Comparison and evaluation of multiple GCMs, statistical downscaling and hydrological models in the study of climate change impacts on runoff

• Published in: Journal of hydrology (Amsterdam) 2012-04, Vol.434-435, p.36-45
• Main Authors: Chen, Hua, Xu, Chong-Yu, Guo, Shenglian
• Format: Article
• Language: English
• Subjects: basins; China; climate; Climate change; climate models; Earth sciences; Earth, ocean, space; emissions; Engineering and environment geology. Geothermics; Exact sciences and technology; GCM; Hanjiang basin; hydrologic models; Hydrological models; Hydrology; Hydrology. Hydrogeology; model validation; Pollution, environment geology; rain; rainfall simulation; runoff; statistical analysis; Statistical downscaling; statistical models; water balance
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2012.02.040

► We compare two statistical downscaling methods (SD) combined with two GCMs and two hydrological models. ► We find runoff simulations vary greatly driven by rainfall simulations from same climate scenario and different SD. ► SDSM performs better than SSVM in simulating rainfall, but simulated runoff driven by SDSM’ simulation worse than by SSVM’s. ► We verified Nash–Sutcliffe efficiency can be used as a key indicator to evaluate the downscaling rainfall’s performance. In this study a rigorous evaluation and comparison of the difference in water balance simulations resulted from using different downscaling techniques, GCMs and hydrological models is performed in upper Hanjiang basin in China. The study consists of the following steps: (1) the NCEP/NCAR reanalysis data for the period 1961–2000 are used to calibrate and validate the statistical downscaling techniques, i.e. SSVM (Smooth Support Vector Machine) and SDSM (Statistical Downscaling Model); (2) the A2 emission scenarios from CGCM3 and HadCM3 for the same period are used as input to the statistical downscaling models; and (3) the downscaled local scale climate scenarios are then used as the input to the Xin-anjiang and HBV hydrological models. The results show that: (1) for the same GCM, the simulated runoffs vary greatly when using rainfall provided by different statistical downscaling techniques as the input to the hydrological models; (2) although most widely used statistics in the literature for evaluation of statistical downscaling methods show SDSM has better performance than SSVM in downscaling rainfall except the Nash–Sutcliffe efficiency (NSC) and root mean square error-observations standard deviation ratio (RSR), the runoff simulation efficiency driven by SDSM rainfall is far lower than by SSVM; and (3) by comparing different statistics in rainfall and runoff simulation, it can be concluded that NSC and RSR between simulated and observed rainfall can be used as key statistics to evaluate the statistical downscaling models’ performance when downscaled precipitation scenarios are used as input for hydrological models.