Six different weather generator models were compared. The first two models (M1 and M2) use a first-order autoregressive daily model and the third model (M3) uses a newly proposed semi-parametric method to reproduce the correlation and autocorrelation of the variables. Three other models (M1-2, M2-2 and M3-2) are the combinations of these models with an adjustment algorithm for the low-frequency variances (SL). The comparison revealed that M1-2 model (daily weather generator with the SL adjustment algorithm) and the M2-2 model (daily weather generator in combination with a monthly weather generator and the SL adjustment algorithm) are the best models in the study area. All the studied models have acceptable performance in relation to the shape of the probability distribution functions. Three first models have deficiencies in relation to the inter-annual standard deviations. The M2 and M3 models, in which the high-frequency standard deviation (SH) is used instead of the total standard deviation values (ST), slightly underestimated these inter-annual variations. But, the performance of the M1 model is considerably poorer than the other two models. The results revealed that the adjustment of the inter-annual standard deviations improves model performance. Moreover, the newly proposed algorithm has the potential for multi-station simulations.
Proposing a new semi-parametric weather generator algorithm and comparison of six algorithms for non-precipitation climatic variables
Behnam Ababaei, Teymour Sohrabi, Farhad Mirzaei; Proposing a new semi-parametric weather generator algorithm and comparison of six algorithms for non-precipitation climatic variables. Journal of Water and Climate Change 1 March 2014; 5 (1): 25–35. doi: https://doi.org/10.2166/wcc.2013.168
Download citation file: