Sample Optimization of Ensemble Forecast to Simulate a Tropical Cyclone Using the Observed Track

Abstract

Ensemble forecasting is widely used in numerical weather prediction. However, the ensemble may not satisfy a perfect Gaussian probability distribution because of a limited number of members, with some members significantly deviating from the true atmospheric state. Such outliers (belonging to low probability events) may downgrade the accuracy of an ensemble forecast. In this study, the observed track of a tropical cyclone (TC) is used to restrict the probability distribution of samples by investigating the evolution of TCs. Unlike data assimilation, the method we employed uses observational data. By restricting the probability distribution, ensemble spread could be decreased through sample optimization. In addition, the prediction results showed that track and intensity errors could be reduced by sample optimization. When the vertical structures of TCs considered in this study were compared, different thermal structures were found. This difference may have been caused by sample optimization, which may affect intensity and track. Nevertheless, it should be noted that the replacement of a large number of inferior samples may inhibit the improvement of simulated results.