A Hybrid Statistical Downscaling Method Based on the Classification of Rainfall Patterns

A hybrid statistical downscaling method based on the classification of rainfall patterns is presented which is capable of overcoming the poor representation of extreme events. The large-scale datasets, which are obtained from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data and the global circulation models (GCMs) outputs, and the local daily rainfall data are analyzed to assess the impacts of climate change on rainfall. The proposed method is composed of two steps. The first step is the classification of daily rainfall patterns. The detrended fluctuation analysis (DFA) is introduced to define the extreme rainfall. Two classification models, extreme rainfall and wet rainfall, are developed to describe the relationship between large-scale weather factors and rainfall patterns using support vector machine (SVM). These two models are able to identify the three rainfall patterns (the extreme, the normal and the dry rainfall) of the daily weather factors. The second step is the estimation of daily rainfall. The improved self-organizing linear output map (ISOLO) is adopted to estimate the rainfall for the aforementioned three different rainfall patterns. The future rainfall changes are calculated for the periods 2046–2065 and 2081–2100 under the A2 and B1 scenarios. An application to Taiwan has shown that the proposed method provides reliable and accurate rainfall-pattern classification. In addition, the improvement of the estimation of daily rainfall is significant, especially for the extreme rainfall. In conclusion, the proposed method is effective to overcome the poor representation of extreme events and the impacts of climate change on rainfall are analyzed.