Abstract
While Stochastic Weather Generators (SWGs) are used intensively in climate and hydrological applications to simulate hydroclimatic time series and estimate risks and performance measures linked to climate variability, there have been few investigations into how many realizations are required for a robust estimation of these measures. Given the computational cost and time necessary to force climate-sensitive systems with multiple realizations, the estimation of the optimal number of synthetic time series to generate with a particular SWG for a predefined accuracy when estimating a particular risk or performance measure is particularly important. In this paper, the required number of realizations of five SWGs coupled with a SWAT model (the Soil and Water Assessment Tool) needed in order to achieve a predefined Relative Root Mean Square Error is investigated. The statistical indices used are the mean, standard deviation, skewness, and kurtosis of four hydroclimatic variables: precipitation, maximum and minimum temperature, and annual streamflow obtained for each observed and model-generated time series. While the results vary somewhat across SWGs, variables and indicators, they overall show that the marginal improvement decreases dramatically after 25 realizations. The results also indicate that the benefit of generating more than 100 realizations of climate and streamflow data is very minimal. The methodology presented herein can be applied in further investigations of other set of risk indicators, SWGs, hydrological models, and watersheds to minimize the required workload.