Extended Statistical Downscaling Model Using Data Mining Methods for Modeling Hydrologic Time Series

摘要

The main purpose of this paper was to develop a hydrologic prediction model based on the statistical downscaling model (SDSM) platform and data mining (DM) methods. The model is an extension data mining statistical downscaling model (EDMSDM) that was applied hydrological time series modeling. The six DM technologies, namely multiple linear regressions, ARESLab, k-nearest neighbor, M5PrimeLab, support vector machine (SVM), as well as GMDH (group method of data handling)-type polynomial neural networks constitute the mathematical kernel of EDMSDM. In the instance study of this paper, the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) databases have been utilized for adjustment and test period. The daily records of fifteen rainfall stations distributed in Jiangxi Yifeng River basins in China are utilized to contrast the efficiency of EDMSDM with SDSM. Used data covered 10 years from 1992 to 2001 for daily rainfall. Contradistinction results between EDMSDM and SDSM in the fifteen rainfall stations demonstrates that EDMSDM can offer daily rainfall computations with closer standard deviation and skew values to the actual records for both adjustment and validation periods. Among the presented methods in EDMSDM, the outcomes of the instance study have indicated that ARESLab and M5PrimeLab have offered better performances both in modeling quantity and occurrence of rainfall. The results of this paper have indicated that EDMSDM can be a helpful tool for statistical downscaling of rainfall.