The Earth Engine version of the Fire Information for Resource Management System (FIRMS) dataset contains the LANCE fire detection product in rasterized form. The near real-time (NRT) active fire locations are processed by LANCE using the standard MODIS MOD14/MYD14 Fire and Thermal Anomalies product. Each active fire location represents the centroid of a 1km pixel that is flagged by the algorithm as containing one or more fires within the pixel. The data are rasterized as follows: for each FIRMS active fire point, a 1km bounding box (BB) is defined; pixels in the MODIS sinusoidal projection that intersect the FIRMS BB are identified; if multiple FIRMS BBs intersect the same pixel, the one with higher confidence is retained; in case of a tie, the brighter one is retained. The data in the near-real-time dataset are not considered to be of science quality. Additional information can be found here. NOTE: VIIRS FIRMS datasets from NOAA20 and SUOMI are also available: NASA/LANCE/NOAA20_VIIRS/C2 NASA/LANCE/SNPP_VIIRS/C2

GEEBAM is an interim product and there is no ground truthing or assessment of accuracy. Fire Extent and Severity Mapping (FESM) data should be used for accurate information on fire severity and loss of biomass in relation to bushfires. The intention of this dataset was to provide a rapid assessment of fire impact.

The Terra and Aqua combined MCD64A1 Version 6.1 Burned Area data product is a monthly, global gridded 500m product containing per-pixel burned-area and quality information. The MCD64A1 burned-area mapping approach employs 500m MODIS Surface Reflectance imagery coupled with 1km MODIS active fire observations. The algorithm uses a burn sensitive vegetation …

Data acquisitionOccurrence data for bee species were downloaded from ALA60 using ALA4R version 1.8.064 in R version 3.6.265.Floral visitation data were obtained from ALA60, Museums Victoria, the Western Australian Museum66,67, and publications (Tables S1 and S2). Floral visitation records were checked for errors and synonymies using the Australian Plant Name Index68. Life-history traits for bee species were sourced, in most cases, from the most recent taxonomic descriptions, or other publications (Tables S1 and S2). A one-hectare resolution Major Vegetation Subgroup (MVS) map was sourced from Geoscience Australia’s National Mapping Division (NMD)61. Fire frequency data from 1988 to 2016 were downloaded from the Department of Environment and Energy (DEE)69, 2019–20 wildfire occurrence data (National Indicative Aggregated Fire Extent Dataset — NIAFED — version 20200623) were sourced from the Department of Agriculture, Water and the Environment (DAWE)36, and 2019–20 wildfire intensity data (Google Earth Engine Burnt Area Map — GEEBAM) were sourced from the Department of Planning, Industry and Environment (DPIE)62. All raster data sources were matched in resolution to the one-hectare MVS map. These GIS data sources may vary in spatial uncertainty or resolution and their caveats can be found at their respective locations online.Data filtering and analysesOccurrence data from ALA were filtered to include only reliable (“preserved specimens”, “machine observations” — e.g., malaise traps, — and data from published datasets) and “present” (compared to “absent”) records. Records without geographic locations or that did not align with base maps were excluded from GIS analyses. Species were then filtered for minimum sample size (n = 30) and minimum number of unique localities (n = 5). However, if there were 15 or more unique localities and a sample size of less than 30, the species was included.The MVS map was reprojected to a world geodetic system (WGS 1984, EPSG:4326) and clipped to the 2019–20 wildfire map in QGIS version 3.1270. The NIAFED and GEEBAM maps were aligned and matched to the resolution of the MVS map using the package raster version 3.0-1271 in R version 3.6.265. Major vegetation subgroups61, 2019–20 wildfire status36, and fire frequency69 were extracted for each ALA record using raster. The proportion of each MVS burnt was calculated by clipping MVS maps with the 2019–20 burn map in ArcMap Version 10.6.172. All map files used in our analyses are available at (html location to be confirmed upon acceptance) for use with our R script.We complemented species distributional data (ALA60 point data) with spatial information on their associated habitat (MVS61), to avoid reliance on the limited data for some species. To determine the potential distribution of each species we buffered the latitudinal and longitudinal extents of the raster datasets (MVS, fire frequency, NIAFED, and GEEBAM) by 20% in each direction. For geographically-restricted species with latitudinal or longitudinal ranges less than one degree (~111 km), we buffered their extent by one degree in each direction along that axis or axes. These values were chosen as conservative estimates of species distributional extents, but we recognize that this treatment may over-inflate the distribution of some species with highly-localized ranges. These data are broken into four files:Map_data — hosts all of the map files used in the analysesBee-plant_point_data — hosts the ALA download data, combined bee dataset, and the life history and plant data spreadsheetWard_comparison_data — hosts some of the data used for the Ward co-analysis using our methodAll_other_R_data — hosts many of the runfiles from our main analysis

