Stochastic weather simulation: overview and analysis of two commonly used models

Two stochastic weather simulation models (USCLIMATE and CLIGEN) were compared for their performance in replicating observed precipitation, temperature, and solar radiation variables at six locations in the United States. Statistical tests of significance were performed on means and standard deviations of a variety of standard and derived daily weather variables over monthly and annual time periods. Model replication of extreme events also was evaluated. In general, mean monthly and annual values were well replicated by both models, but variance replication of temperature and solar radiation were found superior in USCLIMATE. Extreme-value replication, especially over short (24-h) time intervals, was highly a function of climatic element, location, and time of year but, in general, was judged only fair in both models. Suggestions for model improvement and enhancement are given, as are suggestions for model applications.