Fuel Models and Fire Potential from Satellite and
Surface Observations
Robert E. Burgan, retired
USDA Forest Service, Rocky Mountain
Research Station, PO Box 8089, Missoula MT 59807
e-mail:
This e-mail address is being protected from spambots. You need JavaScript enabled to view it
Robert W. Klaver
Science and Applications Branch, USGS EROS
Data Center, Sioux Falls, SD 57198
Tel. 605-594-6067; FAX 605-594-6568;
e-mail:
This e-mail address is being protected from spambots. You need JavaScript enabled to view it
Jacqueline M. Klaver
Science and Applications Branch, USGS EROS
Data Center, Sioux Falls, SD 57198
Tel. 605-594-6961; FAX 605-594-6568;
e-mail:
This e-mail address is being protected from spambots. You need JavaScript enabled to view it
Abstract
A national 1-km resolution fire danger fuel
model map was derived through use of previously mapped land cover
classes and ecoregions, and extensive ground sample data, then
refined through review by fire managers familiar with various
portions of the U.S. The fuel model map will be used in the next
generation fire danger rating system for the U.S., but it also made
possible immediate development of a satellite and ground based fire
potential index map. The inputs and algorithm of the fire potential
index are presented, along with a case study of the correlation
between the fire potential index and fire occurrence in California
and Nevada. Application of the fire potential index in the
Mediterranean ecosystems of Spain, Chile, and Mexico will be
tested.
Keywords
Fire potential; Fire danger; Fuels; Fire
model; Satellite data
Introduction
The need for a method to
rate wildland fire-danger was recognized at least as far back as
1940, in fire control conferences called by the Forest Service,
U.S. Department of Agriculture, in Ogden, Utah. By 1954 several
fire-danger rating systems were in use across the United States. In
1958 John Keetch, Washington Office, Aviation and Fire Management,
headed a team to develop a national system. By 1964 most fire
control organizations in the United States were using a "spread
index" system. In 1968 another research effort was established in
Fort Collins, Colorado to develop an analytical system based on the
physics of moisture exchange, heat transfer and other known aspects
of the problem (Bradshaw et al. 1983). The resulting fire spread
model (Rothermel 1972) was used in the first truly National Fire
Danger Rating System (NFDRS), introduced in 1972 (Deeming et al.
1972, revised in 1974). This system has since been revised twice,
in 1978 (Deeming et al. 1977) and in 1988 (Burgan 1988).
Decisions fire managers must make depend on the
temporal and spatial scales involved as well as management
objectives. Presuppression decisions are often aimed at allocation
of firefighting funds, personnel, and equipment. Such decisions
usually have a large spatial context, encompassing millions of
hectares, and a time scale of 1 to 3 days. Once a fire occurs
initial attack and suppression decisions are directed at attaining
cost-effective management of the fire. This may include a decision
to not suppress the fire if it is burning within predefined
constraints. These decisions have a spatial scale of a few thousand
hectares and a temporal scale of 24 hours or less. Once a decision
has been made to extinguish a fire, decisions are required on a
spatial scale of several hundred hectares or less and a temporal
scale of a few minutes to a few hours. The attitude toward wildland
fire in the United States is changing from that of simply
extinguishment to realization that fire must play a role in
maintaining forest health, thus the need for prescribed fires is
being recognized (Mutch 1994). Methods to assess fire potential
both strategically and tactically must also evolve.
Assessment of fire potential at any scale
requires basically the same information about the fuels,
topography, and weather conditions that combine to produce the
potential fire environment. These factors have traditionally been
measured for specific sites, with the resulting fire potential
estimates produced as alpha-numeric text, and the results applied
to vaguely defined geographic areas and temporal periods, with the
knowledge that the further one is displaced (in time or space) from
the point where such measurements have been taken, the less
applicable the fire potential estimate is. This situation is
rapidly changing because Geographic Information Systems (GIS) and
space-borne observations are greatly improving the capability to
assess fire potential at much finer spatial and temporal
resolution.
Recent improvements to fire potential assessment
technology include both broad scale fire-danger maps and local
scale fire behavior simulations. In the context of local scale fire
behavior, FARSITE (Finney 1994) and BEHAVE (Burgan and Rothermel
1984, Andrews 1986, Andrews and Chase 1989), provide methods to
simulate fire behavior for areas up to several thousand hectares.
