead to more fuels with lower moisture, more fuels in con-
tact with the soil, and increased vegetation burn severity when
ignited (Huang et al., 2020; Keeley et al., 2022). Covariates
related to drought impacts could improve prediction of SBS,
as has been reported for vegetation burn severity (Huang et al.,
2020; Keeley et al., 2022). We attempted to capture some
drought impact to vegetation by utilizing pre-fire NDVI and
NDWI, but a different NDVI (i.e., dNDVI) from apre-drought
period to a post-drought period would be a better metric of
drought impact, as opposed to the general vegetation health
captured with a single pre-fire NDVI/EVI/EVI image (Keeley
et al., 2022).
4.5 Management implications and utility
We report 103 km2greater moderate and high SBS in our
DSMSBS map compared to the FSSBS map. Changes to the
area classified in high or moderate SBS have widespread
impacts to hazard assessment and forest management. For
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

16 WILSON and PRENTICE
example, increases in the area classified as high SBS will
increase the risk of debris flows (Cheung & Giardino,
2023). As a primary input to debris flow hazard models,
improved classification accuracy and area calculations of SBS
are essential to risk assessment and emergency response to
the risk of post-fire debris flows and flooding (Cheung &
Giardino, 2023; Thomas et al., 2023). With respect to manage-
ment, increased SBS may lead to increased runoff and erosion
from post-fire salvage logging, with mitigation and best man-
agement practices to mitigate impacts dependent on accurate
assessment of SBS (Wagenbrenner et al., 2023). We report
91% more area burned at high severity in the DSMSBS map
compared to the FSSBS map. Changes to stream sediment
loads, risks to drinking water supplies, impacts to aquatic
habitat, as well as changes to soil C and nutrient biogeo-
chemistry, are related to the degree of SBS, with the impact
increasing as the area burned at high severity increases (Bixby
et al., 2015; Bladon et al., 2014; Dove et al., 2020; Fernández
& Vega, 2016; McCool et al., 2023; Uzun et al., 2020; Vieira
et al., 2015).
A DSMSBS model trained on many fires could be used
to predict SBS on fires as soon as post-fire imagery is avail-
able and before boots are on the ground, improving post-fire
hazard assessment. This larger DSMSBS model could then
be updated once ground-based data becomes available for
that fire, improving fire-specific accuracy. A key improve-
ment of the DSMSBS map is the generation of prediction
uncertainty. If uncertainty maps could be generated from a
larger model before a site visit, they could guide sampling, so
that areas of high uncertainty could be targeted for sampling,
as additional samples in areas of high uncertainty appear to
improve predictions in that area (Figure 5a–c). For example,
post-fire assessment teams could prioritize sampling in areas
of high uncertainty and high risk, such as in the wildland-
urban interface, over sampling in areas with high risk but low
uncertainty, maximizing limited ground resources. However,
more research is required to test the accuracy of larger models
trained on multiple fires, as well as the utility of SBS uncer-
tainty estimates to guide post-fire SBS sampling and rapid
hazard assessment.
5CONCLUSION
We report, for the first time to the authors’ knowledge, a
DSM-based SBS mapping technique, which we utilized to
classify and map SBS in the Creek megafire which burned
more than 152,000 ha in the Sierra Nevada of California
in 2020. The DSMSBS model utilized ground-based obser-
vations of SBS combined with raster-based proxies for soil
variability (e.g., terrain attributes, climate, and NDVI), rasters
of fire impacts (dNBR), and the RF algorithm to classify
and map the degree of thermal damage to soils. In cross-
validation, the DSMSBS model had an overall accuracy of
48% and good user’s and producer’s accuracy for classifica-
tion of high SBS (62% and 58%, respectively). Additionally,
when the DSMSBS model was fit to all available training
data and used to generate the DSMSBS map, there was nearly
100% accuracy between SBS pixels and ground-based obser-
vations of SBS, far exceeding the accuracy (46%) of the
FSSBS map. However, this is overly optimistic as we sacri-
ficed overfitting at the 169 pixels that contained ground-based
observations in favor of including all training data to better
predict the >5 million pixels that did not have training data
associated with them. We used DSMSBS to generate map
uncertainty estimates, a key improvement over the established
SBS mapping technique, potentially allowing practitioners to
target additional sampling in high-risk watersheds that have
high uncertainty. Furthermore, the DSMSBS map had 13%
more burned area classified as high and moderate SBS and
91% more area classified as high SBS compared to the FSSBS
approach. Changes to the area classified as high and moderate
SBS impact decisions around nearly every post-fire interven-
tion, including risks to life and property, impacts to forest
management, impacts to water quality, and impacts to soil
biogeochemical cycles. Moreover, the DSMSBS methodol-
ogy allows for machine-level local adjustment of SBS based
on proxies for both soil forming factors and fire effects to
vegetation, improving fidelity between SBS maps and SBS
ground-based observations. This may address potential dis-
connects between remotely sensed vegetation burn severity
and SBS.
