An approach to revising the climate forecast system reanalysis rainfall data in a sparsely-gauged mountain basin

Abstract

Accurate spatial rainfall data are a key input parameter for distributed hydrological models and a significant contributor to hydrological model uncertainty when rain gauges are sparsely distributed, especially in complex mountain basins. The rainfall data recorded by gauge stations are considered to be real and accurate but gauged rainfall data are sometimes not representative of the rainfall spatial distribution due to a lack of stations. The Climate Forecast System Reanalysis (CFSR) product is widely used to provide the spatial variability of rainfall, especially in sparsely-gauged basins. However, the CFSR has a critical flaw of a high single-point observation error. Hence, this study developed an approach to revise the daily CFSR rainfall data by combining gauged rainfall data and considering the spatial heterogeneity of the CFSR rainfall. First, Thiessen polygons were generated to define the measurement domain of each rain gauge station. Second, spatial regression equations were developed between the CFSR rainfall data based on the size of the correlation coefficient (R), i.e., the R was ranked to determine the criteria. Third, the CFSR rainfall data in the gauged pixels were replaced by the gauged rainfall data (CFSR pixels containing one gauge station that are termed gauged pixels). Finally, the CFSR rainfall data in the non-gauged pixels was corrected based on the regression equations obtained in the second step. The upstream of the Lancang-Mekong River (transboundary river in Southeast Asia) was served as an example and three types of rainfall data (gauged, CFSR, and corrected CFSR rainfall data) were applied to establish the Soil and Water Assessment Tool (SWAT) model, which was used to simulate runoff in the upstream of the Lancang-Mekong River (UL-MR) at monthly scales. We investigated the difference in the SWAT model results among these three datasets using the relative bias (BIAS), coefficient of determination (R2), and Nash-Sutcliffe efficiency (NSE). The results indicated that the CFSR rainfall data corrected by our proposed method exhibited super performance compared to the gauged rainfall data for discharge simulations based on the SWAT model; the NSE value increased by 18.92% (from 0.74 to 0.88), the R2 value increased by 2.30% (from 0.87 to 0.89), and the BIAS value decreased by 9.48% (from 17.4% to 7.95%) in the validation period at the outlet station of the UL-MR. The proposed correction method takes into account the spatial heterogeneity of rainfall and achieves a better result for the hydrological simulation of discharge than the gauged and CFSR rainfall data, especially in complex mountain basins with sparsely distributed gauges.