Day of burning maps and burn severity landscape metrics in the southwestern United States 2002-2020
Data files
Mar 19, 2025 version files 142.20 MB
-
ExtremeSpread_Data.7z
142.19 MB
-
README.md
11.59 KB
Abstract
Extreme fire spread events rapidly burn large areas with disproportionate impacts on people and ecosystems. Such events are associated with warmer and drier fire seasons and are expected to increase in the future. Our understanding of the landscape outcomes of extreme events is limited, particularly whether or not they burn more severely or produce spatial patterns less conducive to ecosystem recovery. To assess relationships between fire spread rates and landscape burn severity patterns, we used satellite fire detections to create day-of-burning (DOB) maps for 623 fires comprising 4,267 single-day events within forested ecoregions of the southwestern United States. We related satellite-measured burn severity and a suite of high-severity patch metrics to the daily area burned. Extreme fire spread events (defined here as burning >4900 ha/day) exhibited higher mean burn severity, a greater proportion of area burned severely, and increased like adjacencies between high-severity pixels. Further, increasing daily area burned also resulted in greater distances within high-severity patches to live tree seed sources. High-severity patch size and total high-severity core area were substantially higher for fires containing one or more extreme spread events than fires without an extreme event. Larger and more homogenous high-severity patches produced during extreme events can limit tree regeneration and set the stage for protracted forest conversion. These landscape outcomes are expected to be magnified under future climate, accelerating fire-driven forest loss and long-term ecological change.
Authors: Jessika R. McFarland, Jonathan D. Coop, Jared A. Balik, Kyle C. Rodman, Sean A. Parks, and Camille S. Stevens-Rumann
Dryad DOI: https://doi.org/10.5061/dryad.9kd51c5sr
OVERVIEW:
This folder includes the results, statistical models, raster data, and code used to produce our manuscript on extreme fire spread events and their burn severity outcomes on fires in the southwestern US. Descriptions of each facet of archived data are below.
Spatial data overview:
We used a suite of publicly available, remotely sensed data products to do our analysis. Methods are documented in detail in McFarland et al. (2025)
1. Monitoring Trends in Burn Severity (MTBS):
MTBS data were utilized to select large (>404 ha) fire perimeters within our study area using the Burned Area Boundaries Dataset.
These data can be found at: https://www.mtbs.gov/direct-download
2. Moderate Resolution Imaging Spectroradiometer (MODIS) & Visible Infrared Imaging Radiometer Suite (VIIRS)
MODIS and VIIRS data were utilized based on methods developed by Parks (2014) to interpolate daily fire progression (i.e., day of burning (DOB) maps) through the use of active fire-detection data.
These data can be found at: https://firms.modaps.eosdis.nasa.gov/active_fire/
3. Composite Burn Index (CBI)
CBI were calculated using methods developed by Parks et al. (2019)
4. LANDFIRE
LANDFIRE data were used to develop forest filters to ensure analysis of solely forested areas. We used Existing Vegetation Type (EVT) data that preceded the year of the fire event.
These data can be found at: https://landfire.gov/vegetation/evt
FILE STRUCTURE:
Our data are split into four subfolders (CODE, MODELS, RASTERS, RESULTS) as described below.
1) CODE
This subfolder contains the necessary R scripts that analyzed the DOB and CBI data to produce our results.
1a. ExtremeSpread_Thresholds.R
This script gathers data from MTBS perimeters and interpolated MODIS/VIIRS daily fire detections to summarize data on daily area burned (e.g., area burned per day). It then calculates the mean plus 1-2 standard deviations of log10-transformed daily burned area (ha) to obtain thresholds for categorical spread events. These events are classified as follows:
Common: < mean + 1SD
Large: mean + 1SD < X < mean + 2SD
Extreme: > mean + 2SD
1b. ExtremeSpread_Metrics.R
This script utilizes multiple spatial datasets to calculate landscape metrics used in our analyses. It contains steps for calculating composite burn indices, proportion of area burned at high severity, percentage of high-severity like adjacencies, distance to nearest live tree seed source, area-weighted mean patch size, and total high-severity core area. This script also applies a forest filter to ensure we did not sample non-forested areas.