We suggest that the DSMSBS methodology represents a
significant improvement in SBS mapping, away from valida-
tion of remote sensing, and toward a quantitative landscape
model of SBS, with ground-based observations of SBS driv-
ing the direct prediction of SBS. This is an encouraging
advancement in SBS mapping, which may improve the accu-
racy and reproducibility of SBS maps in the United States and
beyond, changing the perception of the spatial extent of fires
effects to soils and natural resources. Finally, the DSMSBS
model is the first to test the utility of incorporating rasters
of soil forming factors with rasters of fire effects to predict
fire impacts to soils, a promising expansion in the applica-
tion of DSM for improving our understanding of fire and soil
interactions.
AUTHOR CONTRIBUTIONS
Stewart G. Wilson: Conceptualization; data curation; for-
mal analysis; funding acquisition; investigation; methodol-
ogy; project administration; resources; software; supervision;
validation; visualization; writing—original draft; writing—
review and editing. Samuel Prentice: Conceptualization;
investigation; writing—review and editing.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 17
ACKNOWLEDGMENTS
We would like to acknowledge the California State Univer-
sity Agricultural Research Institute and the USDA NIFA
McIntire-Stennis Forest Capacity grant for funding.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
ORCID
StewartG.Wilson https://orcid.org/0000-0003-0280-0290
REFERENCES
Beaudette, D., Roudier, P., Beaudette, M. D., & RcolorBrewer, M.
(2023). Package ‘aqp’.https://cran.r-project.org/web/packages/aqp/
index.html
Beaudette, D. E., Roudier, P., & O’Geen, A. T. (2013). Algorithms
for quantitative pedology: A toolkit for soil scientists. Computers &
Geosciences,52, 258–268.
Behrens, T., Zhu, A. X., Schmidt, K., & Scholten, T. (2010). Multi-scale
digital terrain analysis and feature selection for digital soil mapping.
Geoderma,155(3–4), 175–185. https://doi.org/10.1016/j.geoderma.
2009.07.010
Beltrán-Marcos, D., Suárez-Seoane, S., Fernández-Guisuraga, J. M.,
Fernández-García, V., Pinto, R., García-Llamas, P., & Calvo, L.
(2021). Mapping soil burn severity at very high spatial resolution
from unmanned aerial vehicles. Forest s,12(2), 179. https://doi.org/
10.3390/f12020179
Birch, J. D., Dickinson, M. B., Reiner, A., Knapp, E. E., Dailey, S.
N., Ewell, C., Lutz, J. A., & Miesel, J. R. (2023). Heading and
backing fire behaviours mediate the influence of fuels on wildfire
energy. International Journal of Wildland Fire,32(8), 1244–1261.
https://doi.org/10.1071/WF22010
Bixby, R. J., Cooper, S. D., Gresswell, R. E., Brown, L. E., Dahm, C.
N., & Dwire, K. A. (2015). Fire effects on aquatic ecosystems: An
assessment of the current state of the science. Freshwater Science,
34(4), 1340–1350. https://doi.org/10.1086/684073
Bladon, K. D., Emelko, M. B., Silins, U., & Stone, M. (2014). Wildfire
and the future of water supply. ACS Publications.
Bobbe, T., Lachowski, H., Maus, P., Greer, J., & Dull, C. (2001). A
primer on mapping vegetation using remote sensing. International
Journal of Wildland Fire,10(4), 277–287.
Boettinger,J.L.,Ramsey,R.D.,Bodily,J.M.,Cole,N.J.,Kienast-
Brown, S., Nield, S. J., Saunders, A. M., & Stum, A. K. (2008).
Landsat spectral data for digital soil mapping. In A. E. Hartemink, A.
McBratney, & M. Mendonça-Santos (Eds.), Digital soil mapping with
limited data (pp. 193–202). Springer. https://doi.org/10.1007/978-1-
4020-8592-5_16
Breiman, L. (2001). Random forests. Machine learning,45, 5–32. https://
doi.org/10.1023/A:1010933404324
Busse, M. D., Hubbert, K. R., Fiddler, G. O., Shestak, C. J., & Powers,
R. F. (2005). Lethal soil temperatures during burning of masticated
forest residues. International Journal of Wildland Fire,14(3), 267–
276. https://doi.org/10.1071/WF04062
Busse, M. D., Hubbert, K. R., & Moghaddas, E. E. Y. (2014). Fuel
reduction practices and their effects on soil quality (General Technical
Report No. PSW-GTR-241). USDA Forest Service, Pacific Southwest
Research Station.
Cannon, S. H., & DeGraff, J. (2009). The increasing wildfire and
post-fire debris-flow threat in western USA, and implications for
consequences of climate change. In K. Sassa, & P. Canuti (Eds.),
Landslides—Disaster risk reduction (pp. 177–190). Springer.
Certini, G. (2005). Effects of fire on properties of forest soils: A review.
Oecologia,143(1), 1–10. https://doi.org/10.1007/s00442-004- 1788-
8
Chaney, N. W., Minasny, B., Herman, J. D., Nauman, T. W., Brungard,
C. W., Morgan, C. L. S., McBratney, A. B., Wood, E. F., &
Yimam, Y. (2019). POLARIS soil properties: 30-m probabilis-
tic maps of soil properties over the contiguous United States.
Water Resources Research,55(4), 2916–2938. https://doi.org/10.
