Multiple Endmember Spectral Mixture Analysis (MESMA) to map burn severity levels from Landsat images in Mediterranean countries

Abstract

Forest fires are a major hazard in Mediterranean countries, with an average of 45,000 fires per year. Discrimination of different degrees of burn severity is critical to improve management of fire-affected areas. In this work, an unmixing-based methodology was evaluated in three Mediterranean study areas to estimate burn severity from medium spatial resolution optical satellite data. Post-fire Landsat 5 Thematic Mapper (TM) images were unmixed into four fraction images: non-photosynthetic vegetation and ash (NPV–Ash), green vegetation (GV), soil and shade using Multiple Endmember Spectral Mixture Analysis (MESMA). MESMA decomposes each pixel using different combinations of potential endmembers, overcoming the Linear Spectral Mixture Analysis limitation of using the same number of endmembers to model all image pixels. Next, a decision tree was used to classify the shade normalized fraction images into four classes: unburned and low, moderate, and high levels of burn severity. Finally, the burn severity estimates were validated using error matrix, producer and user accuracies per class, and κ statistic. For reference data, we used 50 plots per class defined from a 50cm post-fire orthophotography (proportion of dead tree<50%, low severity; proportion of dead tree between 50 and 90%, moderate severity; and proportion of dead tree>90%, high severity). MESMA-based burn severity estimates showed a high accuracy (0.80, 0.80, and 0.78) for the three test sites. We conclude that the proposed MESMA based methodology is valid to accurately map burn severity in Mediterranean countries from moderate resolution satellite data.