Comparative Analysis of AIFS and IFS Forecasts for Two Heavy Rainfall Events

The forecasting performance and reasons for differences of ECMWF’s Artificial Intelligence Forecasting System (AIFS) and Integrated Forecasting System (IFS) for two heavy rainfall events was analysised. Spatial verification indicates that the IFS precipitation field can be used to identify smaller heavy rainfall objects, with better consistency in precipitation intensity and local distribution characteristics compared to observations, while AIFS performs better in forecasting the spatial location and area of heavy rainfall objects. AIFS exhibits an excessive frequency of precipitation above light rain, resulting in lower Threat Score (TS) and Equitable Threat Scores (ETS) for ≥ 0.1 mm precipitation, while the amplitude of precipitation forecasts is less than that of IFS, which better forecasts the extremity of precipitation. AIFS excels in forecasting the circulation patterns and the position of mesoscale systems generating precipitation for both events compared to IFS, but the convergence intensity of airflow in the mid-to-low layers is lower than IFS. AIFS shows a positive anomaly of Integrated Water Vapor Transport (IVT) on land due to localized mesoscale systems relative to ERA5, whereas IFS exhibits a consistent positive anomaly of IVT from the ocean to the heavy rainfall area, indicating stronger moisture transport by IFS. The standard deviation of IFS precipitable water (PW) is stronger than ERA5, with vertical velocity profiles matching ERA5 very well, while AIFS has a weaker PW standard deviation and fails to depict the profile characteristics of vertical velocity, which may be an important reason for the weaker intensity forecast of heavy rainfall by AIFS.