Seasonal forecasting of tropical storm frequency using a multi-model ensemble

The skill of seven coupled ocean-atmosphere models to predict the frequency of tropical storms from 1987 to 2001 has been assessed using a procedure for tracking model tropical storms. The tropical storm tracker takes account of the difference of atmospheric horizontal resolution between the different models. Results indicate that the models display some skill in predicting the interannual variability of tropical storms over the Atlantic., the eastern North Pacific, the western North Pacific, the Australian basin and the South Pacific. A simple multimodel forecast has been built by adding all the seven ensemble forecasts together after calibration. The skill of the simple multi-model system is overall better than the skill of any individual model. Over a specific basin, combining several models leads to better forecasts than the best individual model. This indicates that the multimodel approach could benefit the dynamical seasonal forecast of tropical storms. This conclusion is also valid for a longer time period (1959-2001). However, the individual models and the simple multi-model display less skill in predicting the interannual variability of tropical storms in the earlier decades.