Suomi NPP Visible Infrared Imaging Radiometer Suite (VIIRS) Active Fire detection product is based on the instrument's 375m nominal resolution data. Compared to other coarser resolution (≥ 1km) satellite fire detection products, the improved 375 m data provide greater response over fires of relatively small areas, as well as improved mapping of large fire perimeters. Consequently, the data are well suited for use in support of fire management (e.g., near real-time alert systems), as well as other science applications requiring improved fire mapping fidelity. The data in the near-real-time dataset are not considered to be of science quality. Additional information can be found here.

The Landscape Fire Scars Database for Chile makes publicly available for the first time a historical high-resolution (~30 m) burned area and fire severity product for the country. The georeferenced database is a multi-institutional effort containing information on more than 8,000 fires events between July 1984 and June 2018. Using Google Earth Engine (GEE), we reconstructed the fire scar area, perimeter, and severity for each fire. We also provide the Landsat mosaic image of pre- and post-fire events, including the NDVI and NBR indexes. In the related paper, we release the GEE code to reproduce our database or enable the international community to reconstruct another individual burned areas and fire severity data, with minimum input requirements. In the summary file is the list of reconstructed fire events. The identification number (ID) relates the initial information of the wildfires with fire scar and severity data.

The California Department of Forestry and Fire Protection's Fire and Resource Assessment Program (FRAP) annually maintains and distributes an historical wildland fire perimeter dataset from across public and private lands in California. The GIS data is developed with the cooperation of the United States Forest Service Region 5, the Bureau of Land Management, California State Parks, National Park Service and the United States Fish and Wildlife Service and is released in the spring with added data from the previous calendar year. Although the dataset represents the most complete digital record of fire perimeters in California, it is still incomplete, and users should be cautious when drawing conclusions based on the data.

This data should be used carefully for statistical analysis and reporting due to missing perimeters (see Use Limitation in metadata). Some fires are missing because historical records were lost or damaged, were too small for the minimum cutoffs, had inadequate documentation or have not yet been incorporated into the database. Other errors with the fire perimeter database include duplicate fires and over-generalization. Additionally, over-generalization, particularly with large old fires, may show unburned "islands" within the final perimeter as burned. Users of the fire perimeter database must exercise caution in application of the data. Careful use of the fire perimeter database will prevent users from drawing inaccurate or erroneous conclusions from the data. This data is updated annually in the spring with fire perimeters from the previous fire season. This dataset may differ in California compared to that available from the National Interagency Fire Center (NIFC) due to different requirements between the two datasets. The data covers fires back to 1878. As of May 2024, it represents fire23_1.

Please help improve this dataset by filling out this survey with feedback:

Historic Fire Perimeter Dataset Feedback (arcgis.com)

Current criteria for data collection are as follows:

CAL FIRE (including contract counties) submit perimeters ≥10 acres in timber, ≥50 acres in brush, or ≥300 acres in grass, and/or ≥3 impacted residential or commercial structures, and/or caused ≥1 fatality.

All cooperating agencies submit perimeters ≥10 acres.

Version update:

Firep23_1 was released in May 2024. Two hundred eighty four fires from the 2023 fire season were added to the database (21 from BLM, 102 from CAL FIRE, 72 from Contract Counties, 19 from LRA, 9 from NPS, 57 from USFS and 4 from USFW). The 2020 Cottonwood fire, 2021 Lone Rock and Union fires, as well as the 2022 Lost Lake fire were added. USFW submitted a higher accuracy perimeter to replace the 2022 River perimeter. Additionally, 48 perimeters were digitized from an historical map included in a publication from Weeks, d. et al. The Utilization of El Dorado County Land. May 1934, Bulletin 572. University of California, Berkeley. Two thousand eighteen perimeters had attributes updated, the bulk of which had IRWIN IDs added. A duplicate 2020 Erbes perimeter was removed. The following fires were identified as meeting our collection criteria, but are not included in this version and will hopefully be added in the next update: Big Hill #2 (2023-CAHIA-001020).

