hide
Free keywords:
-
Abstract:
It is often argued that numerical weather prediction models remain deficient in forecasting specific weather events and that these discrepancies contribute significantly to the overall model error. To clarify these claims, we quantify the forecast error associated with specific weather features in the operational ECMWF Integrated Forecasting System as well as in the ERA5 reanalysis. We detect cyclones, fronts, upper tropospheric jets, and moisture transport axes and quantify contributions to the climatological model error and bias associated with these different features. Furthermore, we identify the contribution of weather features to extreme forecast errors that could classify as forecast busts.
We found that, in specific regions, feature-related errors have twice the amplitude compared to the model error in the absence of weather features during the wintertime. For example, for low level humidity, errors associated with fronts can be twice as large compared to errors when no front is present. We find similar results for moisture transport pathways and show that forecast errors in the upper tropospheric wind are often associated with jets. For some regions and features, the feature-related forecast error is also better than climatology. Considering forecast busts, we assess if the detected features play a prominent role for the top 10% of extreme errors and highlight which feature is the most relevant for different regions. For example, the frequency of occurrence of cyclones is 15-20% along the North Atlantic storm track in the climatology, but frequency of occurrence with respect to the top 10% extreme forecast errors increases to 25-30%.
