Geographically weighted regression of the urban heat island of a small city

Abstract

Despite differences in regional climates, cities world-wide have developed one common characteristic - the urban heat island (UHI). Its magnitude is related to city size, especially under cloudless sky conditions on a regional basis, although individual cities may be impacted by such local factors as proximity to large water bodies or prevailing winds. The UHI pattern in the small city of Ljutomer was examined in order to assess its intensity and morphology and to test the utility of the geographically weighted regression (GWR) method in modeling the regression relationships between mean air temperature and related influence factors in this small-scale urban example. Significant differences in mean air temperature between urban and rural areas were measured. It turned out that built-up areas in Ljutomer are on average 1 °C warmer than the rural surroundings in winter time. The regression analyses confirmed the important role of local non-stationary explanatory variables - distance to urban area, topographic position index and land-cover diversity - and global stationary variables - building volume per area and northness - in explaining spatial variation in mean air temperature. The relationships between mean air temperature and these five explanatory variables produced an overall model fit of 91%, utilizing the semiparametric GWR method, which was tested on the smallest scale so far published.