Wetland burned area extent derived from Sentinel-2 across the southeastern U.S. (2016-2019)
Wildfires and prescribed fires are frequent but under-mapped across wetlands of the southeastern United States . High annual precipitation supports rapid post-fire recovery of wetland vegetation, while associated cloud cover limits clear-sky observations. In addition, the low burn severity of prescribed fires and spectral confusion between fluctuating water levels and burned areas have resulted in wetland burned area being chronically under-estimated across the region. In this analysis, we first quantify the increase in clear-sky observations by using Sentinel-2 in addition to Landsat 8. We then present an approach using the Sentinel-2 archive (2016-2019) to train a wetland burned area algorithm at 20 m resolution. We coupled a Python-derived random forest model with Google Earth Engine to apply the algorithm across the southeastern United States (>290,000 km2). The burned area extent was validated (burned, unburned) using points derived from 27 WorldView-2 and WorldView-3 images. The burned area extent was compared to 555 wetland fire perimeters compiled from state and federal agencies. On an annual timestep, combining the Sentinel-2 and Landsat 8 data increased the mean observation count from 17 to 46 in 2016 and from 16 to 78 in 2019. When validating single-scene burned area extent, the Sentinel-2 output had 29% and 30% omission and commission error rates, respectively. We compared this to the U.S. Geological Survey’s Landsat 8 Burned Area Product (L8 BA), which had 47% and 8% omission and commission error rates, respectively. Across the four-year period, by count the Sentinel-2 burned area detected 78% of the wetland fire perimeters, compared to the L8 BA which detected 60% of the wetland fire perimeters. By area, Sentinel-2 burned area mapped 48% of the perimeter area as burned, compared to the L8 BA which mapped 32% of the perimeter area as burned. This analysis demonstrated the potential of Sentinel-2 to support efforts to track burned area extent even across challenging ecosystem types, such as wetlands.
Find Related Datasets
Search by Tags
Click any tag below to search for similar datasets
Complete Metadata
| @type | dcat:Dataset |
|---|---|
| accessLevel | public |
| bureauCode |
[ "010:12" ] |
| contactPoint |
{ "fn": "Melanie Vanderhoof", "@type": "vcard:Contact", "hasEmail": "mailto:mvanderhoof@usgs.gov" } |
| description | Wildfires and prescribed fires are frequent but under-mapped across wetlands of the southeastern United States . High annual precipitation supports rapid post-fire recovery of wetland vegetation, while associated cloud cover limits clear-sky observations. In addition, the low burn severity of prescribed fires and spectral confusion between fluctuating water levels and burned areas have resulted in wetland burned area being chronically under-estimated across the region. In this analysis, we first quantify the increase in clear-sky observations by using Sentinel-2 in addition to Landsat 8. We then present an approach using the Sentinel-2 archive (2016-2019) to train a wetland burned area algorithm at 20 m resolution. We coupled a Python-derived random forest model with Google Earth Engine to apply the algorithm across the southeastern United States (>290,000 km2). The burned area extent was validated (burned, unburned) using points derived from 27 WorldView-2 and WorldView-3 images. The burned area extent was compared to 555 wetland fire perimeters compiled from state and federal agencies. On an annual timestep, combining the Sentinel-2 and Landsat 8 data increased the mean observation count from 17 to 46 in 2016 and from 16 to 78 in 2019. When validating single-scene burned area extent, the Sentinel-2 output had 29% and 30% omission and commission error rates, respectively. We compared this to the U.S. Geological Survey’s Landsat 8 Burned Area Product (L8 BA), which had 47% and 8% omission and commission error rates, respectively. Across the four-year period, by count the Sentinel-2 burned area detected 78% of the wetland fire perimeters, compared to the L8 BA which detected 60% of the wetland fire perimeters. By area, Sentinel-2 burned area mapped 48% of the perimeter area as burned, compared to the L8 BA which mapped 32% of the perimeter area as burned. This analysis demonstrated the potential of Sentinel-2 to support efforts to track burned area extent even across challenging ecosystem types, such as wetlands. |
| distribution |
[ { "@type": "dcat:Distribution", "title": "Digital Data", "format": "XML", "accessURL": "https://doi.org/10.5066/P9S8SLEM", "mediaType": "application/http", "description": "Landing page for access to the data" }, { "@type": "dcat:Distribution", "title": "Original Metadata", "format": "XML", "mediaType": "text/xml", "description": "The metadata original format", "downloadURL": "https://data.usgs.gov/datacatalog/metadata/USGS.603d2d6ed34eb1203117ef8f.xml" } ] |
| identifier | http://datainventory.doi.gov/id/dataset/USGS_603d2d6ed34eb1203117ef8f |
| keyword |
[ "Alabama", "Everglades", "Florida", "Georgia", "Google Earth Engine", "Jupyter Notebook", "South Carolina", "USGS:603d2d6ed34eb1203117ef8f", "drought", "prescribed fire", "wetland", "wildland fire" ] |
| modified | 2021-08-26T00:00:00Z |
| publisher |
{ "name": "U.S. Geological Survey", "@type": "org:Organization" } |
| spatial | -89.067582, 24.995905, -78.594114, 34.405121 |
| theme |
[ "geospatial" ] |
| title | Wetland burned area extent derived from Sentinel-2 across the southeastern U.S. (2016-2019) |