YEAR_ field changed to a short integer type. San Diego CAL FIRE UNIT_ID changed to SDU (the former code MVU is maintained in the UNIT_ID domains). COMPLEX_INCNUM renamed to COMPLEX_ID and is in process of transitioning from local incident number to the complex IRWIN ID. Perimeters managed in a complex in 2023 are added with the complex IRWIN ID. Those previously added will transition to complex IRWIN IDs in a future update.

Includes separate layers filtered by criteria as follows:

California Fire Perimeters (All): Unfiltered. The entire collection of wildfire perimeters in the database. It is scale dependent and starts displaying at the country level scale.

Recent Large Fire Perimeters (≥5000 acres): Filtered for wildfires greater or equal to 5,000 acres for the last 5 years of fires (2019-2023), symbolized with color by year and is scale dependent and starts displaying at the country level scale. Year-only labels for recent large fires.

California Fire Perimeters (1950+): Filtered for wildfires that started in 1950-present. Symbolized by decade, and display starting at country level scale.

Detailed metadata is included in the following documents:

Wildland Fire Perimeters (Firep23_1) Metadata

For any questions, please contact the data steward:

Kim Wallin, GIS Specialist

CAL FIRE, Fire & Resource Assessment Program (FRAP)

kimberly.wallin@fire.ca.gov

NOAA-20 (JPSS-1) Visible Infrared Imaging Radiometer Suite (VIIRS) Active Fire detection product is based on the instrument's 375m nominal resolution data. Compared to other coarser resolution (≥ 1km) satellite fire detection products, the improved 375 m data provide greater response over fires of relatively small areas, as well as improved mapping of large fire perimeters. Consequently, the data are well suited for use in support of fire management (e.g., near real-time alert systems), as well as other science applications requiring improved fire mapping fidelity. The data in the near-real-time dataset are not considered to be of science quality. Additional information can be found here.

Please refer to README file.

The Crisis Mapping Toolkit (CMT) is a collection of tools for processing geospatial data (images, satellite data, etc.) into cartographic products that improve understanding of large-scale crises, such as natural disasters. The cartographic products produced by CMT include flood inundation maps, maps of damaged or destroyed structures, forest fire maps, population density estimates, etc. CMT is designed to rapidly process large-scale data using Google Earth Engine and other geospatial data systems.

Although fire is a fundamental ecological process in western North American forests, climate warming and accumulating forest fuels due to fire suppression have led to wildfires that burn at high severity across larger fractions of their footprint than were historically typical. These trends have spiked upwards in recent years and are particularly pronounced in the Sierra Nevada-southern Cascades ecoregion of California, USA and neighboring states. We assessed annual area burned and percentage of area burned at high and low-to-moderate severity for seven major forest types in this region from 1984 to 2020. We compared values for this period against estimates for the pre-Euro-American settlement (EAS) period prior to 1850 and against a previous study of trends from 1984–2009. Our results show that total average annual area burned remained below pre-EAS levels, but that gap is decreasing (i.e., c. 14% of pre-EAS for 1984–2009, but 39% for 2010–2020 [including c. 150% in 2020]). Although average annual area burned has remained low compared to pre-EAS, both the average annual area burned at high severity and the percentage of wildfire area burned at high severity have increased rapidly. The percentage of area burned at high severity – which was already above pre-EAS average for the 1984–2009 period – has continued to rise for five of seven forest types. Notably, between 2010 and 2020, the average annual area burned at high severity exceeded the pre-EAS average for the first time on record. By contrast, percentage of area that burned at low-to-moderate severity decreased, particularly in the lower elevation oak and mixed conifer forest types. These findings underline how forests historically adapted to frequent low-to-moderate-severity fire are being reshaped by novel proportions and extents of high-severity burning. The shift toward a high severity-dominated fire regime is associated with ecological disruptions, including changes in forest structure, species composition, carbon storage, wildlife habitat, ecosystem services, and resilience. Our results underscore the importance of finding a better balance between the current management focus on fire suppression and one that puts greater emphasis on proactive fuel reduction and increased forest resilience to climate change and ecological disturbance. Methods The primary source of fire severity data came from the "Vegetation Burn Severity – 1984 to 2017” geospatial data layer (USDA 2018). For the area of analysis, we the same perimeter used by Mallek et al. (2013) and Miller et al. (2009), and is based on the Sierra Nevada ecoregion as defined by the Sierra Nevada Ecosystem Project (SNEP 1996) and the Sierra Nevada Forest Plan Amendment (SNFPA; USDA 2004). We included fires greater than or equal to 80 ha in size (see Methods for specifics). For the 2018–2020 fire years, we estimated burn severity using Google Earth Engine following Parks et al. (2018c, 2021). For the 2018–2020 fires we used the attached zip file of fire perimeters.

