Comparison of spatial vine copula and ordinary kriging interpolation methods of daily rainfall with radar information

In recent years, grid-based hydrological modelling of catchments has become commonplace. But with limited scattered gauge data that serve as the basic input, such modelling efforts can be under threat. For this reason, methods for interpolating gauged data over catchment grid nodes have received much attention. Copula and Kriging based methods have been identified as outperforming other methods such as nearest neighbour and thin plate splines. The focus of this paper is to enhance the capability of Spatial Vine Copula (SVC) and Ordinary Kriging (OK) interpolation methods by extracting the relevant spatial structures from radar rainfall records. A 128 km x 128 km region within the range of the Mt Stapylton (near Brisbane) weather radar field is selected for the case study. The region hosts 117 daily rainfall gauged stations. Data for the first 4 months of 2014 were analysed, and zeros of the data of the day were treated as left-censored when fitting two-parameter distributions. Leave-One-Out-Cross-Validation (LOOCV) was used to compare the two competing interpolation methods over 1 km2 grid resolution. Both methods performed very well and similarly. OK marginally outperformed SVC in 58% of cases in root-mean-square-error (RMSE) and 83% of cases in mean-absolute-bias (MAB). However, for days with gauge wetness in excess of 80%, SVC marginally outperformed OK in RMSE and mean-inter-quartile-range (MIQR). © 2015, Engineers Australia. All rights reserved.