Seasonal prediction of typhoons approaching the Korean Peninsula using several statistical methods

Typhoon is a devastating weather system, and the typhoon forecast is one of the most important issues to minimize their damages. In this study, we developed multiple linear regression, backpropagation neural network, support vector machine (SVM) and regression tree models and compared their results. To make a choice of reasonable predictors for statistical seasonal prediction of the number of typhoons, the climatology of typhoon activity over the Korean Peninsula and correlation between circulation indices in preceding winter and the number of typhoon approaching the Korean Peninsula (NTY-KP) were analyzed. The main findings were drawn as follows. (1) The interannual variability in NTY-KP for June–July–August–September during the period of 1949–2020 has a large variability, whose mean value is 2.51 and standard deviation is 1.55. (2) The lag correlation maps of NTY-KP with the area-averaged sea level pressure and geopotential height, air temperature, zonal and meridional wind anomalies at various isobaric levels in preceding winter over the area 10° S–90° N, 60° E–60° W were analyzed, and 22 indices described by difference between area-averaged climate variable anomalies over any two areas in preceding winter were chosen as potential predictors, and their own circulation linkages well related to NTY-KP were statistically identified. (3) In seasonal prediction experiments for NTY-KP, the prediction skills of several models were evaluated with correlation coefficient (R), root-mean-square-error (RMSE) and mean absolute error (MAE). The results show that SVM model had an obvious advantage over the other three models, with 0.9 of RMSE, 0.85 of R and 0.55 of MAE.