A regression tree approach to investigate the nonlinear relationship between land surface temperature and vegetation abundance

Urban heat environment is a key ecological and environmental parameter in urban system. Despite the importance of structure of landscape in urban of deriving urban heat environment, their nonlinear relationship and seasonal variation of impacts have seldom been examined. In this paper, we took Guangzhou core urban area as an example, and investigated the nonlinear relationship between LST and vegetation abundance represented by NDVI with the use of regression tree approach across four different dates in 2005. Both of LST and NDVI were derived from Landsat TM5. Results found that LST was negatively related to NDVI in all dates, and the relationship between LST and NDVI was obviously nonlinear and strongly affected by season. Regression tree preformed more suitable in predicting LST variation compared with traditional linear regression. Additionally, those rules derived from regression tree could serve as the indicators of specific landscape types. This study provide a great insight of the complicated relationship and modeling of UHI in heterogeneous urban area.