Uncertainty Propagation of Regional Climate Model Precipitation Forecasts to Hydrologic Impact Assessment

Abstract

A Monte Carlo framework is adopted for propagating uncertainty in dynamically downscaled seasonal forecasts of area-averaged daily precipitation to associated streamflow response calculations. Daily precipitation is modeled as a mixture of two stochastic processes: a binary occurrence process and a continuous intensity process, both exhibiting serial correlation. The parameters of these processes (e.g., the proportion of wet days and the average wet-day precipitation intensity in a month) are derived from the forecast record. Parameter uncertainty is characterized via an empirical Bayesian model, whereby such parameters are modeled as random with a specific joint probability distribution. The hyperparameters specifying this probability distribution are derived from historical precipitation records at the study basin. Simulated parameter values are then generated using the Bayesian model, leading to alternative synthetic daily precipitation records simulated via the stochastic precipitation model. The set of such synthetic precipitation records is finally input to a physically based deterministic hydrologic model for propagating uncertainty in forecasted precipitation to hydrologic impact assessment studies. The stochastic simulation approach is applied for generating an ensemble (set) of synthetic area-averaged daily precipitation records at the Hopland basin in the northern California Coast Range for the winter months (December through February: DJF) of 1997/98. The parameters of the stochastic precipitation model are derived from a seasonal precipitation forecast based on the Regional Climate System Model (RCSM), available at a 36-km2 grid spacing. The large-scale forcing input to RCSM for dynamical downscaling was a seasonal prediction of the University of California, Los Angeles, Atmospheric General Circulation Model. A semidistributed deterministic hydrologic model (?TOPMODEL?) is then used for calculating the streamflow response for each member of the area-averaged precipitation ensemble set. Uncertainty in the parameters of the stochastic precipitation model is finally propagated to associated streamflow response, by considering parameter values derived from historical (DJF 1958?92) area-averaged precipitation records at Hopland.