17:15 〜 19:15
[U04-P05] High Spatial Resolution Binary Burned Area Maps for Rapid Community Recovery from Wildfires
キーワード：WIldfire, Geospatial Analysis, California, Remote Sensing
Wildland fire is a natural and integral force on our landscapes. However, uncontained wildfires can devastate communities, threaten our health, and result in substantial economic losses. Between 2017-2021 wildfire insurance claims increased more than $50B. Parametric insurance for wildfire, where claims payment is contingent on the ‘triggering’ of measurable, pre-defined metrics, can protect home and business owners from the financial impacts of wildfire.
Our objective is to develop a high-confidence grid-based wildfire burn product using multiple sources of satellite data to determine whether a ‘conflagration’ (fire larger than 999+ acres) has ‘breached’ a grid cell. This product will improve parametric wildfire insurance by expediting the verification of the parametric trigger at a finer-spatial resolution . Working with our partners at Kettle reinsurance, we use VIIRS (Visible Infrared Imaging Radiometer Suite) 375 m fire detections and Sentinel-2 10 m satellite imagery to create a 20-m gridded fire burn product.
We develop a methodology to create high resolution, low latency binary burned area maps that can be used for parametric based wildfire insurance policies and for other applications that require high resolution burned area, including disaster response. With a focus on quantifying only burned area and not fire severity, we employ relative differenced NBR (RdNBR) from Sentinel-2 combined with active fire detections from VIIRS to determine burned area at a 20m spatial resolution. RdNBR provides a more consistent definition of severity between landscapes and allows for higher accuracies in high severity burn in heterogeneous landscapes. A mosaic function is used on pre- and post-fire Sentinel-2 images to minimize the temporal sensitivity of image selection.
Our product has been validated using the California Department of Forestry and Fire Protection Damage Inspection Database (DINS), which provides high resolution geospatial information on structure damage within fire perimeters in California. We compare what was reported damaged on the ground to our burn maps to estimate our error matrix. We have also verified our fire burn product against MODIS/ASTER Airborne Simulator (MASTER) Infrared (IR) data from the Fire Influence on Regional to Global Environments Experiment - Air Quality (FIREX-AQ) 2019 campaign. These points of active fire at high resolution (~25 m) are compared to our burn maps and we found an overall 91% agreement. The first iteration has been developed for assessing wildfires in California, with the possibility to expand this methodology nationwide and globally.
Our objective is to develop a high-confidence grid-based wildfire burn product using multiple sources of satellite data to determine whether a ‘conflagration’ (fire larger than 999+ acres) has ‘breached’ a grid cell. This product will improve parametric wildfire insurance by expediting the verification of the parametric trigger at a finer-spatial resolution . Working with our partners at Kettle reinsurance, we use VIIRS (Visible Infrared Imaging Radiometer Suite) 375 m fire detections and Sentinel-2 10 m satellite imagery to create a 20-m gridded fire burn product.
We develop a methodology to create high resolution, low latency binary burned area maps that can be used for parametric based wildfire insurance policies and for other applications that require high resolution burned area, including disaster response. With a focus on quantifying only burned area and not fire severity, we employ relative differenced NBR (RdNBR) from Sentinel-2 combined with active fire detections from VIIRS to determine burned area at a 20m spatial resolution. RdNBR provides a more consistent definition of severity between landscapes and allows for higher accuracies in high severity burn in heterogeneous landscapes. A mosaic function is used on pre- and post-fire Sentinel-2 images to minimize the temporal sensitivity of image selection.
Our product has been validated using the California Department of Forestry and Fire Protection Damage Inspection Database (DINS), which provides high resolution geospatial information on structure damage within fire perimeters in California. We compare what was reported damaged on the ground to our burn maps to estimate our error matrix. We have also verified our fire burn product against MODIS/ASTER Airborne Simulator (MASTER) Infrared (IR) data from the Fire Influence on Regional to Global Environments Experiment - Air Quality (FIREX-AQ) 2019 campaign. These points of active fire at high resolution (~25 m) are compared to our burn maps and we found an overall 91% agreement. The first iteration has been developed for assessing wildfires in California, with the possibility to expand this methodology nationwide and globally.