matbdev/waterscript-ey-data-science-challenge

A multi-modal machine learning solution to forecast water quality parameters in South Africa rivers using Landsat satellite imagery (STAC API), TerraClimate meteorological data, and Snowflake (Snowpark ML). Developed for the EY AI & Data Challenge 2026. # 💧 Optimizing South Africa's Clean Water Supply — EY AI & Data Challenge 2026 --- ## Introduction ### The Problem Access to clean water is one of the most critical challenges of our time. In South Africa, managing and predicting water quality across sprawling, geographically diverse river systems is essential to optimizing the clean water supply and protecting ecosystems. Traditionally, monitoring water quality requires expensive, manual ground-level sampling that lacks spatial and temporal continuity. By bridging the gap between sporadic physical water sampling and continuous satellite/climate observation, machine learning can offer a scalable, cost-effective, and robust solution to predict and manage water resources. ### Program & Project Description This repository contains a full-scale machine learning and data engineering solution built for the **EY AI & Data Challenge 2026**. The core objective is to **develop a machine learning model capable of forecasting three key water quality parameters** across various river locations in South Africa: 1. **Total Alkalinity** 2. **Electrical Conductance** (Salinity/Conductivity) 3. **Dissolved Reactive Phosphorus** (Phosphate levels) The dataset spans from **2011 to 2015** across approximately **200 river locations**. To build a model that generalizes well, the solution must predict these parameters for a separate validation dataset containing river locations from entirely different regions not present in the training set. To achieve high-accuracy predictions, this project leverages a multi-modal data pipeline that enriches physical water samples with: - **Satellite Imagery (Landsat)**: Extracted via the Microsoft Planetary Computer STAC API to capture surface reflectance and turbidity. - **Climate & Weather Datasets (TerraClimate)**: Providing historical monthly meteorological and water balance data. --- ## Technology Stack This project leverages a modern data-science and cloud data warehouse stack …