In the broad area fire danger context, spot measurements of fire
danger, calculated using the NFDRS at specific weather stations,
are being interpolated and mapped on a national basis (Figure 1)
through the Wildland Fire Assessment System (Burgan et al. 1997)
(http://www.fs.fed.us/land/wfas/welcome.html).

Figure 1. National Fire Danger Rating System indexes are
calculated for each weather station, then the indicated staffing
levels are interpolated and mapped on a national basis
(http://www.fs.fed.us/land/wfas/fd_class.gif)
The Canadians publish similar maps for their
fire danger system on the internet
(http://www.nofc.forestry.ca/fire/cwfis) (Lee 1995) (Stocks
et al. 1989). The U.S. maps are produced using an inverse distance
squared weighting of staffing levels. Staffing level defines the
readiness status of the suppression organization. It is based on
comparison of current fire danger index values with historical
values. The staffing (or readiness) level increases as the current
index approaches historically high values. Because fire managers
across the United States have not been consistent in their
selection of an NFDR index on which to base staffing levels,
staffing level itself is the only common parameter with which to
map fire danger. Staffing level normalizes all indexes against
their historical values so it does not matter which of the several
fire danger indexes a fire manager selected. However this method
neither addresses the effect of topography on fire potential, nor
provides fire potential estimates for specific locations or
landscape resolutions.
An operational process that does provide 1
km2 landscape resolution is the Oklahoma Fire Danger
Rating System (Carlson et al. 1996)
(http://radar.metr.ou.edu/agwx/fire/intro.html), although it
still does not recognize the effect of topography. The Oklahoma
Fire Danger Rating System represents the direction of future
fire-danger systems research for the United States, but the
intensive weather network it relies upon could make this type of
system difficult for others to apply.
A wildland fuel map, terrain data, and a
reasonable sampling of weather are inputs to most fire danger
systems. This paper discusses development of a national 1
km2 fuel model map for the United States and describes a
Fire Potential Index (FPI) model that can be used to assess fire
hazard at 1 km2 resolution.
The NFDR Fuel Model Map
Traditionally 1
to 4 fire danger fuel models (Deeming et al. 1977) have been
assigned to each fire weather station. These fuel models represent
the most common or most hazardous vegetation types occurring in the
vicinity of the weather station. The exact geographic location
represented by each fuel model has not been well defined. Progress
in assessing fire potential across the landscape obviously requires
much better fuels information.
In 1991, the U.S. Geological Survey's Earth
Resources Observation Systems (EROS) Data Center, Sioux Falls,
South Dakota, prepared a 159 class, 1 km2 resolution,
land cover characteristics database (Loveland et al. 1991) that
portrayed vegetation patterns across the conterminous United
States. The initial vegetation map was produced by unsupervised
clustering of eight monthly composites of Normalized Difference
Vegetation Index (NDVI) (Goward et al. 1990) data for 1990. A
postclassification refinement was accomplished through use of
several ancillary data layers, however ground truth data was not
used. It was obvious this map could provide the basis for a
national fire danger fuel model map for the next generation
National Fire Danger Rating System. However, because the vegetation
map was designed to satisfy a wide range of applications, it was
necessary to obtain ground sample data specifically for the purpose
of developing an NFDRS fuel model map.
The first author and Colin Hardy of the
Intermountain Fire Sciences Laboratory collaborated with the EROS
Data Center to collect ground sample data for numerous locations
across the U.S. Help was enlisted from numerous federal and state
land management agencies to collect the ground data. (Burgan et
al.1999). A total of 3500 1 km2 ground sample plots were
located on seven hundred 7½ minute USGS quadrangle maps
(1:24000) (Figure 2).

Figure 2. Ground sample data was collected from 2560
plots on these 7.5 minute USGS quandrangle maps. There were up to 5
plots per quadrangle map.
Data was obtained from 2560 of these plots.
Percent cover, height, and diameter data were recorded on the four
major tree and shrub species, and percent cover and depth were
recorded for subshrubs, forbs, mosses and grass. Shrub and grass
morphology and density classes were also recorded. Up to four 35 mm
slides were taken for many of the plots. All data were entered into
a database for analysis, and the slides and graphical analysis
summaries were recorded on a CDROM and are available for viewing
with a standard browser (Burgan et al. 1997).
