Random Forest–Predicted 1 km² Monthly Surface Ozone over Sub-Saharan Africa, 2005–2025

Abstract

Monthly 1 km² surface ozone predictions from the Random Forest model developed in “Machine Learning-Based Prediction of Monthly Surface Ozone Over Sub-Saharan Africa Using Satellite-Derived Precursors, Meteorology, and Surface Measurements.” The model was trained primarily on INDAAF observations from predominantly rural and semi-savannah environments, with a few sites near urban centers. Therefore, predicted ozone magnitudes in urban areas should be interpreted with caution. The predictions are provided on a 1 km² grid; however, this fine prediction grid does not imply 1 km² native spatial information for all predictors, as some input datasets have coarser native spatial resolutions. Consequently, the ability of the dataset to resolve fine-scale urban ozone gradients may be limited, and the machine-learning model may smooth extreme ozone concentrations. For details on the model, training data, predictors, validation, and methodology, please refer to the accompanying paper.