Performance of stochastic weather generators LARS-WG and AAFC-WG for reproducing daily extremes of diverse Canadian climates

摘要

ABSTRACT: Stochastic weather generators are widely used for generating synthetic weather data, and constitute one of the techniques for developing local climate scenarios from large-scale climate changes simulated by global climate models. Since climate change impact models may be more sensitive to changes in climate extremes than to changes in climate means, there is a need to know the capability of stochastic weather generating techniques in reproducing climate extremes. An evaluation of 2 stochastic weather generators, namely LARS-WG and AAFC-WG, is presented in this study, from the perspective of reproducing observed climate extremes. Extreme daily values (highest daily maximum temperature, lowest daily minimum temperature and maximum daily precipitation) were analysed on a monthly and annual basis, as well as for the growing season (1 May to 30 September), for 9 stations across Canada. The evaluation was based on statistical tests for basic statistical properties and return values derived from fitted generalised extreme value distributions. Results showed that the weather generators LARS-WG and AAFC-WG could reproduce statistical properties of the extreme values of daily precipitation satisfactorily, including 10, 20 and 50 yr return values. However, the performance of the weather generators in reproducing extreme daily values of temperatures was not as good as for precipitation, especially that of LARS-WG. The deficiency in reproducing extremely low temperatures was also more noticeable than extremely high temperatures for LARS-WG. The mismatches of return values were often caused by a more extreme value estimated from synthetic data than that derived from observations.