1029/2018WR022797
Chaney, N. W., Wood, E. F., McBratney, A. B., Hempel, J. W., Nauman,
T. W., Brungard, C. W., & Odgers, N. P. (2016). POLARIS: A 30-
meter probabilistic soil series map of the contiguous United States.
Geoderma,274, 54–67. https://doi.org/10.1016/j.geoderma.2016.03.
025
Chen, G., Metz, M. R., Rizzo, D. M., & Meentemeyer, R. K. (2015).
Mapping burn severity in a disease-impacted forest landscape using
Landsat and MASTER imagery. International Journal of Applied
Earth Observation and Geoinformation,40, 91–99. https://doi.org/
10.1016/j.jag.2015.04.005
Cheung, D. J., & Giardino, J. R. (2023). Debris flow occurrence
under changing climate and wildfire regimes: A southern California
perspective. Geomorphology,422, 108538. https://doi.org/10.1016/j.
geomorph.2022.108538
Chuvieco, E., Martin, M. P., & Palacios, A. (2002). Assessment of dif-
ferent spectral indices in the red-near-infrared spectral domain for
burned land discrimination. International Journal of Remote Sensing,
23(23), 5103–5110. https://doi.org/10.1080/01431160210153129
Clark, J., Parsons, A., Zajkowski, T., & Lannom, K. (2003). Remote
sensing imagery support for burned area emergency response teams
on 2003 southern California wildfires. USFS Remote Sensing
Applications Center BAER Support Summary.
Cole, R. P., Bladon, K. D., Wagenbrenner, J. W., & Coe, D. B. R.
(2020). Hillslope sediment production after wildfire and post-fire
forest management in northern California. Hydrological Processes,
34(26), 5242–5259. https://doi.org/10.1002/hyp.13932
Collins, L., Griffioen, P., Newell, G., & Mellor, A. (2018). The util-
ity of random forests for wildfire severity mapping. Remote Sensing
of Environment,216, 374–384. https://doi.org/10.1016/j.rse.2018.07.
005
Dahlgren, R. A., Boettinger, J. L., Huntington, G. L., & Amundson, R. G.
(1997). Soil development along an elevational transect in the western
Sierra Nevada, California. Geoderma,78(3–4), 207–236. https://doi.
org/10.1016/S0016-7061(97)00034- 7
Daly, C., Halbleib, M., Smith, J. I., Gibson, W. P., Doggett, M. K., Taylor,
G. H., Curtis, J., & Pasteris, P. P. (2008). Physiographically sensitive
mapping of climatological temperature and precipitation across the
conterminous United States. International Journal of Climatology:
A Journal of the Royal Meteorological Society,28(15), 2031–2064.
https://doi.org/10.1002/joc.1688
Dillon, G. K., Holden, Z. A., Morgan, P., Crimmins, M. A., Heyerdahl, E.
K., & Luce, C. H. (2011). Both topography and climate affected forest
and woodland burn severity in two regions of the western US, 1984 to
2006. Ecosphere,2(12), 1–33. https://doi.org/10.1890/ES11-00271.1
Dolan, J. (2018). Highway 101 in Santa Barbara County reopens, nearly
two weeks after a massive mudslide. Los Angeles Times.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

18 WILSON and PRENTICE
Dove, N. C., Safford, H. D., Bohlman, G. N., Estes, B. L., & Hart, S. C.
(2020). High-severity wildfire leads to multi-decadal impacts on soil
biogeochemistry in mixed-conifer forests. Ecological Applications,
30(4), e02072. https://doi.org/10.1002/eap.2072
Duval, J. S., Carson, J. M., Holman, P. B., & Darnley, A. G. (2005).
Terrestrial radioactivity and gamma-ray exposure in the United States
and Canada (Open-File Report 2005–1413). USGS.
Fallon, K., Wheelock, S. J., Sadegh, M., Pierce, J. L., McNamara, J. P.,
Cattau, M., & Baker, V. R. (2024). Post-fire hydrologic analysis: A
tale of two severities. Hydrological Sciences Journal,69(1), 139–148.
https://doi.org/10.1080/02626667.2023.2284306
Fernández, C., & Vega, J. A. (2016). Modelling the effect of soil burn
severity on soil erosion at hillslope scale in the first year following
wildfire in NW Spain. Earth Surface Processes and Landforms,41(7),
928–935. https://doi.org/10.1002/esp.3876
Fernandez-Garcia, V., Santamarta, M., Fernandez-Manso, A., Quintano,
C., Marcos, E., & Calvo, L. (2018). Burn severity metrics in fire-prone
pine ecosystems along a climatic gradient using Landsat imagery.
Remote Sensing of Environment,206, 205–217. https://doi.org/10.
1016/j.rse.2017.12.029
Fernández-Guisuraga, J. M., Fernandes, P. M., Marcos, E., Beltrán-
Marcos, D., Sarricolea, P., Farris, M., & Calvo, L. (2023). Caution
is needed across Mediterranean ecosystems when interpreting wall-
to-wall fire severity estimates based on spectral indices. Forest
Ecology and Management,546, 121383. https://doi.org/10.1016/j.
foreco.2023.121383
Fettig, C. J., Hood, S. M., Runyon, J. B., & Stalling, C. M. (2021). Bark
beetle and fire interactions in western coniferous forests: Research
findings. Fire Managment Today,79(1), 14–23.
