A new method for improving the performance of weather generators in reproducing low-frequency variability and in downscaling

Weather generators (WGs) are models fitted to a series of observed weather data in order to generate long synthetic series of these data, which is important for risk assessment. WGs are also among the main downscaling tools for climate change impact assessments. Despite their many capabilities, WGs cannot properly reproduce low-frequency variability (LFV). Moreover, when using WGs for the downscaling of future climate scenarios, a change in rainfall occurrence will have unintended effects on secondary variables such as minimum temperature (T-min) and maximum temperature (T-max). This paper presents an approach based on the Quantile Perturbation Method for correcting the LFV of rainfall outputs pertaining to WGs by perturbing the monthly means of the daily generated series. A multivariate monthly generator was used to improve LFV inT(min)andT(max). An advantage of the proposed method is the preservation of the correlation between daily and monthly timescales as well as the cross-correlation between variables during the LFV correction. This method further reduces the unintended effects of changes in rainfall occurrence on the secondary variables during downscaling. After applying the method to the LARS-WG model, the performance of the modified model was tested both directly and indirectly. Concerning the direct validation, the statistical characteristics of the simulated and observed weather data related to the four stations were compared over a diverse range of climatic conditions. Regarding the indirect validation, a comparison was made between the statistical characteristics of the runoffs simulated with a daily hydrologic model fed with the generated and observed weather data. The results showed that the method improved the performance of LARS-WG regarding the reproduction of some of the observed weather characteristics, particularly LFV. The method was also shown to be well capable of reducing the unintended effects of the changes in rainfall occurrence on secondary variables during downscaling.