Cost-Effective ECMWF AIFS Ensemble Inference for Subseasonal Forecasting in East Africa 

In Eastern Africa, subseasonal forecasts are critical for early warning systems as climate extremes severely impact food security and livelihoods. ECMWF Artificial Intelligence Forecasting System (AIFS ENS v1.0) , an ensemble-based probabilistic data-driven forecast model developed by ECMWF offers unprecedented opportunities for regional applications through AI-driven weather prediction, but GPU compute costs and data access challenges limit deployment. As participants in the ECMWF AI Weather Quest, we developed solutions enabling cost-effective, cloud-based AIFS ensemble forecasting tailored for regional climate centers.   We implemented a workflow (github.com) leveraging Google Cloud Platform infrastructure. Initial conditions are accessed via ECMWF's IFS data stored at AWS (Amazon Web Service) open data program at S3 Cloud storage using GRIB index-kerchunk, and VirtualiZarr methods for efficient data streaming without local storage overhead. The workflow employs experimental FP16 (half-precision) inference on AIFS ensemble models along with the standard FP32, evaluating GPU memory requirements and enabling deployment on cost-effective T4/L4 GPUs rather than expensive A100 instances.   Verification results from the SON (September-October-November) 2025 season as part of the AI Weather Quest demonstrates that Team Fahamu's submission using AIFS ensemble forecasts for temperature and mean sea-level pressure outperforms climatology benchmarks. Regional evaluation over East Africa reveals promising subseasonal skill for temperature at lead times of 2-4 weeks—critical timescales for agricultural planning and anticipatory drought/flood action—while evaluation of precipitation forecasts is ongoing. This method provides a scalable template for regional climate centers globally to operationalize state-of-the-art AI weather models cost-effectively, advancing the democratization of advanced forecasting capabilities.