Mallek, C., H. Safford, J. Viers, and J. Miller. 2013. Modern departures in fire severity and area vary by forest type, Sierra Nevada and southern Cascades, California, USA. Ecosphere 4. Parks, S. A., L. M. Holsinger, M. A. Voss, R. A. Loehman, and N. P. Robinson. 2018c. Mean Composite Fire Severity Metrics Computed with Google Earth Engine Offer Improved Accuracy and Expanded Mapping Potential. Remote Sensing 10. Parks, S. A., L. M. Holsinger, M. A. Voss, R. A. Loehman, and N. P. Robinson. 2021. Correction: Mean Composite Fire Severity Metrics Computed with Google Earth Engine Offer Improved Accuracy and Expanded Mapping Potential. Remote Sens. 10, 879, 2018. Remote Sensing 13. SNEP. 1996. Sierra Nevada Ecosystem Project: Final report to Congress.in C. f. W. a. W. Resources, editor. University of California, Davis, California, USA. USDA. 2004. Sierra Nevada Forest Plan Amendment, Record of Decision.in P. S. Region, editor. USDA Forest Service, Vallejo, California, USA. USDA. 2018. R5 VegBurnSeverity - Metadata.in P. S. R. USDA Forest Service, editor., USDA Forest Service, Pacific Southwest Region, California.

This data package includes 481 geospatial vector polygons of historic fire perimeters associated with National Wildlife Refuges in Alaska. These polygons, originally held within the Alaska Large Fire Database (ALFD), were reviewed and found to have geospatial inaccuracies with respect to the fire they represent. They were corrected and updated based on a variety of remote sensing resources; the Monitoring Trends in Burn Severity (MTBS) database (https://www.mtbs.gov/viewer/index.html), geospatial rasters of historic fire activity derived from Landsat 1-9 imagery in the Google Earth Engine(https://earthengine.google.com) environment, and historical air photos acquired through EarthExplorer (https://earthexplorer.usgs.gov/). Wildfire records occurring within Alaska Wildlife Refuge units and considered for necessary updates, spanned the 1943-2022 time period and comprised 1,229 recorded fires. After reviewing all fire records, 400 fire perimeters were updated and 81 previously unrecorded fires were added to the database. These updated records spanned 1954-2021.

Wildfire is globally an important ecological disturbance affecting biochemical cycles, and vegetation composition, but also puts people and their homes at risk. Suppressing wildfires has detrimental ecological effects and can promote larger and more intense wildfires when fuels accumulate, which increases the threat to buildings in the Wildland Urban Interface (WUI). Yet, when wildfires occur, typically only a small proportion of the buildings within the fire perimeter are lost, and the question is what determines which buildings burn. Our goal was to examine which factors are related to building loss when a wildfire occurs throughout the United States. We were particularly interested in the relative roles of vegetation, topography, and the spatial arrangement of buildings, and how their respective roles vary among ecoregions. We analyzed all fires that occurred within the conterminous U.S. from 2000 to 2010 and digitized which buildings were lost and which survived according to Google Earth historical imagery. We modeled the occurrence as well as the percentage of buildings lost within clusters using logistic and linear regression. Overall, variables related to topography and the spatial arrangement of buildings were more frequently present in the best 20 regression models than vegetation-related variables. In other words, specific locations in the landscape have a higher fire risk, and certain development patterns can exacerbate that risk. Fire policies and prevention efforts focused on vegetation management are important, but insufficient to solve current wildfire problems. Furthermore, the factors associated with building loss varied considerably among ecoregions suggesting that fire policy applied uniformly across the US will not work equally well in all regions and that efforts to adapt communities to wildfires must be regionally tailored.

