Properties of extreme point rainfall III: Identification of spatial inter-site correlation structure

A general stochastic model that describes the observed variability of extreme rainfall properties estimated at different sites in a region is developed. The observed variability is divided into sampling variability and variability due to regional heterogeneity. A method for estimating the inter-site correlation structure of the at-site sampling errors based on partial duration series theory is developed, and geostatistical theory is introduced to estimate the spatial correlation structure of the true, but unknown, variables. On this basis the variance of the regional mean value can be determined and used in a regional design approach. The theory is applied to extreme values of rainfall depth and maximum 10 minutes intensity per rainfall event measured at 43 gauging stations in Denmark. Adding one design standard deviation to the regionally averaged T-year event estimates yields increased design values between 8 and 18%, depending on the return period and the rainfall characteristic considered. For small return periods this compares to results obtained from traditional non-parametric statistics, but for large return periods the suggested approach yields significantly lower and more accurate design values.