Three Approaches for Statistical Prediction of Non-Gaussian Climate Extremes

Prediction of climate extremes is challenging, especially for non-Gaussian extremes, since the Gaussian assumption used in traditional linear regression is violated. Three approaches are introduced for statistical prediction of non-Gaussian climate extremes in this presentation. (1) The first one uses a multiple linear regression model after transforming the non-Gaussian predictant to a quasi-Gaussian variable when the predictant does not deviate from Gaussian distribution too much, and uses Pearson’s correlation test to identify potential predictors (Qian et al. 2015, JC). (2) The second one uses a generalized linear model when the transformation is difficult and uses a nonparametric Spearman’s correlation test to identify potential predictors (Qian et al. 2015, JC). (3) With the help of the first-order difference (year-to-year increment), the difference series is more likely a Gaussian distribution than it was in the original series and is thus used as the predictant to find predictors and to construct a prediction model by using traditional linear regression. The difference is first predicted and is then added to the observed value of the target variable at the preceding time to obtain the final prediction result. This method can take the urbanization effect into account and is thus suggested for statistical prediction of climate extremes in urban areas (Qian et al. 2017). The non-Gaussian annual occurrence of hot days and hot nights at Macau and Hong Kong are used to illustrate the three approaches. The possible physical mechanisms on how the atmospheric rivers are related to the hot extremes in Macau and Hong Kong are also discussed.

Qian, C., W. Zhou, S. K. Fong, and K. C. Leong, 2015: Two approaches for statistical prediction of non-Gaussian climate extremes: a case study of Macao hot extremes during 1912−2012. J. Climate, 28(2), 623−636
Qian, C., W. Zhou and Xiu-Qun Yang, 2017: Statistical prediction of non-Gaussian climate extremes in urban areas based on the first-order difference method （submitted）