jameskoero/nyando-flood-ai

AI-powered flood risk prediction for Nyando Basin, Kenya. XGBoost + SHAP + FastAPI + React. AUC-ROC 0.9717. 100% open data — Kenya DPA 2019 compliant. # 🌊 Nyando Basin Flood Risk Prediction System **An open-source, AI-powered flood early warning system for Nyando River Basin, Kisumu County, Kenya.** Ward-level flood susceptibility mapping at 100m resolution with 72-hour prediction lead time. > Trained on **real Google Earth Engine satellite data** — NASA NASADEM, CHIRPS v2, Sentinel-1 SAR, SoilGrids, HydroSHEDS, ESA WorldCover. --- ## 🔴 Live Deployments | Service | URL | Status | |---|---|---| | **Prediction API** | nyando-flood-api.onrender.c… | ✅ Live — Python on Render | | **Health Check** | nyando-flood-api.onrender.c… | ✅ `model_loaded: true` | | **Donor Dashboard** | nyando-flood-ai.vercel.app | ✅ Live — React + Vite on Vercel | > ⚠️ The API runs on Render's free tier — first request after idle may take 30–60s to cold-start. Subsequent requests return in Model: **GradientBoostingClassifier** trained on 2,308 real GEE satellite observations. > Features: 6 real satellite variables. Labels: physics-calibrated with 2 Sentinel-1 SAR-confirmed flood anchors. > Evaluation: stratified 80/20 split + 5-fold spatial cross-validation. | Metric | Score | Interpretation | |---|---|---| | **AUC-ROC** | **0.9717** | Near-perfect flood/no-flood discrimination | | **F1-Score** | **0.9022** | High balance — minimises false alarms and missed floods | | **Precision** | 0.8830 | 88.3% of HIGH/CRITICAL alerts are genuine flood events | | **Recall** | 0.9222 | 92.2% of real flood zones correctly identified | | **Brier Score** | 0.0736 | Well-calibrated probability estimates | | **CV AUC (5-fold)** | 0.9727 ± 0.0040 | Stable — generalises well across spatial folds | | **Training points** | 2,308 real GEE | Real satellite feature values from Nyando Basin | | **Resolution** | 100m grid | Ward-level mapping | | **CI Tests** | 41 passing ✅ | GitHub Actions — all green | ### Model Comparison | Model | AUC-ROC | F1 | Notes | |---|---|---|---| | Logistic Regression | 0.82 | 0.74 | Baseline | | Random F …