[M-AG33] Satellite Land Physical Processes Monitoring at Medium/High/Very High Resolution

[MAG33-P05] Burned Area Mapping with CubeSat PlanetScope and Active Fire Data from MODIS and VIIRS

*HENG LI¹, Wei Yang^1,2 (1.Department of Remote Sensing, Graduate School of Science and Engineering, Chiba University, 2.Center for Environmental Remote Sensing, Chiba University)

Keywords:Burned Area, PlanetScope, Deep Learning, Active Fire

Wildfires not only cause human casualties but also lead to deforestation, loss of forest resources, and substantial greenhouse gas emissions. Understanding the spatial and temporal trends of burned areas (BA) on a large scale is crucial for calculating carbon dioxide emissions, assessing wildfire risks, and monitoring environmental recovery. Many previous studies have utilized different spatial and temporal resolution satellite data to generate burned area mapping (BAM) products, including MODIS, Sentinel-2, and PlanetScope. Among these satellites, the CubeSat constellation PlanetScope showed the best performances for BAM results. However, due to the limitations of the spectral range of the bands, all PlanetScope-based BAM use deep learning, with most methods focusing on spectral and spatial changes to identify burned areas. Surface changes that resemble burned areas both spectrally and spatially, such as shadows, harvests, and varying image quality, may be misclassified as burned areas. Furthermore, research using PlanetScope for BAM is almost entirely based on cases where wildfires have been confirmed, which may lead to errors when applied to other regions. Additionally, active fire data from MODIS and VIIRS is a critical tool for monitoring ongoing fire activity on Earth’s surface. This study suggests that active fire data can help PlanetScope determine the approximate extent of the burned area and filter out misclassified fake fire events. Therefore, this research employed a deep learning method (U-NET) with PlanetScope data to extract burned areas and then refined these results by using active fire products. Two study areas in Peru were selected to produce monthly BAM products in 2023. We also compared the BAM products of MODIS(MCD64A1) and VIIRS(VNP64A1) with those of PlanetScope. The results demonstrated that PlanetScope exhibited exceptional sensitivity in detecting wildfires. Incorporating active fire data significantly enhanced the accuracy of the PlanetScope BAM products, enabling the most accurate extractions of wildfire areas and effective identification of the majority of fire events. Furthermore, the PlanetScope BAM products, refined by active fire data, provided wildfire area estimates that were the closest to the actual ground truth data. In contrast, BAM products from MODIS and VIIRS displayed lower accuracy and consistently failed to detect wildfires that areas smaller than 1 km².