CSIRO PUBLISHING
International Journal of Wildland Fire 2016, 25, 10–24
http://dx.doi.org/10.1071/WF15092
Pre-fire and post-fire surface fuel and cover measurements
collected in the southeastern United States for model
evaluation and development – RxCADRE 2008,
2011 and 2012
Roger D. Ottmar A,D, Andrew T. Hudak B, Susan J. Prichard C, Clinton S. Wright A,
Joseph C. Restaino C, Maureen C. Kennedy C and Robert E. Vihnanek A
A
USDA Forest Service, Pacific Northwest Research Station, Pacific Wildland Fire Sciences
Laboratory, 400 North 34th Street, Suite 201, Seattle, WA 98103, USA.
B
USDA Forest Service, Rocky Mountain Research Station, 1221 South Main Street, Moscow,
ID 83843, USA.
C
School of Environmental and Forest Sciences, University of Washington, Box 352100,
Seattle, WA 98195, USA.
D
Corresponding author. Email: rottmar@fs.fed.us
Abstract. A lack of independent, quality-assured data prevents scientists from effectively evaluating predictions
and uncertainties in fire models used by land managers. This paper presents a summary of pre-fire and post-fire fuel, fuel
moisture and surface cover fraction data that can be used for fire model evaluation and development. The data were
collected in the southeastern United States on 14 forest and 14 non-forest sample units associated with 6 small replicate
and 10 large operational prescribed fires conducted during 2008, 2011, and 2012 as part of the Prescribed Fire Combustion
and Atmospheric Dynamics Research Experiment (RxCADRE). Fuel loading and fuel consumption averaged 6.8 and
4.1 Mg ha1 respectively in the forest units and 3.0 and 2.2 Mg ha1 in the non-forest units. Post-fire white ash cover
ranged from 1 to 28%. Data were used to evaluate two fuel consumption models, CONSUME and FOFEM, and to
develop regression equations for predicting fuel consumption from ash cover. CONSUME and FOFEM produced similar
predictions of total fuel consumption and were comparable with measured values. Simple linear models to predict pre-fire
fuel loading and fuel consumption from post-fire white ash cover explained 46 and 59% of variation respectively.
Additional keywords: ash, fire effects, fuel consumption, fuel loading, longleaf pine, prescribed fire.
Received 13 September 2014, accepted 11 August 2015, published online 13 October 2015
Introduction
Fuel consumption is defined as the amount of biomass that is
fully combusted during a wildland fire. It is one of the critical
components for estimating fire behaviour, amount of heat
released and smoke produced, effectiveness of fire at reducing
fuel or exposing mineral soil, and many other fire effects such
as carbon reallocation, tree mortality and soil heating (Agee
1993; Hardy et al. 2001; Agee and Skinner 2005; Peterson
et al. 2005; Urbanski et al. 2011; Ottmar 2013; Wright
2013a, 2013b; Parsons et al. in press). For example, millions of
hectares are prescribed-burned each year in the southern
United States and these projects require estimates of fuel
consumption to meet smoke management, fuel treatment and
ecological restoration guidelines (Waldrop and Goodrick
2012; Ryan et al. 2013).
To assist managers in planning for wildland fire, consumption studies of shrubs, forbs, grasses, woody fuel, litter and duff
in forests and rangelands have been conducted in temperate,
Journal compilation Ó IAWF 2016
tropical and boreal regions of the world and offer datasets that
include fuel characteristics, fuel moisture, fuel consumption and
environmental variables from both wildfires and prescribed fires
(Sandberg 1980; Brown et al. 1991; Scholl and Waldrop 1999;
Ottmar and Sandberg 2003; Sullivan et al. 2003; Hollis et al.
2010; Ottmar et al. 2013; Wright 2013a). In most cases, these
datasets have been used to develop fuel consumption models
found in software systems in use today such as CONSUME
(Prichard et al. 2007), FOFEM (Reinhardt et al. 1997), CanFIRE
and BORFIRE (de Groot et al. 2007, 2009) or in the early
development of new techniques to retrospectively predict surface fuel loadings and consumption based on residual white ash
cover, the first-order product of complete combustion (Stronach
and McNaughton 1989; Hudak et al. 2013a, 2013b). Although
many of these models are mainstays of fire effects modelling,
most of the systems and processes have not been thoroughly
quantitatively evaluated for uncertainties and potential error
because independent, fully documented, quality-assured fuel
www.publish.csiro.au/journals/ijwf
RxCADRE fuel and ash measurements
Int. J. Wildland Fire
Host land
manager
Lead
coordinator
11
Data manager
Observational data collection disciplines
Fuel
Meteorology
Fire
behaviour
Energy
Emissions
Fire effects
Study plan and
incident management
plan
Data collection,
reduction, and analysis
Final report, data
repository, and papers
Fig. 1. Organisational diagram of the Prescribed Fire Combustion Atmospheric Dynamics Experiment
(RxCADRE) and the six discipline teams.
consumption data are lacking (Cruz and Alexander 2010;
Alexander and Cruz 2013). The one exception is the evaluation
of CONSUME and the First Order Fire Effects Model (FOFEM)
provided by Prichard et al. (2014) using data collected in the
eastern and southeastern United States (Ottmar et al. 2012;
Reid et al. 2012).
The current paper presents ground-based measurements of
fuel loading, fuel moisture content, fuel consumption and postfire cover fractions of white ash and other surface materials. The
data were collected as part of the Prescribed Fire Combustion
and Atmospheric Dynamics Research Experiment (RxCADRE)
(Joint Fire Science Program 2014) to provide novel and critical
observational data within six discipline areas necessary for
building and validating fire models (Fig. 1). The comprehensive
fuels dataset presented here will be useful for future fire and fuel
consumption model development and evaluation (Ottmar and
Restaino 2014) and will support the fuel information needs of
other scientists participating in the project (Butler et al. 2015;
Clements et al. 2015; Hudak et al. 2015; O’Brien et al. 2015;
Rowell et al. 2015; Strand et al. 2015). To demonstrate the
usefulness of these datasets, we compared fuel consumption
measurements with predicted consumption from CONSUME
and FOFEM, two models that are commonly used by fire and
fuel managers in the United States. We also assessed whether
ground cover fractions of white ash and other surface materials
correlate significantly with pre-fire fuel loading and fuel consumption, thereby providing an alternative approach to predicting fuel loading and consumption in the absence of pre-fire
fuels data.
Methods
Study areas
Twenty-eight sample units were established within 6 small
replicate and 10 large operational prescribed fires at the Joseph
W. Jones Ecological Research Center (JJERC) and Eglin Air
Force Base (Eglin AFB) in the southeastern USA in 2008, 2011
and 2012 (Table 1, Fig. 2). The 16 prescribed fires ranged in size
from 2 to 828 ha and the sample units ranged in size from 0.04 to
19 ha.
The JJERC is located in the Lower Coastal Plain and Flatwood Province of southwestern Georgia, with elevations ranging from 35 to 45 m above sea level (McNab and Avers 1994).
The climate of this region is characterised as humid subtropical
with a mean annual precipitation of 131 cm distributed evenly
throughout the year, and mean daily temperatures ranging from
21 to 348C in summer and from 5 to 178C in winter (Goebel et al.
1997). The province is characterised by flat, weakly dissected
alluvial deposits over Ocala Limestone (Florea and Vacher
2009). Parent materials consist of marine and continental sand
and clay deposits formed during the Mesozoic and Cenozoic
eras (Wilson et al. 1999). The longleaf pine (Pinus palustris)–
wiregrass (Aristida stricta) ecosystems have been maintained
using regular understorey prescribed burning (average return
interval of 2 or 3 years) since the 1930s (Hendricks et al. 2002).
The woodlands are characterised by an overstorey of longleaf
pine and an understorey dominated by wiregrass, but with many
other species of perennial grasses, forbs and hardwood sprouts
(Goebel et al. 1997, Drew et al. 1998).
12
Int. J. Wildland Fire
R. D. Ottmar et al.
Table 1. Sample unit information including location (Eglin Air Force Base (EAFB) or Joseph W. Jones Ecological Research Center (JJERC)),
burn date, cover type (forest or non-forest), size and associated prescribed burn unit
Sample unit ID
Location
Burn date
Cover type
Size (ha)
2008_DubignonEast
2008_NorthBoundary
2008_TurkeyWoods
2008_307B
2008_608A
2011_608A_NW
2011_608A_SW
2011_608A_SE
2011_703C_W
2011_703C_E
2012_L2F
2012_L2F_HIP1
2012_L2F_HIP2
2012_L2F_HIP3
2012_L1G
2012_L1G_HIP1
2012_L1G_HIP2
2012_L1G_HIP3
2012_L2G
2012_L2G_HIP1
2012_L2G_HIP2
2012_L2G_HIP3
2012_S3
2012_S4
2012_S5
2012_S7
2012_S8
2012_S9
JJERC
JJERC
JJERC
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
EAFB
06-Mar-08
05-Mar-08
03-Mar-08
02-Mar-08
01-Mar-08
09-Feb-11
09-Feb-11
09-Feb-11
06-Feb-11
06-Feb-11
11-Nov-12
11-Nov-12
11-Nov-12
11-Nov-12
4-Nov-12
4-Nov-12
4-Nov-12
4-Nov-12
10-Nov-12
10-Nov-12
10-Nov-12
10-Nov-12
1-Nov-12
1-Nov-12
1-Nov-12
7-Nov-12
7-Nov-12
7-Nov-12
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
Non-forest
1.52
1.52
1.52
1.52
1.52
0.16
0.16
0.29
0.16
0.16
19.00
0.04
0.04
0.04
19.00
0.04
0.04
0.04
19.00
0.04
0.04
0.04
2.00
2.00
2.00
2.00
2.00
2.00
Eglin AFB, in the panhandle of western Florida, is characterised by large areas of xeric longleaf pine sandhill forest and
grass, and grass- and shrub-dominated military training areas
kept in a tree-free state. Elevation ranges from 52 to 85 m above
sea level, the land is flat, and soils are characterised by
unconsolidated, well-drained sand deposits of Quartzipsamments of the Lakeland series (Overing et al. 1995).The climate
is subtropical, with warm, humid summers and mild winters.
Mean annual temperature is 19.88C, with a mean annual precipitation of 158 cm, most of which falls from June through
September (Overing et al. 1995). Experimental prescribed fires
were conducted in both forest and non-forest areas at Eglin AFB.
