Downscaling transient climate change using a Neyman-Scott Rectangular Pulses stochastic rainfall model

The future management of hydrological systems must be informed by climate change projections at relevant time horizons and at appropriate spatial scales. Furthermore, the robustness of such management decisions is dependent on both the uncertainty inherent in future climate change scenarios and the natural climate system. Addressing these needs, we present a new transient rainfall simulation methodology which combines dynamical and statistical downscaling techniques to produce transient (i.e. temporally non-stationary) climate change scenarios. This is used to generate a transient multi-model ensemble of simulated point-scale rainfall time series for 1997-2085 for the polluted Brevilles spring in Northern France. The recovery of this previously potable source may be affected by climatic changes and variability over the next few decades. The provision of locally-relevant transient climate change scenarios for use as input to hydrological models of both water quality and quantity will ultimately provide a valuable resource for planning and decision making. Observed rainfall from 1988-2006 was characterised in terms of a set of statistics for each calendar month: the daily mean, variance, probability dry, lag-1 autocorrelation and skew, and the monthly variance. The Neyman-Scott Rectangular Pulses (NSRP) stochastic rainfall model was fitted to these observed statistics and correctly simulated both monthly statistics and extreme rainfall properties. Multiplicative change factors which quantify the change in each statistic between the periods 1961-1990 and 2071-2100 were estimated for each month and for each of 13 Regional Climate Models (RCMs) from the PRUDENCE ensemble. To produce transient climate change scenarios, pattern scaling factors were estimated and interpolated from four time-slice integrations of two General Circulation Models which condition the RCMs, ECHAM4/OPYC and HadCM3. Applying both factors to the observed statistics provided projected transient rainfall statistics (PTRS) to which piece-wise smoothly varying transient rainfall model parameterizations were fitted. These fits provided good representations of the PTRS for each RCM. An ensemble of 100 continuous daily rainfall time series, with steadily varying stochastic properties which model these projections of transient climate change, was then simulated using a new transient NSRP simulation methodology for each RCM. Together the ensembles form a 1300 member transient multi-model ensemble of rainfall time series. The simulated transient ensemble properties were investigated, identifying RCMs giving rise to unusual behaviour. For the Brevilles, annual rainfall is projected to decrease until 2085 but the change is highly sensitive to General Circulation Model forcing: ECHAM4-driven RCMs project larger annual decreases than HadCM3/HadAM3H/P driven RCMs. All RCMs project an increase in winter rainfall and a larger summer decrease. An increase of similar to 10% in the 10-year return period annual maximum rainfall is projected by 2085, however both strong increasing trends and a slight decreasing trend are found for individual RCMs. Compared with transient RCMs, the new methodology provides a number of advantages: reduced biases, point scale scenarios relevant for local-scale impact studies, improved representation of natural variability and improved representation of extremes. (C) 2009 Elsevier B.V. All rights reserved.