SullenSchemer/ey-ai-data-challenge

EY AI & Data Challenge 2026 - Predicting water quality parameters for South African rivers using ML # EY AI & Data Challenge 2026 — Water Quality Prediction Predicting water quality in South African rivers using Landsat satellite imagery and TerraClimate climate data. **Final score: 0.313 R²** | Baseline: 0.20 | Certificate threshold: 0.40 | Leaderboard top: 0.8529 --- ## The Problem Predict three water quality parameters at river monitoring sites: | Parameter | Unit | Healthy Range | |---|---|---| | Total Alkalinity | mg/L | 20–200 | | Electrical Conductance | µS/cm | **Note:** EY-provided datasets (water quality labels, pre-extracted Landsat and TerraClimate features, submission template) are excluded from this repository per the competition Terms & Conditions. To reproduce results, register for the challenge and download from the EY Challenge portal. --- ## Project Structure ``` ├── data/ │ ├── terraclimate_training_multi.csv # Self-extracted climate variables (ppt, soil, def, tmax, q) │ ├── terraclimate_validation_multi.csv # Self-extracted climate variables for validation │ ├── water_quality_training_dataset.csv # ⛔ Not included (EY confidential) │ ├── landsat_features_training.csv # ⛔ Not included (EY confidential) │ ├── landsat_features_validation.csv # ⛔ Not included (EY confidential) │ ├── terraclimate_features_training.csv # ⛔ Not included (EY confidential) │ └── terraclimate_features_validation.csv# ⛔ Not included (EY confidential) ├── submissions/ │ ├── submission_final.csv # Best predictions (v8 — 0.313 R²) │ └── submission_template.csv # ⛔ Not included (EY confidential) ├── .gitignore ├── extract_terraclimate.py # TerraClimate API extraction script ├── ey_water_quality_model.py # Final prediction model ├── README.md └── requirements.txt ``` --- ## Usage ```bash # Install dependencies pip install -r requirements.txt # Run the prediction model (requires EY datasets in data/ folder) python ey_water_quality_model.py # (Optional) Re-extract TerraCl …