Fettig, C. J., Mortenson, L. A., Bulaon, B. M., & Foulk, P. B. (2019).
Tree mortality following drought in the central and southern Sierra
Nevada, California, US. Forest Ecology and Management,432, 164–
178. https://doi.org/10.1016/j.foreco.2018.09.006
Fraser, A. M., Chester, M. V., & Underwood, B. S. (2020). Wildfire risk,
post-fire debris flows, and transportation infrastructure vulnerability.
Sustainable and Resilient Infrastructure,7(3), 188–200. https://doi.
org/10.1080/23789689.2020.1737785
Fraser, A. M., Chester, M. V., & Underwood, B. S. (2022). Wildfire risk,
post-fire debris flows, and transportation infrastructure vulnerability.
Sustainable and Resilient Infrastructure,7(3), 188–200. https://doi.
org/10.1080/23789689.2020.1737785
Gao, B.-C. (1996). NDWI—A normalized difference water index for
remote sensing of vegetation liquid water from space. Remote Sens-
ing of Environment,58(3), 257–266. https://doi.org/10.1016/S0034-
4257(96)00067-3
García-Llamas, P., Suárez-Seoane, S., Fernández-Guisuraga, J. M.,
Fernández-García, V., Fernández-Manso, A., Quintano, C., Taboada,
A., Marcos, E., & Calvo, L. (2019). Evaluation and comparison of
Landsat 8, Sentinel-2 and Deimos-1 remote sensing indices for assess-
ing burn severity in Mediterranean fire-prone ecosystems. Interna-
tional Journal of Applied Earth Observation and Geoinformation,80,
137–144. https://doi.org/10.1016/j.jag.2019.04.006
Giger, D. R., & Schmitt, G. J. (1983). Soil survey of the Sierra National
Forest area, California. USDA Forest Service.
Graham, R., Rossi, A., & Hubbert, R. (2010). Rock to regolith conver-
sion: Producing hospitable substrates for terrestrial ecosystems. GSA
Today,20, 4–9. https://doi.org/10.1130/GSAT57A.1
Grimm, R., Behrens, T., Märker, M., & Elsenbeer, H. (2008). Soil
organic carbon concentrations and stocks on Barro Colorado Island—
Digital soil mapping using random forests analysis. Geoderma,
146(1–2), 102–113. https://doi.org/10.1016/j.geoderma.2008.05.008
Harvey, B. J., Andrus, R. A., & Anderson, S. C. (2019). Incorporat-
ing biophysical gradients and uncertainty into burn severity maps
in a temperate fire-prone forested region. Ecosphere,10(2), e02600.
https://doi.org/10.1002/ecs2.2600
Hengl, T., Leenaars, J. G. B., Shepherd, K. D., Walsh, M. G., Heuvelink,
G., Mamo, T., Tilahun, H., Berkhout, E., Cooper, M., & Fegraus, E.
(2017). Soil nutrient maps of Sub-Saharan Africa: Assessment of soil
nutrient content at 250 m spatial resolution using machine learning.
Nutrient Cycling in Agroecosystems,109(1), 77–102. https://doi.org/
10.1007/s10705-017- 9870-x
Hengl, T., & MacMillan, R. A. (2019). Predictive soil mapping with R.
Lulu.
Hengl, T., Mendes de Jesus, J., Heuvelink, G. B. M., Ruiperez Gonzalez,
M., Kilibarda, M., Blagotić, A., Shangguan, W., Wright, M. N., Geng,
X., & Bauer-Marschallinger, B. (2017). SoilGrids250m: Global grid-
ded soil information based on machine learning. PLoS ONE,12(2),
e0169748. https://doi.org/10.1371/journal.pone.0169748
Hengl, T., Miller, M. A. E., Križan, J., Shepherd, K. D., Sila, A.,
Kilibarda, M., Antonijević, O., Glušica, L., Dobermann, A., &
Haefele, S. M. (2021). African soil properties and nutrients mapped
at 30 m spatial resolution using two-scale ensemble machine learn-
ing. Scientific Reports,11(1), Article 6130. https://doi.org/10.1038/
s41598-021-85639-y
Hicke, J. A., Johnson, M. C., Hayes, J. L., & Preisler, H. K. (2012).
Effects of bark beetle-caused tree mortality on wildfire. Fore st E co l-
ogy and Management,271, 81–90. https://doi.org/10.1016/j.foreco.
2012.02.005
Hijmans, R. (2024). raster: Geographic data analysis and modeling
(R package version 3.6-27) [Computer software]. https://rspatial.org/
raster
Hohner, A. K., Rhoades, C. C., Wilkerson, P., & Rosario-Ortiz, F. L.
(2019). Wildfires alter forest watersheds and threaten drinking water
quality. Accounts of Chemical Research,52(5), 1234–1244. https://
doi.org/10.1021/acs.accounts.8b00670
Holden, Z. A., Morgan, P., & Evans, J. S. (2009). A predictive model of
burn severity based on 20-year satellite-inferred burn severity data in
a large southwestern US wilderness area. Forest Ecology and Man-
agement,258(11), 2399–2406. https://doi.org/10.1016/j.foreco.2009.
