Dry Season Monthly Rainfall Forecast for Tseng-Wen Reservoir Catchment

Pei-Min  Zhao

This study aims to develop a monthly rainfall forecast model based on k nearest neighbor (KNN) for Tseng-Wen reservoir catchment. The study integrated KNN with various meteorological variables, such as rainfall, temperature, sea level pressure and sea surface temperature, for developing the rainfall forecast model. The results of proposed rainfall forecast model were compared with seasonal weather outlook provided by Central Weather Bureau, Taiwan. The study assessed the model performance in terms of precision and accuracy for various combinations of meteorological variables. The results indicate that the proposed rainfall forecast model can (1) give better predictions with lower mean absolute error (higher accuracy) (2) significantly reduce standard deviation of prediction values (higher precision). A drought event hit Taiwan during the period from November 2014 to April 2015 due to lack of rainfalls. The study selected the event as a case study to inspect the performance of the proposed model. The results show that (1) all the historical observations fall within the prediction internals; (2) the proposed rainfall forecast model gives a smaller prediction interval then seasonal weather outlook product.

Key words: rainfall forecast model, k nearest neighbor, seasonal weather outlook