New paper: Using machine learning to improve rainfall forecasts in East Africa

Weather forecasts over tropical areas are typically not very good at predicting how heavy rain will be and exactly when it will fall. This is particularly a problem for parts of East Africa, where heavy rainfall can lead to negative impacts such as flooding. We investigate whether forecasts over East Africa can be improved using a combination of machine learning (ML) and more traditional techniques, where forecast corrections are learned from satellite observations. We show these methods substantially improve forecasts errors, particularly in the timing of rainfall. The ML model performs well even when forecasting over an extremely wet season.

 This work complements the joint project between Oxford AOPP, the World Food Programme, and meteorological agencies in east Africa (https://www.physics.ox.ac.uk/news/ai-led-science-innovation-protects-communities-hit-climate-change), in which similar techniques are being used to improve early warning systems in the region.

Read the paper here: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024MS004796