RainSim: A spatial–temporal stochastic rainfall modelling system

RainSim V3 is a robust and well tested stochastic rainfall field generator used successfully in a broad range of climates and end-user applications. Rainfall fields or multi-site time series can be sampled from a spatial–temporal Neyman–Scott rectangular pulses process: storm events occur as a temporal Poisson process; each triggers raincell generation using a stationary spatial Poisson process; raincells are clustered in time lagging the storm event; each raincell contributes rainfall uniformly across its circular extent and throughout its lifetime; raincell lag, duration, radius and intensity are random variables; orographic effects are accounted for by non-uniform scaling of the rainfall field. Robust and efficient numerical optimization schemes for model calibration are identified following the evaluation of five schemes with optional log-transformation of the parameters. The log-parameter Shuffled Complex Evolution (lnSCE) algorithm with a convergence criterion is chosen for single site applications and an effort limited restarted lnSCE algorithm is selected for spatial applications. The new objective function is described and shown to improve model calibration. Linear and quadratic expressions are identified which can reduce the bias between the fitted and simulated probabilities of both dry hours and dry days as used in calibration. Exact fitting of mean rainfall statistics is also implemented and demonstrated. An application to the Dommel catchment on the Netherlands/Belgian border illustrates the ability of the improved model to match observed statistics and extremes.