High-Resolution Forest Loss Maps Utilizing Transformer Models

Abstract

Although high-resolution multispectral images from satellites such as Sentinel-2 and Landsat has become easily accessible, accurately detecting and monitoring deforestation and land-cover change over time remains challenging. Conventional approaches often fail to capture the complex temporal dynamics, subtle vegetation transitions, and smallholder-driven deforestation patterns due to their limited ability to model long-term spatial-temporal dependencies. So, to fill this gap our research is focusing mainly in developing a Transformer-based framework for multi-temporal land-cover and deforestation monitoring by the usage of multispectral satellite imagery of Sentinel-2. Our proposed model is aiming to utilize the temporal dependencies and spatial context within satellite time-series data to accurately detect forest loss, land-use change, and smallholder-driven deforestation over time.