Wildfires

Identification

1.	Indicator Description

This indicator tracks wildfire frequency, total burned acreage, burn severity, and the seasonality of
wildfire activity in the United States from 1983 to 2020. Evidence suggests that the incidence of large
forest fires in the western United States and Alaska has increased since the 1980s, and it is projected to
increase further in those regions as the climate changes (USGCRP, 2017). Although wildfires occur
naturally and play a long-term role in the health of ecosystems, climate change threatens to increase the
frequency, extent, and severity of fires through increased temperatures and drought. Earlier spring
melting and reduced snowpack result in decreased water availability during hot summer conditions,
which in turn contributes to an increased risk of wildfires, allowing fires to start more easily and burn
hotter. The literature suggests that wildfires have undergone an increase in not just severity but also
season length. Due to such factors as low fuel moisture, earlier warmer temperatures, and earlier
snowmelt, the wildfire season has lengthened in many systems, and this trend is predicted to continue
(Westerling, 2016; Fill et al., 2019).

Thus, while climate change is not the only factor that influences patterns in wildfire, the many
connections between wildfire and climate make this indicator a useful tool for examining a possible
impact of climate change on ecosystems and human well-being. Wildfires are also relevant to climate
because they release carbon dioxide into the atmosphere, which in turn contributes to additional
climate change.

Components of this indicator include:

•	Wildfire frequency (Figure 1).

•	Burned acreage from wildfires (Figure 2).

•	Wildfire burn severity (Figure 3).

•	Burned acreage from wildfires by state over time (Figures 4 and 5).

•	Monthly burned acreage from wildfires over time nationwide (Figure 6).

•	Monthly burned acreage from wildfires over time in the East and West (Figure 7).

2.	Revision History

May 2014:	Indicator published.

June 2015:	Updated Figures 1 and 2 with data through 2014. Updated Figures 3 and 4 with

April 2016:
August 2016:
April 2021:

data through 2013. Split Figure 4 into Figures 4 and 5.

Updated Figures 1 and 2 with data through 2015.

Updated Figures 3, 4, and 5 with data through 2014.

Updated Figures 1 and 2 with data through 2020 and Figures 3, 4, and 5 with
data through 2018. Added Figures 6 and 7 (monthly patterns).

Technical Documentation: Wildfires

1


-------
Data Sources

3.	Data Sources

Wildfire data come from three sources:

1.	Summary data for wildfire frequency and burned acreage from 1983 through 2020 (Figures 1
and 2) are provided by the National Interagency Coordination Center (NICC), housed within the
National Interagency Fire Center (NIFC).

2.	For comparison in Figures 1 and 2, EPA obtained a data set called the United States Department
of Agriculture (USDA) Forest Service Wildfire Statistics, which provides annual frequency and
burned acreage totals through 1997 based on a different counting approach.

3.	Burn severity (Figure 3), state-by-state burn acreage (Figures 4 and 5), and monthly burned
acreage (Figures 6 and 7) data were obtained from the Monitoring Trends in Burn Severity
(MTBS) project, sponsored by the Wildland Fire Leadership Council (WFLC). The MTBS is a joint
project of the USDA Forest Service Remote Sensing Applications Center (RSAC) and the United
States Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center. Other
collaborators include the National Park Service, other USGS and USDA research facilities, and
various academic institutions. The project provides data on individual wildfire incidents that
meet certain size criteria (> 1,000 acres in the western United States or > 500 acres in the
eastern United States). These data were available from 1984 to 2018, although the analysis in
Figures 6 and 7 only extends through 2017 due to the timing of data processing.

The analysis in Figures 4 and 5 normalizes wildfire extent by the land area of each state. Land areas
come from the U.S. Census Bureau.

4.	Data Availability

NIFC data for annual trends in wildfire frequency and acreage are available from the NIFC website at:
www.nifc.gov/firelnfo/firelnfo statistics.html. These NIFC data are also mirrored in the Wildland Fire
Annual Reports from 2000 through 2020 at:

www.predictiveservices.nifc.gov/intelligence/intelligence.htm. NIFC totals are based on raw fire
incidence data reported via the Incident Command System (ICS) Incident Status Summary Reports (ICS-
209 forms). Some raw ICS-209 forms are available for individual viewing at: https://fam.nwcg.gov/fam-
web/hist 209/report list 209.

