392481 - Combination of ANN ensemble for improving performance of monthly rainfall forecast
Wednesday, June 6
10:30 AM - 12:00 PM
Location: Greenway IJ
Ju-Young Shin, Seoul – Yonsei University; Kyoungwon Joo, Seoul – Yonsei University; Jun-Haeng Heo, Seoul – Yonsei University
Accurate and outstanding monthly rainfall forecast is still a challenging task for efficient water resources management and supply under the nonstationary climate conditions. Until now, various techniques and materials with different models have been applied to forecast the monthly rainfall and the combined results from different models show improved performance than that of single model. In this study, we aim to improve the performance of the monthly rainfall forecast by applying new concept of combining the ANN ensemble obtained from the various types of ANN model. For this, the climate indices which can consider the effect of climate variability in monthly rainfall were used as input data for the ANN model. The ANN ensembles were classified into three categories to consider the initial weight conditions, model structure, and input variables. After generating ANN ensembles, the concept of multi-model combination techniques for forecast combination was employed to improve their performance. For application, monthly rainfall from South Korea and climate indices from NOAA/ESRL were used. The results suggest that the combination of ANN ensemble gives improved and reliable forecast result than one single ANN model. Especially, it showed enhanced results in the forecast of extreme values of summer rainy season which showed limitations in the previous studies.