Mohammed-el-kassoiri/Mapping-Water-Stress-and-Agricultural-Adaptation-Potential

Water scarcity is a critical challenge in the Souss-Massa region of Morocco. Traditional monitoring methods (field visits) are slow, and standard indices (NDVI) often fail to capture root-zone moisture stress. This project introduces a Multi-Modal Attention U-Net that fuses 16 spectral and environmental bands to precisely segment agricultural area # Mapping-Water-Stress-and-Agricultural-Adaptation-Potential # 🌍 Mapping Water Stress & Agricultural Adaptation using Attention U-Net ### 📍 Case Study: Taroudant, Morocco > **A Deep Learning approach to precision agriculture:** Combining multi-source satellite data (Optical, Radar, Climate) to map water stress in taroudant. --- ## 📖 Project Overview Water scarcity is a critical challenge in the Souss-Massa region of Morocco. Traditional monitoring methods (field visits) are slow, and standard indices (NDVI) often fail to capture root-zone moisture stress. This project introduces a **Multi-Modal Attention U-Net** that fuses **16 spectral and environmental bands** to precisely segment agricultural areas into varying levels of water stress. By leveraging **Weak Supervision** (using a synthetic "Smart Index" as ground truth), we overcome the lack of dense physical labels. ### 🎯 Key Objectives * **High-Resolution Mapping:** Downscaling coarse environmental data (SMAP 9km) to field-level precision (10m). * **Multi-Source Fusion:** Integrating Sentinel-2 (Optical), Sentinel-1 (Radar), and SMAP (Soil Moisture). * **Adaptation Planning:** Identifying specific zones requiring immediate irrigation intervention. --- ## 🚀 Key Features * **🏗️ Attention U-Net Architecture:** Utilizes Attention Gates to focus on relevant features (stressed crops) while suppressing noise (urban areas). * **🛰️ 17-Band Input Stack:** A comprehensive feature set including Raw Bands, Indices (NDVI, NDWI, SAVI), Radar Backscatter, Climate (LST, Rainfall), and socio-economic data. * **🧠 Weak Supervision Workflow:** A practical approach to training deep learning models using a calculated "Smart Health Index" rather than manual pixelwise labels. * **⚡ Google Earth Engine Integration:** Automated data collection and preprocessing pipeline (scripts provided / configurable). --- ## 🌐 Google Earth Engine Data Collection This project uses a complete remote sensing data pipeline built in **Google E …