A Dynamical-Statistical-Analog Ensemble Forecast Model: Theory and an Application to Heavy Rainfall Forecasts of Landfalling Tropical Cyclones

Abstract

Abstract Combining dynamical models with statistical algorithms is an important way to improve weather and climate prediction. In this study, a concept of a perfect model, whose solutions are from observations, is introduced, and a dynamical-statistical-analog ensemble forecast (DSAEF) model is developed as an initial-value problem of the perfect model. This new analog-based forecast model consists of the following three steps: (i) construct generalized initial value (GIV), (ii) identify analogs from historical observations, and (iii) produce an ensemble of predictands. The first step includes all appropriate variables, not only at an instant state but also during their temporal evolution, that play an important role in determining the accuracy of each predictand. An application of the DSAEF model is illustrated through the prediction of accumulated rainfall associated with 21 landfalling typhoons occurring over South China during the years of 2012–16. Assuming a reliable forecast of landfalling typhoon track, two different experiments are conducted, in which the GIV is constructed by including (i) typhoon track only; and (ii) both typhoon track and landfall season. Results show overall better performance of the second experiment than the first one in predicting heavy accumulated rainfall in the training sample tests. In addition, the forecast performance of both experiments is comparable to the operational numerical weather prediction models currently used in China, the United States, and Europe. Some limitations and future improvements as well as comparisons with some existing analog ensemble models are also discussed.