Forest burn units were characterised by an overstorey dominated
by longleaf pine mixed with various deciduous oaks (turkey oak
(Quercus laevis), sand post oak (Q. margaretta Ashe), blue jack
oak (Q. incana Bartram), sand live oak (Q. germinate Small) and
laurel oak (Q. laurifolia)) (Hiers et al. 2007). Non-forest burn
units were selected for homogeneity in the coverage and density
of their herbaceous fuel although various shrubs were also
present (e.g. woody goldenrod (Chrysoma pauciflosculosa),
lowbush huckleberry (Gaylussacia dumosa), gopher apple
(Licania michauxii), saw palmetto (Serenoa repens), persimmon (Diospyros virginiana) and hawthorne (Crataegus spp.)).
All of the 2008 and 2011 prescribed burn sites at both
locations were forested. Longleaf pine dominated the overstorey, with turkey oak or saw palmetto (Serenoa repens)
occurring in an understorey matrix of wiregrass and other
Associated prescribed burn
Dubignon East
North Boundary
Turkey Woods
307B
608A
608A
608A
608A
703C
703C
L2F
L2F
L2F
L2F
L1G
L1G
L1G
L1G
L2G
L2G
L2G
L2G
S3
S4
S5
S7
S8
S9
grasses (Fig. 3a). Of the nine prescribed burns conducted in
2012, one was in a longleaf pine forest (Fig. 3b) and the eight
remaining burns were non-forest with a mix of grass and shrubs,
predominantly turkey oak (Fig. 3c). All prescribed burns
had been treated with fire every 2 to 4 years to meet several
management objectives including fuel reduction to mitigate fire
hazard, longleaf pine ecosystem and associated wildlife habitat
maintenance, and for maintaining sites for military training
operations. One unit (L1G) also had been treated with a herbicide (hexazinone) to reduce hardwood recruitment. The 10
operational burns were either ignited with all-terrain vehicles
carrying drip torches or helicopter-ignited with incendiary
spheres and typically burned using a strip heading or flanking
fire. The six replicate burns were hand-ignited with a single strip
of fire using a drip torch producing a single head fire. All burns
followed established prescription criteria for meeting landmanagement objectives; no burning occurred under extremely
dry or wet conditions. Prescription weather parameters included
10-h fuel moisture between 4 and 20%, 1000-h fuel moisture
between 15 and 40% and wind ,8.9 m s1.
Field measurements
Pre-fire and post-fire fuel loading and consumption
Fuel sampling protocols were developed from earlier fuel
studies in the southeastern United States (Scholl and Waldrop
1999; Ottmar et al. 2012; Wright 2013a) and were modified
RxCADRE fuel and ash measurements
Int. J. Wildland Fire
(a)
Prescribed fire year
Turkey
Woods
Administrative boundary
2008
Eglin Air Force Base
2011
Joseph W. Jones
Ecological Research Center
2012
13
North
Boundary
Dubignon
East
Georgia
608A (2008 and 2011)
Florida
N
0
20
40
80
km
(b)
Fig. 2. (a) Location of the 16 RxCADRE experimental prescribed fires conducted in 2008, 2011 and 2012.
(b) Small replicate (S) and large operational (L) prescribed buns were selected for the 2012 RxCADRE
research project located on the B70 bombing range at Eglin Air Force Base, Florida. Only large operational
prescribed burns were selected for the RxCADRE research burns in 2008 and 2011.
throughout 2012 after reviewing previous datasets to more
accurately capture fuel characteristics and consumption and to
accommodate requirements of RxCADRE scientists. In 2008,
destructive-sample fuel plots were established in one 5-ha
systematic grid pattern in each of the five operational prescribed
burn sites at Eglin AFB (.125 ha each) and JJERC (.40 ha
each). A total of 20 pre-fire and 20 post-fire fuel plots (1 1 m)
were alternately located at 20-m intervals along two parallel
transects 40 m apart. In February 2011, the two operational
prescribed burns were conducted at Eglin AFB (.125 ha each),
with two widely separated 5-ha sampling sites in one burn and
three widely separated sampling sites in the other. One sampling
site in the latter case had 20 pre-fire and 20 post-fire fuel plots
(1 1 m) alternately situated at 5-m intervals along two parallel
transects 30 m apart (similar to the 2008 sampling design). The
other four sampling sites consisted of 20 pre-fire and 20 post-fire
fuel plots (1 1 m) distributed at 5-m intervals around the
periphery of a 40 40-m highly instrumented plot (HIP) to
better characterise the fuel and consumption near the fire
behaviour and fire effects instrumentation. In November 2012,
six small replicate prescribed fires (2 ha each) and three larger
operational prescribed fires (.125 ha each, comparable in size
with the 2008 and 2011 prescribed burns) were conducted at
Eglin AFB. The small replicate prescribed burns were surrounded by 25 pre-fire and 25 post-fire fuel plots alternately
situated at 10-m intervals. In each of the large operational
prescribed burns, 30 pre-fire and 30 post-fire fuel plots were
alternately located at 50-m intervals along three approximately
parallel transects ,100 m apart (similar to the 2008 sampling
design, but covering a much larger portion of the burn unit).
Three HIPs were located within each large operational prescribed burn. Each HIP consisted of 9 (L1G and L2G) or 12
(L2F) pre-fire and 9 (L1G and L2G) or 12 (L2F) post-fire plots
alternately situated at 2.5-m intervals around the periphery of a
20 20-m area (similar to the 2011 sampling design). Clip plots
in the 2012 non-forest sample units were 1 1 m as in 2008 and
2011. Clip plots in the 2012 forest sample units were reduced to
0.5 0.5 m to speed up sampling. Fuel from within all clip plots
was collected and consolidated into four general categories:
shrub, herbaceous (grasses and forbs), down-and-dead fine wood
(#7.6 cm in diameter) and litter. Samples were oven-dried at
708C for 48 h, then weighed to determine loading.
14
Int. J. Wildland Fire
(a)
(b)
(c)
Fig. 3. (a) Typical forest prescribed burn area at JJERS in 2008, and (b, c)
forest and non-forest prescribed burn areas at Eglin Air Force Base in 2012.
Measureable large woody fuels (.7.6 cm) were only
encountered at two of the 28 sample units (307B and L2F);
therefore, large woody fuel was not included in the loading and
consumption totals considered for FOFEM and CONSUME
evaluation. However, woody fuel is the primary contributor to
post-fire white ash cover (Hudak et al. 2013a). Consequently,
the large-woody-fuel component was included in the surface
fuel loading and consumption totals at the 307B and L2F sample
units for predicting total fuel loading and consumption
R. D. Ottmar et al.
retrospectively by using post-fire white ash cover across all 28
sample units. Planar intersect transects 22 m long (Brown 1974)
originating at each fuel sample plot were used to measure woody
fuels .7.6 cm in diameter in these two sample units.
Fuel moisture sampling for each fuel bed category (herb,
shrub, fine wood, litter) was conducted 30 min before ignition
for each burn unit. Samples (5 # n # 10) of each fuel bed
category were collected inside the perimeter of each prescribed
burn unit near sample units and sealed in airtight 6-L plastic
bags. All moisture samples were weighed as soon as possible in
the field, generally within 6 h of ignition. Samples were ovendried at 708C for 48 h and then weighed to determine moisture
content as a fraction of dry weight.
Post-fire surface cover fractions
Post-fire surface cover fractions at all sample units were
visually estimated by the same observer. Estimates were made
from fuels within a square quadrat equal in size to those used for
clipping pre-fire fuels, and before they were disturbed during
post-fire fuel collection. The four cover fractions of green
vegetation, litter (including dead vegetation and woody debris),
white ash and mineral soil were estimated under the constraint
that they must sum to one (unity). Char cover was estimated
outside the sum-to-unity constraint and primarily represents the
combined percentage of litter but also some soil that was
blackened by the fire.
Fuel consumption model parameterisation
To represent fuel and environmental conditions for each sample
unit, we used sample data to run CONSUME version 4.2 and
FOFEM version 6.0. Inputs included herbaceous, shrub, 1-, 10-,
100- and 1000-h downed wood biomass, and 10-h, 1000-h and
duff fuel moisture (FM) content. Additional CONSUME inputs
include an estimate of the percentage of the area burned, litter
depth (cm), coverage of litter (percentage), litter arrangement,
duff derivation (from needles and leaves) and coverage of duff
(percentage). We used CONSUME’s southeastern natural fuel
consumption equations. Additional FOFEM inputs include
forest cover type, season of burn (winter or spring), duff depth,
duff biomass, litter biomass and percentage of logs that are
rotten. Default FOFEM settings include region (southeast), fire
type (moderate) and consumption (natural fuel).
There were missing input data for both models, which
required use of representative inputs for some model runs.
Measurements of 1000-h FM were not collected for the nonforest sites because there was no downed woody material
.7.6 cm in diameter; moisture content of 41% collected from
the nearest forest site (L2F) was used. As little or no duff was
evident at any prescribed burn sites except L2F, a moderate
moisture content of 70% was assigned (Prichard et al. 2007).
Post-burn photos were reviewed to estimate percentage area
burned and a median value of 80% assigned for all sites.
Data analysis
Fuel loading and fuel consumption
Fuel loading by fuel category was measured and calculated
as the mean component loading of all pre-fire or post-fire clip
plots. Fuel consumption was calculated for each fuel bed
RxCADRE fuel and ash measurements
Int. J. Wildland Fire
category by subtracting the mean post-fire loading from the
mean pre-fire loading for each set of plots. Because pre-fire and
post-fire fuel plot locations must always differ when sampling
methods are destructive, absolute consumption (Mg ha1) and
relative consumption (%) could be calculated only after aggregating plot measures within each sample unit.
To assess the distribution of fuel loading data, we used a
normal quantile–quantile plot (Q-Q) to visually test the normality of the data. We then used a Shapiro–Wilk test to confirm
the significance of the non-normality of the data distribution.
Levene’s test was used to examine within-site and across-site
variance due to the non-normal distribution of the data. To
identify potential outliers and protect against the misinterpretation of fuel loading data, sample variance was calculated for
each vegetation type and burn unit.
Fuel model comparison
We compared measured and predicted consumption of the
following fuel categories: herbaceous vegetation, shrubs, fine
downed wood (1-, 10- and 100-h size classes) and litter. We did
not compare large woody fuel or duff fuel consumption predicted by CONSUME or FOFEM because of insufficient observations for validation; measureable amounts of large woody
fuels only existed at two sample units (307B and L2F), and of
duff at only one sample unit (L2F). For each of the four fuel bed
categories considered, we plotted predicted consumption
against measured consumption and conducted ordinary leastsquares regression to evaluate goodness of fit and trends in
model residuals. Model evaluation is based on model residuals,
which we express as predicted values minus measured values.