The USDA Forest Service Wildfire Statistics represent a complementary approach to compiling fire
occurrence and extent data. These statistics come from annual Forest Service reports officially known as
annual "Wildland Fire Statistics," but more commonly called "Smokey Bear Reports." These compilation
reports are based on reports submitted to the Forest Service by individual state and federal agencies,
covering land within each agency's jurisdiction. Smokey Bear Reports were provided to EPA by Forest
Service researcher Karen Short. The Smokey Bear Report extent totals that appear in Figure 2 have also
been published in Short (2015).

MTBS project analyses use raw ICS-209 form data from 1984 to 2017 as the basis for further processing.
Summary data are publicly available at: https://mtbs.gov/direct-download. This online database search

Technical Documentation: Wildfires

2


-------
tool provides detailed and comprehensive records, including the location of initial fire occurrence as a
GIS point file, burned area as a GIS polygon shape file, and burned area severity classification as a raster
file for each individual fire incident. Detailed records for this indicator were downloaded from the MTBS
website.

The U.S. Census Bureau has published official land areas for each state in the Statistical Abstract of the
United States, available online at: www.census.gov/librarv/publications/time-
series/statistical abstracts.html.

Methodology

5. Data Collection

This indicator presents four measures of wildfires over time reported on an annual basis: (1) the total
number of wildfires, (2) acreage burned by wildfires, (3) the burn severity of those fires, and (4) acreage
burned by month. For the purposes of this indicator, a wildfire is a fire that fits the following nearly
synonymous definitions:

"An unplanned, unwanted wildland fire including unauthorized human-caused fires, escaped
wildland fire use events, escaped prescribed fire projects, and all other wildland fires where the
objective is to put the fire out" (www.nifc.gov/BAER/definitions.html)

"A wildland fire originating from an unplanned ignition, such as lightning, volcanos,
unauthorized and accidental human caused fires, and prescribed fires that are declared
wildfires" (NWCG, 2020)

NWCG (2020) defines a wildland as "an area in which development is essentially non-existent, except for
roads, railroads, powerlines, and similar transportation facilities." Fire severity is defined as the "degree
to which a site has been altered or disrupted by fire; loosely a product of fire intensity and residence
time." The data for this indicator cover all 50 states.

Figures 1 and 2. Wildfire Frequency and Acreage in the United States, 1983-2020

Wildfire frequency and burn acreage data are based upon local-, state-, and national-level reporting of
wildland fire incidents submitted to NIFC via the ICS-209 form (Fire and Aviation Management and
Predictive Services, 2009). The data captured in these forms can also be submitted to NIFC using the
Incident Management Situation (SIT)-209 reporting application. The ICS-209 guidelines require that large
fires (100+ acres in timber and 300+ acres in grasslands) must be reported, but they do not set a
minimum fire size for reporting. Thus, the data set includes small fires, including some that may have
burned just a few acres or less.

Supplementary data come from annual Smokey Bear Reports, which are based on annual reports
submitted to the Forest Service by individual state and federal agencies. These original reports describe
fires taking place on land within each reporting agency's fire protection jurisdiction. The USDA Forest
Service stopped compiling Smokey Bear Reports after 1997.

Technical Documentation: Wildfires

3


-------
Figure 3. Damage Caused by Wildfires in the United States, 1984-2018

MTBS uses satellite imagery to map burn severity and perimeters of large fires (> 1,000 acres in the
western United States or > 500 acres in the eastern United States). These thresholds are applied based
on the "West" and "East" regions shown in Figure TD-1.