08.017
Holden, Z. A., Smith, A. M. S., Morgan, P., Rollins, M. G., & Gessler, P.
E. (2005). Evaluation of novel thermally enhanced spectral indices for
mapping fire perimeters and comparisons with fire atlas data. Inter-
national Journal of Remote Sensing,26(21), 4801–4808. https://doi.
org/10.1080/01431160500239008
Huang, Y., Jin, Y., Schwartz, M. W., & Thorne, J. H. (2020). Intensi-
fied burn severity in California’s northern coastal mountains by drier
climatic condition. Environmental Research Letters,15(10), 104033.
https://doi.org/10.1088/1748-9326/aba6af
Hudak, A. T., Morgan, P., Bobbitt, M. J., Smith, A. M. S., Lewis, S.
A., Lentile, L. B., Robichaud, P. R., Clark, J. T., & McKinley, R. A.
(2007). The relationship of multispectral satellite imagery to imme-
diate fire effects. Fire Ecology,3, 64–90. https://doi.org/10.4996/
fireecology.0301064
Hudson, B. D. (1992). The soil survey as paradigm-based science. Soil
Science Society of America Journal,56(3), 836–841. https://doi.org/
10.2136/sssaj1992.03615995005600030027x
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 19
Huntington, G. L. (1954). The effect of vertical zonality on clay content
in residual granitic soils of the Sierra Nevada mountains.University
of California.
James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduc-
tion to statistical learning (Vol. 112). Springer.
Kasischke, E. S., Turetsky, M. R., Ottmar, R. D., French,N. H. F., Hoy, E.
E., & Kane, E. S. (2008). Evaluation of the composite burn index for
assessing fire severity in Alaskan black spruce forests. International
Journal of Wildland Fire,17(4), 515–526.
Kean, J. W., Staley, D. M., Lancaster, J. T., Rengers, F. K., Swanson,
B. J., Coe, J. A., Hernandez, J. L., Sigman, A. J., Allstadt, K. E., &
Lindsay, D. N. (2019). Inundation, flow dynamics, and damage in the
9 January 2018 Montecito debris-flow event, California, USA: Oppor-
tunities and challenges for post-wildfire risk assessment. Geosphere,
15(4), 1140–1163. https://doi.org/10.1130/GES02048.1
Keane, R. E. (2016). Spatiotemporal variability of wildland fuels in US
northern Rocky Mountain forests. Fore st s,7(7), 129. https://doi.org/
10.3390/f7070129
Keeley, J. E., Brennan, T. J., & Syphard, A. D. (2022). The effects
of prolonged drought on vegetation dieback and megafires in south-
ern California chaparral. Ecosphere,13(8), e4203. https://doi.org/10.
1002/ecs2.4203
Kempen, B., Brus, D. J., Heuvelink, G. B. M., & Stoorvogel, J. J. (2009).
Updating the 1: 50,000 Dutch soil map using legacy soil data: A multi-
nomial logistic regression approach. Geoderma,151(3–4), 311–326.
https://doi.org/10.1016/j.geoderma.2009.04.023
Key, C. H., & Benson, N. C. (2006). Landscape assessment (LA). In
D. C. Lutes, R. E. Keane, J. F. Caratti, C. H. Key, N. C. Benson, S.
Sutherland, & L. J. Gangi (Eds.), FIREMON: Fire effects monitoring
and inventory system (Vol. 164, pp. LA1–51). USDA Forest Service,
Rocky Mountain Research Station.
Kreye, J. K., Brewer, N. W., Morgan, P., Varner, J. M., Smith, A. M. S.,
Hoffman, C. M., & Ottmar, R. D. (2014). Fire behavior in masticated
fuels: A review. Forest Ecology and Management,314, 193–207.
https://doi.org/10.1016/j.foreco.2013.11.035
Kuhn, M. (2008). Building predictive models in R using the caret pack-
age. Journal of Statistical Software,28(5), 1–26. https://doi.org/10.
18637/jss.v028.i05
Lee, J. M., Mirocha, J. D., Lareau, N. P., Whitney, T., To, W., Kochanski,
A., & Lassman, W. (2023). Sensitivity of pyrocumulus convection to
tree mortality during the 2020 Creek fire in California. Geophysical
Research Letters,50(16), e2023GL104193. https://doi.org/10.1029/
2023GL104193
Levi, M. R., & Bestelmeyer, B. T. (2018). Digital soil mapping for
fire prediction and management in rangelands. Fire Ecology,14(2),
Article 11. https://doi.org/10.1186/s42408-018-0018-4
Littell, J. S., Peterson, D. L., Riley, K. L., Liu, Y., & Luce, C. H. (2016).
A review of the relationships between drought and forest fire in the
United States. Global Change Biology,22(7), 2353–2369. https://doi.
org/10.1111/gcb.13275
Loudermilk, E. L., O’Brien, J. J., Mitchell, R. J., Cropper, W. P., Hiers, J.
K., Grunwald, S., Grego, J., & Fernandez-Diaz, J. C. (2012). Linking
complex forest fuel structure and fire behaviour at fine scales. Inter-
national Journal of Wildland Fire,21(7), 882–893. https://doi.org/10.
1071/WF10116
Lukashov, S. G., Lancaster, J. T., Oakley, N. S., & Swanson, B. J.