We characterised model uncertainty using the paired t-test for
equivalence (Robinson and Froese 2004; Robinson 2013) on the
model residuals to estimate a ‘region of indifference’ for the
predicted consumption relative to measured consumption
(Prichard et al. 2014). The calculated regions of indifference
(Mg ha1) represent the range in which predicted values fail to
reject the null hypothesis that predicted and measured values are
not equivalent. In other words, the regions of indifference
represent the range in which predicted and measured values
can be considered statistically equivalent.
Retrospective fuel load and consumption predictions
Plot-level measurements (loading, consumption and cover)
were aggregated to the sample unit level for analysis. Owing
to typically log-normal data distributions, Spearman correlations were used to test the strength of relationships between
post-fire surface cover fractions and surface fuel loading or
consumption measurements and across all sample units. Natural
log-transformations were used to normalise log-normal distributions before developing simple linear regression models to
predict fuel load and consumption from highly correlated postfire cover fractions. Back-transformation of the model predictions from the geometric scale to the arithmetic scale introduces
bias, but a bias correction factor (cb) can be calculated based on
the mean square error (MSE) of the model residuals as per
Baskerville (1972):
cb ¼ expð0:5MSEÞ
ð1Þ
15
Table 2. Forest sample unit (n 5 14) and non-forest sample units
(n 5 14) mean and standard deviation (s.d.) of preburn fuel loading,
measured and predicted consumption and relative consumption by
fuel bed category including total, herb, shrub, fine wood (combined
1-, 10-, 100-h time-lag classes) and litter
Fuel bed
category
Method
Preburn
loading
Mean s.d.
(Mg ha1)
Total
Herb
Shrub
Fine wood
Litter
Total
Herb
Shrub
Fine wood
Litter
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
Measured
CONSUME
FOFEM
6.8
2.4
0.6
0.4
1.1
1.3
1.6
0.8
3.5
1.7
3.0
1.0
1.7
0.3
0.7
0.8
0.1
0.2
0.5
0.3
Consumption
Mean s.d.
(Mg ha1)
Forest units
4.1
1.7
5.1
2.0
4.6
2.1
0.6
0.4
0.5
0.4
0.6
0.4
0.5
0.6
0.8
0.9
0.5
0.8
0.5
0.6
1.1
0.6
0.3
0.6
2.7
1.7
2.3
1.0
3.2
1.7
Non-forest units
2.2
1.0
2.4
0.7
2.9
1.0
1.4
0.3
1.5
0.2
1.7
0.2
0.4
0.7
0.5
0.6
0.7
0.7
0.1
0.1
0.1
0.1
0.1
0.1
0.4
0.3
0.2
0.1
0.5
0.3
Mean s.d.
(%)
60.8
73.4
65.0
92.2
87.6
93.7
40.9
77.9
32.0
29.4
71.5
17.6
76.5
69.4
92.6
18.8
12.2
16.6
15.3
11.9
12.4
24.3
4.0
36.4
35.3
18.5
24.2
22.5
18.8
19.4
75.3
81.2
97.1
86.4
93.0
99.9
48.2
68.0
88.9
41.1
77.2
28.7
81.4
31.5
99.6
15.9
5.7
2.7
8.5
0.6
0.3
35.2
20.0
26.5
37.1
29.4
24.9
26.6
2.6
0.9
where the back-transformed predictions are multiplied by cb.
Data were analysed using R statistical software (R Development
Core Team 2012).
Results
Fuel loading and fuel consumption
The 28 sample units were divided into two groups (‘forest’
and ‘non-forest’) because of differences in fuel categories and
pre-fire fuel loads (Table S1 in the Supplementary material,
available online only). Pre-fire surface fuels in the forest
units (mean standard deviation) were often dominated by
litter (3.5 1.7 Mg ha1). Fine wood (1.6 0.8 Mg ha1),
shrubs (1.1 1.3 Mg ha1) and herbaceous biomass
(0.6 0.4 Mg ha1) were also present (Table 2). Pre-fire surface
fuels in the non-forest units were dominated by herbaceous
16
Int. J. Wildland Fire
biomass (1.7 0.3 Mg ha1), whereas shrub loading displayed
the greatest pre-fire variance (0.7 0.8 Mg ha1) (Table 2). Prefire loading for all fuel categories in both groups was universally
more variable than post-fire fuel loading. Pre-fire variance in
surface fuel loading was greatest in the fine wood and shrub
categories. Pre-fire loading and variance of surface fuels in
forest units were both generally two to three times greater than
surface fuels measured in non-forest units.
Examination of the data distribution for pre-fire surface fuel
loading revealed a potentially non-normal pattern in the forest
and non-forest units (Fig. S1). The null hypothesis that pre-fire
surface fuel loading is normally distributed in non-forest units
was rejected (Shapiro–Wilk W ¼ 0.8219, P ¼ 0.0094). The same
null hypothesis for pre-fire loading in forest units was also
rejected (W ¼ 0.8375, P ¼ 0.0151).
Due to the non-normality of the data, Levene’s tests were
used to explore patterns of variance in the pre-fire surface
loading data. In both forest (P ¼ 0.4734) and non-forest
(P ¼ 0.0735) units, pre-fire variance in surface fuel loading
was not significantly different (a ¼ 0.05) for burn units with
different sampling intensity (i.e. n ¼ 9–30). This suggests that
the evolution of the sampling procedure did not have a significant effect on inferences from the data. Pre-fire variance of
surface fuel loading measured in units at JJERC relative to units
in Eglin AFB was also not significantly different (a ¼ 0.05)
(P ¼ 0.4493), suggesting that inferences may be adequately
made across both study areas.
There were large differences in the total mass of fuel
consumed between the forest and non-forest units, ranging from
1.7 to 8.9 Mg ha1 in forest units and from 1.3 to 5.3 Mg ha1 in
non-forest units (Table S1). Most of the fuel consumed in the
forest units was in the litter category, whereas consumption in
the non-forest units was dominated by the herbaceous category,
followed by litter and shrubs (Table 2). Higher pre-fire fuel
loading in the forest units generally led to greater absolute but
lower relative fuel consumption than in the non-forest units. In
forest units, percentage consumption by fuel bed category was
greatest in herbaceous fuels (92%), followed by litter (77%),
shrubs (41%) and fine wood (29%). Percentage consumption in
non-forest units followed a similar pattern and was greatest in
herbaceous fuels (86%), followed by litter (81%), shrubs (48%)
and fine wood (41%) (Table 2).
Day-of-burn FM content ranged widely across all fuel bed
categories (Table 3). Relative to other fuel bed categories,
variance was smallest for litter fuels in both forest and nonforest units. Shrubs had the largest variance in forest units,
whereas herbs had the largest variance in non-forest units.
Fuel model comparison
In the forest sample units, total consumption averaged
61 19% as compared with the 73 12% predicted by
CONSUME and 65 17% predicted by FOFEM (Table 2). In
the non-forest sample units, total consumption averaged
75 16% compared with the 81 6% predicted by CONSUME
and 97% 3% predicted by FOFEM (Table 2). Overall,
CONSUME and FOFEM made similar predictions of total fuel
consumption (Fig. 4) with no significant bias in model residuals
(Table 4). Based on calculated regions of indifference, predicted
R. D. Ottmar et al.
Table 3. Fuel moisture content (% of dry weight) and standard
deviation (s.d.) of pre-fire fuel bed categories including herb, shrub,
fine wood (combined 1-, 10-, 100-h time-lag classes) and litter
One set of fuel moisture samples was associated with more than one sample
unit if time to collect samples was limited
Sample unit ID
Fuel moisture (%)
Herb
Mean
2008_Dubignon East
2008_North Boundary
2008_Turkey Woods
2008_307B
2008_608A
2011_608A_NW
2011_608A_SW
2011_608A_SE
2011_703CA
2012_L2FB
2012_L1GC
2012_L2GD
2012_S3
2012_S4
2012_S5
2012_S7
2012_S8
2012_S9
s.d.
Shrub
Mean
s.d.
Fine wood
Mean
Forest units
39.4 8.5 12.2 1.0 44.2
54.7 9.6
57.2
48.9 18.5
–
22.5
31.1 9.0 64.1 63.3 23.0
18.6 4.6 32.0 19.8 22.6
15.3 5.5 12.6 0.9 50.2
–
12.6 0.9 56.4
39.3 11.1 79.0 66.6 78.3
30.7 6.2 83.2 59.0 61.4
92.7 58.5 118.6 43.2 18.5
Non-forest units
93.4 54.8 131.6 35.0
82.2 54.7 131.0 22.9
108.8 61.3 142.8 22.1
109.7 61.3 144.7 31.3
105.1 57.2 167.7 55.6
100.7 55.7 124.6 46.1
102.7 65.4 124.3 27.4
106.7 71.7 123.0 34.3
Litter
s.d.
Mean
s.d.
6.7
25.9
10.4
9.8
8.9
13.1
23.0
22.4
25.2
2.9
18.8
19.8
11.7
9.5
16.4
12.6
15.1
22.7
27.7
11.4
2.3
4.3
1.3
4.2
4.7
5.0
3.9
5.6
13.3
4.1
10.7
8.5
6.4
8.2
8.6
13.3
9.8
7.7
5.2
2.5
1.0
6.4
5.6
3.2
1.2
0.2
A
Includes 2011_703C_E and 2011_703C_W.
Includes 2012_L2F_HIP1, HIP2, and HIP3.
C
Includes 2012_L1G_HIP1, HIP2, and HIP3.
D
Includes 2012_L2G_HIP1, HIP2, and HIP3.
B
values can be considered to be statistically equivalent
within 1 Mg ha1.
CONSUME and FOFEM accurately predicted herbaceous
fuel consumption in the combined forest and non-forest sample
units with R2 values of 0.92 (P , 0.01) in each simple linear
regression model. CONSUME predictions had no significant
bias whereas FOFEM predictions had a significant positive bias
(Table 4, Fig. 5). Regions of indifference were narrow; predicted
values can be considered equivalent to measured values within
0.10 Mg ha1 in CONSUME and 0.20 Mg ha1 in FOFEM
(Fig. 4).
Predicted shrub consumption in both CONSUME and
FOFEM was significantly related to measured consumption
(R2 ¼ 0.68 and 0.71 respectively). However, models significantly overpredicted shrub consumption and had a significant
positive bias (Fig. 5). Regions of indifference were wide;
predicted and measured shrub consumption can be considered
statistically equivalent within 0.45 Mg ha1 for CONSUME
and 0.35 Mg ha1 for FOFEM predictions (Fig. 4).
