Quantifying Forest Loss in Legal Amazon Settlements through AI-Driven Remote Sensing

Abstract

The Amazon rainforest, a critical component of the Earth's climate system, faces increasing deforestation, particularly in settlement areas established through agrarian reform programs. This study investigates forest loss in the Alcobaça Settlement and Juma Tract in the Legal Amazon from 2018 to 2022 using remote sensing and artificial intelligence (AI) techniques. Sentinel-2 satellite imagery was analyzed using machine learning algorithms, including Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Convolutional Neural Networks (CNNs), to classify land cover and quantify deforestation trends. The results demonstrate that CNN outperformed the other classifiers, achieving the highest accuracy and better identifying deforestation patterns over time. The trained model was then applied to the Juma Tract to assess its generalization capability. Although the CNN approach proved effective, it overestimated deforestation by 8.32% in 2022 compared to manual classification, highlighting challenges in transferring machine learning models to different regions without additional calibration. The findings emphasize the potential of AI-driven remote sensing for large-scale environmental monitoring while underscoring the necessity of localized training and validation to improve classification accuracy. This research contributes to the development of automated methods for forest loss assessment, providing valuable insights for environmental management and policy-making in the Amazon.