Figure TD-1. Region Boundaries for MTBS Size Threshold Application and Monthly Burned Acreage

Analysis

MTBS starts primarily from ICS-209 reports and solicits additional data from the states if inclusion in ICS-
209 is unclear. Other sources for fire occurrence data include federal data, the National Fire Plan
Operations and Reporting System (NFPORS), and InciWeb. These records are compiled into a
standardized project database. MTBS identifies corresponding imagery using the Global Visualization
Image Selection (GLOVIS) browser developed by the USGS EROS Center. ArcGIS shapefiles and scene-
specific Advanced Very High Resolution Radiometer (AVHRR) greenness plots are incorporated into the
viewer to aid scene selection and determination of peak periods of photosynthetic activity. Pre-fire and
post-fire images are selected for each incident. Wildfires are analyzed on the scale of individual
incidents, but the data can also be aggregated at other spatial scales for analytical purposes.

Figures 4 and 5. Average Annual Burned Acreage and Change in Burned Acreage by State, 1984-2018

Figures 4 and 5 are based on acreage data for large fires as compiled by the MTBS program through the
analytical steps described above for Figure 3. These numbers are based on ICS-209 reports and
additional state data compiled by MTBS.

Figures 6 and 7. Comparison of Monthly Burned Area Due to Wildfires in the United States Between
1984-2000 and 2001-2017

Figures 6 and 7 are based on the same MTBS acreage data for large fires that were used for Figures 3, 4,
and 5. The MTBS data set identifies each fire's date of ignition, which this indicator uses to assign each
fire to a month.

West

\

Technical Documentation: Wildfires

4


-------
6. Indicator Derivation

Figures 1 and 2. Wildfire Frequency and Acreage in the United States, 1983-2020

NIFC compiles local, state, and national reports to create the annual summary statistics published
online. Data are aggregated to provide national, state and local statistics. EPA aggregated state-by-state
totals in the annual Smokey Bear Reports to generate additional measures of annual wildfire frequency
and extent.

Figure 3. Damage Caused by Wildfires in the United States, 1984-2018

Burn severity is a qualitative measure describing the degree to which a site has been altered by fire
(NWCG, 2020). MTBS uses the Normalized Burn Ratio (NBR) to measure burn severity. NBR is a
normalized index that uses satellite imagery from Landsat 5 and/or Landsat 7 TM/ETM bands 4 (near-
infrared) and 7 (mid-infrared) to compare photosynthetically healthy and burned vegetation. Pre- and
post-fire NBR are calculated to compare vegetation conditions before and after each wildfire.

The difference between pre- and post-fire NBRs is the Differenced Normalized Burn Ratio (dNBR).
Calculated dNBR values are grouped into severity classes to give a qualitative assessment of the effects
of fire damage. Exact classification thresholds may vary by site, based on site-specific ecological
characteristics and expert judgment. The table at:

http://gsp.humboldt.edu/OLM/Courses/GSP 216 Online/lesson5-l/NBR.html (from Humboldt State
University, reprinting information originally from USGS) offers a default starting point for these
classification thresholds. Further discussion of the NBR and dNBR calculation methodology can be found
at: www.mtbs.gov/mapping-methods and the glossary links therein.

Selected satellite images are also filtered through a complex sequence of data pre-processing, perimeter
delineation, and other data quality assurance techniques. These procedures are documented in full on
the MTBS website at: www.mtbs.gov/mapping-methods.

The timing of the satellite imagery selected for analysis depends on the type of assessment that is
conducted for a given fire. The optimal assessment type is selected based on the biophysical setting in
which each fire occurs. MTBS conducts two main types of assessments:

•	Initial Assessments compare imagery from shortly before and shortly after the fire, typically
relying on the first available satellite data after the fire—on the scale of a few days. These
assessments focus on the maximum post-fire data signal and are used primarily in ecosystems
that exhibit rapid post-fire vegetation response (i.e., herbaceous and particular shrubland
systems).

•	Extended Assessments compare "peak green" conditions in the subsequent growing season with
"peak green" conditions in the previous growing season, prior to the fire. These assessments are
designed to capture delayed first-order effects (e.g., latent tree mortality) and dominant
second-order effects that are ecologically significant (e.g., initial site response and early
secondary effects).

MTBS occasionally conducts a Single Scene Assessment, which uses only a post-fire image (either
"initial" or "extended"), when limited by factors such as data availability.

Technical Documentation: Wildfires

5


-------
See: www.mtbs.gov/glossarv for a glossary of MTBS assessment terms.

Figure 3 was created by filtering MTBS's database output to remove any fires not meeting MTBS's size
criteria—although most such fires would not have been processed by MTBS anyway—and removing
fires classified as "prescribed," "wildland fire use/' or "unknown." The resulting analysis is therefore
limited to fires classified as true "wildfires."

The total acreage shown in Figure 3 (the sum of the stacked burn severity sections) does not match the
total acreage in Figure 2 because the burn severity analysis in Figure 3 is limited to fires above a specific
size threshold (> 1,000 acres in the western United States and > 500 acres in the eastern United States)
and because the graph does not include acreage classified as "outside perimeter" or "non-processing
area mask/' the latter of which denotes areas within the fire perimeter that could not be assessed for
burn severity because the imagery was affected by clouds, cloud shadows, or data gaps. The "Key
Points" text that describes the proportion of high severity acreage is based on high severity as a
percentage of total assessed acreage (i.e., the total acreage after non-processing area has been
excluded). The size threshold resulted in five states that did not have any reported fires for the time
period: Connecticut, Illinois, New Hampshire, Rhode Island, and Vermont. Puerto Rico is included in
Figure 3 totals.

Figure 4. Average Annual Burned Acreage by State, 1984-2018

To create this map, EPA divided the annual acreage burned in each state by the state's total land area.
After doing this for all years during the period of record (1984-2018), EPA calculated an average value
and plotted it on the map. The same five states without fires meeting the size threshold for Figure 3 are
colored gray to indicate insufficient data.

Figure 5. Change in Annual Burned Acreage by State Between 1984-2001 and 2002-2018

To create this map, EPA calculated each state's average annual acreage burned per square mile for the
first half of the record (1984-2001) and the average for the second half (2002-2018). EPA found the
difference between these values and expressed it as a percentage difference (e.g., average annual
acreage during the second half of the record was 10 percent higher than average annual acreage burned
during the first half). The same five states without fires meeting the size threshold for Figure 3 are
colored gray to indicate insufficient data. Changes have been characterized using this method rather
than measuring a slope over time (e.g., a linear regression) because of the length and shape of the data
set. Visual inspection of the NIFC line in Figure 2 (burned acreage across all states) suggests periods of
relative stability punctuated by a noticeable jump in acreage during the late 1990s. This jump coincides
with a period of transition in certain natural climate oscillations that tend to shift every few decades—
notably, a shift in the Pacific Decadal Oscillation (PDO) around 1998 (Peterson and Schwing, 2003;
Rodionov and Overland, 2005). This shift—combined with other ongoing changes in temperature,
drought, and snowmelt—may have contributed to warmer, drier conditions that have fueled wildfires in
parts of the western United States (Kitzberger et al., 2007; Westerling, 2016). With approximately 30
years of data punctuated by a phase transition, and with research strongly suggesting that the PDO and
other decadal-scale oscillations contribute to cyclical patterns in wildfires in the western United States,
EPA determined that linear regression is not an appropriate method of describing changes over time in
this particular indicator. Instead, EPA chose to simply compare two sub-periods in a manner that
considers all years of data and avoids inferring an annual rate of change. Without a nuanced statistical

Technical Documentation: Wildfires

6


-------
analysis to define a break point between two sub-periods, EPA chose to simply break the record into
two periods of approximately equal length: 1984-2001 and 2002-2018 (the former being longer by one
year due to the odd number of total years). The fact that the break point currently lands around the
same time as the PDO shift by this method is a coincidence. As more data are added in future years, the
"halfway" break point will move accordingly.

EPA plans to investigate opportunities for a more robust interpretation of state-level trends over time in
future editions of this indicator.