(2019). Post-fire debris flows of 9 January 2018, Thomas fire, south-
ern California: Initiation areas, precipitation and impacts. In Seventh
international conference on debris-flow hazards mitigation, associa-
tion of environmental and engineering geologists special publication
28 (pp. 774–781). Association of Environmental and Engineering
Geologists.
Malone, B. P., Minasny, B., & McBratney, A. B. (2017). Using R for
digital soil mapping (Vol. 35). Springer.
McBratney, A. B., Santos, M. L. M., & Minasny, B. (2003). On digi-
tal soil mapping. Geoderma,117(1–2), 3–52. https://doi.org/10.1016/
S0016-7061(03)00223-4
McCool, K. D., Holub, S. M., Gao, S., Morrissette, B. A., Blunn, J. E.,
Gallo, A. C., & Hatten, J. A. (2023). Quantifying impacts of forest
fire on soil carbon in a young, intensively managed tree farm in the
western Oregon Cascades. Soil Science Society of America Journal,
87(6), 1458–1473. https://doi.org/10.1002/saj2.20582
McHugh, M. L. (2012). Interrater reliability: the kappa statistic.
Biochemia Medica,22(3), 276–282.
Morales-Barquero, L., Lyons, M. B., Phinn, S. R., & Roelfsema, C. M.
(2019). Trends in remote sensing accuracy assessment approaches
in the context of natural resources. Remote Sensing,11(19), 2305.
https://doi.org/10.3390/rs11192305
Morgan, P., Keane, R. E., Dillon, G. K., Jain, T. B., Hudak, A. T., Karau,
E. C., Sikkink, P. G., Holden, Z. A., & Strand, E. K. (2014). Chal-
lenges of assessing fire and burn severity using field measures, remote
sensing and modelling. International Journal of Wildland Fire,23(8),
1045–1060. https://doi.org/10.1071/WF13058
Morgan, P., Moy, M., Droske, C. A., Lewis, S. A., Lentile, L. B.,
Robichaud, P. R., Hudak, A. T., & Williams, C. J. (2015). Vegetation
response to burn severity, native grass seeding, and salvage log-
ging. Fire Ecology,11(2), 31–58. https://doi.org/10.4996/fireecology.
1102031
Morresi, D., Marzano, R., Lingua, E., Motta, R., & Garbarino, M. (2022).
Mapping burn severity in the western Italian Alps through pheno-
logically coherent reflectance composites derived from Sentinel-2
imagery. Remote Sensing of Environment,269, 112800. https://doi.
org/10.1016/j.rse.2021.112800
Odion, D. C., Frost, E. J., Strittholt, J. R., Jiang, H., Dellasala, D. A., &
Moritz, M. A. (2004). Patterns of fire severity and forest conditions
in the western Klamath Mountains, California. Conservation Biology,
18(4), 927–936. https://doi.org/10.1111/j.1523-1739.2004.00493.x
Parks, S. A., Holsinger, L. M., Voss, M. A., Loehman, R. A., & Robinson,
N. P. (2018). Mean composite fire severity metrics computed with
Google Earth Engine offer improved accuracy and expanded map-
ping potential. Remote Sensing,10(6), 879. https://doi.org/10.3390/
rs10060879
Parsons, A. (2003). Burned area emergency rehabilitation (BAER) soil
burn severity definitions and mapping guidelines (Draft). USDA
Forest Service, Rocky Mountain Research Station.
Parsons, A., Robichaud, P. R., Lewis, S. A., Napper, C., & Clark, J.
T. (2010). Field guide for mapping post-fire soil burn severity (Vol.
243, p. 49). General technical report RMRS-GTR-243. USDA Forest
Service, Rocky Mountain Research Station.
Poggio, L., De Sousa, L. M., Batjes, N. H., Heuvelink, G., Kempen, B.,
Ribeiro, E., & Rossiter, D. (2021). SoilGrids 2.0: Producing soil infor-
mation for the globe with quantified spatial uncertainty. Soilless,7(1),
217–240.
Ramcharan, A., Hengl, T., Nauman,T., Brungard, C., Waltman, S., Wills,
S., & Thompson, J. (2018). Soil property and class maps of the con-
terminous United States at 100-meter spatial resolution. Soil Science
Society of America Journal,82(1), 186–201. https://doi.org/10.2136/
sssaj2017.04.0122
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

20 WILSON and PRENTICE
Rengers, F. K., McGuire, L. A., Oakley, N. S., Kean, J. W., Staley, D.
M., & Tang, H. (2020). Landslides after wildfire: Initiation, magni-
tude, and mobility. Landslides,17(11), 2631–2641. https://doi.org/10.
1007/s10346-020- 01506-3
Restaino, C., Young, D. J. N., Estes, B., Gross, S., Wuenschel, A.,
Meyer, M., & Safford, H. (2019). Forest structure and climate mediate
drought-induced tree mortality in forests of the Sierra Nevada, USA.
Ecological Applications,29(4), e01902. https://doi.org/10.1002/eap.
1902
Robichaud, P. R., Bone, E. D., Lewis, S. A., Brooks, E. S., & Brown, R. E.
