Seasonal prediction of tropical cyclone activity over the East China Sea using the multivariate least absolute deviation regression method

In present study, we develop and validate a statistical model to predict tropical cyclone (TC) activity over the East China Sea (25°N-35°N, 120°E-130°E) during July, August, and September using the least absolute deviation regression model. Through a lag correlation analysis between the seasonal TC occurrence number in target region and various pre-seasonal environmental parameters during past 20 years (1979-1998), some physically related environmental signals are found. Using screening predictors method based on step-wise regression, the final three predictors, i.e. sea surface temperature, outgoing long-wave radiation, 850-hPa and relative vorticity are selected. The three predictors are related with the El Nino/Southern Oscillation, the western North Pacific summer monsoon and activity of tropical convection, respectively. The correlation coefficient between the leave-one-out cross validation results and the observed seasonal TC occurrence number is 0.69 for the 1979-1998. If the prediction results for 1999-2006 are included, the correlation coefficient becomes 0.70. The results suggest that the model is skillful in predicting the seasonal TC occurrence number over the East China Sea.