Figures 6 and 7. Comparison of Monthly Burned Area Due to Wildfires in the United States Between
1984-2000 and 2001-2017

EPA created Figures 6 and 7 by starting from the filtered data set described above for Figure 3 (i.e.,
excluding any fires not meeting MTBS's size criteria). EPA divided the period of record into two equal
halves (1984-2000 and 2001-2017) to support analysis of change over time. Changes have been
characterized using this comparative method rather than measuring a slope over time (e.g., a linear
regression) because of the variability in the data set. The timing of the burn acreage was highly variable
from year to year (even more so when viewed within specific regions of the country). Instead, EPA chose
to simply compare two sub-periods in a manner that considers all years of data and avoids inferring an
annual rate of change. Without a nuanced statistical analysis to define a break point between two sub-
periods, EPA chose to simply break the record into two 17-year periods. As more data are added in
future years, the "halfway" break point will move accordingly.

To create the graphs, EPA calculated the average annual acreage burned each month across the United
States for each half of the period of record. The date of fire occurrence used in this analysis corresponds
to the ignition date of the fire as defined in the MTBS data set. For this figure, fires are aggregated by
month. For Figure 7, the calculation was conducted separately for fires in the East and West, as defined
by the MTBS program and shown in Figure TD-1.

Comparison of Sources

Figure TD-2 compares total wildfire extent estimates from NIFC, Smokey Bear Reports, and MTBS. This
graph shows that MTBS estimates follow the same pattern as the NIFC data set but are always
somewhat lower than NIFC's totals because MTBS excludes small fires. The graph also shows how the
most recent MTBS estimates compare with the MTBS data release used in the previous update to this
indicator. As expected, the data show evidence of revisions to historical data, but the changes are not
extensive.

Technical Documentation: Wildfires

7


-------
Note: These MTBS totals exclude small fires and RX, WFU, and UNKfire types. They include acreage classified as "non-
processing area mask," so they will differ from the totals implied by Figure 3.

Indicator Development

EPA added Figures 6 and 7 to this indicator in 2021 as a result of a new indicator development effort
that focused on wildfire season characteristics. EPA used wildfire activity data because it was nationally
available with a reasonable data record length (consistently collected since the early 1980s). However,
there are many analytical approaches and definitions for examining the wildfire season that have been
used in the peer-reviewed scientific literature, most of which have been specific to regional conditions
including ecosystem type, vegetation, and fire regimes. EPA plans to continue to research methods and
data for examining long-term changes in the seasonality of wildfires (onset, peak, end, and duration).

NIFC's website provides data from 1960 through 2020, and Smokey Bear Reports are available from
1917 to 1997. The data available prior to the early 1980s, however, provide incomplete geographic
coverage, as fire statistics at the time were not compiled from the full extent of "burnable" lands. Thus,
Figures 1 and 2 of this indicator begin in 1983, which was the first year of nationwide reporting via ICS-
209 reports. Figures 3 through 7 begin in 1984, which was the first year for which the MTBS project
conducted its detailed analysis. MTBS depends on aerial imagery and the level of detail captured
consistently in ICS reports. Thus, while a longer period of record would be desirable when analyzing
long-term changes in a climatological context, EPA could not extend this indicator with pre-1983 data
without introducing inconsistencies and gaps that would preclude meaningful comparisons over time
and space.

For more discussion regarding the availability, coverage, and reliability of historical wildfire statistics, see
the authoritative discussions in Short (2013) and Short (2015). Based on these sources, Table TD-1
summarizes the available data sets, their coverage, and their underlying sources. NIFC's pre-1983
estimates actually derive from the Smokey Bear Reports (Short, 2015); therefore, in reality, the Smokey
Bear Reports are the only underlying nationwide source of pre-1983 wildfire statistics.

Technical Documentation: Wildfires

8


-------
Table TD-1. Comparison of Historical Wildfire Data Sources

Data set

Variables

Temporal
range

Resolution

Geographic
coverage

Underlying sources

NIFC

(Figures 1 and 2)

Acreage and
incidence
(number of
fires)

1983-2020

Annual

National

ICS incident reports

NIFC pre-1983

Acreage and
incidence

1960-1982

Annual

National
with gaps

Smokey Bear Reports,
which are based on
estimates submitted
by various agencies

Smokey Bear
Reports

(recent data in
Figures 1 and 2)

Acreage and
incidence

1917-1997

Annual

National
with gaps

Estimates submitted
by various agencies

MTBS

(Figures 3
through 7)

