Unmixing PRISMA hyperspectral images by multiple endmember spectral mixture analysis (MESMA) to assess fire severity in Mediterranean forest ecosystems

The high spectral dimension of hyperspectral data justifies its great potential to accurately estimate fire severity. In this work, we validated spaceborne hyperspectral data from the Italian Precursore IperSpettrale della Missione Applicativa (PRISMA) mission to estimate fire severity in one of the largest forest fires ever recorded in the western Mediterranean, the megafire (28,046 ha) that occurred in the Sierra de la Culebra (northwestern Iberian Peninsula) between 15 and 19 June 2022. To take advantage of the high spectral dimensionality, Multiple Endmember Spectral Mixture Analysis (MESMA) was used to transform an original PRISMA Level 2D 1A scene into three fraction images of the basic components present in the image, namely CHAR, photosynthetic vegetation (PV), and non-photosynthetic vegetation and soil (NPVS). MESMA decomposed each pixel using different combinations of potential endmembers, overcoming the limitation of Linear Spectral Mixture Analysis of using the same number of endmembers to model all image pixels. We field measured initial fire severity in 70 plots using a slightly modified version of the Composite Burn Index (CBI). Shade-normalized char fraction image and CBI values showed a relatively strong linear relationship (R2 = 0.70). As a benchmark, we used the Delta Normalized Burn Ratio (dNBR) index from Sentinel 2A data. Linear regression between dNBR and CBI values obtained an R2 = 0.36. These results allow to conclude the suitability of the PRISMA hyperspatial data and the MESMA processing technique for the accurate assessment of fire severity in Mediterranean forest ecosystems.