Because a major objective of the
ground sampling was to relate fire danger fuel models to the EROS
Land Cover Classes, a fuel model assignment was required for each
plot. The fuel model assignments were not made in the field
however, because it was felt the diversity of people involved would
produce large inconsistencies in making these assignments. Instead,
one knowledgeable person was asked to review the data sheets and
plot photographs to make the fuel model assignments, which were
then added to the database. The Land Cover Characteristics Database
also contained a map of Omernick Ecoregions (Figure 3) of the
conterminous U.S. (Omernick 1987), so the ecoregion for each plot
was also recorded. With this data, a frequency count of fuel model
by Omernick Ecoregion and Land Cover Class was obtained through a
contract with Statistical Sciences Incorporated, 1700 Westlake Ave.
N., Seattle, WA 98109. The purpose of including ecoregion data was
to permit regionalizing fuel model assignments. The fuel
model/ecoregion/landcover associations were manually inspected and
entered into a computer program that produced a 1 km2
resolution fuel model map for the conterminous U.S. The program
built the NFDR fuel model map by using the ecoregion and landcover
class values read from separate binary data files. With these
inputs a table lookup method was used to determine the fuel model
assignment for each 1 km square pixel. This became the "first
draft" NFDR fuel model map.

Figure3. Omernick ecoregions were used to localize
refinements to the NFDRS fuel model map.
Because the ground data sample size was small
for many fuel model/ecoregion/landcover combinations, some fuel
model assignments were made with inadequate data, thus it was felt
that review by fire managers from throughout the U.S. was
necessary. This was accomplished by having individual fire managers
come to the Intermountain Fire Sciences Laboratory to use the GRASS
(U.S. Army Construction Engineering Research Laboratory 1988) GIS
software for detailed review of the fuel model map within their
area of knowledge. This process permitted alteration of fuel models
by Land Cover Class within individual ecoregions by modifying the
lookup table based on ecoregions and landcover class. Although
there were changes, they were surprisingly limited considering the
sparseness of the ground sample data. Fire danger fuel models E, I,
J, and K (Deeming et al. 1977) were not used. Satellite observation
of seasonal changes in vegetation greenness eliminates the need for
using model E as a winter season subsititue for model R, and the
slash models I, J, and K don't cover sufficient area to be
considered. The NFDR Fuel Model map (Figure 4) may undergo future
revisions, but the most current version is on the Forest Service
home page (http://www.fs.fed.us/land/wfas/welcome.html).
The EROS Data Center has completed a
1-km resolution land cover database for the world (Belward 1996)
(Loveland et al. In press). These
data will provide the key to development of fuel model maps for
many countries.

Figure4. The 1-km resolution fire danger fuel model map
will be used in the next generation fire danger rating system
(http://www.fs.fed.us/land wfas/nfdr_map.htm )
The Fire Potential Index Model
Justification and Inputs
The Fire
Potential Index (FPI) model was developed to incorporate both
satellite and surface observations in an index that correlates well
with fire occurrence and can be used to map fire potential from
national to local scales through use of a GIS. The primary reasons
for developing the model were: 1) to produce a method to depict
fire potential at continental scale and at 1 km resolution, 2)
provide a method of estimating fire potential that was simpler to
operate than the current U.S. National Fire Danger Rating
System.
The assumptions of the FPI model
are: 1) fire potential can be assessed if the proportion of live
vegetation is defined, and it is known how close the dead fine fuel
moisture is to the moisture of extinction, 2) vegetation greenness
provides a useful parameterization of the quantity of high moisture
content live vegetation, 3) ten hour timelag fuel moisture should
be used to represent the dead vegetation because the moisture
content of small dead fuels is critical to determination of fire
spread, and 4) wind should not be included because it is so
transitory. Thus the inputs to the FPI model are a 1-km resolution
fuel model map, a Relative Greenness (RG) map (Burgan and Hartford
1993) that indicates current vegetation greenness compared to
historical maximum and minimum values, a maximum vegetation greenness map, and 10 hour
timelag dead fuel moisture (Fosberg and Deeming 1971) . Ten hour
timelag fuels are defined as dead woody vegetation in the size
range of 0.6 to 2.5 cm in diameter. These inputs must be in raster
format and provided as byte data representing 1-km pixels. The
output is a national scale,1-km resolution map that presents FPI
values ranging from 1 to 100.
Fuel Models
In the
traditional sense, fuel models are a set of numbers that describe
vegetation in terms that are required by the Rothermel fire model.
Thus fuel models used in the U.S. National Fire Danger Rating
System have numerous parameters that define live and dead fuel
loads by size class, surface area to volume ratios of the various
size classes, heat content, dead fuel moisture of extinction, wind
reduction factors, and mineral and moisture damping coefficients.
