Generating spatial precipitation ensembles: impact of temporal correlation structure

摘要

Sound spatially distributed rainfall fields including a proper spatial andtemporal error structure are of key interest for hydrologists to forcehydrological models and to identify uncertainties in the simulated andforecasted catchment response. The current paper presents a temporally coherenterror identification method based on time-dependent multivariate spatialconditional simulations, which are conditioned on precedingsimulations.A sensitivity analysis and real-world experiment are carried out within thehilly region of the Belgian Ardennes. Precipitation fields are simulated forpixels of 10 km × 10 km resolution. Uncertainty analyses in thesimulated fields focus on (1) the number of previous simulation hours on whichthe new simulation is conditioned, (2) the advection speed of the rainfall event,(3) the size of the catchment considered, and (4) the rain gauge density withinthe catchment. The results for a sensitivity analysis show for typical advectionspeeds >20 km h?1, no uncertainty is added in terms of across ensemblespread when conditioned on more than one or two previous hourly simulations. However, forthe real-world experiment, additional uncertainty can still be added when conditioningon a larger number of previous simulations. This is because for actual precipitationfields, the dynamics exhibit a larger spatial and temporal variability. Moreover, bythinning the observation network with 50%, the added uncertainty increases onlyslightly and the cross-validation shows that the simulations at the unobserved locationsare unbiased. Finally, the first-order autocorrelation coefficients show clear temporalcoherence in the time series of the areal precipitation using the time-dependentmultivariate conditional simulations, which was not the case using the time-independentunivariate conditional simulations. The presented work can be easily implemented withina hydrological calibration and data assimilation framework and can be used as animprovement over currently used simplistic approaches to perturb the interpolatedpoint or spatially distributed precipitation estimates.