This project builds a robust, end-to-end Machine Learning pipeline to predict water quality parameters (Total Alkalinity, Electrical Conductance, Dissolved Reactive Phosphorus) in South Africa using satellite imagery and climate data.

Core Engineering Stack (核心技术栈):

• Data Engineering: Microsoft Planetary Computer API (pystac-client, odc-stac) for pure, temporally-bounded (±30 days) optical band extraction with strict bit-wise Cloud/Shadow masking.
• Feature Engineering: Domain-specific synthesis of Remote Sensing Water Indices (SABI, NDWI, MNDWI, WRI, NDVI) and TerraClimate data integration.
• Modeling Engine: Ensembled Gradient Boosting framework utilizing XGBoost, LightGBM, and CatBoost for heterogeneous tabular approximation.
• MLOps & Validation: Automated CI/CD (GitHub Actions) hooked to a custom Spatial GroupKFold local evaluation script, ensuring strict geographical extrapolation testing without coordinate leakage.

Official Website: EY Open Science Data Challenge 2026

本项目采用现代 MLOps 与数据工程最佳实践，构建了一条端到端的机器学习流水线。技术栈涵盖了从微软行星计算机 API 的纯净光谱抓取、去云处理，到基于环境物理构造遥感水体指数的特征工程区，并最终交由深度调优的异构集成树模型（XGB/LGBM/CatBoost）进行拟合验证。同时项目内置了严格的防御性自动化空间交叉验证框架。

Please refer to the detailed Bilingual Project Manual covering the entire optimization journey: 👉 EY_Challenge_2026_Report.pdf (Located in doc/dist/)

1. Clone the repo:

   git clone https://github.com/EUR-UN/EY_Challenge_2026.git
   cd EY_Challenge_2026

2. Install Python Dependencies:

   pip install -r resources/code/general/requirements.txt
   pip install xgboost lightgbm catboost optuna pystac-client odc-stac scikit-learn pandas numpy

The project has evolved far beyond the official baseline notebook into a modular Python pipeline:

• Data Fetching: python src/data/fetch_planetary_data.py (Downloads pristine optical bands and cloud masks from MS Planetary Computer)
• Model Training: python src/models/ensemble_model.py (Trains the Gradient Boosting ensemble)
• Local Validation: python src/evaluation/evaluate_local.py (Calculates Spatial K-Fold CV mapped to the LB)

EY_Challenge_2026/
├── doc/                 # LaTeX Documentation Source
│   ├── dist/            # Compiled PDF (stable name)
│   └── chapters/        # Bilingual Content
├── resources/
│   ├── code/            # Cleaned Code Packages (Snowflake/General)
│   ├── data/            # Training Data & Templates
│   └── media/           # Images & Tutorials
└── .github/             # Governance templates

We welcome contributions! Please see CONTRIBUTING.md for details on our Fork & Pull Request workflow.

This project is licensed under the Business Source License 1.1 (BUSL-1.1).

⚠️ Restriction / 限制: You may NOT use this code to submit an entry to the EY Open Science Data Challenge 2026. 您不得使用此代码提交 2026 EY 开放科学数据挑战赛 的参赛作品。

🔓 Open Source Date / 开源日期: On 2026-05-07 (after the challenge ends), this restriction lifts and the license automatically converts to MIT, allowing full free use. 在 2026年5月7日（挑战结束后），此限制将解除，许可证自动转换为 MIT，允许完全免费使用。