Detección del cambio de cobertura arbórea mediante el uso de imágenes satelitales Sentinel 2 para dos zonas de estudio de la Amazonía ecuatoriana y colombiana

Abstract

Deforestation and forest degradation are advancing at an alarming rate across the globe. According to the United Nations, the highest net loss rates are occurring in South America. This study evaluated the use of Sentinel satellite imagery to detect and analyze tree cover changes in two Amazonian micro-watersheds: the Caño Blanco River (Colombia) and the Pan de Azúcar River (Ecuador), over the period 2018 - 2024. Land cover classes were defined according to the context of each country: gallery forest, heterogeneous agricultural areas, water bodies, and open areas for Colombia; and forest, pasture, regenerating pasture, water bodies, and populated areas for Ecuador. Image classifications were performed using the Random Forest algorithm, and validation was conducted using confusion matrices based on field data and high-resolution PlanetScope imagery, yielding overall accuracies ranging from 91.33% to 97.92% and Kappa indices between 0.84 and 0.96. In Colombia, gallery forest showed a slight expansion (from 36.08% to 39.40%) while agricultural areas slightly decreased. In Ecuador, forest cover increased from 49.58% to 53.63%, while pastures declined from 48.52% to 43.03%. These trends indicate processes of natural regeneration, partly driven by the presence of protected areas. The classification model demonstrated high robustness in detecting land cover change patterns, validating the effectiveness of optical sensors and machine learning algorithms as tools for forest monitoring in densely wooded regions.