valofils/food-security-gee-madagascar

Monthly district-level NDVI and rainfall dataset for food security modeling # 🌵 District-Level Drought Risk Prediction in Southern Madagascar ### Machine Learning + Remote Sensing Early Warning System --- ## 📋 Table of Contents 1. Project Overview 2. Study Area 3. Data 4. Drought Definition 5. Machine Learning Models 6. Early Warning System Outputs 7. Dashboard 8. Project Structure 9. Key Results 10. Limitations 11. Next Steps 12. Potential Applications 13. Author --- ## 1. Project Overview This project develops a **district-level early warning system** to predict next-month drought risk in the **Grand Sud of Madagascar** using Earth observation data and machine learning. **Objectives:** - 🍽️ Food security early warning - 🤝 Humanitarian planning - 🌍 Climate risk monitoring - 🛡️ Drought preparedness > The system predicts the **probability of drought for the next month** for each district. --- ## 2. Study Area The project focuses on the **Grand Sud of Madagascar**, one of the most drought-prone regions in Sub-Saharan Africa. **Key districts:** | District | | |---|---| | Ambovombe-Androy | Tsihombe | | Amboasary-Atsimo | Bekily | | Betioky Atsimo | Taolagnaro | | Ampanihy Ouest | | These areas are highly vulnerable to recurrent drought, food insecurity, agricultural shocks, and climate variability. --- ## 3. Data The dataset is built at **district** spatial level and **monthly** temporal resolution, using remote sensing and climate data. | Variable | Description | |---|---| | NDVI mean | Vegetation index | | NDVI anomaly | Vegetation anomaly | | Rainfall total | Monthly rainfall | | Rainfall anomaly | Rainfall anomaly | | LST | Land surface temperature | | Evapotranspiration | Water loss | | Rainfall lag | Rainfall from previous months | | NDVI lag | NDVI from previous months | | Temperature lag | Temperature from previous months | | Rolling means | 3-month averages | | Seasonal features | Month sine/cosine | --- ## 4. Drought Definition (Target Variable) A drought proxy was defined using **environmental anomalies**. A di …