A Bayesian joint probability approach for flood record augmentation

At-site flood frequency prediction is often subject to large uncertainties because of the limited length of record available at the gauging station of interest. In many cases, however, one can find a nearby gauging station that has a considerably longer record. In this paper, we introduce a two-site joint probability approach for the transfer of flood information between two stations. We assume that the annual maximum floods at the two stations follow a bivariate generalized extreme value distribution. We use a Bayesian procedure to estimate the parameters of the distribution by jointly using all the flood data available at the two stations. Efficient transfer of information is achieved by simultaneously carrying out (1) the inference of correlation of floods at the two stations, (2) the transfer of information between the two stations, and (3) the inference of the frequency distributions for the two stations.