Broadly, the steps of this analysis are as follows:
Step 1: Burn Severity Raster Pre-Processing & daily continuous CBI sampling. (Figure 3A/3B)
Step 2: Calculate patch metrics that are ultimately summarized by daily spread class across the fire. (Figure 3C-3H)
Step 3: Calculate patch metrics that are summarized by individual fire events. (Figure 4)
Step 4/5: Assemble datasets across fires
R packages utilized for this analysis include:
Terra: https://cran.r-project.org/web/packages/terra/index.html
dplyr: https://cran.r-project.org/web/packages/dplyr/index.html
landscapemetrics: https://cran.r-project.org/web/packages/landscapemetrics/index.html
foreach: https://cran.r-project.org/web/packages/foreach/index.html
doParallel: https://cran.r-project.org/web/packages/doParallel/index.html
glmmTMB: https://cran.r-project.org/web/packages/glmmTMB/index.html
2) MODELS
This subfolder contains an R script used to run statistical models on the data generated from our R code. We ran a series of linear mixed effects models using the glmmTMB package.
Two models were run for each landscape metric to assess the relationships between landscape metrics and fire spread rates; a) categorical models and b) continuous models. Fire ID was modeled as a fixed effect for daily-scale landscape metrics. For subsampled metrics (CBI, D2Forest), we incorporated the day of burning (DOB) as a random nested intercept term.
3) RASTERS
This subfolder contains all the original raster data used to calculate our landscape metrics. Data are organized by fire year, and then by individual MTBS fire ID subfolders. Raster data are masked to only include forested areas (“masked_forest_GCB”).
Each fire ID subfolder contains the following data:
3A. “_cbi_bc_masked_FOREST_GCB.tif”: continuous CBI raster
3B. “_cbi_cat_masked_FOREST.GCB.tif”: categorical CBI raster (high-severity and non-high severity)
3C. “_dob_FOREST_GCB.tif”: day of burning (DOB) raster
3D. “_HS_patches_FOREST.GCB.gpkg”: high-severity patch data
3E. “_D2ForestGCB.gpkg”: distance to nearest live seed source data
Additionally, each subfolder has unique landscape metric CSV output files generated from the analysis. These files are summarized in section 4 (Results).
3F. “_continuous_CBI_subsample_FOREST_GCB_v2.csv”: CBI output (Fig. 3A/B)
3G. “_Daily_Prop_HS_FOREST_GCB.csv”: Proportion burned at high severity output (Fig.3C/D)
3H. “_Daily_HS_PLADJ_FOREST_GCB.csv”: Percentage of high-severity like adjacencies output (Fig.3E/F)
3I. “_D2ForestGCB_subsample.csv”: Distance to nearest live tree seed source output (Fig.3G/H)
3J. “_Fire_MPS_TCA_Forest_GCB.csv”: Mean high-severity patch size and total high-severity core area output (Fig.4A-D)
4) RESULTS
Overview:
This subfolder contains the final outputs of processed data from our analyses. Specifically, each CSV contains data for each landscape metric we calculated for the southwestern US. These files are named as follows, and are described in further detail below:
CBI.csv
PropHS.csv
D2Forest.csv
PLADJ.csv
MPS.TCA.csv
File format:
CBI, PropHS, D2Forest, and PLADJ were analyzed on a daily scale, while MPS.TCA was analyzed on a per-fire basis. CSVs for daily scale metrics (CBI, PropHS, D2Forest, PLADJ) differ in row length due to sampling differences, but share the same file format in terms of column headers, and are described below. Unique outputs for each metric will then be described through sections 4.1 - 4.4. MPS.TCA were analyzed per-fire, and thus have a different file structure. This structure is described in section 4.5.
