Development of Teleconnection-based Seasonal Rainfall Forecasting Models for Reservoir Watersheds in Taiwan

  • 吳 政諺

Student thesis: Doctoral Thesis


In order to support decision making for preparing drought-resistance actions in advance during the dry period the study developed the teleconnection-based seasonal rainfall forecasting models for the 21 reservoir/dam watersheds in Taiwan The correlations between the seasonal rainfalls for each watershed and the monthly climatic teleconnection indices (i e Circulation Sea Surface Temperature Wind Field Ni?o Sea Surface Temperature Southern Oscillation index Pacific Decadal Oscillation and the previous monthly rainfall of watershed) from 1982 to 2018 were first investigated It is found that these climatic teleconnection indices can be used as the candidates of model input variables Further 127 combinations of these climatic teleconnection indices were used as the model input variables to construct the forecasting models for predicting the seasonal rainfall of watershed for each forecast-beginning month (form October to March) by using the leave-one-out cross-validation For example the forecasting model for the forecast-beginning month October was constructed for forecasting the seasonal rainfall from November to January Three model-construction methods multiple regression analysis (MRA) support vector machine (SVM) and random forest (RF) were adopted for comparison For each reservoir/dam watershed and each forecast-beginning month the optimal teleconnection-based seasonal rainfall forecasting model (i e the model with the optimal input-variable combination) was developed by the three methods Comparison results show that the RF-based forecasting models are better than the MRA-based and SVM-based forecasting models for most cases For simultaneously considering the application convenience and forecasting accuracy the 21 reservoir/dam watersheds were grouped into the northern central and southern regions by their geographic locations For each region the “regional” optimal teleconnection-based seasonal rainfall forecasting model based on RF was developed for each forecast-beginning month Comparison results show that the proposed regional teleconnection-based seasonal rainfall forecasting models perform better than the tradition forecasting models which only use the antecedent monthly rainfall as model input variable
Date of Award2020
Original languageEnglish
SupervisorPao-Shan Yu (Supervisor)

Cite this