Tropical Cyclone Predictions from the Met Office and ECMWF Global Models – Relative Performance and Error Diagnosis

For many years the Met Office global model had lower tropical cyclone track forecast errors than the ECMWF model at short lead times (up to 72 hours). Due to a lower rate of error growth, the ECMWF model predictions at longer lead times (up to 144 hours) had similar or lower errors than the Met Office model. However, between 2004 and 2008 there was a rapid drop in track forecast errors from the ECMWF model not seen in the Met Office model. This has resulted in the ECMWF model outperforming the Met Office model for tropical cyclone track prediction at all lead times from 24 hours onwards in recent years.

In order to diagnose the cause of differences in tropical cyclone track prediction errors between the two models a series of experiments were run using the Met Office global model where the operational global analysis was replaced by the ECMWF analysis before running a forecast. In the majority of cases these experiments showed that it was the ECMWF analysis that was responsible for the better forecast. Further experiments were undertaken involving the transplant of the ECMWF analysis over a restricted area around the tropical cyclone and transplanting selected fields only from the ECMWF analysis. This has enabled further diagnosis of the cause of the differences between the two models' predictions and has shown that analysis of the lower tropospheric wind structure around the tropical cyclone is crucial to producing an accurate track forecast.