Air pollution is the second leading cause of death worldwide, and the problem is especially serious in places like sub-Saharan Africa, where reliable air quality data is limited. In this study, we use satellite data, weather datasets and machine learning to derive daily levels of particulate air pollution (PM_2.5) in Ghana at a high resolution (1 km × 1 km) over the past 20 years (2005–2024). For model training, we gather model input data from NASA satellites, European Space Agency satellites and European ReAnalysis v5 data along with surface air quality measurements from a network of over 100 monitors across Ghana. Our best-performing model is a neural network-based machine learning model, which predicts pollution levels more accurately than other existing methods. Finally, we used a World Health Organization tool (AirQ+) to quantify effects of particulate air pollution on people’s health and life expectancy in Ghana. This is one of the first efforts to produce a high-resolution, high-accuracy map of PM_2.5 across Africa. The dataset will be shared on a public platform to help policymakers address pollution and protect public health through evidence-based policies.