The FPI algorithm uses just the dead fuel extinction moisture
parameter for the mapped NFDR fuel models (Table 1). Dead fuel moisture of extinction is defined as
the fine dead fuel (0.6 to 2.5 cm dia) moisture content at which
fires will no longer spread.
NFDR
Ext
Vegetation
Model Mois (%)
Represented
A
|
15
|
Western annual grasses
|
B
|
15
|
California mixed chaparral
|
C
|
20
|
Pine grass savanna
|
D
|
30
|
Southern rough
|
E
|
----
|
Hardwoods (winter)
|
F
|
15
|
Intermediate brush
|
G
|
25
|
Short needle conifers with heavy dead load
|
H
|
20
|
Short needle conifers with normal dead
load
|
I
|
----
|
Heavy logging slash1
|
J
|
----
|
Intermediate logging slash1
|
K
|
----
|
Light logging slash1
|
L
|
15
|
Western perennial grasses
|
M
|
----
|
Agricultural land
|
N
|
25
|
Sawgrass or other thick stemmed grasses
|
O
|
30
|
High pocosin
|
P
|
30
|
Southern pine plantation
|
Q
|
25
|
Alaskan black spruce
|
R
|
25
|
Hardwoods (summer)
|
S
|
25
|
Alpine tundra
|
T
|
15
|
Sagebrush-grass mixture
|
U
|
20
|
Western long-needle conifer
|
V
|
----
|
Water1
|
W
|
----
|
Barren1
|
X
|
----
|
Water1
|
1 Fire Potential Index
not calculated for this case.
Table 1. Extinction moistures used in calculating the
Fire Potential Index.
Maximum Live Ratio Map
In the original formulation of the FPI
algorithm, maximum live ratios were determined as a function of the
live and dead loads assigned to each fuel model. However, this
resulted in similar live ratios for fuel models that represent very
different vegetation types - not a realistic situation. The effect
was to overestimate the FPI in the eastern U.S. during summer, when
the vegetation is normally very green. This dilemma was resolved by
deriving a maximum live ratio map from the maximum NDVI map of the
conterminous United States, under the assumption of a direct
relationship between the two. The algorithm used is:
LRmx = 35 + 40 * (NDmx - 100)/ 80
where
LRmx = Live ratio for a given pixel when the vegetation
is at maximum greenness
NDmx = historical maximum NDVI for a given pixel.
NDVI values were scaled to range from a minimum of
100 by multiplying the standard fractional NDVI data values by 100,
then adding 100. This keeps NDVI
within the range of binary byte data (0-255), making for efficient
data compression. The value 35 is
used as a the lowest maximum percent green, even for arid areas of
the west. That is, whatever amount of
vegetation does exist, will be at least 35 percent green at its
greenest, the remainder being dead vegetation from previous years
growth. The value 40 scales the
maximum live ratio from 35% to 75% as the maximum NDVI ranges from
100 to 180, the highest value recorded for the conterminous
U.S.
Figure 5. Maximum live ratio map for the conterminous
U.S.
The live ratios for the current date are
determined as a function of the current Relative Greenness for each
pixel, thus seasonally modifying the live/dead ratio. The 1-km fuel
model map of the U.S. provides a key to the dead fuel extinction
moisture value for each pixel.
Relative Greenness
Relative
greenness is derived from the Normalized Difference Vegetation
Index (NDVI) (Goward et al. 1990) which is calculated from data
obtained by the Advanced Very High Resolution Radiometer (AVHRR) on
board the National Oceanic and Atmospheric Administration's TIROS-N
series of polar-orbiting weather satellites. The basis for
calculating RG is historical NDVI data (1989 to present) that
defines the maximum and minimum NDVI values observed for each
pixel. Thus RG indicates how green each pixel currently is in
relation to the range of historical NDVI observations for it. RG
values are scaled from 0 to 100, with low values indicating the
vegetation is at or near its minimum greenness. Specifically the
algorithm is:
RG = (NDo - NDmn) /
(NDmx - Ndmn) * 100
where
NDo = highest observed NDVI
value for the 1 week composite period
NDmn = historical minimum NDVI value for a given
pixel
NDmx = historical maximum NDVI value for a given
pixel
The purpose of
using relative greenness in the FPI model is to define the
proportion of live and dead vegetation. The RG map has a 1-km
resolution and is registered with the fuels map.
