Sakwimba/Climatilogical-Rainfall-Regoinisation-in-zambia

An Unsupervised Machine Learning Approach to Rainfall Pattern Clustering. Climatological Regionalization of Zambia An Unsupervised Machine Learning Approach to Rainfall Pattern Clustering Project Overview This project applies K-Means clustering to 20 years (2003–2023) of district-level daily rainfall data from Zambia to objectively classify districts into homogeneous rainfall zones. The pipeline replaces the static Agro-Ecological Zone (AEZ) framework with a data-driven, reproducible classification that captures onset dates, cessation dates, dry spell dynamics, and seasonal totals. Repository Structure zambia_rainfall_clustering/ │ ├── data/ │ ├── raw/ # Original ZMD/NASA POWER data (do not modify) │ ├── processed/ # Cleaned daily data, engineered features, standardized features │ ├── reference/ # District shapefiles, AEZ mappings, station metadata │ └── outputs/ # Cluster assignments, maps, reports, validation metrics │ ├── src/ │ ├── data_ingest.py # Module 1: Data acquisition with ZMD/NASA POWER merge-switch logic │ ├── preprocess.py # Module 2: Missing value handling, outlier detection, standardization │ ├── feature_engineer.py # Module 3: 13-feature extraction from daily time-series │ ├── cluster.py # Module 4: PCA + K-Means with Elbow/Silhouette optimal-K selection │ ├── validate.py # Module 5: Internal metrics + AEZ external comparison │ ├── visualize.py # Module 6: GIS maps, cluster profiles, risk zones │ └── pipeline.py # Orchestrator: runs all modules end-to-end │ ├── tests/ │ ├── test_ingest.py │ ├── test_preprocess.py │ ├── test_feature_engineer.py │ ├── test_cluster.py │ └── test_validate.py │ ├── notebooks/ │ └── exploratory_analysis.ipynb │ ├── docs/ │ └── methodology_notes.md │ ├── dashboard.py # Streamlit interactive dashboard (Iteration 4 deliverable) ├── requirements.txt # Exact pinned versions for pip ├── environment.yml # Conda environment s …