Research Published in Frontiers in Climate: Deep Learning Improves Seasonal Precipitation Forecasts for the Blue Nile Basin

A new peer-reviewed study by SPS Blue Nile researchers at the Karlsruhe Institute of Technology (KIT), published in Frontiers in Climate, presents a deep learning approach to substantially improve seasonal precipitation forecasts for the Blue Nile Basin.

The paper, authored by Rebecca Wiegels, Christian Chwala, Julius Polz, Luca Glawion, Christof Lorenz, Tanja C. Schober, and Harald Kunstmann, introduces Seasonal AFNOCast – a novel deep learning architecture based on an Adaptive Fourier Neural Operator (AFNO) – to bias-correct and downscale SEAS5 precipitation forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) for the transboundary Blue Nile Basin in Ethiopia and Sudan.

Seasonal precipitation forecasts are essential for climate-sensitive sectors such as agriculture, hydropower, and water management in the region. However, global forecasting systems suffer from systematic biases and insufficient spatial resolution to capture local characteristics. Seasonal AFNOCast addresses these limitations by learning spatial and temporal dependencies across the full ensemble, generating high-resolution, physically plausible precipitation fields.

Evaluated against the established statistical post-processing method Bias Correction and Spatial Disaggregation (BCSD) over the period 2017–2023, results show that both methods substantially improve precipitation distributions and spatial patterns. Seasonal AFNOCast demonstrates particular strengths at longer lead times and in the representation of high-intensity rainfall events. Crucially, it generates forecasts 5–20 times faster than BCSD, making it highly suitable for operational forecasting contexts.

Both methods show clear skill enhancements during the March–May pre-season – a highly variable yet operationally critical period for agricultural decision-making in the basin.

The study was conducted in collaboration with ICPAC, with the explicit aim of supporting operational seasonal forecasting services in the Greater Horn of Africa.

👉 Read the full article: https://doi.org/10.3389/fclim.2026.1691030