Utilization of the Copula-Based Composite Likelihood Approach to Improve Design Precipitation Estimates Accuracy

Coupling of the bivariate copula function and likelihood function was utilized for parameter estimation in precipitation frequency analysis. In this method, data of various lengths in a homogeneous region were incorporated into a multivariate framework. Therefore, the information contained in one series can be employed to improve the accuracies of the design values of other series. Annual precipitation data collected from 36 stations in the Guanzhong region, China, were used as a case study. To obtain more reliable estimates, the stationarity and serial independence of each series were tested first. Then, the entire study region was divided into four homogeneous sub-regions, and frequency analyses were carried out in each sub-region. Two measures, the standard error of fit (SEF) and the maximum absolute relative deviation (MARD), were used to compare the goodness-of-fit of the copula-based composite likelihood approach with the univariate method. Results showed that the SEF and MARD values of the design precipitation values obtained from the copula-based composite likelihood approach were lower than those from the univariate method, and the decrease became more significant for stations with a shorter series in the bivariate composite case. Therefore, this study indicates the advantage of the use of a copula-based composite likelihood approach for improving the accuracy of design precipitation values.