Ten Hour Timelag Fuel Moisture
Given an
ignition source, the probability that a wildland fire will ignite
and spread is strongly dependent on the moisture content of small
dead vegetation. The U.S. National Fire Danger Rating System
separates dead fuel moisture response into timelag classes of 1,
10, 100, and 1000 hours (Deeming et al. 1977), meaning that their
moisture content will change about 2/3 of the difference between
initial and final conditions in one timelag period. Anderson
(Anderson 1985) has shown that most dead wildland vegetation
primarily involved in determining fire spread rate is in the 1 to
10 hour timelag response category, with only very fine fuels such
as cheatgrass having response times of 1 hour or less. On this
basis 10 hour timelag fuel moisture was selected to represent the
moisture content of all dead vegetation in the 1 to 10 hour timelag
size classes.
Ten hour fuel moisture is calculated
from temperature, relative humidity, and state of the weather
(cloudiness and occurrence of precipitation). These data are
measured at surface weather stations and must be extrapolated
across the landscape to meet the FPI model input requirement of
1-km resolution byte data. The process currently used to
extrapolate this point data to a 1-km grid is an inverse distance
squared algorithm. The advantage of this process is that it is
convenient and simple to perform. The disadvantage is that it does
not account for the influence of topography on fuel moisture. If
the weather station network is reasonably dense, with weather
stations at both high and low elevations, the resulting
interpolations are quite useable. But if the weather station
network is too sparse or all the weather stations are at low
elevations, the interpolations are not adequate. Improvement of the
process for calculating 10-h TLFM is the subject of further
work.
The Model
The FPI model
uses the proportion of the vegetation
that is live, and the ratio of ten hour timelag dead fuel moisture
to the moisture of extinction, for estimating relative fire
potential. The fuel model map is used to reference the dead fuel
extinction moisture for each pixel, and Relative Greenness is used
to determine the proportion of the surface vegetation that is live
(Fig 6a).

Figure 6a. Relative greenness, 10-hour fuel moisture maps,
and NFDR fuel model (fig. 4) and the maximum live ratio (fig.5)
maps are inputs to the FPI map calculation.
The FPI index is scaled from
1-100. The specific process for each
pixel is to obtain the inputs from the 1-km fuel model, Relative
Greenness, 10-h TLFM, and maximum live ratio maps, then perform the
following calculations:
Set the
FPI to a "no data" value greater than 100
(1) FPI = 105
Convert RG to a
fractional value
(2) RGf = RG/100
Relative greenness
fraction is used to determine the current live fuel ratio (LR) for
the pixel.
(3) LR = RGf *
LRmx / 100
Fractional 10-h
TLFM is normalized on dead fuel moisture of extinction
(MXd) for the fuel model, expressed as a percent (Table
1). Dead fuel moisture of extinction is defined as the dead fuel
moisture at which a fire will not spread (Rothermel 1972). It
varies from one vegetation or fuel type to another and is generally
higher for moist climates such as the southeastern U.S. Ten hour
fuel moisture (percent of dry weight) is normalized to the
moisture of extinction to produce a fractional ten hour moisture
scaled the same as fractional relative greenness (0-1). Ten
hour fuel moisture is limited to a minimum of 2 percent, thus
subtracting 2 from both the 10 hour moisture and the extinction
moisture allows TNf to reach zero when the ten
hour moisture is at its minimum value and provides a convient
method of scaling the FPI from 0 to 100. The fractional ten
hour moisture is smoothed near its minimum and maximum limits (0
and 1) to avoid discontinuities.
(4) TNf = (FM10 - 2)/(
MXd - 2)
where
TNf = fractional ten hour fuel
moisture
FM10 = ten hour moisture (percent)
MXd = dead fuel extinction moisture (percent)
The FPI
calculation is performed only if the this pixel represents a vaild
fuel model, i.e. not agriculture, barren, etc. The live ratio (LR) defines the proportion of
live vegetation, and inversely the proportion of dead vegetation
(proportion dead equals 1 minus proportion live). Because live
vegetation is green, it is assumed to have a high moisture content,
thus reducing fire potential. The
dead vegetation, as calculated from current weather data, has a
relatively low moisture content -- less than 30%. Thus the FPI can be thought of as a "dryness"
fraction times a "deadness" fraction.
(5) FPI = (1 - TNf ) * (1 - LR) * 100
where
FPI = fire potential index
Equation (5) produces FPI values
that can range from 0 to 100. The FPI will equal 0 when the
TNf is 1 (the dead fuel moisture equals the moisture of
extinction) or the LR value is 1 (the vegetation is fully green).