CONSUME predictions of fine wood consumption were
weakly related to measured consumption, with R2 values of
0.27 (P , 0.01) and with a significant positive bias in model
residuals (Figs. 4, 5). Even though FOFEM predictions were not
correlated with measured consumption, they provided a closer
RxCADRE fuel and ash measurements
Int. J. Wildland Fire
Total
Herb
17
Shrub
3.5
2.0
3.0
8
2.5
1.5
6
2.0
1.0
1.5
4
1.0
0.5
Measured (Mg ha1)
0.5
2
0
0
2
4
6
8
0
0.5
1.0
Fine wood
1.5
2.0
0
0.5 1.0 1.5 2.0 2.5 3.0 3.5
Litter
3.0
8
2.5
6
2.0
1.5
CONSUME
FOFEM
4
1.0
2
0.5
0
0
0
0.5
1.0
1.5
2.0
2.5
3.0
0
4
2
6
8
Predicted (Mg ha1)
Fig. 4. Predicted versus measured fuel consumption (Mg ha1) for total, herb, shrub, fine wood (,7.6 cm in diameter) and litter categories. Open
symbols represent CONSUME predictions, and closed symbols represent FOFEM predictions. Regions of indifference, for which predicted and
measured consumption can be considered statistically equivalent, are represented by solid lines for CONSUME and hatched lines for FOFEM.
Table 4. Model fit, bias and equivalence of comparisons between measured consumption values and predicted values from
CONSUME and FOFEM from the combined forest and non-forest units
P-values for all model fits were ,0.01
Fuel bed category
Total
Herb
Shrub
Fine wood
Litter
Model fit (R2)
Regions of indifference (Mg ha1)
Bias
CONSUME
FOFEM
CONSUME
FOFEM
CONSUME
FOFEM
0.5118
0.9213
0.6762
0.2665
0.7084
0.5133
0.9211
0.7100
0.0000
0.8010
No bias P ¼ 0.0794
No bias P ¼ 0.4702
Positive P ¼ 0.0056
Positive P ¼ 0.0071
No bias P ¼ 0.0653
No bias P ¼ 0.0776
Positive P ¼ 0.0089
Positive P ¼ 0.0370
No bias P ¼ 0.4430
No bias P ¼ 0.0614
1.00
0.10
0.45
0.60
0.65
0.95
0.20
0.35
0.30
0.60
estimate to measured values than CONSUME based on calculated regions of indifference (Table 4).
CONSUME and FOFEM predictions of litter consumption
are both significantly related to measured litter consumption
(R2 ¼ 0.71 and 0.80 respectively; P , 0.01). Neither model
prediction was significantly biased, but CONSUME tended to
underpredict consumption whereas FOFEM tended to overpredict consumption (Fig. 5). Models were comparable in estimated
18
Int. J. Wildland Fire
R. D. Ottmar et al.
CONSUME
FOFEM
Predicted – Measured (Mg ha1)
4
2
0
2
4
Total
Herb
Shrub
Fine wood
Litter
Total
Herb
Shrub
Fine wood
Litter
Fig. 5. Boxplots of residuals (predicted – measured) in Mg ha1 for comparisons with CONSUME and FOFEM.
equivalence with measured values; regions of indifference
were 0.65 and 0.60 Mg ha1 for CONSUME and FOFEM
respectively (Table 4).
Post-fire surface cover fractions
Post-fire surface cover was primarily composed of litter
(53 17%) and mineral soil (mean 39 18%) with minor
contributions of green vegetation (mean 3 4%) and white ash
(4 3%). Variability between sample units was high for all
cover fractions: mineral soil cover ranged from 4 to 81%, white
ash cover from 1 to 28%, litter cover from 14 to 93%, green
vegetation cover from 0 to 10%, and the black char fraction in
the post-fire plots ranged from 13 to 90%. Litter was the dominant post-fire cover fraction in the forest units, given the higher
fuel loads than in the non-forest units, whereas mineral soil was
more prevalent in the non-forest units (Fig. 6). Although a
median of 80% of the plot area burned, slightly less than half
(48 19%) of the post-fire litter and soil components were
charred because most exposed mineral soil remained uncharred.
White ash cover was greater in the forest than in the non-forest
units (Fig. 6) owing to the greater availability of mostly fine
woody fuels to consume.
All post-fire surface cover fractions were significantly correlated with pre-fire and post-fire fuel loadings, except green
vegetation (Table 5). All cover fractions were significantly
correlated with absolute consumption, but only soil cover was
significantly correlated with relative consumption. Exposed
mineral soil correlated highly with fuel loading measured prefire (^
r ¼ 0.79, P , 0.001) and post-fire (^
r ¼ 0.77, P , 0.001)
(Table 5). However, unlike white ash, mineral soil cover varied
highly before and after the fire (Fig. 6). If the pre-fire soil cover
fraction is subtracted from the post-fire cover fraction to calculate fractional cover change, then the Spearman (^
r ) correlations
with pre-fire and post-fire fuel loadings decrease to slightly
less-than-significant correlations of 0.69 (P ¼ 0.07) and
0.71 (P ¼ 0.06) respectively. The increase in mineral soil
cover caused by the fire correlates highly with relative consumption (^
r 5 0.79, P ¼ 0.03), particularly relative consumption of the woody (^
r 5 0.90, P ¼ 0.005) and litter (^
r 5 0.92,
P ¼ 0.001) fuel bed categories. Measured pre-fire and post-fire
fuel loadings were indicators of consumption: absolute fuel
consumption was significantly correlated with pre-fire fuel
loads (^
r ¼ 0.83, P , 0.001), especially in the litter category
(^
r ¼ 0.86, P , 0.001), whereas relative fuel consumption was
significantly correlated with post-fire fuel loading (^
r ¼ 0.87,
P , 0.001) but not to any particular fuel bed category. From
among the surface cover materials, white ash cover was most
significantly correlated with absolute consumption (Spearman’s
^ ¼ 0.76, P , 0.001) and was nearly as highly correlated with
r
pre-fire fuel loading (^
r ¼ 0.73, P , 0.001) (Table 5). White ash
cover was most strongly correlated with the herbaceous, woody
and litter categories of pre-fire fuel loads, and the herb and litter
categories of consumption (Table 6).
Retrospective fuel load and consumption predictions
Given these high correlations and the fact that white ash is the
first-order product of complete combustion, we selected white
ash as the surface cover fraction from which to retrospectively
predict fuel loading and consumption. Because the variables
were not normal according to the Shapiro–Wilk normality test
(ash: W ¼ 0.6081, P , 0.001; fuel load: W ¼ 0.8709, P ¼ 0.003;
consumption: W ¼ 0.8913, P ¼ 0.007), we applied natural
RxCADRE fuel and ash measurements
Int. J. Wildland Fire
Ash
Soil
Litter
Vegetation
19
Char
Dubignon East (F)
North Boundary (F)
Turkey Woods (F)
307B (F)
608A (F)
608A – NW (F)
608A – SW (F)
608A – SE (F)
703C – W (F)
703C – E (F)
L2F (F)
L2F – HIP 1 (F)
L2F – HIP 2 (F)
L2F – HIP 3 (F)
L1G (N)
L1G – HIP 1 (N)
L1G – HIP 2 (N)
L1G – HIP 3 (N)
L2G (N)
L2G – HIP 1 (N)
L2G – HIP 2 (N)
L2G – HIP 3 (N)
S3 (N)
S4 (N)
S5 (N)
S7 (N)
S8 (N)
S9 (N)
0
100 0
50
50
Surface material (%)
100
Charred material (%)
Fig. 6. Mean (s.e. bars) percentage cover of surface materials ocularly estimated post-fire in 28 sample units. Surface cover fractions of mineral
soil, white ash, litter and green vegetation materials (left) were constrained to sum to 100%, whereas the cover fraction of charred material (right)
is the percentage of the surface that was charred.
Table 5. Spearman (^
r) correlations between surface fuel loading or consumption and post-fire surface cover fractions
Significant correlations are indicated as follows: ***, P , 0.001; **, P , 0.01; *, P , 0.05
Green vegetation (%)
Litter (%)
Black char (%)
White ash (%)
Mineral soil (%)
0.39*
0.04
0.46*
0.27
0.64***
0.61***
0.42*
0.37
0.68***
0.42*
0.58**
0.06
0.73***
0.53**
0.76***
0.15
0.79***
0.77***
0.52**
0.48*
1
Pre-fire loading (Mg ha )
Post-fire loading (Mg ha1)
Consumption (Mg ha1)
Consumption (%)
Table 6. Spearman (^
r) correlations between post-fire white ash fraction (%) and herb, shrub, woody and litter categories of pre-fire fuel
loadings and consumption
Significant correlations are indicated as: ***, P , 0.001
Pre-fire loading
(Mg ha1)
Consumption
(Mg ha1)
Herb
(Mg ha1)
Shrub
(Mg ha1)
Woody
(Mg ha1)
Litter
(Mg ha1)
0.71***
0.27
0.72***
0.80***
0.66***
0.29
0.08
0.86***
logarithm (ln) transformations to normalise the data distributions (ash: W ¼ 0.9364, P ¼ 0.09; fuel load: W ¼ 0.9554,
P ¼ 0.27; consumption: W ¼ 0.9876, P ¼ 0.98). Simple linear
regression models to predict pre-fire fuel loading and consumption from percentage white ash cover were both highly
significant (Fig. 7a, b). To adjust for logarithmic bias, correction
factors of 1.07 and 1.06 calculated by Eqn 1 were applied while
back-transforming the fuel load and consumption predictions
respectively from geometric to arithmetic scales. The resulting
predictions were within 2 Mg ha1 of observed fuel loadings
and 1 Mg ha1 of observed consumption and highly significant
for both response variables (Fig. 7c, d ).
20
Int. J. Wildland Fire
R. D. Ottmar et al.
(a)
(b)
In(Consumption (Mg ha1))
In(Pre-fire loading (Mg ha1))
2.5
2.0
1.5
1.0
0.5
0
y 0.44103x 1.00786
Adj. R 2 0.52
P 0.0001
1:1
0
0.5
1.0
1.5
2.0
2.5
2.0
1.5
1.0
0.5
0
1:1
0
2.5
y 0.40432x 0.61661
Adj. R 2 0.51
P 0.0001
0.5
In(Ash (%))
1.5
2.0
2.5
In(Ash (%))
(c)
(d )
10
8
6
4
2
0
RMSE 1.97
Adj. R 2 0.46
P 0.0001
1:1
0
2
4
6
8
10
12
1
Predicted fuel loading (Mg ha )
Observed consumption (Mg ha1)
12
Observed fuel loading (Mg ha1)
1.0
8
6
4
2
0
RMSE 1.02
Adj. R 2 0.59
P 0.0001
1:1
0
2
4
6
8
1
Predicted consumption (Mg ha )
Fig. 7. Simple linear regression models predicting (a) pre-fire fuel loading, and (b) fuel consumption from post-fire white ash cover (%).
