Forecast skill is unevenly distributed across regions and scales, with major implications for the design of sub-seasonal to seasonal (S2S) services. A recent study (Keane et al., accepted) demonstrates a clear crossover in model skill at 5–7 days lead time: mid-latitudes favour short-term, fine-scale forecasts, while tropical regions exhibit higher skill at longer lead times and coarser spatial scales. This pattern holds for both traditional numerical models and machine learning systems (which surprised us), suggesting that it reflects fundamental atmospheric dynamics rather than modelling artefacts.
A companion study (Dunn-Sigouin et al., accepted) shows that spatial aggregation improves forecast accuracy and extends usable lead time. This is especially true for daily rather than weekly accumulations. I will illustrate these findings using Storm Hans (2023), likely the most expensive weather-related disaster in Norwegian history, to show how aggregation would have improved the usefulness of early warnings for extreme rainfall.
Together, these papers point to a need for region-specific forecast strategies. In tropical regions—especially in Africa—this means investing both in improving short-term capabilities and in harnessing the relative strength of longer-range probabilistic forecasts. Yet these efforts face growing headwinds. U.S. budget cuts have disrupted FEWS NET and threaten CHIRPS and key model data streams such as GFS and CFS, on which many African meteorological services depend.
In this talk, I will reflect on how these scientific and infrastructural challenges intersect in practice, as well as why ECMWF and European institutions may now be called upon to fill emerging global data gaps and support more equitable access to forecast information.