These circumstances do occur, but the FPI is limited to a minimum
value of 1 so that areas outside the United States can be
identified as the value 0 (no data). The FPI will attain a value of
100 if the LRis 0 (all the vegetation is cured) and the 10 hour
timelag fuel moisture is at its minimum value of 2 percent.
Fuel model map pixels that
indicate agricultural lands are assigned an FPI value of 101.
The RG image for the current composite period is processed by the
EROS Data Center in a manner to indicate clouds, so pixels
appearing cloudy in the RG map can be mapped as cloudy (102) in the
FPI map. Pixels indicated as barren lands in the fuel model map are
assigned an FPI value of 103, and marsh land pixels are assigned a
value of 104. Water pixels are assigned a value of 255. A "C"
program to perfom these calculations is available from the author.
The resulting output is a gridded raster file that can be displayed
and analyzed using a GIS, or from which a graphics image can be
prepared. Figure 6b illustrates the relationship
between the FPI map and the standard NFDR map for October 4,
1998.

Figure 6b. The standard NFDRS map is provided for comparison with
the fire potential index map
(http://www.fs.fed.us/land/wfas/exp_fp_4.gif).
Model Application
Fire Potential
Maps derived from this model were first introduced to fire managers
in California and Nevada in 1996. Their response was very
favorable, but anecdotal. In the fall of 1996 we required a simple
method to assess fire potential in Mediterranean environments as
part of a project sponsored by The Pan American Institute for
Geography and History (PAIGH) (Klaver et al. 1997). PAIGH, in
cooperation with the U.S. Geological Survey EROS Data Center, the
Instituto Geografico Nacional, Spain, the Instituto Geografico
Militar de Chile, and the Instituto Nacional De Estatistica
Geografia e Informatica, Mexico is supporting the project "Digital
Imagery for Forest Fire Hazard Assessment for the Mediterranean
Regions of Chile, Mexico, Spain, and the U.S." In support of this
effort we calculated daily FPI maps for mid-March to late October
for the years 1990-1995, and performed statistical analyses of the
correlation between fire occurrence and the FPI. The California
Division of Forestry supplied the required weather data and the
fire location data. We looked at the distribution of FPI values for
1990 -1994 in two contexts: 1) FPI for only those pixels in which a
fire occurred (Fig. 7), and 2) FPI for all the pixels within the
study area (Fig 8), which was basically California and Nevada.

Figure 7. For only those pixels in which fires occurred, in
the years 1990 to 1994, the frequency of FPI index values is
shown.

Figure 8. The frequency of pixels in the entire study area
is shown for Fire Potential Index values calculated for 1990 to
1994.
For the first case the frequency distribution of
FPI values was very similar for all years, indicating that in spite
of fire season variability the relationship between fire occurrence
and the FPI remains relatively constant. For the second case the
frequency distribution of FPI values for all pixels varied between
years, indicating that the FPI can discriminate fire season
severity in the broad geographical sense. Correlation between the
FPI and fire occurrence was very high, with r2 values by
year of: 1990, 0.44; 1991, 0.85; 1992, 0.87; 1993, 0.90; and 1994,
0.88. The r2 value for all years combined was 0.72. The
reason for the low correlation for 1990 is unknown, but could be
due to changes in calibration of the AVHRR sensor, accuracy of fire
location, or the two week rather than one week compositing
period.
Annual comparisons show that the linear
equations for the FPI and fire density were statistically identical
for 1991, 1993, and 1994 (r2=0.825, df=1 and 318, F=375.05, p=0.0).
The linear equation for 1990 was different from these years in both
slope and intercept. The linear equation for 1992 had a greater
intercept than the other years but the same slope (Figure 9). That
is, fire occurrence was greater for a given FPI value in 1992 than
for 1991, 1993, and 1994.

Figure 9. The slopes of the regression lines are very
similar for all years except 1990.
The FPI map is also being tested, along with
several NFDR indexes, for application to the problem of assessing
seasonal fire severity for the United States. This is an important
and difficult problem for which there is no standard procedure at
this time. The problem is important because millions of dollars are
made available to those Forest Service Regions that can show they
expect to experience a fire season that is considerably more severe
than average, and difficult because the decision of where to place
the additional funds must be made 2-4 weeks in advance of the
expected fire problems. The accuracy of these decisions depends on
the accuracy of long range weather forecasting, so making the
process simple in terms of weather requirements is important.
