Applicability of kriging to regional flood estimation problem in Victoria

Kriging is seldom used for regional flood frequency analysis (RFFA) in Australia. RFFA is usually applied when no or insufficient data are available to assess the probability of exceedance of a flood event for a site of interest. A geostatistical technique, namely ordinary kriging, can interpolate the unknown values and provide the estimation using regional transposition or regionalisation. In this paper, we applied ordinary kriging to estimate design floods at ungauged sites. A data set of 186 stations located in Victoria (Australia) was used to illustrate the approach and its usefulness in regional flood frequency estimation. In order to evaluate the performance of the method, the leave-one-out (LOO) cross validation technique was applied to estimate flood quantiles for selected annual exceedance probabilities (AEP) (50%, 20%, 10%, 5%, 2% and 1%). The results show that the relative error values for estimating the flood quantiles are in the range of 36 to 39%. Results also indicate that the relative error increases as events become rare. In comparison to the RFFE approach (as recommended in Australian Rainfall and Runoff 2016), kriging yields satisfactory results, performing better in more than half the sites.