Natural log-transforms were used to normalise the log-normal distributions. Back-transformation with bias correction yielded the predicted vs
observed relationships for (c) pre-fire fuel loading, and (d ) fuel consumption.
Discussion
Ground measurements of fuel and fuel consumed supported
other science disciplines participating in the RxCADRE study
(Butler et al. 2015; Clements et al. 2015; Hudak et al. 2015;
O’Brien et al. 2015; Rowell et al. 2015; Strand et al. 2015) and
providing a valuable dataset for fuel consumption model evaluation and advancement. The relative percentage consumption
was similar to values reported by Prichard et al. (2014) and
suggests that this validation set will be a good representation of
southern pine sites. We observed the expected variation of fuel
and fuel consumption between the forest and non-forest sites.
Fuel loading and fuel consumption on forest sites were twice
that of non-forest sites because the forest litter loading was
larger and mostly consumed during the experimental fires
(Table 2).
Variability in measured moisture content was observed and
expected because the research burns ranged from November to
March in three different years. This variability provides a range
of moisture content inputs for evaluating fuel consumption and
other fire models.
To demonstrate the use of these datasets in model validation,
pre-fire loading and fuel consumption data were used to evaluate
the fuel consumption model results in CONSUME and FOFEM.
Although a similar evaluation of CONSUME and FOFEM fuel
consumption predictions was conducted by Prichard et al.
(2014) and Wright (2013a) on southeastern pine sites, data
collected for the RxCADRE provide additional independent
points of comparison with which to evaluate the same models.
Overall, CONSUME and FOFEM offer similar predictions of
total fuel consumption. Both models tended to overpredict total
consumption (Fig. 5), but this bias was non-significant (Table 4).
Assessing individual fuel bed components, CONSUME generally overpredicted herb, shrub and fine wood consumption
whereas FOFEM generally overpredicted herb, shrub and litter
RxCADRE fuel and ash measurements
consumption. The overprediction may be partially explained by
CONSUME and FOFEM assuming homogeneous fire spread
across a continuous fuelbed. Furthermore, fuel consumption
models within CONSUME and FOFEM were derived using
more data collected from the western United States where
burning conditions, fuel loading and fuel arrangement can be
much different than in the southern United States (Ottmar 2013).
The overprediction may lead to overestimation of smoke for
smoke management planning and other potential fire effects
such as tree mortality and mineral soil exposure.
We found strong correlations between predicted and measured herb, shrub and litter consumption, and weak correlations
between predicted and measured fine wood consumption in both
models (Fig. 4). Because prescribed burns generally consume a
high proportion of herbaceous biomass, measured and predicted
herb consumption correlations are both generally high across
sites. FOFEM assumes 100% consumption (Reinhardt et al.
1997), and CONSUME uses a value of 92.7% derived from
prescribed burning experiments in grasslands (Prichard et al.
2007). Based on narrow regions of indifference in statistical
comparisons of both modelled values, predicted herb consumption would closely approximate measured values in similar sites
and burning conditions.
Predicted and measured shrub consumption values were
significantly correlated in the present study, but both models
tended to overpredict shrub consumption (Fig. 5). The wide
regions of indifference and overprediction bias in CONSUME
and FOFEM outputs indicate that actual consumption might be
considerably lower than predicted consumption, which could be
of particular concern in sites with high shrub loading. Both
model predictions of litter consumption were highly correlated
with measured values. Although neither model had significant
bias in comparison with measured values, CONSUME generally
underpredicted litter consumption whereas FOFEM’s assumption of 100% tended to overpredict consumption. In either case,
calculated regions of indifference suggest that predicted values
will be within ,0.6 Mg ha1 of measured values.
Our study highlights a need for better predictive models of
fine-wood consumption in the southeastern United States. There
was little correlation between predicted and measured fuel
consumption of fine wood. Although CONSUME predictions
were statistically correlated with measured consumption, the
model predicted very high consumption (72 and 77%) versus
actual consumption (29 and 41%) for the forest and non-forest
sample units, respectively. FOFEM predictions (18 and 29%)
have narrower regions of indifference and are probably more
suitable for providing estimates of fine wood consumption.
Post-fire surface cover fraction estimates have utility for
retrospective inferences, as we demonstrated by using white ash
fraction, the direct result of complete combustion (Smith and
Hudak 2005), and therefore the surface material most highly
correlated with consumption (Table 5), as a predictor of consumption. This result demonstrates the potential utility of white
ash cover for retrospective estimates of fuel loading and consumption on which emissions estimates are based (Jenkins
et al. 1998), especially in wildfire situations where pre-fire
fuel loading is typically unknown. The root-mean-square error
from our retrospective prediction of consumption based on
white ash cover matched that estimated with FOFEM and
Int. J. Wildland Fire
21
CONSUME (1 Mg ha1) but for much less time and effort.
White ash, however, is a minor post-fire cover constituent that
can quickly dissipate, making it important to quantify immediately post-fire, or as soon as possible before rain or wind events
disturb and redistribute it (Hudak et al. 2013a; Bodı́ et al. 2014).
Litter, soil and black char persist much longer than white ash and
have greater areal coverage, making it more feasible to quantify
them not just on the ground but remotely (Hudak et al. 2007).
This idea of scalable variables is supported by Smith et al.
(2007), who found black char fraction to be a good indicator of
tree mortality in ponderosa pine forests. Black char is indicative
of incomplete combustion and is more resistant to environmental degradation than the original biomass. This makes char
production an effective means to sequester carbon (Santı́n
et al. 2015), especially in fuel bed types with a sizeable large
woody or duff fuel component. Lewis et al. (2011) found
significant correlations between fuel consumption and post-fire
cover materials (green vegetation, litter, black char, white ash
and exposed mineral soil or rock) estimated on the ground and
by remote sensing at the 2004 Taylor Complex wildfires in
interior Alaska. In that study, the post-fire cover measure
strongly correlated with fuel consumption was green vegetation
(or the lack thereof), rather than black char or white ash. Further
research is needed to improve our understanding of the utility of
post-fire surface cover fractions for retrospective assessments of
fire effects, especially for fire severity applications (Morgan
et al. 2014).
The present research project also provided an opportunity
to modify and calibrate fuel loading and fuel consumption
inventory techniques on sites with a fairly flat and homogeneous
fuel bed. Although we used a standard sampling protocol of
destructive-sample plots, a higher concentration of plots should
be considered to reduce the error associated with fuel variability.
The next step will be to extend this research to more complex
fuel beds with greater fuel loading and spatial variability and to
burn them under more extreme conditions to provide a more
robust dataset for evaluating fire models. The fuel beds of this
study were characteristic of longleaf pine understories that are
regularly burned and therefore lacked large, dead and down
woody fuels or duff. These fuel bed components leave a thick
layer of white ash if fully combusted, or black char if not. In
types of fuel beds with large woody fuels or duff components, it
would be advisable to measure the depth of white ash and black
char, in addition to coverage, to allow the volume of white ash
and black char to be estimated. Further, bulk density measures of
white ash and black char are also recommended to convert
volumetric units to mass quantities.
The quality-assured datasets of pre-fire fuel loading, FM,
fuel consumption and surface cover fractions collected as part of
the RxCADRE project in the southeastern United States in
2008, 2011 and 2012 provide critical observational data within
six discipline areas necessary for building and evaluating fire
models. These fuel datasets are publicly available from the US
Forest Service Research Data Archive (Hudak 2014; Ottmar and
Restaino 2014).
Conclusion
These validation datasets of ground-based measurements of
fuel loading, FM content, fuel consumption and surface cover
22
Int. J. Wildland Fire
fractions provide opportunities to evaluate and advance operational fire models. The RxCADRE fuel datasets are available on
a globally accessible repository maintained by the US Forest
Service Research Data Archive (Hudak 2014; Ottmar and
Restaino 2014) for use in testing and evaluation of fuel, fire and
fire effects models.
Our datasets were used to evaluate how accurately CONSUME and FOFEM predict fuel consumption in forest and
non-forest types of the southeastern United States. Overall,
CONSUME and FOFEM offer reasonable predictions of total
fuel consumption and, in similar sites and prescription windows,
could be expected to be within 1 Mg ha1 of actual fuel consumption. Models differed only slightly in comparison with
measured values. These findings are similar to those in Prichard
et al. (2014) and suggest that either CONSUME or FOFEM
would be suitable for predicting total fuel consumption in
southeastern longleaf pine forests with similar fuel characteristics. The models are less reliable for estimating fuel consumption
for individual fuel bed components including fine wood and
shrubs. Relationships between surface fuel loadings and total
consumption (laborious measurements) and post-fire surface
cover fractions (easy measurements) produced mostly significant
correlations. In particular, we demonstrated that white ash cover
is a strong predictor of pre-fire fuel loading and surface fuel
consumption, justifying the quantification of white ash in
retrospective assessments of fuel consumption and fire severity.
Although the datasets presented can be used for model
validation, the amount of fuel consumption was quite small
for both the forest and non-forest sites because of limited fuel
loading and the narrow range of environmental conditions.
Additional fuel and fuel consumption datasets will be needed
that characterise heavier fuel loads in complex terrain burned
under a range of fuel moisture and environmental conditions in
other parts of the country. This will provide a more robust
dataset for improved model validation.
Acknowledgements
We acknowledge funding from the Joint Fire Science Program (project no.
11-2-1-11), with additional support from the Pacific Northwest, Rocky
Mountain, Northern and Southern research stations of the US Forest Service
and from the National Fire Plan. We thank James Furman, J. Kevin Hiers,
Brett Williams and the entire staff at Jackson Guard, Eglin AFB, and Lindsey
Boring and Mark Melvin at the JJERC for hosting the RxCADRE and
providing the logistical and planning support to complete the research burns.
We also thank Jon Dvorak for assistance with the fuel data collection and
reduction, along with Eva Strand, Donovan Birch, Benjamin Bright and
Ben Hornsby.
References
Agee JK (1993) ‘Fire ecology of Pacific Northwest forests.’ (Island Press:
Washington, DC)
Agee JK, Skinner CN (2005) Basic principles of forest fuel reduction
treatments. Forest Ecology and Management 211, 83–96. doi:10.1016/
J.FORECO.2005.01.034
Alexander ME, Cruz MG (2013) Are applications of wildland fire behaviour
models getting ahead of their evaluation again? Environmental Modelling & Software 41, 65–71. doi:10.1016/J.ENVSOFT.2012.11.001
Baskerville GL (1972) Use of logarithmic regression in the estimation
of plant biomass. Canadian Journal of Forest Research 2, 49–53.
doi:10.1139/X72-009
R. D. Ottmar et al.
Bodı́ MB, Martin DA, Balfour VN, Santı́n C, Doerr SH, Pereira P, Cerdá A,
Mataix-Solera J (2014) Wildland fire ash: production, composition and
eco-hydro-geomorphic effects. Earth-Science Reviews 130, 103–127.
doi:10.1016/J.EARSCIREV.2013.12.007
Brown JK (1974) Handbook for inventorying downed woody material.
