资源环境科技发展态势分析平台

In this study, a recursive dissimilarity and similarity inferential climate classification (ReDSICC) approach is developed to provide an alternative tool for climate classification. Based on incorporation of a discrete distribution transformation (DDT) method and integration of advanced statistical inferential methods, a recursive framework of dissimilarity and similarity inferences is proposed for stepwise grouping multi-dimensional climate-variable observations. ReDSICC is capable of eliminating the restriction of samples being normally distributed, enabling classification of regional climates under data uncertainties and multivariate dependencies, identifying the most desired climate classification result, and avoiding subjective judgments in the classification process. To verify methodological effectiveness and facilitate related studies, ReDSICC is applied to climate classification in the Athabasca River Basin (ARB), Canada. It is revealed that the complicated dissimilarities and similarities of climatic conditions among all grids over the ARB are effectively reflected in the results of ReDSICC. A reversible transformation between an abnormal distribution and a normal distribution is achieved by DDT. The effectiveness of climate classification which is represented as the Nash coefficient for climatic features over any grid and the corresponding climate class is decreased if DDT is not employed. In comparison with daily minimum temperature, the spatial heterogeneity of daily maximum temperature is higher while that of daily cumulative precipitation is lower over the ARB. The classification result of ReDSICC varies with changes of representative climate variables and parameter values. These advantages and revelations are helpful for enhancing the reliability of climate classification results, improving the effectiveness of existing climate classification methods, and providing scientific support for the related studies in the ARB or neighbouring regions.