Burn
severity;
acreage by
state

1984-2018

Annual

National

ICS incident reports

A fundamental shift in wildfire reporting took place in the early 1980s with the onset of the ICS reporting
system. Prior to this time, reports were submitted to the USDA Forest Service by selected state and
federal agencies, covering land within each agency's jurisdiction. Many of these reports were limited to
fires on land with "protected" status (i.e., land designated for cooperative fire control). Fires occurring
on "unprotected" land would not necessarily be fought, and they would not be counted in the statistics
either. Figure TD-3 below, based on data obtained from the USDA Forest Service (Short, 2015),
demonstrates how the reporting area was well below the total nationwide "burnable" acreage until the
1980s. Increases in the reporting area occurred when additional agencies joined the reporting program.
For example, the Bureau of Land Management began reporting around 1964, which accounts for the
noticeable jump in the blue line in Figure TD-3.

Technical Documentation: Wildfires

9


-------
Figure TD-3. "Smokey Bear Reports" Reporting Area Over Time

2,000

1,600

J 1,200

1


-------
processed following existing USGS-EROS protocols. Data obtained from MTBS were also cross-checked
prior to conducting analyses for this indicator.

Analysis	

8.	Comparability Over Time and Space

NIFC methods and statistics have not changed since 1983, and they can be compared on an annual basis
at national scales. The sole exception is the NIFC fire count and burned acreage data points for the year
2004, which are missing totals from state lands in North Carolina. Thus, these two points slightly
compromise the comparability over both time and space. Smokey Bear Reports also used consistent
methods from 1983 to 1997, and they covered the full extent of "burnable" U.S. lands throughout this
period. MTBS has used consistent methods to classify burn severity from 1984 through 2018, allowing
for annual comparisons through time and allowing for spatial comparisons among states. MTBS is based
on a type of satellite imagery that has been collected consistently with sufficient resolution to support
this analysis throughout the period of record.

Figures 3 through 7 were derived from an MTBS data set that uses different size thresholds for different
states. This data set includes fires > 1,000 acres in the western United States and > 500 acres in the
eastern United States. Thus, the indicator might undercount small fires in the West, compared with the
East. However, because total burned area is dominated by large fires, this is not expected to heavily
affect the calculations underlying Figures 3 through 7. In addition, these thresholds have held consistent
for each region over time, which lends validity to the analysis of regional trends over time.

9.	Data Limitations

Factors that may impact the confidence, application, or conclusions drawn from this indicator are as
follows:

1.	Wildfire activity can be influenced by a variety of other factors besides climate. Examples
include changes in human activities and land management strategies over time, particularly
changes in fire suppression and fire management practices, which (among other things) can
potentially contribute to more damaging fires in the future if they result in a buildup of fuel in
the understory. Resources available to fight and manage wildfires can also influence the amount
of area burned over time. Grazing activities can also influence the amount and type of
vegetation in the landscape, and changes in land cover and land use—for example, forest to
non-forest conversion—can affect the extent and type of "burnable" land. Thus, further analysis
is needed before an apparent change in wildfire activity can necessarily be attributed to climate
change.

2.	The dominant drivers of wildfire activity can vary by region. Contributing factors may include
(but are not limited to) temperatures in specific seasons (particularly spring), drought, and
precipitation that contributes to vegetation growth. As described in Section 6, wildfire trends in
some regions have been linked with certain phases of multi-year and multi-decadal climate
oscillations (Kitzberger et al., 2007; Westerling et al., 2006). Climate patterns that lead to more
wildfire activity in some parts of the United States may lead to a simultaneous decrease in
activity in other regions (e.g., the Northwest versus the Southwest). Reconstructions based on

Technical Documentation: Wildfires

11


-------
tree rings can provide hundreds of years of context for understanding such patterns and how
they vary regionally (e.g., Swetnam and Betancourt, 1998).

3.	While this indicator is officially limited to "wildland" fires, it includes fires that encroach on—or
perhaps started in—developed areas at the wildland-urban interface (WUI). Encroachment of
the WUI over time into previously wild lands could influence trends in wildfire frequency and
extent (Radeloff et al., 2005).