USDA Forest Service Intermountain Forest and Range Experiment
Station General Technical Report INT-16. (Ogden, UT)
Brown JK, Reinhardt ED, Fischer WC (1991) Predicting duff and woody
fuel consumption in northern Idaho prescribed fires. Forest Science
37(6), 1550–1566.
Butler B, Teske C, Jimenez D, O’Brien J, Sopko P, Wold C, Vosburgh M,
Hornsby B, Loudermilk E (2015) Observations of energy transport
and rate of spreads from low-intensity fires in longleaf pine habitat –
RxCADRE 2012. International Journal of Wildland Fire 25, 76–89.
doi:10.1071/WF14154
Clements CB, Lareau N, Seto D, Contezac J, Davis B, Teske C, Zajkowski TJ,
Hudak AT, Bright BC, Dickinson MB, Butler B, Jimenez D, Hiers JK
(2015) Fire weather conditions and fire–atmosphere interactions
observed during low-intensity prescribed fires – RxCADRE 2012.
International Journal of Wildland Fire 25, 90–101. doi:10.1071/
WF14173
Cruz MG, Alexander ME (2010) Assessing crown fire potential in coniferous forests of western North America: a critique of current approaches
and recent simulation studies. International Journal of Wildland Fire
19, 377–398. doi:10.1071/WF08132
de Groot WJ, Landry R, Kurz WA, Anderson KR, Englefield P, Fraser RH,
Hall RI, Banfield E, Raymond DA, Decker V, Lynham TJ, Pritchard
JM (2007) Estimating direct carbon emissions from Canadian wildland
fires. International Journal of Wildland Fire 16, 593–606. doi:10.1071/
WF06150
de Groot WJ, Pritchard J, Lynham TJ (2009) Forest floor fuel consumption
and carbon emissions in Canadian boreal forest fires. Canadian Journal
of Forest Research 39, 367–382. doi:10.1139/X08-192
Drew MB, Kirkman LK, Gholson AK (1998) The vascular flora of
Ichauway, Baker County, Georgia: a remnant longleaf pine/wiregrass
ecosystem. Castanea 63, 1–24.
Florea LJ, Vacher HL (2009) The southeastern coastal plain: an overview. In
‘Caves and karst of the USA’. (Ed. A. N. Palmer) pp. 181–184. Available at
http://digitalcommons.wku.edu/geog_fac_pub/22 [Verified 24 July 2015]
Goebel PC, Palik BJ, Kirkman LK, West L (1997) Field guide: landscape
ecosystem types of Ichauway. Technical Report 97–1. (Joseph W. Jones
Ecological Research Center: Newton, GA)
Hardy CC, Ottmar RD, Peterson JL, Core JE, Seamon P (2001) Smoke
management guide for prescribed and wildland fire: 2001 edn. (National
Wildfire Coordination Group: Boise ID) Available at http://www.nwcg.
gov/sites/default/files/products/pms420-2.pdf [Verified 1 September 2015]
Hendricks JJ, Wilson CA, Boring LR (2002) Foliar litter position and
decomposition in a fire-maintained longleaf pine–wiregrass ecosystem.
Canadian Journal of Forest Research 32, 928–941. doi:10.1139/
X02-020
Hiers JK, O’Brien JJ, Will RE, Mitchell RJ (2007) Forest floor depth
mediates understory vigor in xeric Pinus palustris ecosystems. Ecological Applications 17, 806–814. doi:10.1890/06-1015
Hollis JJ, Matthews S, Ottmar RD, Prichard SJ, Slijepcevic A, Burrows
ND, Ward B, Tolhurst KG, Anderson WR, Gould JS (2010) Testing
woody fuel consumption models for application in Australian southern
eucalypt forest fires. Forest Ecology and Management 260, 948–964.
doi:10.1016/J.FORECO.2010.06.007
Hudak AT, Morgan P, Bobbitt MJ, Smith AMS, Lewis SA, Lentile LB,
Robichaud PR, Clark JT, McKinley RA (2007) The relationship of
multispectral satellite imagery to immediate fire effects. Fire Ecology 3,
64–90. doi:10.4996/FIREECOLOGY.0301064
Hudak AT, Ottmar R, Vihnanek B, Brewer N, Smith AMS, Morgan P
(2013a) The relationship of post-fire white ash cover to surface fuel
RxCADRE fuel and ash measurements
consumption. International Journal of Wildland Fire 22, 780–785.
doi:10.1071/WF12150
Hudak AT, Ottmar RD, Vihnanek RE, Wright CS (2013b) Relationship of
post-fire ground cover to surface fuel loads and consumption in longleaf
pine ecosystems. In ‘Proceedings of the 4th Fire Behavior and Fuels
Conference’, 18–22 February 2013, Raleigh, NC. (Eds DD Wade,
RL Fox) International Association of Wildland Fire, pp. 235–240.
(Missoula, MT)
Hudak AT (2014) RxCADRE 2008, 2011, and 2012: ground cover fractions.
USDA Forest Service, Research Data Archive. doi:10.2737/RDS-20140029. Available at http://www.fs.usda.gov/rds/archive/Product/RDS2014-0029/ [Verified 8 September 2015]
Hudak A, Dickinson M, Bright B, Kremens R, Loudermilk L, O’Brien J,
Ottmar R, Hornsby B, Ottmar RD (2015) Measurements to relate fire
radiative energy density and surface fuel consumption – RxCADRE 2011
and 2012. International Journal of Wildland Fire 25, 25–37. doi:10.1071/
WF14159
Jenkins BM, Baxter LL, Miles TR, Jr, Miles TR (1998) Combustion
properties of biomass. Fuel Processing Technology 54, 17–46.
doi:10.1016/S0378-3820(97)00059-3
Joint Fire Science Program (2014) Data set for fuels, fire behavior, smoke,
and fire effects model development and evaluation – the RxCADRE
Project. Available at http://www.firescience.gov/projects/11-2-1-11/
project/11-2-1-11_final_report.pdf [Verified 24 July 2015]
Lewis SA, Hudak AT, Lentile LB, Ottmar RD, Cronan JB, Hood SM,
Robichaud PR, Morgan P (2011) Using hyperspectral imagery to
estimate forest floor consumption from wildfire in boreal forests of
Alaska, USA. International Journal of Wildland Fire 20, 255–271.
doi:10.1071/WF09081
McNab WH, Avers PE (1994) Ecological subregions of the United States:
section descriptions. USDA Forest Service, Administrative Publication
WO-WSA-5. (Washington, DC)
Morgan P, Keane RE, Dillon GK, Jain TB, Hudak AT, Karau EC, Sikkink
PG, Holden ZA, Strand EK (2014) Challenges of assessing fire and burn
severity using field measures, remote sensing and modelling. International Journal of Wildland Fire 23, 1045–1060. doi:10.1071/WF13058
O’Brien J, Loudermilk L, Hornsby B, Hudak A, Bright B, Dickinson M,
Hiers JK, Teske C, Ottmar RD (2015) High-resolution infrared thermography for capturing wildland fire behavior – RxCADRE 2012. International Journal of Wildland Fire 25, 62–75. doi:10.1071/WF14165
Ottmar RD (2013) Wildland fire emissions, carbon, and climate: modeling
fuel consumption. Forest Ecology and Management 317, 4–50.
doi:10.1016/J.FORECO.2013.06.010
Ottmar RD, Restaino JC (2014) RxCADRE 2008, 2011, and 2012: ground fuel
measurements from prescribed fires. USDA Forest Service, Research Data
Archive. doi:10.2737/RDS-2014-0028. Available at http://www.fs.usda.
gov/rds/archive/Product/RDS-2014-0028/ [Verified 1 September 2015]
Ottmar RD, Sandberg DV (2003) Predicting forest floor consumption from
wildland fire in boreal forests of Alaska – preliminary results. Tall
Timbers Research Station Miscellaneous Publication 13, pp. 218–224.
(Tall Timbers Research Station: Tallahassee, FL)
Ottmar RD, Dickinson MB, Reinhardt E (2012) Validation of fuel consumption models for smoke management planning in the eastern regions of
the United States. Available at http://www.firescience.gov/projects/
08-1-6-01/project/08-1-6-01_final_report.pdf [Verified 24 July 2015]
Ottmar RD, Vihnanek RE, Wright CS, Hudak AT (2013) Ground measurements of fuel and fuel consumption from experimental and operational
prescribed fires at Eglin Air Force Base, Florida. In ‘Proceedings of the
4th Fire Behavior and Fuels Conference’, 18–22 February 2013, Raleigh,
NC. (Eds DD Wade, RL Fox) International Association of Wildland Fire,
pp. 227–234. (Missoula, MT)
Overing JD, Weeks HH, Wilson JP, Sullivan J, Ford RD (1995) ‘Soil survey
of Okaloosa County, Florida.’ (USDA Natural Resource Conservation
Service: Washington, DC)
Int. J. Wildland Fire
23
Parsons R, Jolly MW, Hoffman C, Ottmar R (in press) The role of fuels in
extreme fire behavior. USDA Forest Service, Pacific Northwest Research
Station General Technical Report PNW-GTR-891. (Portland, OR)
Peterson DL, Johnson MC, Agee JK, Jain TB, McKenzie D, Reinhardt ER
(2005) Forest structure and fire hazard in dry forests of the western
United States. USDA Forest Service, Pacific Northwest Research
Station General Technical Report PNW-GTR-628. (Portland, OR)
Prichard SJ, Ottmar RD, Anderson GK (2007) CONSUME user’s guide and
scientific documentation. Available at http://www.fs.fed.us/pnw/fera/
research/smoke/consume/consume30_users_guide.pdf [Verified 24
July 2015]
Prichard SJ, Karau EC, Ottmar RD, Kennedy MC, Cronan JB, Wright CS,
Keane RE (2014) Evaluation of the CONSUME and FOFEM fuel
consumption models in pine and mixed hardwood forests of the eastern
United States. Canadian Journal of Forest Research 44, 784–795.
doi:10.1139/CJFR-2013-0499
R Development Core Team (2012) R: A language and environment for
statistical computing. (R Foundation for Statistical Computing: Vienna,
Austria) Available at http://www.Rproject.org [Verified 24 July 2015]
Reid AM, Robertson KM, Hmielowski TL (2012) Predicting litter and live
herb fuel consumption during prescribed fires in native and old-field
upland pine communities of the southeastern United States. Canadian
Journal of Forest Research 42, 1611–1622. doi:10.1139/X2012-096
Reinhardt ED, Keane RE, Brown JK (1997) First Order Fire Effects Model:
FOFEM 4.0, users guide. USDA Forest Service, Intermountain Research
Station General Technical Report INT-GTR-344. (Ogden, UT)
Robinson AP (2013) R package ‘equivalence’. Available at http://cran.
r-project.org/web/packages/equivalence/equivalence.pdf [Verified 24
July 2015]
Robinson AP, Froese RE (2004) Model validation using equivalence tests.
