Résumé
In this paper, we focus on forecasting Land Use Land Cover (LULC) changes in Senegal's Great Green Wall (GGW), using Remote Sensing Images (RSIs). The Sahel region of Senegal faces significant challenges, including land degradation, climate change, drought, and loss of biodiversity. By analyzing historical LULC and predicting future developments, this study aims to support the Sustainable Development Goals (SDGs) by preserving biodiversity conservation, food security, and resilience to environmental and socioeconomic challenges. In this study, we present DeepCascade GGW-Forecast, a deep learning framework that leverages Sentinel-2 and corresponding Dynamical World data for 2019, 2021, and 2023 to forecast future trends. DeepCascade GGW-Forecast integrates the SatForecaster model to forecast future satellite imagery and U-Net for semantic segmentation to classify LULC classes, including Trees, Crops, Shrub-and-Scrub and Bare ground. SatForecaster uses ConvLSTM layers to forecast future satellite imagery achieved a high R2 score of 93%. The U-Net segmentation task, utilizing pretrained encoders (Inception-ResNet-v2 and ResNet-50), accurately classify LULC with 89% Accuracy and 80% IoU. The LULC change analysis quantifies spatiotemporal dynamics, projecting significant increases in Shrub-Scrub, and Tree cover by 2025, while Crop and Bare ground are expected to decrease, likely due to ongoing restoration efforts. This insight is crucial for effective environmental management in the region.