sophia14324/East-African-Monsoon-ML-project

# Kenya Long-Rains ▸ ML-powered Climate Diagnostics 🌧️🔍 This project is an end-to-end, fully reproducible pipeline that ingests CHIRPS satellite rasters, cleans and aggregates them into March-April-May (MAM) rainfall totals for Kenya, then applies statistical and machine-learning methods (SPI-3 drought index, ARIMA forecasting, K-means clustering, random-forest classification). A single CLI command handles everything, raw GeoTIFF download to plotted insights, making it trivial to extend the workflow to any East-African country by only changing a bounding box. ## 🔑 Key features | Module | What it does | Tech highlights | |--------|--------------|-----------------| | **1. Data ingest** | Downloads >130 monthly CHIRPS v3 files (or optional NASA POWER / ERA5-Land) | resilient retry & on-disk cache | | **2. Pre-processing** | Raster → bbox clip → mm averages | `rasterio`, vectorised NumPy | | **3. Climate indices** | SPI-3, %-of-normal, z-scores | Gamma fit with `scipy.stats` | | **4. Forecasting** | ARIMA(1,0,0) + uncertainty ribbon | `statsmodels` | | **5. Pattern mining** | K-Means clustering of MAM months | `scikit-learn` | | **6. Classification** | Random-Forest predicts dry / normal / wet terciles | lag-features + grid-search | | **7. Outputs** | PNG charts, CSV tables, pickled models | one-command reproducibility | ## 🚀 Quick start ### 1. clone and create a clean env (conda or venv) pip install -r requirements.txt ### 2. end-to-end run for Kenya 1981-2025 python east_africa_monsoon_project.py --run-all --start 1981 --end 2025 ### 🌍 Want Tanzania instead? Just change the bounding-box (The ETL, SPI-3, clustering and plots adapt automatically): python east_africa_monsoon_project.py --bbox -11.7 -0.9 29.3 40.4 --run-all ### **Further Reading: Medium**