Estimation of extreme rainfall quantiles at ungauged sites in the Loess Plateau, China by regional frequency analysis

Abstract

Abstract An increase in the frequency of extreme rainfall events has caused more severe floods than before under the influence of climate change, and is receiving a lot of attention in data‐scarce areas like the Loess Plateau, especially in some small basins without hydrometeorological data, and to avert the risk of disaster needs the estimation of extreme rainfall quantiles. In this study, generalized additive models (GAM) are combined with a nonlinear canonical correlation analysis (NLCCA) procedure to estimate extreme rainfall quantiles at ungauged sites in the Wei River basin on the Loess Plateau. Also, the single inverse distance weighting (IDW) interpolation, NLCCA‐backpropagation (BP), and NLCCA‐radial basis function (RBF) models were used as comparative methods for testing the efficiency of the NLCCA‐GAM model. Because meteorological data are generally lacking in rainfall‐ungauged basins, data at ungauged sites was interpolated by IDW interpolation when NLCCA delineated homogeneous regions. Results of validation indicated that maximum daily rainfall quantiles in the Loess Plateau were well estimated by NLCCA‐GAM‐based regional frequency analysis, implying the NLCCA‐GAM is a useful tool for estimating design extreme rainfall in ungauged basins. Results of comparison demonstrated that NLCCA‐GAM was more robust and better reflected the nonlinear relationship between explanatory and response variables in the region than did comparative approaches. In addition, it was found that the NLCCA‐GAM approach made full use of the data from sites in homogeneous regions to mine the valuable nonlinear correlation between ungauged and gauged sites and estimate extreme rainfall quantiles at ungauged sites.

Keywords

• delineation of homogeneous regions
• maximum daily rainfall quantiles
• NLCCA‐GAM
• regional frequency analysis
• ungauged sites