(2021). Effectiveness of post-fire salvage logging stream buffer man-
agement for hillslope erosion in the US Inland Northwest Mountains.
Hydrological Processes,35(1), e13943. https://doi.org/10.1002/hyp.
13943
Robichaud, P. R., Lewis, S. A., Brown, R. E., Bone, E. D., & Brooks,
E. S. (2020). Evaluating post-wildfire logging-slash cover treat-
ment to reduce hillslope erosion after salvage logging using ground
measurements and remote sensing. Hydrological Processes,34(23),
4431–4445. https://doi.org/10.1002/hyp.13882
Robichaud, P. R., Lewis, S. A., Laes, D. Y. M., Hudak, A. T., Kokaly,
R. F., & Zamudio, J. A. (2007). Postfire soil burn severity mapping
with hyperspectral image unmixing. Remote Sensing of Environment,
108(4), 467–480. https://doi.org/10.1016/j.rse.2006.11.027
Ryan, K. C., & Opperman, T. S. (2013). LANDFIRE—A national
vegetation/fuels data base for use in fuels treatment, restoration,
and suppression planning. Forest Ecology and Management,294,
208–216. https://doi.org/10.1016/j.foreco.2012.11.003
Saatchi, S., Halligan, K., Despain, D. G., & Crabtree, R. L. (2007).
Estimation of forest fuel load from radar remote sensing. IEEE
Transactions on Geoscience and Remote Sensing,45(6), 1726–1740.
https://doi.org/10.1109/TGRS.2006.887002
Saberi, S. J., Agne, M. C., & Harvey, B. J. (2022). Do you CBI what I
see? The relationship between the Composite Burn Index and quanti-
tative field measures of burn severity varies across gradients of forest
structure. International Journal of Wildland Fire,31(2), 112–123.
https://doi.org/10.1071/WF21062
Saberi, S. J., & Harvey, B. J. (2023). What is the color when black is
burned? Quantifying (re)burn severity using field and satellite remote
sensing indices. Fire Ecology,19(1), Article 24. https://doi.org/10.
1186/s42408-023- 00178-3
Safford, H. D., Miller, J., Schmidt, D., Roath, B., & Parsons, A. (2008).
BAER soil burn severity maps do not measure fire effects to vegeta-
tion: A comment on Odion and Hanson (2006). Ecosystems,11, 1–11.
https://doi.org/10.1007/s10021-007- 9094-z
Smith, H. G., Sheridan, G. J., Lane, P. N. J., Nyman, P., & Haydon,
S. (2011). Wildfire effects on water quality in forest catchments:
A review with implications for water supply. Journal of Hydrol-
ogy,396(1–2), 170–192. https://doi.org/10.1016/j.jhydrol.2010.
10.043
Sobrino, J. A., Llorens, R., Fernández, C., Fernández-Alonso, J. M.,
& Vega, J. A. (2019). Relationship between soil burn severity in
forest fires measured in situ and through spectral indices of remote
detection. Forest s,10(5), 457. https://doi.org/10.3390/f10050457
Soverel, N. O., Perrakis, D. D. B., & Coops, N. C. (2010). Estimating
burn severity from Landsat dNBR and RdNBR indices across western
Canada. Remote Sensing of Environment,114(9), 1896–1909. https://
doi.org/10.1016/j.rse.2010.03.013
Staley,D.M.,Negri,J.A.,Kean,J.W.,Laber,J.L.,Tillery,A.C.,&
Youberg, A. M. (2016). Updated logistic regression equations for the
calculation of post-fire debris-flow likelihood in the western United
States (Report No. 2016–1106). US Geological Survey.
Stanley, P., Spertus, J., Chiartas, J., Stark, P. B., & Bowles, T. (2023).
Valid inferences about soil carbon in heterogeneous landscapes.
Geoderma,430, 116323. https://doi.org/10.1016/j.geoderma.2022.
116323
Stephens, S. L., Bernal, A. A., Collins, B. M., Finney, M. A.,
Lautenberger, C., & Saah, D. (2022). Mass fire behavior created
by extensive tree mortality and high tree density not predicted by
operational fire behavior models in the southern Sierra Nevada. For-
est Ecology and Management,518, 120258. https://doi.org/10.1016/
j.foreco.2022.120258
Stephens, S. L., Burrows, N., Buyantuyev, A., Gray, R. W., Keane, R.
E., Kubian, R., Liu, S., Seijo, F., Shu, L., & Tolhurst, K. G. (2014).
Temperate and boreal forest mega-fires: Characteristics and chal-
lenges. Frontiers in Ecology and the Environment,12(2), 115–122.
https://doi.org/10.1890/120332
Stevens, J. T., Collins, B. M., Miller, J. D., North, M. P., & Stephens, S. L.
(2017). Changing spatial patterns of stand-replacing fire in California
conifer forests. Forest Ecology and Management,406, 28–36. https://
doi.org/10.1016/j.foreco.2017.08.051
Thomas, M. A., Kean, J. W., McCoy, S. W., Lindsay, D. N., Kostelnik, J.,
Cavagnaro, D. B., Rengers,F. K., East, A. E., Schwartz, J. Y., & Smith,
D. P. (2023). Postfire hydrologic response along the Central Califor-
nia (USA) coast: Insights for the emergency assessment of postfire
debris-flow hazards. Landslides,20, 2421–2436. https://doi.org/10.
