Diagnostics of systematic intensity errors for tropical cyclones in ECMWF forecasts

Wednesday, 20 April 2016: 1:30 PM
Ponce de Leon C (The Condado Hilton Plaza)
Linus Magnusson, ECMWF, Reading, United Kingdom

Forecasting the intensity of tropical cyclones is still a challenge in numerical weather prediction, from both modelling and data assimilation aspects. The dominant intensity error in global models is for cyclones to be too weak, as is generally the case for the ECMWF ensemble forecasts. In contrast, several cyclones in recent years have been developing too deeply in the ECMWF high-resolution (HRES) forecasts, most frequently when approaching Japan.

Studying the lead-time dependence of intensity bias, we find in general too weak cyclones at short lead times and for HRES a drift towards too intense cyclones at longer lead times (5 days or more). However, the bias for long lead-times is connected to the issue that the verification methodology used here only samples mature cyclones and this could influence the results. Preliminary results from coupling the TL1279 (16 km) resolution atmospheric model with a high resolution (0.25° instead of the operational 1°) NEMO ocean model suggests that the over-deepening is at least partly due to a deficiency in the atmosphere-ocean coupling. For young cyclones, which are usually smaller in horizontal scale, the model resolution is still a limiting factor, and with increased model resolution the intensity of these cyclones is shown to improve.

In this presentation we will show examples of tailored diagnostics for tropical cyclones in the context of observation usage in the data assimilation, sensitivity to model resolution and air-sea coupling in the forecasts, applied to a selection of tropical cyclone cases. The results presented will be in the context of coming ECMWF model upgrades that include substantially increased horizontal resolution to 9 km in HRES and 18 km for the ensemble.

- Indicates paper has been withdrawn from meeting
- Submission entered in competition