LANDFIRE (LF), Landscape Fire and Resource Management Planning Tools, is a shared program between the wildland fire management programs of the U.S. Department of Agriculture's Forest Service, U.S. Department of the Interior's Geological Survey, and The Nature Conservancy. Landfire (LF) Historical fire regimes, intervals, and vegetation conditions are mapped using …

LANDFIRE (LF), Landscape Fire and Resource Management Planning Tools, is a shared program between the wildland fire management programs of the U.S. Department of Agriculture's Forest Service, U.S. Department of the Interior's Geological Survey, and The Nature Conservancy. LANDFIRE (LF) layers are created using predictive landscape models based on extensive field-referenced data, satellite imagery and biophysical gradient layers using classification and regression trees. LANDFIRE's (LF) Existing Vegetation Cover (EVC) represents the vertically projected percent cover of the live canopy layer for a 30-m cell. EVC is generated separately for tree, shrub, and herbaceous cover lifeforms using training data and other geospatial layers. Percentage tree, shrub, and herbaceous canopy cover training data are generated using plot-level ground-based visual assessments and lidar observations. * Once the training data are developed, relationships are then established separately for each lifeform between the training data and combination of Landsat and ancillary data. Each of the derived data layers (tree, shrub, herbaceous) has a potential range from 0-100 percent which are merged into a single composite EVC layer. * Disturbance data were used to develop LF Remap products for LFRDB plot filtering and to ensure 2015 and 2016 disturbances were included that were not visible in the source imagery. * The EVC product is then reconciled through QA/QC measures to ensure life-form is synchronized with both Existing Vegetation Height and Existing Vegetation Type products. LF uses EVC in several subsequent layers, including the development of the fuel layers. The LANDIFRE Vegetation datasets include: Biophysical Settings (BPS) Environmental Site Potential (ESP) Existing Vegetation Canopy Cover (EVC) Existing Vegetation Height (EVH). Existing Vegetation Type (EVT) These layers are created using predictive landscape models based on extensive field-referenced data, satellite imagery and biophysical gradient layers using classification and regression trees.

The Unpublished Digital Post-Hurricane Sandy (2015) Geomorphological Map of Fire Island National Seashore and Vicinity, New York is composed of GIS data layers and GIS tables in a 10.0 file geodatabase (fiis_post-sandy_geology.gdb), a 10.0 ArcMap (.MXD) map document (fiis_post-sandy_geology.mxd), and individual 10.0 layer (.LYR) files for each GIS data layer, an ancillary map information (.PDF) document (fiis_geomorphology.pdf) which contains source map unit descriptions, as well as other source map text, figures and tables, metadata in FGDC text (.TXT) and FAQ (.HTML) formats, and a GIS readme file (fiis_gis_readme.pdf). Please read the fiis_gis_readme.pdf for information pertaining to the proper extraction of the file geodatabase and other map files. To request GIS data in ESRI 10.0 shapefile format contact Stephanie O’Meara (stephanie.omeara@colostate.edu; see contact information below). The data is also available as a 2.2 KMZ/KML file for use in Google Earth, however, this format version of the map is limited in data layers presented and in access to GRI ancillary table information. Google Earth software is available for free at: http://www.google.com/earth/index.html. Users are encouraged to only use the Google Earth data for basic visualization, and to use the GIS data for any type of data analysis or investigation. The data were completed as a component of the Geologic Resources Inventory (GRI) program, a National Park Service (NPS) Inventory and Monitoring (I&M) Division funded program that is administered by the NPS Geologic Resources Division (GRD). Source geologic maps and data used to complete this GRI digital dataset were provided by the following: Rutgers University, Institute of Marine and Coastal Sciences. Detailed information concerning the sources used and their contribution the GRI product are listed in the Source Citation section(s) of this metadata record (fiis_post-sandy_metadata_faq.html; available at http://nrdata.nps.gov/geology/gri_data/gis/fiis/fiis_post-sandy_metadata_faq.html). Users of this data are cautioned about the locational accuracy of features within this dataset. Based on the source map scale of 1:12,000 and United States National Map Accuracy Standards features are within (horizontally) 6.1 meters or 20 feet of their actual _location as presented by this dataset. Users of this data should thus not assume the _location of features is exactly where they are portrayed in Google Earth, ArcGIS or other software used to display this dataset. All GIS and ancillary tables were produced as per the NPS GRI Geology-GIS Geodatabase Data Model v. 2.1. (available at: http://science.nature.nps.gov/im/inventory/geology/GeologyGISDataModel.cfm). The GIS data projection is NAD83, UTM Zone 18N, however, for the KML/KMZ format the data is projected upon export to WGS84 Geographic, the native coordinate system used by Google Earth. The data is within the area of interest of Fire Island National Seashore.

