Estimación de la severidad de incendios, usando imágenes satelitales Landsat para el Parque Nacional Natural El Tuparro entre los años 2013 a 2023.

Abstract

This study aimed to assess the severity of forest fires and analyze their impact on vegetation and surface temperature in El Tuparro National Natural Park, Colombia, during the period 2013-2023. Landsat satellite images were used to generate five spectral indices (NDVI, NBR, NBR2, SAVI, and NDWI), identifying the characteristics of the land cover in the study area. A total of 17 mosaics were generated. Using this information and the best of three supervised machine learning algorithms (Random Forest, SVM, and KNN), each mosaic was classified with the most suitable algorithm to quantify the land cover types. Severity indices (dNBR) were also calculated, and time series were analyzed in GEE for eight affected polygons, evaluating vegetation recovery. Anomalies in each time series were identified using the Isolation Forest algorithm. Finally, surface temperature data obtained from GEE (MODIS-Terra) were analyzed to identify trends and anomalies. The results show that high accuracy was achieved in estimating areas affected by fires. Additionally, a significant impact on biomass and vegetation health was observed in each polygon. However, covers such as grasslands exhibited rapid and effective recovery after the fires. Regarding surface temperature, no significant upward trend was detected during the study period.