Column headers for daily landscape metrics (cols 1-17):
DOB: the calendar day of burning for each individual fire ID.
fire.id: unique MTBS fire perimeter IDs.
fire.year: the year the fire took place.
pixel.count: count of 30-m resolution pixels per day.
area.ha: the area burned per day in hectares (ha).
Country: the country the fire took place in. (USA)
Perim.Source: the source of the fire perimeter data. (MTBS)
area.m2: the area burned per day in meters (m) squared.
area.m2.cum: the cumulative area burned per unique fire ID in meters (m) squared.
days.burned: the total days a unique fire ID burned.
Chrono.DOB: the chronological day of burning.
Common: whether or not a fire was considered a categorically ‘common’ fire spread event. (TRUE/FALSE)
Large: whether or not a fire was considered a categorically ‘large’ fire spread event. (TRUE/FALSE)
Extreme: whether or not a fire was considered a categorically ‘extreme’ fire spread event. (TRUE/FALSE)
extreme.any: whether or not a unique fire ID contained one or more extreme fire spread events. (TRUE/FALSE)
fire.type: which categorical spread event type the day of burning qualified as. (Common, Large, Extreme)
These data are featured in 4.1-4.4, but can also be found in a standalone format in this folder under Categorical_Spread_Thresholds.csv.
4.1: CBI
This file contains data on the Composite Burn Index (CBI) per day of burning. This metric was calculated as a random subsample at 0.01% within each day and contains 308404 observations. The unique column for this data is described as:
CBI: the Composite Burn Index value of the subsampled pixel (continuous: 0-3).
4.2: PropHS
This file contains data on the proportion of area burned at high severity per day of burning and contains 4267 observations. The unique columns for this data are described as:
total.HS.area.ha: the total high-severity area of each day of burning in hectares (ha).
dob.forest.area.ha: the total forested area of each day of burning in hectares (ha).
prop.HS.forest: the proportion of total.HS.area.ha / dob.forest.area.ha. This ensures we were not capturing areas in the proportion that could not burn severely.
4.3: D2Forest
This file contains data on the distance to the nearest live tree seed source (i.e., the nearest forested pixel that was not burned at high severity (CBI >= 2.25)) within a high severity patch per day of burning. This metric was calculated as a random subsample within daily high-severity patches at 0.01% and contains 87677 observations. The unique columns are described as:
HS.patch.ID: the unique ID of the high-severity patch calculated.
HS.patch.area.ha: the area of high-severity patch in hectares (ha).
point.ID: the unique ID of the randomly subsampled point within the high-severity patch.
point.D2Conifer.m: the distance from point.ID to the nearest forested pixel that was either unburned or burned below CBI < 2.25.
4.4: PLADJ
This file contains data on the percentage of like adjacencies (PLADJ) of high-severity pixels for a given day of burning and contains 4267 observations. The unique column for this data is described as:
HS.PLADJ: Percentage of like adjacencies summarizes the percentage of high-severity 30-m cells adjacent to other high-severity cells, ranging from 0% (uniformly spaced; all pixels in different patches) to 100% (highly aggregated; all pixels in one patch).
4.5: MPS.TCA
This file contains data on the area-weighted mean high-severity patch size (MPS) and the mean total high-severity core area (TCA) for each individual fire that had high-severity patches. This file contains 522 observations. The unique columns of this data are described below:
Column headers for individual fire metrics:
fire.id: unique MTBS fire perimeter IDs.
fire.year: the year the fire took place. (2002:2020)
Country: the country the fire took place in. (USA)
Perim.Source: the source of the fire perimeter data. (MTBS)
days.burned: the total days a unique fire ID burned.
extreme.any: whether or not a unique fire ID contained one or more extreme fire spread events. (TRUE/FALSE)
total.HS.area.ha: the total area burned at high severity within the unique fire event (hectares).
mps.ha: the average area-weighted mean high-severity patch size (hectares).
tca.ha: the average total high-severity core area (hectares).
total.area.burned.ha: total area burned within the unique fire event (hectares).
mean.area.burned.ha: the mean daily area burned for each unique fire event (hectares).
This data was summarized from the datasets featuring daily burned area data.