Hyperspectral wavelengths analysis for burnt area derivation

Kurzfassung

We identified globally distributed wildfire events with available post-fire EnMAP scenes by intersecting thermal anomaly information from the NASA FIRMS hotspot products with the entire EnMAP archive. We selected fire events using stratified sampling accounting for the size, fire season, fire-affected land cover / land use (LULC) type and climate zone. For each identified fire event, a cloud-free post-fire EnMAP L2A product was downloaded. As the NIR and SSWIR domain is well known to yield best results for the burnt area mapping task, we used each individual EnMAP NIR/SSWIR band as well as several band combinations as input data within the burnt area processing chain incorporated in the DLR ZKI wildfire monitoring system to sequentially generate burnt area classifications. Accuracy assessment was conducted by calculating the Intersection over Union (IoU) between each classified burnt area perimeter and a manually digitized reference burnt area. Finally, identification of the best performing bands was done by computing mean IoU values. Preliminary results indicate that the optimal wavelength for burnt area derivation using hyperspectral data is located around 820 nm (NIR). However, this is dependent on various factors, such as the intensity of the burning, the affected vegetation type and the seasonality.