General Description

A manually annotated dataset, consisting of the video frames and segmentation masks, for segmentation of forest fire burned area based on a video captured by a UAV. A detailed explanation of the dataset generation is available in the open-access article "Burned area semantic segmentation: A novel dataset and evaluation using convolutional networks".

Data Collection

The BurnedAreaUAV dataset derives from a video captured at the coordinates' latitude 41° 23' 37.56" and longitude -7° 37' 0.32", at Torre do Pinhão, in northern Portugal in an area characterized by shrubby to herbaceous vegetation. The video was captured during the evolution of a prescribed fire using a DJI Phantom 4 PRO UAV equipped with an FC6310S RGB camera.

Video Overview

The video captures a prescribed fire where the burned area increases progressively. At the beginning of the sequence, a significant portion of the UAV's sensor field of view is already burned, and the burned area expands as time goes by. The video was collected by an RGB sensor installed on a drone while keeping the drone in a nearly static stationary stance during the data collection duration.

The video has about 15 minutes and a frame rate of 25 frames per second, amounting to 22500 images. It was collected by an RGB sensor installed on a drone while keeping the drone in a nearly static stationary stance during the data collection duration. Throughout this period, the progression of the burned area is observed. The original video has a resolution of 720×1280 and is stored in H.264 (or MPEG-4 Part 10) format. No audio signal was collected.

Manual Annotation

The annotation was done every 100 frames, which corresponds to a sampling period of 4 seconds. Two classes are considered: burned_area and unburned_area. This annotation has been done for the entire length of the video. The training set consists of 226 frame-image pairs and the test set of 23. The training and test annotations are offset by 50 frames.

We plan to expand this dataset in the future.

File Organization (BurnedAreaUAV_v1.rar)

The data is available in PNG, JSON (Labelme format), and WKT (segmentation masks only). The raw video data is also made available.

Concomitantly, photos were taken that allow to obtain metadata about the position of the drone, including height and coordinates, the orientation of the drone and the camera, and others. The geographic data regarding the location of the controlled fire are represented in a KML file that Google Earth and other geospatial software can read. We also provide two high-resolution orthophotos of the area of interest before and after burning.

The data produced by the segmentation models developed in "Burned area semantic segmentation: A novel dataset and evaluation using convolutional networks", comprising outputs in PNG and WKT formats, is also readily available upon request

BurnedAreaUAV_dataset_v1.rar
MP4_video (folder)
-- original_prescribed_burn_video.mp4

PNG (folder)
train (folder)
frames (folder)
-- frame_000000.png (raster image)
-- frame_000100.png
-- frame_000200.png
…
msks (folder)
-- mask_000000.png
-- mask_000100.png
-- mask_000200.png
…

test (folder)
frames (folder)
-- frame_020250.png
-- frame_020350.png
-- frame_020350.png
…
msks (folder)
-- mask_020250.png
-- mask_020350.png
-- mask_020350.png
…
JSON (folder)
-- train_valid_json (folder)
-- frame_000000.json (Labelme format)
-- frame_000100.json
-- frame_000200.json
-- frame_000300.json
…
-- test_json (folder)
-- frame_020250.json
-- frame_020350.json
-- frame_020450.json
…
WKT_files (folder)
-- train_valid.wkt (list of masks polygons)
-- test.wkt

UAV photos (metadata)
-- uav_photo1_metadata.JPG
-- uav_photo2_metadata.JPG

High resolution ortophoto files
-- odm_orthophoto_afterBurning.png
-- odm_orthophoto_beforeBurning.png

Keyhole Markup Language file (area under study polygon)
-- pinhao_cell_precribed_area.kml

Acknowledgements

This dataset results from activities developed in the context of partially projects funded by FCT - Fundação para a Ciência e a Tecnologia, I.P., through projects MIT-EXPL/ACC/0057/2021 and UIDB/04524/2020, and under the Scientific Employment Stimulus - Institutional Call - CEECINST/00051/2018.

The source code is available here.

