Regionalization plays an important role in climatology and hydrology research, as it can be used to understand the changing pattern of a hydroclimate variable better and explore its underlying drivers effectively by identifying homogeneous regions which are determined by topography and climatic conditions. However, as climatic drivers change over time, homogeneous regions should vary when the intra-annual patterns are of interest. In this paper we investigate the regionalization and changing patterns of maximum daily temperature in China based on the maximum air temperature (Tmax) derived from the CN05.1 daily surface air temperature gridded data from 1981 to 2010. We also propose an objective variable-based regionalization approach by combining the rotated empirical orthogonal function (REOF) method and K-means clustering analysis, to determine homogeneous regions effectively for such a large volume of daily grid data. The REOF approach is not able to identify discrete regional boundaries easily, but it is efficient at determining the number of regions and their rough spatial location. Furthermore, while clustering analysis suffers from convergence when the number of observations is large, it is able to provide clean boundaries between regions. The hybrid method tends to overcome their individual disadvantages while maintaining their individual advantages. The results show that the patterns of homogeneous regions have distinct features in both summer and winter seasons. The reasonableness of regionalization is then examined by the first principal component within regions, which shows the homogeneity within regions, and the scatterplots of Tmax time series between regions, which show the heterogeneity between regions. Furthermore, the boxplots of normalized Tmax within regions for both summer and winter seasons also clearly show the different characteristics of inter-annual variability in each region. In order to reveal detailed process of the change in the number and pattern of regions, we further conducted the regionalization for individual months and explored the potential linkages between regionalization and climatic conditions over time. This research is noteworthy in promoting research on regionalization with consideration of seasonal and intra-annual variation as well as extreme temperature events. The proposed hybrid regionalization method is broadly applicable to other regionalization studies.