Conclusions
The FPI appears to be
strongly correlated with fire occurrence and is well adapted to
portraying fire potential across both large geographic areas and
for local areas down to a few square kilometers. It is not a
physically based model and thus requires enough historical data to
develop the statistical relationships that can provide fire
probability given a specific FPI value. Use of the FPI requires a
fuel model map, a maximum live ratio map, access to current RG maps
as calculated from AVHRR/NDVI data, and a reasonably dense network
of surface weather stations. The 10-h timelag fuel moistures must
be calculated from the weather station data and interpolated for
all 1-km pixels. Efforts are underway to improve the interpolation
procedure. The results of FPI tests for California and Nevada
indicate that it may be a valuable tool for fire managers in other
countries. This will be determined by future tests in the
Mediterranean ecosystems of Spain, Chile, Argentina and Mexico.
Acknowledgements.
Partial funding for collection of field data
required to develop the NFDR fuel map was provided by the U.S.
Department of Interior, Bureau of Land Management. The U.S.
Forest Service, the Bureau of Land Management, the U.S. Fish
and Wildlife Department, and several state forest or land
management agencies contributed time and money to collect the field
data. Without their help this effort would not have been
possible.
Update (5/2000).
The authors thank Yakov Pachepsky of RSML-USDA
for the algorithm to smooth fractional ten hour moisture to its
upper and lower limits, and R. Andres Ferreyra, Area de Sensores
Remotos, CEPROCOR (Cordoba, Argentina), for performing a
sensitivity analysis of the FPI and for suggesting a simpler method
for scaling the FPI from 0 to 100. The statistical
correlations reported here did not include this
simplification. The statistics reported here are from
an earlier algorithm that erroneously limited the maximum FPI to
about 80. Nevertheless, the authors feel the current
algorithm should be presented here. The first author (Burgan)
should be contacted for the "C" program containing the final
alogrithm.
References
Anderson, H.E. 1985. Moisture and fine forest
fuel response. In: Proceedings Eighth Conference of Fire and
Forest Meteorology (edited by L.R. Donoghue and R.E. Martin),
Detroit, Michigan, April 29-May 2, 1985. Society of American
Foresters, Bethesda, Maryland. pages 192-199.
Andrews, P.L. 1986. BEHAVE: fire behavior prediction and fuel
modeling system-BURN subsystem, Part 1. United States Department of
Agriculture, Forest Service, General Technical Report INT-194,
Intermountain Forest and Range Experiment Station, Ogden, Utah. 130
pages.
Andrews, P.L.; Chase, C.H. 1989. BEHAVE: fire behavior
prediction and fuel modeling system-BURN subsystem, Part 2. United
States Department of Agriculture, Forest Service, General Technical
Report INT-260, Intermountain Forest and Range Experiment Station,
Ogden, Utah. 93 pages.
Belward, A.S., ed., 1996. The IGBP-DIS global 1 km land cover
data set (DISCover) - proposal and implementation plans: IGBP-DIS
Working Paper No. 13, Toulouse, France, 61p.
Bradshaw, L.S.; Deeming, J.E.; Burgan, R.E.; Cohen, J.D. 1983.
The 1978 National Fire-Danger Rating System: Technical
Documentation. United States Department of Agriculture, Forest
Service, General Technical Report INT-169, Intermountain Forest and
Range Experiment Station, Ogden, Utah. 44 pages.
Burgan, R.E.; Rothermel, R.C. 1984. BEHAVE: fire behavior
prediction and fuel modeling system-FUEL subsystem. United States
Department of Agriculture, Forest Service, General Technical Report
INT-167, Intermountain Forest and Range Experiment Station, Ogden,
Utah. 126 pages.
Burgan, R.E. 1988. 1988 revisions to the 1978 National
Fire-Danger Rating System. United States Department of Agriculture,
Forest Service, Research Paper SE-273, Southeastern Forest
Experiment Station, Asheville, North Carolina. 39 pages.
Burgan, R.E.; Hartford, R.A. 1993. Monitoring vegetation
greenness with satellite data. United States Department of
Agriculture, Forest Service, General Technical Report INT-297,
Intermountain Forest and Range Experiment Station, Ogden, Utah. 13
pages.
Burgan, R.E.; Andrews, P.L.; Bradshaw, L.S.; Chase, C.H.;
Hartford, R.A.; Latham, D.J. 1997. WFAS:wildland fire assessment
system, Fire Management Notes, 57(2):14-17; 1997.
Burgan, R.E.; Hardy, C.C.; Ohlen, D.O.; Fosnight, G. 1997.
Landcover Ground Sample Data. United States Department of
Agriculture, Forest Service, General Technical Report
INT-GTR-368CD,. 1997.
