The role of regional information in estimation of extreme point rainfalls

Previous analysis has shown that inclusion of regional information improves at-site estimation of point rainfalls and makes it possible to obtain estimates at non-monitored sites. The basis for this analysis was a partial duration series (PDS) modelling of individual rainfall observations and use of regional prior distributions for the PDS-parameters in a Bayesian estimation procedure. The advantages of this theoretically satisfactory, but also somewhat complicated procedure are evaluated by means of a comparison with simplified procedures. These include modelling based on regional pooling of all point rainfall data into one sample from a common parent distribution and modelling with disregard of either the dependence between stations or the regional heterogeneity. The different models are analysed and compared with respect to the uncertainty of the predicted extreme events.