Spring rainfall is one of the main sources for reservoir storage in Taiwan. If Mei-Yu and typhoon rainfall was weak in the previous season, insufficient spring rain in the ensuing season will lead to evident water shortage. To reduce the impact of water shortage, accurate predictions of spring rainfall can support the decision making for preparing drought-resistance actions in advance. Therefore, the study analyzed the relationships between spring rainfall and climatic teleconnection indices (i.e., the large-scale climatic factors and El Niño-Southern Oscillation indexes) for constructing a spring-rainfall prediction model. It is found that the climatic teleconnection indices including Niño3, circulation, sea surface temperature, and, wind field near Taiwan are the dominant factors influencing Taiwan’s spring rainfall. Based on the multivariate regression and machine learning techniques (i.e., artificial neural network, support vector machine, and random forests), the spring-rainfall prediction models were established by using the aforementioned dominant factors as the input variables and the average spring rainfall of the six meteorological stations in western Taiwan (i.e., Taipei, Hsinchu, Taichung, Chiayi, Tainan, and Kaohsiung stations) as the output variable. The preliminary results show that the random-forests-based model has the best prediction performance in the case study.
All Science Journal Classification (ASJC) codes
- Agricultural and Biological Sciences(all)