4.	NIFC data, which are derived from government entities of varying scope or jurisdiction, can be
limited by inconsistencies across how data are reported through ICS-209 forms. With
aggregation from so many potential sources, wildfire incidence data, particularly historical data,
may be redundant or erroneous. Data aggregation among sources may result in variability in
reporting accuracy and protocol.

5.	The MTBS program depends on certain conditions to make accurate measurements of burn
severity:

•	Accurate fire location coordinates that match burn scars visible via satellite.

•	Accurate fire size information that ensures that fires meeting the MTBS size criteria are
properly included.

•	Accurate date of ignition and out date that guide the appropriate selection of imagery,
particularly for baseline assessments.

•	Pre-fire and post-fire images that are cloud-free to avoid visual obscuration of the fire area.

6.	Some fires of very low severity may not be visible in the satellite imagery and therefore
impossible to delineate or characterize. Cloud cover, cloud shadow, or data gaps can also
preclude damage assessment. To account for all of these limitations, the MTBS project includes
a burn severity classification of "non-processing area mask." This classification accounts for
approximately 5.0 percent of the total wildfire acreage from 1984 through 2018.

10.	Sources of Uncertainty

Uncertainties in these data sets have not been quantified. The most likely sources of uncertainty relate
to initial data collection methods. Federal land management agencies have varying standards for
content, geospatial accuracy, and nomenclature. Duplicate records occur due to reporting of a given
incident by multiple agencies, such as redundant reports from local, state, or federal entities. In any
given year, as much as three-quarters (or more) of all fire incidents are reported by non-federal state
and local agencies (for example, see NICC [2020]). Cases of gross geospatial inaccuracies may also occur.
Similar inconsistencies occur within state databases; however, the MTBS project addresses issues such
as duplicates and nomenclature during pre-processing. Given the comprehensive reporting
infrastructure that has been in place over the past several decades, it is unlikely that any large fires
would have been missed.

11.	Sources of Variability

Forest conditions, and therefore wildfire incidents, are highly affected by climate conditions. In addition
to year-to-year variations, evidence suggests that wildfire patterns in the western United States are
influenced by multi-year and multi-decadal climate oscillations such as the Pacific Decadal Oscillation

Technical Documentation: Wildfires

12


-------
(www.ncdc.noaa.gov/teleconnections/pdo) and the Atlantic Multidecadal Oscillation
(www.aoml.noaa.gov/phod/amo faq.php). For example, see Kitzberger et al. (2007) and Westerling
(2016) for discussion of warmer, drier conditions that have contributed to increases in wildfire activity in
certain regions.

Changes in the frequency of wildfire triggers (e.g., lightning, negligent or deliberate human activity)
could also affect wildfire frequency and the timing of wildfire activity. Burn severity is affected by local
vegetation regimes and fuel loads. Finally, improvements or strategic changes in firefighting and fire
management may affect wildfire prevalence, size, and resulting damages. Forest management practices
have changed over time from complete fire suppression to controlled burns. These varied approaches
and the extent to which they are applied on state or regional levels can influence the wildfire data
presented in this indicator.

12. Statistical/Trend Analysis

As described in Section 6, the nature of this topic and the length and shape of the time series suggest
that linear regression is not a suitable tool for characterizing long-term wildfire trends and their
significance. Thus, the figures and key points do not report regression results. Ordinary least-squares
linear regressions from the NIFC data set have been calculated here, just for reference. Regression
slopes and p-values are indicated in Table TD-2 below.

Table TD-2. Wildfire Regression Statistics

Indicator component

Regression slope

P-value

NIFC fire frequency (Figure 1)

+63 fires/year

0.809

NIFC burn acreage (Figure 2)

+180,496 acres/year

<0.001

EPA used a two-tailed paired t-test to compare the monthly burned acreage time series (the lines in the
graphs in Figures 6 and 7) between the two time periods. This testing treats each period as a series of 12
monthly means (the 12 months—each representing the mean from a sample set of 17 years' worth of
data); the paired aspect means that the test compares the January mean from the first period to the
January mean from the second period, then does the same for February and so on. The paired t-test
offers a simple assessment of whether wildfire extent (burned acreage) over the course of the year has
increased or decreased to a statistically significant degree. Differences were significant to a 95 percent
level (p < 0.05) in the United States as a whole (Figure 1; p-value of 0.024 for the comparison) and the
West (Figure 2; p = 0.025). The East had an insignificant difference between the two periods (p = 0.320).