Burgan, R.E.; Hardy, C.C.; Ohlen, D.O.; Fosnight, G.; Treder, R.
1999. Ground sample data for the national land cover
characteristics database. United States Department of Agriculture,
Forest Service, General Technical Report RMRS-GTR-41, Rocky
Mountain Research Station,Ogden, Utah. 12 pages.
Carlson, J.D.; Burgan, R.E.; Engle, D.M. 1996. Using the
Oklahoma mesonet in developing a near-real- time, next generation
fire danger rating system. In: 22nd Conference on Agricultural
& Forest Meteorology with Symposium on Fire & Forest
Meteorology and the 12th Conference on Biometeorology and
Aerobiology, Atlanta, Georgia, January 28-February 2, 1996.
American Meteorological Society, Boston, Massachusetts, Pages
249-252.
Deeming, J.E.; Lancaster, J.W.; Fosberg, M.A.; Furman, W.R.;
Schroeder, M.J. 1974. The National Fire- Danger Rating System.
United States Department of Agriculture, Forest Service, Research
Paper RM-84, Rocky Mountain Forest and Range Experiment Station,
Fort Collins, Colorado. 165 pages.
Deeming, J.E.; Burgan, R.E.; Cohen, J.D. 1977. The National
Fire-Danger Rating System-1978. United States Department of
Agriculture, Forest Service, General Technical Report INT-39,
Intermountain Forest and Range Experiment Station, Ogden, Utah. 66
pages.
Finney, M.A. 1994. FARSITE:a fire area simulator for fire
managers, The Biswell Symposium, February 15- 17, 1994. Walnut
Creek, California.
Fosberg, M.A.; Deeming, J.E. 1971. Derivation of the 1- and
10-hour timelag fuel moisture calculations of fire-danger. United
States Department of Agriculture, Forest Service, Research Note
RM-207, Rocky Mountain Forest and Range Experiment Station, Fort
Collins, Colorado. 8 pages.
Goward, S.N.; Markham, B.; Dye, D.G.; Dulaney, W.; Yang, J.
1990. Normalized difference vegetation index measurements from the
advanced very high resolution radiometer. Remote Sensing
Environment 35:257-277; 1990.
Klaver, J.W.; Klaver, R.W.; Burgan, Robert E. 1997. Using GIS to
assess forest fire hazard in the Mediterranean region of the U.S.
In: 17th Annual ESRI Users Conference, San Diego, CA, July 8- 11,
1997.
Lee, B.S. 1995. The Canadian Wildland Fire Information System.
In: 9th Annual Symposium on Geographic Information Systems in
Forestry, Environment and Natural Resource Management, Vancouver,
B.C., March 27-30, 1995. GIS World Inc. Fort Collins, Colorado.
Pages 639-646.
Loveland, T.R.; Merchant, J.W; Ohlen, D.O.; Brown, J.F. 1991.
Development of a land-cover characteristics database for the
conterminous U.S. Photogrammetric Engineering and Remote Sensing
57(11):1453-1463.
Loveland, T.R.; Ohlen, D.O.; Brown, J.F.; Reed, B.C.; Zhu, Z.;
Merchant, J.W.; Yang, L. In press. Western hemisphere land
cover-progress toward a global land cover characteristics database,
In: Proceedings, Pecora 13, Human Interventions with the
Environment: Perspectives From Space.
Mutch, R.W. 1994. A return to ecosystem health. Journal of
Forestry. 92(11): 31-33.
Omernick, J.M. 1987. Ecoregions of the Conterminous United
States. Annuals of the Association of American Geographers.
77(1):118-125.
Rothermel, R.C. 1972. A mathematical model for predicting fire
spread in wildland fuels. United States Department of Agriculture,
Forest Service, Research Paper INT-115, Intermountain Forest and
Range Experiment Station, Ogden, Utah. 40 pages.
Stocks, B.J.; Lawson, B.D.; Alexander, M.E.; Van Wagner, C.E.;
McAlpine, R.S.; Lynham, T.J.; Dube, D.E. 1989. The Canadian Forest
Fire Danger Rating System: An Overview. The Forestry Chronicle
65(6):450-457.
U.S. Army Construction Engineering Research Laboratory. 1988.
Geographic resource analysis support system (GRASS)-users and
programmers manual. USA-CERL ADP Report N-87/22, Environmental
Division, U.S. Army Construction Engineering Research Laboratory,
Champain, Illinois. 563 pages.