Tropical cyclones in the GEOS S2S forecasts

This work examines the climatology, prediction skill, and predictability of tropical cyclones (TCs) in NASA's Global Earth Observing System Subseasonal to Seasonal (GEOS-S2S) forecast system versions 2 and 3. The representation of TCs and the associated precipitation (TCP) in both GEOS versions are compared to observations as well as to each other. The skill of GEOS subseasonal TC and TCP forecasts are compared with models participating in the WMO S2S project. GEOSv2, like most S2S models (except the ECMWF), is not skillful at predicting TC activity or TCP beyond week 1 lead times, depending on the basin. Forecast calibration improves skill such that calibrated GEOSv2 forecasts are skillful up to week 2 in most basins. The model-based predictability analysis of TC activity shows how TC forecasts are modulated by the Madden-Julian Oscillation (MJO) phase; there is, however, a gap between MJO-related prediction skill and predictability due to the model biases, in particular due to the spatial distribution of TCs.

TCs in GEOSv3, the latest version of the GEOS-S2S system, differ from those in GEOSv2 in many aspects. For example, GEOSv2 simulates too few TCs in the North Atlantic while GEOSv3 simulates slightly too many, compared to observations. TCs in GEOSv3 are, on average, stronger than in GEOSv2. Ongoing process-oriented diagnostics aims to understand these climatological differences. Furthermore, differences in skill due to increased ensemble members in GEOSv 3 (5 members, or 15 if including ocean perturbations) relative to GEOSv2 (4 members) will also be discussed.