Rice cultivation in Rosso, Mauritania, a critical agricultural hub along the Senegal River, is threatened by diseases such as rice blast, bacterial leaf blight, and sheath blight, exacerbated by climate variability and limited ground monitoring. This study develops a novel deep learning (DL) framework for early disease detection and prediction using multi-source satellite imagery to enhance crop health management in resource-constrained regions. Using Google Earth Engine, we analyzed Sentinel-1 (SAR), Sentinel-2 (10m resolution), and Landsat 8/9 (30m) imagery from January to August 2025, identifying May–June as optimal for vegetation indices due to peak rice growth. Thirteen indices (NDVI, SAVI, NDWI, GNDVI, NDRE, MCARI, EVI, VARI, ARVI, MSI, NBR, CIgreen, CIrededge) were computed, and TIFF images were exported and labeled in QGIS for rice vs. non-rice classification. DeepLabV3 and U-Net++ models achieved 95% accuracy in segmenting rice fields. Pre-trained ResNet and MobileNet models classified disease types using indices and ~300 geo-referenced samples from public datasets (e.g., PlantVillage). An LSTM model forecasted disease risk with >85% accuracy. Strong correlations (Pearson r > 0.78) between DL predictions and indices, enhanced by SAR’s rainy-season capability, enabled precise disease hotspot mapping. This proof-of-concept integrates open-source DL and satellite data, offering scalable solutions for West African rice farming. Future work should focus on field validation and the development of farmer-accessible mobile tools.