1007/s10346-023- 02106-7
Thompson, J. A., Kienast-Brown, S., D’Avello, T., Philippe, J., &
Brungard, C. (2020). Soils2026 and digital soil mapping—A founda-
tion for the future of soils information in the United States. Geoderma
Regional,22, e00294. https://doi.org/10.1016/j.geodrs.2020.e00294
Uzun, H., Dahlgren, R. A., Olivares, C., Erdem, C. U., Karanfil, T., &
Chow, A. T. (2020). Two years of post-wildfire impacts on dissolved
organic matter, nitrogen, and precursors of disinfection by-products
in California stream waters. Water Research,181, 115891. https://doi.
org/10.1016/j.watres.2020.115891
Vakili, E., Hoffman, C. M., Keane, R. E., Tinkham, W. T., & Dickinson,
Y. (2016). Spatial variability of surface fuels in treated and untreated
ponderosa pine forests of the southern Rocky Mountains. Interna-
tional Journal of Wildland Fire,25(11), 1156–1168. https://doi.org/
10.1071/WF16072
Van Wagtendonk, J. W., Benedict, J. M., & Sydoriak, W. M. (1998).
Fuel bed characteristics of Sierra Nevada conifers. Western Journal of
Applied Forestry,13(3), 73–84. https://doi.org/10.1093/wjaf/13.3.73
Viedma, O., Quesada, J., Torres, I., De Santis, A., & Moreno, J. M.
(2015). Fire severity in a large fire in a Pinus pinaster forest is highly
predictable from burning conditions, stand structure, and topography.
Ecosystems,18, 237–250. https://doi.org/10.1007/s10021-014-9824-
y
Vieira, D. C. S., Fernández, C., Vega, J. A., & Keizer, J. J. (2015).
Does soil burn severity affect the post-fire runoff and interrill erosion
response? A review based on meta-analysis of field rainfall simulation
data. Journal of Hydrology,523, 452–464. https://doi.org/10.1016/j.
jhydrol.2015.01.071
Wagenbrenner, J., Coe, D., & Olsen, W. (2023). Mitigating potential sed-
iment delivery from post-fire salvage logging. Department of Forestry,
State of California, The Natural Resources Agency.
Westerling, A. L. (2016). Increasing western US forest wildfire activ-
ity: Sensitivity to changes in the timing of spring. Philosophical
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 21
Transactions of the Royal Society B: Biological Sciences,371(1696),
20150178. https://doi.org/10.1098/rstb.2015.0178
Wilford, J. (2012). A weathering intensity index for the Australian
continent using airborne gamma-ray spectrometry and digital ter-
rain analysis. Geoderma,183, 124–142. https://doi.org/10.1016/j.
geoderma.2010.12.022
Williams, J. N., Safford, H. D., Enstice, N., Steel, Z. L., & Paulson, A. K.
(2023). High-severity burned area and proportion exceed historic con-
ditions in Sierra Nevada, California, and adjacent ranges. Ecosphere,
14(1), e4397. https://doi.org/10.1002/ecs2.4397
Woods, S. W., & Balfour, V. N. (2010). The effects of soil texture and
ash thickness on the post-fire hydrological response from ash-covered
soils. Journal of Hydrology,393(3–4), 274–286. https://doi.org/10.
1016/j.jhydrol.2010.08.025
How to cite this article: Wilson, S. G., & Prentice,
S. (2024). Digital soil mapping of soil burn severity.
Soil Science Society of America Journal, 1–23.
https://doi.org/10.1002/saj2.20702
APPENDIX
FIGURE A Graphical illustration of mapped soil types, soil development properties, and ecological zones along Creek Fire burn scar elevation
transect. Parent material is granodiorite of varying mica content. Elevation data from 30 m United States Geologic Survey (USGS) digital elevation
model (DEM). MAP profile from isohyet map of California along transect. Dashed blue lines represent areas in which actual MAP diverges from
MAP values predicted from weather station derived data.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

22 WILSON and PRENTICE
FIGURE B (a) High soil burn severity (SBS) class probabilities from random forest generated (digital soil mapping technique) SBS
classifications, Creek fire (2020), and SBS ground-based (soil) observations; (b) moderate SBS class probabilities from random forest generated
(digital soil mapping technique) SBS classifications, Creek fire (2020), and SBS ground-based (soil) observations; (c) low SBS class probabilities
from random forest generated (digital soil mapping technique) SBS classifications, Creek fire (2020), and SBS ground-based (soil) observations; and
(d) unburned SBS class probabilities from random forest generated (digital soil mapping technique) SBS classifications, Creek fire (2020), and SBS
ground-based (soil) observations.
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

WILSON and PRENTICE 23
FIGURE C Map of Shannon entropy (log base 2) for the probability of high, moderate, low, and unburned soil burn severity (SBS). Points are
field observation in each SBS class as recorded by soil scientists on the ground immediately after fire. The inset map is Upper Chiquito Creek area;
Creek Fire (2020).
14350661, 0, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/saj2.20702 by California Polytechnic State Univ Kennedy Library, Wiley Online Library on [17/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License