Ecological Modelling 176, 349–358. doi:10.1016/J.ECOLMODEL.
2004.01.013
Rowell EM, Seielstad CA, Ottmar RD (2015) Development and validation
of fuel height models for terrestrial lidar – RxCADRE 2012. International Journal of Wildland Fire 25, 38–47. doi:10.1071/WF14170
Ryan KC, Knapp EE, Varner JM (2013) Prescribed fire in North American
forests and woodlands: history, current practice, and challenges.
Frontiers in Ecology and the Environment 11, e15–e24. doi:10.1890/
120329
Sandberg DV (1980) Duff reduction by prescribed underburning in Douglasfir. USDA Forest Service, Pacific Northwest Research Station Research
Paper PNW-272. (Portland, OR)
Santı́n CA, Doerr SH, Preston CM, González-Rodrı́guez G (2015)
Pyrogenic organic matter production from wildfires: a missing sink in
the global carbon cycle. Global Change Biology 21, 1621–1633.
doi:10.1111/GCB.12800
Scholl ER, Waldrop TA (1999) Photos for estimating fuel loading before and
after prescribed burning in the upper coastal plain of the southeast.
USDA Forest Service, Southern Research Station General Technical
Report GTR-SRS-26. (Asheville, NC)
Smith AMS, Hudak AT (2005) Estimating combustion of large downed
woody debris from residual white ash. International Journal of Wildland
Fire 14, 245–248. doi:10.1071/WF05011
Smith AMS, Lentile LB, Hudak AT, Morgan P (2007) Evaluation of linear
spectral unmixing and dNBR for predicting post-fire recovery in a North
American ponderosa pine forest. International Journal of Remote
Sensing 28, 5159–5166. doi:10.1080/01431160701395161 [Letter].
Strand T, Gullet B, Urbanski S, O’Neill S, Potter B, Aurell J, Holder A,
Larkin N, Moore M, Rorig M (2015) Grassland and forest understorey
biomass emissions from prescribed fires in the south-eastern United
States – RxCADRE 2012. International Journal of Wildland Fire 25,
102–113. doi:10.1071/WF14166
Stronach NRH, McNaughton SJ (1989) Grassland fire dynamics in the
Serengeti ecosystem, and a potential method of retrospectively
24
Int. J. Wildland Fire
R. D. Ottmar et al.
estimating fire energy. Journal of Applied Ecology 26, 1025–1033.
doi:10.2307/2403709
Sullivan BT, Fettig CJ, Otrosina WJ, Dalusky MJ, Berisford CW (2003)
Association between severity of prescribed burns and subsequent activity of conifer-infesting beetles in stands of longleaf pine. Forest Ecology
and Management 185, 327–340. doi:10.1016/S0378-1127(03)00223-8
Urbanski SP, Hao WM, Nordgren B (2011) The wildfire emissions
inventory: western US emissions estimates and an evaluation of
uncertainty. Atmospheric Chemistry and Physics 11, 12 973–13 000.
doi:10.5194/ACP-11-12973-2011
Waldrop TA, Goodrick SL (2012) Introduction to prescribed fires in the
Southern ecosystems. USDA Forest Service, Southern Research Station
Science Update SRS-054. (Asheville, NC)
Wilson CA, Mitchell RJ, Hendricks JJ, Boring LR (1999) Patterns and
controls of ecosystem function in longleaf pine–wiregrass savannas: II.
Nitrogen dynamics. Canadian Journal of Forest Research 29, 752–760.
doi:10.1139/X99-050
Wright CS (2013a) Fuel consumption models for pine flatwoods fuel types
in the southeastern United States. Southern Journal of Applied Forestry
37(3), 148–159. doi:10.5849/SJAF.12-006
Wright CS (2013b) Models for predicting fuel consumption in sagebrushdominated ecosystems. Rangeland Ecology and Management 66,
254–266. doi:10.2111/REM-D-12-00027.1
www.publish.csiro.au/journals/ijwf
International Journal of Wildland Fire 25(1) 48–61
doi: 10.1071/WF15090_AC
© IAWF 2016
Supplementary material
Measuring radiant emissions from entire prescribed fires with ground, airborne,
and satellite sensors – RxCADRE 2012
Matthew B. DickinsonA,M, Andrew T. HudakB, Thomas ZajkowskiC,K, E. Louise LoudermilkD,
Wilfrid SchroederE, Luke EllisonF,G, Robert L. KremensH, William HolleyI,L, Otto MartinezI,L,
Alexander PaxtonI, Benjamin C. BrightB, Joseph J. O’BrienD, Benjamin HornsbyD, Charles
IchokuF, Jason FaulringH, Aaron GeraceH, David PetersonJ and Joseph MauceriH
A
USDA Forest Service, Northern Research Station, 359 Main Road, Delaware, OH 43015, USA.
B
USDA Forest Service, Rocky Mountain Research Station, Forestry Sciences Laboratory, 1221
South Main Street, Moscow, ID 83843, USA.
C
USDA Forest Service, Remote Sensing Applications Center, 2222 W. 2300 South Salt Lake
City, UT 84119, USA.
D
USDA Forest Service, Center for Forest Disturbance Science, Southern Research Center, 320
Green Street, Athens, GA 30602, USA.
E
Department of Geographical Sciences, University of Maryland, College Park, MD 20742, USA.
F
NASA Goddard Space Flight Center, 8800 Greenbelt Road, Greenbelt, MD 20771, USA.
G
Science Systems and Applications, Inc., 10210 Greenbelt Road, Suite 600, Lanham, MD 20706,
USA.
H
Rochester Institute of Technology, Center for Imaging Science, 54 Lomb Memorial Drive,
Rochester, NY 14623, USA.
I
US Air Force, 96th Test Wing, Eglin Air Force Base, Niceville, FL 32542, USA.
J
National Research Council, 7 Grace Hopper Avenue, Monterey, CA 93943, USA.
K
Present address: Institute for Transportation Research and Education, North Carolina State
University, Raleigh, NC 27695, USA.
L
Deceased.
M
Corresponding author. Email: mdickinson@fs.fed.us
Page 1 of 15
Accessory publication 1:
Calibration procedure for single-band WASP LWIR data - incorporating spectral
sensor response and atmospheric transmission
Matthew B. DickinsonA,C and Robert L. KremensB
A
USDA Forest Service, Northern Research Station, 359 Main Road, Delaware, OH
43105, USA.
B
Rochester Institute of Technology, Center for Imaging Science, 54 Lomb Memorial
Drive, Rochester, NY 14623, USA.
C
Corresponding author: Email: mdickinson@fs.fed.us
We describe a calibration approach providing total ground-leaving radiative excitance (W
m-2), also termed fire radiated flux density (FRFD), from the response of a limitedbandpass infrared sensor. The resulting calibration relationships are specific to individual
fires because spectral atmospheric transmission data are incorporated. Here we report
calibration relationships for the RxCADRE 2011 and 2012 WASP data collections.
Calibration of the WASP longwave infrared (LWIR) sensor (8 to 9.2 μm nominal
bandpass) (Fig. AP1-1) involves relating total ground-leaving radiance (W m-2 sr-1) from
0.1 to 20 µm (accounting for almost all fire radiation) to WASP raw output (digital
number, DN) through a number of linked steps. The general calibration equation is
(1)
Page 2 of 15
where LT (W m-2 sr-1) is total ground-leaving radiance; LLWIR is detector-reaching radiance
in the passband of the WASP LWIR detector during flights over wildland fires; and b and
M are parameters relating restricted-bandpass to total radiance. The inclusion of a unit
solid angle in steradians for dimensional consistency is implied (see Palmer and Grant
2009, eq. 2.32). The WASP Indigo Phoenix LWIR camera (model IA126 LWIR) was
built by Cantronic Systems Incorporated and has quantum-well, cooled detectors. The
WASP LWIR camera has 14 bits of dynamic range and is cooled by a Stirling cooler to
about 60°K.
The first step in the calibration process is to relate DN to calculated blackbody
radiance in the bandpass of the sensor. For calibration measurements, the blackbody and
WASP sensor were placed well off the floor and surrounding areas wrapped in foil to
avoid background effects on calibration measurements. The distance between the
blackbody and sensor was set so that the blackbody filled more than 40 central pixels. In
this near extended–source configuration (NES) (Palmer and Grant 2009, section 7.6.4),
the blackbody is close enough to the sensor so that pixels are much smaller than the
heated area and radiance reaching the front of the lens is equal to blackbody radiance.
The average of pixel DN for a given blackbody temperature is calculated for only the
central region of the blackbody where temperatures are uniform. A swing-in blackbody
calibrator that is now used during WASP operation and shows that there is minimal drift
in DN associated with camera lens temperature variation during flights. The image of the
blackbody was flat-fielded to account for known variation in sensor response across the
field of view. A 2-ms integration time was used in WASP flight operations in order to
Page 3 of 15
avoid saturation yet provide as much background radiance information as possible during
fire imaging missions.
Radiance leaving the blackbody and causing a response by the detector (detector-reaching
radiance, LLWIR) is a function of blackbody temperature and emissivity along with
properties of the sensor. LLWIR is determined over a range of blackbody temperatures
(280–1601°K) through integration of Planck’s radiation law,
,
,
(2)
where the integral is evaluated from 0 µm to λmax (20 µm), ε is blackbody emissivity
(0.95), tL is proportional transmission of infrared radiation through the lens (0.98), T is
blackbody temperature (K), Lλ is spectral radiance (W m-2 µm-1 sr-1), and Rλ is
proportional sensor spectral response (Fig. AP1-1). LLWIR is then related to DN, which is a
second-order polynomial in the case of the WASP longwave infrared sensor (Fig. AP12):
2 10
0.0176
(3)
The parameters b and M relating LT to LLWIR in Eqn 1 are estimated from the output of
10,000 simulations of total radiative excitance from mixed-temperature fire pixels
(Kremens and Dickinson 2014) (Fig. AP1-3). Estimates of total and restricted bandpass
(i.e., LWIR) excitance were based on randomly generated assemblages of 30 sub-pixel
areas representing the pre-frontal fuel bed, the flaming front, and the zone of post-frontal
combustion and cooling. Different sub-pixel areas are defined by their temperatures and
emissivities. Total excitance from these sub-pixel areas was calculated from the StefanBoltzmann Law and summed to give LT. LLWIR is calculated as in Eqn 2, but with the
Page 4 of 15
additional effect of atmospheric absorption so that it represents WASP LWIR detectorreaching radiance during overflights:
,
,
,
(4)
where Aλ is atmospheric spectral transmission calculated from MODTRAN (Table AP11, Fig. AP1-4).