The increasing frequency and severity of human-caused fires likely have deleterious effects on species distribution and persistence. In 2020, megafires in the Brazilian Pantanal burned 43% of the biome’s unburned area and resulted in mass mortality of wildlife. We investigated changes in habitat use or occupancy for an assemblage of eight mammal species in Serra do Amolar, Brazil, following the 2020 fires using a pre- and post-fire camera trap dataset. Additionally, we estimated density for two naturally marked species, jaguars Panthera onca and ocelots Leopardus pardalis. Of the eight species, six (ocelots, collared peccaries Dicotyles tajacu, giant armadillos Priodontes maximus, Azara’s agouti Dasyprocta azarae, red brocket deer Mazama americana, and tapirs Tapirus terrestris) had declining occupancy following fires, and one had stable habitat use (pumas Puma concolor). Giant armadillo experienced the most precipitous decline in occupancy from 0.431 ± 0.171 to 0.077 ± 0.044 after the fires. Jaguars were the only species with increasing habitat use, from 0.393 ± 0.127 to 0.753 ± 0.085. Jaguar density remained stable across years (2.8 ± 1.3, 3.7 ± 1.3, 2.6 ± 0.85 / 100km2), while ocelot density increased from 13.9 ± 3.2 to 16.1 ± 5.2 / 100km2. However, the low number of both jaguars and ocelots recaptured after the fire period suggests that immigration may have sustained the population. Our results indicate that the megafires will have significant consequences for species occupancy and fitness in fire affected areas. The scale of megafires may inhibit successful recolonization, thus wider studies are needed to investigate population trends. Methods We used camera traps (Bushnell 119876, Panthera V4 and Cuddeback 1279) to survey the study area in December 2019 (session 1; year 1 – pre-fires) and December 2020 (session 2, year 2 – 2 months post-fires). Due to logistical constraints, we installed cameras in February 2022 (session 3, year 3 – 15 months post-fires) for an average duration of 53 trap nights (range 1-136, see SI_1 for complete details). All three surveys took place in the rainy season. Thirty-five stations were active in session 1, 43 stations in session 2, and 31 stations in session 3. Cameras were placed at a distance of 1.5 ± .5 km between stations and were located in different land covers (primary, secondary and gallery forest, savannah). Minimum convex polygons for each survey were 189.68 km2 in year 1, 272.26 km2 in year 2, and 245.95 km2 in year 3. We placed double stations to enable photographing both sides of each passing individual, thus enabling identification for naturally marked species like jaguars and ocelots. Each sampling station had 24-hour motion-triggered camera operation with a period of 30 seconds between photograph triggers. Geographic coordinates, camera serial number, date and time of camera installation, canopy cover, habitat, and whether the camera was on or off trail were recorded. Our survey design complied with methodological assumptions to estimate jaguar (Foster et al., 2020; Tobler et al., 2013) and ocelot densities (Boron et al., 2021; Satter, Augustine, Harmsen, Foster, Sanchez, et al., 2019; Wolff et al., 2019), and we kept a discrete distance between stations to obtain data for the wider mammal community (Boron et al., 2021; de Martins et al., 2006; Rovero et al., 2020; Rovero & Ahumada, 2017). Our survey was limited to less than 100 days per year, and fulfills overall capture-recapture model assumptions: a) the population needs to be considered closed and stable, and b) all individuals should have a chance of being captured (Otis et al., 1978; White, 1982). Covariate selection and extraction We selected a set of covariates to test our hypotheses related to pre- and post-fire habitat use/occupancy as well as density. Covariates were Normalized Difference Vegetation Index (NDVI), often used to assess habitat quality for mammals (Pettorelli et al., 2005; White et al., 2022), area burned derived from Normalized Burn Ratio (ΔNBR) which measures fire severity (Escuin et al., 2008), and distance from water (Boron et al., 2019). We additionally included effort as the total of trap nights per station; year, included to account for differences related to time variation as field staff and camera type on p (Gutiérrez-González et al., 2015; Kotze et al., 2012); and whether the camera station was on a trail or not for the probability of detection (p). Year or session was also used as a way to account for the heterogeneity of the detection probability, like seasonal activity of species and the possible loss of camera quality (Kotze et al., 2012; MacKenzie et al., 2003; Tobler et al., 2015). NDVI was obtained for each study session from Copernicus-Sentinel-II sensors via Google Earth Engine (code here). NDVI calculates vegetation greenness on a normalized scale with denser vegetation approaching one and barren areas or water bodies closer to a value of zero (Pettorelli et al., 2005). Annual NDVI rasters were obtained on days with less than 10 percent cloud cover during the period of one month before camera installation with a grain size of 