s-oeder-ds/n2o-emissions-africa-ml

# N₂O Emissions in Africa — Machine Learning Portfolio Project ## Overview This repository contains a machine learning project for modeling nitrous oxide (`N₂O`) emissions across Sub-Saharan African ecosystems. The project uses open scientific data from PANGAEA and follows a complete applied data science workflow: data acquisition, raw-data validation, preprocessing, exploratory data analysis, feature documentation, and model preparation. The final modeling target is the prediction of soil-surface N₂O flux from spatial, temporal, meteorological, soil-related, and management-related features. The project is intentionally structured as a portfolio project. The focus is not only on producing a model score, but also on building a workflow that is readable, reproducible, and suitable for presentation. --- ## Research Context Nitrous oxide is a highly relevant greenhouse gas in agricultural and ecological systems. Its emissions are influenced by nitrogen availability, soil moisture, soil temperature, precipitation events, land use, and management practices. The dataset behind this project combines observations from multiple African land-use systems, including croplands, forests, plantations, and grasslands. This makes the project a realistic tabular machine learning task with spatial structure, heterogeneous feature groups, missing values, and potentially non-linear process relationships. --- ## Project Goals The main goals are: 1. build a reproducible workflow for downloading and preparing the PANGAEA data, 2. inspect and document the dataset structure, 3. analyze N₂O emission patterns across land-use systems and geography, 4. prepare machine-learning-ready features, 5. train an XGBoost regression model, 6. evaluate prediction quality with regression metrics and residual diagnostics, 7. interpret important model drivers in a scientifically plausible way. --- ## Data Source The data comes from the PANGAEA bundled publication: **Agredazywczuk, P. et …