Examining El Nino-Southern Oscillation Effects in the Subtropical Zone to Forecast Long-Distance Total Rainfall from Typhoons: A Case Study in Taiwan

Abstract

Typhoon rainfall predictions provide critical information that can be used for flood control and advanced disaster prevention preparations. However, total rainfall nowcasts (i.e., several days ahead) are not available in Taiwan when typhoons are distant. This paper proposes a long-distance total rainfall forecast (LTRF) model and presents a real-time forecasting process that can use the LTRF model to determine the formation and possible approach of typhoons in the future. The LTRF model was formulated using two designed climate scenarios. Scenario 1 considered El Nino-Southern Oscillation (ENSO) effects, whereas scenario 2 did not. Various raw sensor data, comprising climatological characteristics, sea surface temperature, satellite brightness temperatures, and total rainfall, were collected; moreover, attributes of the ENSO indices, including the Southern Oscillation index and the Nino-3.4 sea surface temperature anomaly, were reviewed. The scenario models were constructed using the C4.5 and random forest tree-based algorithms. Typhoon events occurring during 2001-13 and 2014-15 (specifically, Typhoons Matmo and Fung-Wong in 2014 and Soudelor and Dujuan in 2015) were examined for training and testing purposes, respectively. The Hualien Weather Station in Taiwan was selected as a study site, and the forecasting horizon was set at 6 h. Finally, the model simulations, observations, and Central Weather Bureau (Taiwan) nowcasts were compared. The simulation results showed that the proposed LTRF model, when ENSO effects were accounted for, can efficiently forecast total typhoon rainfall when typhoons are distant from Taiwan.