10 meters. We then extracted the mean NDVI value for a 500-meter buffer around each station. We calculated the area burned (AB) as the area within a 500-meter buffer of each camera station that presented moderate-low severity or higher according to ΔNBR. For AB, we included only ΔNBR values that represent moderate-low burn severity and higher (ΔNBR = 270+) (Keeley, 2009; Key & Benson, 2006) in order to differentiate from areas that may have had low affectation from the fires or that may have presented false-positive values where fires may have burned due to the lack of an NBR system for the region. We obtained surface water data from MapBiomas (https://brasil.mapbiomas.org) and calculated the Euclidean distance of each camera station from surface water. All geoprocessing was conducted in ArcMap Desktop 10.8 (ESRI Inc., 2020). We used Spearman’s correlation test to check for highly correlated covariates (>0.6) with the function ggcorr in the package “GGally v2.1.2” (Schloerke et al., 2022) in Program R v 4.2.2 (R Core Team, 2022). As covariates AB and ΔNBR were correlated (>|0.6|), we fit two global models , each including one of this covariates and selected the best model using Akaike Information Criteria corrected (AICc) for small sample sizes (Burnham & Anderson, 2002). The model with AB covariate performed better than ΔNBR for most species and thus was used in the analysis. Dynamic occupancy We determined the habitat use or initial occupancy probability of eight mammal species in the study area: jaguars (Panthera onca), ocelots (Leopardus pardalis), pumas (Puma concolor), giant armadillo (Priodontes maximus), lowland tapir (Tapirus terrestris), red brocket deer (Mazama americana), collared peccaries (Dicotyles tajacu), and Azara’s agouti (Dasyprocta azarae). We determined initial occupancy probability when we could assume closure (individual’s home range is less than the radius between camera trap stations) between camera trap locations (MacKenzie et al., 2002), for ocelots (Crawshaw & Quigley, 1989), red brocket deer (Varela et al., 2010), Azara’s agouti (Cid et al., 2013) and giant armadillo (Desbiez et al., 2020). And determined habitat use for jaguars (Kantek et al., 2021; Soisalo & Cavalcanti, 2006), pumas (Silveira, 2004), collared peccaries (Desbiez et al., 2009) and tapirs (Medici et al., 2022), whose home range surpassed the distance (1.5 km) between stations. Detection histories were created for each species, grouping camera data into a 7-21 days survey occasions based on the results of goodness of fit (GOF) tests (MacKenzie & Bailey, 2004) (SI_3,4). Dynamic occupancy models (DOM) estimate the probability of occupancy and detection and are particularly useful for monitoring changes in occupancy status over time (MacKenzie et al., 2018), allowing us to detect if certain variables were influencing the colonization (Ɣ) and extinction (Ɛ) trends. We scaled covariates before analysis for interpretability. We used “unmarked” package v 1.2.5 (Fiske & Chandler, 2011) in Program R v 4.2.2 (R Core Team, 2022) for all occupancy analysis. The parameters used in DOM were Ψ = initial probability of a site being occupied; p = probability of a species being detected if it is present, Ɣ = probability of a new area to pass from unoccupied to occupied (or to unused to used) in the next year, Ɛ = probability that a species stops occupying an area, or to pass from used to unused. We fit models for species individually, and selected models according to AICc (Burnham & Anderson, 2002). We used a stepwise method (Doherty et al., 2012) for model selection. We first fit models for detection (p) with all other parameters constant. We included survey effort, whether cameras were on a trail, and year as covariates for detection, and selected the best detection model based on AICc value. We proceeded with this best detection model and subsequently fit models using covariates describing occupancy, colonization, and, finally, extinction. We included NDVI and distance to water as covariates for occupancy, whereas area burned was applied to colonization and extinction. We considered there to be satisfactory statistical evidence for an effect if the 95% confidence interval of logit scale coefficient estimates did not include zero (Muff et al., 2022). The β estimates were back-transformed to obtain model parameter estimates (MacKenzie & Bailey, 2004). We tested model fit by using a parametric bootstrap GOF test based on Pearson’s X2 where p>0.05 indicates adequate model fit (Fiske & Chandler, 2011) (SI_3). Finally, we derived annual probability for each year, and calculated standard errors for the derived values using a bootstrap method (Kéry & Chandler, 2012). To assess whether differences in occupancy were

The Burn Area Index (BAI) is generated from the Red and Near-IR bands, and measures the spectral distance of each pixel from a reference spectral point (the measured reflectance of charcoal). This index is intended to emphasize the charcoal signal in post-fire images. See Chuvieco et al. (2002) for details. This product is generated from the MODIS/006/MYD09GA surface reflectance composites.