Self-Adapting Analog Ensemble Predictions of Tropical Cyclone Tracks

Abstract

An analog model is used to predict the tropical cyclone tracks in the Atlantic and east Pacific basins. The model is self-adapting in its search of ensembles of optimal historic analogs by creating a norm that minimizes the forecast error depending on the model parameters and the kind of prediction. Comparison with the Climatology Persistence (CLIPER) reference model shows different results in the Atlantic and east Pacific basins using the best track data as an independent verification dataset. In the Atlantic, the self-adapting analog model achieves a great circle error of same order as the reference but improves the forecasts by 15%?20% in the east Pacific. In another trial, based on simulated operational data, the performance of both models measured by absolute errors deteriorates compared to the best track data forecasts. However, the self-adapting analog scheme, which is less sensitive to noise, shows positive skill against CLIPER for all lead times in both basins.