Stephen-Austine/rhea-soil-challenge

Reproducible soil nutrient prediction pipeline for Africa using tabular soil chemistry, remote sensing, and environmental covariates. # Rhea Soil Nutrient Prediction Challenge A reproducible machine learning pipeline for predicting 13 soil nutrient concentrations across Africa using tabular soil chemistry, remote sensing, and environmental covariates. ## Project Summary - **Goal:** Predict 13 soil nutrient concentrations for test locations using tabular soil data, remote sensing features, and environmental covariates. - **Targets:** `Al`, `B`, `Ca`, `Cu`, `Fe`, `K`, `Mg`, `Mn`, `N`, `Na`, `P`, `S`, `Zn` - **Metric:** RMSE averaged across all nutrients. - **Masking rule:** `TargetPred_To_Keep.csv` identifies which nutrient values may be submitted; masked entries must be forced to zero before submission. ## What’s Included - `data/raw/`: original competition CSVs - `data/external/`: downloaded Earth observation and soil grid data - `outputs/eda_plots/`: EDA visualizations - `src/`: project scripts for data download, feature engineering, model training, and prediction - `notebooks/`: exploratory analysis notebooks - `submissions/`: generated submission files ## Repository Structure | Path | Purpose | |---|---| | `data/raw/` | Competition data inputs (train, test, sample dates, masks) | | `data/external/` | EO and SoilGrids data caches | | `src/download_eo_data.py` | Download SoilGrids / WorldClim / Sentinel2-derived features | | `src/feature_engineering.py` | Create engineered model features from raw + external data | | `src/train.py` | Train models, generate predictions, and apply submission mask | | `notebooks/` | Exploratory notebooks and analysis scripts | | `submissions/` | Final submission files | ## Setup ```bash pip install -r requirements.txt ``` If you have a Python environment manager available, create and activate a virtual environment first. ## Workflow ### 1. Exploratory Data Analysis ```bash python notebooks/01_EDA.py ``` Purpose: understand nutrient distributions, missing values, depth patterns, and spatial coverage. ### 2. Download External Data ```bash python src/d …