Use of a simple t-test inevitably requires some assumptions about the shape and underlying
characteristics of the data. This test is used to provide initial approximations about the statistical
attributes of this indicator. One strength of the testing described here is that it identifies regions where
there has been an increase in fire activity through much of the year. One weakness is that it may fail to
identify differences in regions where fire activity is concentrated in just a few months of the year,
because most months are zero or near-zero and thus show very little difference between the two time
periods.

Technical Documentation: Wildfires

13


-------
References

Dennison, P.E., S.C. Brewer, J.D. Arnold, and M.A. Moritz. 2014. Large wildfire trends in the western
United States, 1984-2011. Geophys. Res. Lett. 41(8):2928-2933.

Fill, J.M., C.N. David, and R.M. Crandall. 2019. Climate change lengthened southeastern USA lightning-
ignited fire seasons. Glob. Change Biol. 25(10):3562-3569.

Fire and Aviation Management and Predictive Services. 2009. Interagency wildland fire incident
information reporting application (SIT-209). www.predictiveservices.nifc.gov/intelligence/SIT-
209 Bus Req Final vla.pdf.

Kitzberger, T., P.M. Brown, E.K. Heyerdahl, T.W. Swetnam, and T.T. Veblen. 2007. Contingent Pacific-
Atlantic Ocean influence on multicentury wildfire synchrony over western North America. P. Natl. Acad.
Sci. USA 104(2):543-548.

NICC (National Interagency Coordination Center). 2020. Wildland fire summary and statistics annual
report, www.predictiveservices.nifc.gov/intelligence/2020 statssumm/annual report 2020.pdf.

NWCG (National Wildfire Coordinating Group). 2020. NWCG glossary of wildland fire, PMS 205. Updated
December 2020. www.nwcg.gov/glossarv/a-z.

Peterson, W.T., and F.B. Schwing. 2003. A new climate regime in northeast Pacific ecosystems. Geophys.
Res. Lett. 30(17).

Radeloff, V.C., R.B. Hammer, S.I. Stewart, J.S. Fried, S.S. Holcomb, and J.F. McKeefry. 2005. The wildland-
urban interface in the United States. Ecol. Appl. 15:799-805.

Rodionov, S., and J.E. Overland. 2005. Application of a sequential regime shift detection method to the
Bering Sea ecosystem. ICES J. Mar. Sci. 62:328-332.

Short, K.C. 2013. A spatial database of wildfires in the United States, 1992-2011. Earth Syst. Sci. Data
Discuss. 6:297-366. www.fs.fed.us/rm/pubs other/rmrs 2013 short k001.pdf.

Short, K.C. 2015. Sources and implications of bias and uncertainty in a century of U.S. wildfire activity
data. Int. J. Wildland Fire 24(7):883-891.

Swetnam, T.W., and J.L. Betancourt. 1998. Mesoscale disturbance and ecological response to decadal
climatic variability in the American Southwest. J. Climate 11:3128-3147.

USGCRP (U.S. Global Change Research Program). 2017. Climate science special report: Fourth National
Climate Assessment, volume I. Wuebbles, D.J., D.W. Fahey, K.A. Hibbard, D.J. Dokken, B.C. Stewart, and
T.K. Maycock (eds.). https://science2017.globalchange.gov. doi:10.7930/J0J964J6

Westerling, A.L. 2016. Increasing western U.S. forest wildfire activity: Sensitivity to changes in the timing
of spring. Phil. Trans. R. Soc. B. 371:20150178.

Technical Documentation: Wildfires

14


-------
Westerling, A.L., H.G. Hidalgo, D.R. Cayan, and T.W. Swetnam. 2006. Warming and earlier spring
increase western U.S. forest wildfire activity. Science 313(5789):940-943.

Technical Documentation: Wildfires

15


-------