Comparison study of design rainfall mapping using ordinary kriging and kriging with external drift

Abstract

Early development of Extreme Value Theory (EVT) was motivated by a need to quantify the probability of unusually large (or small) events in hydrology and climatology. The current challenge in modelling extremes is how to adequately account for variation of the processes in both space and time. The statistic of interest in EVT is the 1-in-N year return level or design value as it is known in hydrology. This is an important quantity as it helps to identify areas where flood protection infrastructure needs to be erected or improved as well as identifying communities located in high floodrisk areas. This necessitates a regional, rather than an in-situ estimate of the N-year design rainfall value. In this study we focus on estimating the N-year design rainfall surface in a case where the number of sampled sites is small using inverse distance weighting, ordinary kriging and kriging with external drift methods. Does incorporating additional explanatory information, given a spatially sparse sample, lead to any improvement to an estimate of the design rainfall surface?