Combining Equations 1 and 3 and converting to excitance (W m-2) by multiplication
of both sides of the question by , the form of the final calibration equation for groundleaving excitance is:
(5)
where b and M are given in Table 1 for individual fires.
References
Kremens RL, Dickinson MB (2014) Estimating radiated flux density from wildland fires
from the raw output of restricted-bandpass detectors. International Journal of
Wildland Fire, in review.
Palmer JM, Grant BG (2010) ‘The art of radiometry.’ (SPIE Press: Bellingham, WA).
doi:10.1117/3.798237
Page 5 of 15
Table AP1-1. Average atmospheric absorption estimated from MODTRAN for the large 2011 and 2012 burns
Atmospheric profile data used in MODTRAN are from soundings collected from balloons launched before ignition. Relative humidity
is averaged from the surface to 3000 m. Flight altitude used was representative of the VIIRS and MODIS overpass times. Parameters b
and M from Eqn 1 are estimated from a power-law fit to untransformed data because this approach, in contrast to linear regression on
log-log transformed data, resulted in the least bias in background fire radiative flux density (FRFD)
Launch
time
Air
Relative
Ignition time
Flight
Average
(UTC)
altitude
transmission
temperature humidity
Fire
Date
(UTC)
(°K)
(%)
Start
End
(m)
(8-9.2 µm)
b
M
703C
06 February 2011
15:00
280
28
18:24
19:02
3030
0.88
5.216
1.374
608A
08 February 2011
14:50
285
37
18:25
19:55
2250
0.90
5.138
1.374
L1G
04 November 2012 20:31
300
48
18:30
19:46
3155
0.71
7.282 1.393
L2G
10 November 2012 20:10
296
47
18:03
21:00
3160
0.75
7.006 1.380
L2F
11 November 2012 21:49
297
50
18:23
19:05
1550
0.76
6.718 1.385
Page 6 of 15
Fig. AP1-1. Spectral response of the WASP longwave infrared sensor. The nominal bandpass of
the sensor is 8–9.2 µm, approximately the full width of the spectral response at 50% response
(full width at one-half maximum) (FWHM).
Page 7 of 15
Sensor Reaching Radiance (W m-2 sr-1)
350
300
250
y = 2x10-6x2 + 0.0176x
R² = 1
200
150
100
50
0
0
2000
4000
6000
8000
10000
DN
Blackbody calibration
Field calibration
Fig. AP1-2. Calibration relationship for radiance in the WASP longwave infrared passband
determined from laboratory blackbody calibration (closed circles) and three field measurements
(open circles). Field measurements are average WASP background DN during three fires and
detector-reaching radiance estimated for the WASP passband at the observed air temperature
(see above). The polynomial regression fit only included blackbody data.
Page 8 of 15
35000
30000
LT (W m-2 sr-1)
25000
y = 3.753x1.409
R² = 0.92
20000
15000
10000
5000
0
0
100
200
300
400
500
600
LLWIR (W m-2 sr-1)
Fig. AP1-3. Power-law relationship between untransformed (total) ground-leaving and detectorreaching radiance for L2F from 10,000 fire pixel simulations. Results for other fires were similar
(Table 1).
Page 9 of 15
1
Transmission
0.8
0.6
0.4
0.2
0
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20
Wavelength (µm)
Fig. AP1-4. MODTRAN spectral atmospheric transmission for L2F based on an atmospheric
profile from a mid-morning sounding conducted prior to ignition (see Table AP1-1).
Page 10 of 15
Accessory publication 2:
Alternative methods for estimating fire radiated power from MODIS observations when
fire boundaries are known
Luke EllisonA,B,C and Charles IchokuA
A
NASA Goddard Space Flight Center, 8800 Greenbelt Road, Greenbelt, MD 20771, USA.
B
Science Systems and Applications, Inc., 10210 Greenbelt Road, Suite 600, Lanham, MD 20706,
USA.
C
Corresponding author: Email: luke.ellison@ssaihq.com
MODIS fire detections were obtained from the MYD14 active fire product (e.g., Giglio et al.
2003) and, of the four coincident MODIS overpass events for RxCADRE 2012 burns (S6, L1G,
L2G and L2F), only two (L2G and L2F) were represented in MYD14. MODIS detected two fire
pixels for the L2G burn with fire radiated power (FRP) totaling 130.9 MW, and three fire pixels
from one complete scan for the L2F burn totaling 151.9 MW (see Table AP2-1, row 1). The S6
burn signal was too small to be detected in the significantly-off-nadir MODIS pixels as can be
seen in Fig. AP2-1. Although the MYD14 algorithm detected an elevated signal for the L1G
burn, it classified it as cloud due to the significant cloud cover over or near the fire (see Fig.
AP2-1). However, in spite of the extensive cloud cover, knowing that the detected signal was
indeed from the L1G fire provided the rationale to utilize the available data to retrieve FRP. This
was done and a cumulative total FRP of 110.8 MW was found over four pixels for the L1G burn
(Table AP2-1, row 1). The extent of attenuation of FRP by cloud cover is unknown.
Page 11 of 15
In an effort to get the most realistic estimates of FRP possible, different methodologies for
calculating FRP from the MODIS data were implemented for the fires. For each burn, FRP was
calculated in two modes: by summation of the individually retrieved fire detections, and by
clustering together all of the MODIS pixels touching the burn blocks prior to FRP retrieval
(Table AP2-1, column ‘Mode’). In addition to using the MYD14 default background
characterizations, the small number of burns in this experiment allowed us the flexibility to
manually inspect the background pixels of the MODIS fire detections. This was done to ensure
that any non-clear background pixels were properly excluded from the analysis (Table AP2-1,
column ‘Background’). The fact that the differences between rows 1 and 3 and between 2 and 4
in Table AP2-1 are relatively small corroborates the clustering methodology. These small
differences are due to the different order of calculations and the slightly different selection of
background pixels. Thus, for L1G that has a high level of cloud contamination in the
background, the difference between the pixel and cluster modes using the default background
characterization is greater than the others. Having corroborated the clustering technique, more
complete FRP values could be obtained by clustering all the pixels containing any portion of the
burn plot on the ground (Table AP2-1, column ‘Cluster Size’).
The official MYD14 product corresponds to the first row of Table AP2-1, although the value
for L1G was not available in the product but was calculated in this study based on the MYD14
algorithm. When a manual inspection of the background was done to ensure that none of the
selected background pixels were contaminated (by clouds, water, smoke, significant shadows,
etc.), the FRP values for L2G and L2F remained in close agreement with the MYD14 product
(Table AP2-1, row 2). However, the stricter manual implementation of cloud detections yields a
noticeable decrease in FRP for L1G to 94.4 MW. This value is an underestimate of L1G FRP
Page 12 of 15
because, although the background was properly classified, the fire pixels still contain many
clouds that lower the fire signal. Because we have confidence in the clustering method (see
above), we used this method to calculate FRP for all pixels that overlap the burn blocks (see Fig.
AP2-1). The clustering method yields values shown in Table AP2-1, rows 5 and 6, that are
significantly higher than the corresponding prior estimates. Note that the cluster analysis was not
successful for L1G in this case due to the great variability in brightness temperatures of pixels
covering the fire because of the extensive cloud cover in that area. Therefore, the inclusion of
non-detected parts of a fire can mitigate satellite underestimation of the whole-fire FRP output.
We have the greatest confidence in the FRP estimates that are derived from the cluster method
and that manually select background pixels. These FRP estimates are 151.4 MW for L2G and
174.6 MW for L2F. Due to the increased uncertainty in the L1G case from cloud attenuation of
the fire signal that prevented FRP estimation using the whole-fire clustering technique, we can
state with confidence only that FRP from L1G was greater than 94.9 MW.
References
Giglio L, Descloitres J, Justice CO, Kaufman YJ (2003) An enhanced contextual fire detection
algorithm for MODIS. Remote Sensing of Environment 87, 273–282. doi:10.1016/S00344257(03)00184-6
Page 13 of 15
Table AP2-1. FRP values generated using different methodologies from MODIS data for
the L1G, L2G and L2F burns
Each method can be described by its FRP-retrieval ‘mode’, background characterization, and by
cluster size (if applicable). Under the ‘Mode’ column, ‘pixels’ denote hot spot determination of
individual pixels followed by aggregation of their FRP values, whereas ‘cluster’ denotes pixel
aggregation covering the fire followed by a single FRP retrieval for the whole cluster. Under the
‘Background’ column, ‘default’ denotes when the default MYD14 characterizations are used to
select the background, and ‘manual’ denotes when the background pixels are manually selected.
Under the ‘Cluster Size’ column, ‘default’ refers to the use of only pixels flagged as fire in the
‘pixels’ Mode, and ‘all’ refers to the use of all pixels that include any portion of the burn block
on the ground (see Fig. AP2-1)
Mode
Methods of generating FRP
Fire radiated power
Background
Burn unit
Cluster size
L1G
L2G
L2F
(MW)
Pixels
Default
–
110.8
130.9
151.9
Pixels
Manual
–
94.4
130.1
155.6
Cluster
Default
Default
123.8
134.7
160.8
Cluster
Manual
Default
94.9
133.7
158.5
Cluster
Default
All
149.6
152.6
179.1
Cluster
Manual
All
–
151.4
174.6
Page 14 of 15
Fig. AP2-1. Diagrams of the MODIS 1-km pixels superimposed on MODIS 250-m imagery for
units S6, L1G, L2G and L2F. The burn blocks are outlined in yellow, and the MODIS pixels that
cover all or a portion of the burn block, keeping in mind the MODIS triangular response function
that reaches halfway into the neighboring pixels along-scan, are outlined in red (Table AP2-1,
rows 5 and 6). Individual pixels whose signals were strong enough to be deemed as fire
detections are shown with a red dot (Table AP2-1, rows 1 and 2). Pixels excluded from the
background characterization are ‘X’ed out: clouds are shown in purple, water is shown in blue,
and user-selected contaminated pixels are shown in thick black.
Page 15 of 15