Assessment and Improvement of Stochastic Weather Generators in Simulating Maximum and Minimum Temperatures

摘要

Stochastic weather generators are commonly used to generate time series of weather variables to drive agricultural and hydrologic models. One of their most appealing features is the ability to rapidly generate the very long time series used in agricultural and hydrological impact studies. However, they also have various problems, such as the inability to represent the inter annual variability of the climate system, and it is difficult for them to accurately preserve the auto- and cross-correlation ofmaximum and minimum temperatures (T_(max) and T_(min)). This research aims to merge two widely used weather generators (CLIGEN (v5.22564) and WGEN) into a hybrid method that combines the strengths of each (referred to as the conditional method) for generating T_(max) and T_(min) and apply an approach to correct the inter annual variability of T_(max) and T_(min)(referred to as the spectral correction method). The results show that CLIGEN reproduced mean daily T_(max) and T_(min) very well. WGEN also produced mean daily T_(max) reasonably well but slightly underestimated mean daily T_(min). Moreover, CLIGEN was better than WGEN at producing standard deviations of daily T_(max) and T_(min). The conditional and spectral correction methods resulted in a weather generator that accurately produced means, standard deviations, and extremes of daily T_(max) and T_(min). The auto- and cross-correlations of and between daily T_(max) and T_(min) were well reproduced and much better than those of CLIGEN- and WGEN-generated data. Moreover, the spectral correction approach successfully reproduced the observed interannual variability of T_(max) and T_(min).