Enhancing Deforestation Detection: Leveraging SAR and Optical Image Fusion Under Diverse Cloud Conditions with Uncertainty Metrics

This study investigates the effectiveness of data fusion and uncertainty analysis in identifying deforestation in the Amazon rainforest. Using Synthetic Aperture Radar (SAR) and optical satellite imagery, we assessed two base architectures, ResUnet and SWIN-Unet, in different cloud cover conditions. Our results show that combining SAR with optical data in cloud-diverse conditions can achieve detection accuracy similar to models using only cloud-free imagery. The uncertainty metric predictive entropy was used to measure prediction confidence and conduct a simulated audit process. Using this strategy, we could further examine high-uncertainty regions, which enhanced the model's performance after the audit. Our results demonstrate the importance of combining uncertainty analysis with SAR-optical fusion to improve the accuracy and robustness of deforestation detection in cloud-prone areas, providing an enhanced approach to improve environmental monitoring.