A statistical downscaling model, based on the outputs of general circulation models (GCMs) as predictors, was proposed to simulate the daily precipitations in the Shih-Men reservoir catchment in Taiwan. The structure of the proposed downscaling model is composed of two parts: classification and regression. Predictors of classification and regression models were selected from the large-scale weather variables in the National Centers for Environmental Prediction (NECP) reanalysis data based on statistical tests. Discriminant analysis and support vector regression (SVR) were applied to build the classification and regression models. The outputs of five atmosphere-ocean GCMs, which are reported to have properly considered tropical cyclone information and East Asian Monsoon modelling, were used for projecting future precipitations. Data from four grids covering Taiwan were used for developing the downscaling model. The potential of the downscaling models in simulating local precipitations was evaluated, and downscaling results reveal that the proposed downscaling model can reproduce local daily precipitations from large-scale weather variables. Projected local precipitations under two emission scenarios show that the precipitations in the study area tend to decrease.
All Science Journal Classification (ASJC) codes
- Water Science and Technology