Although research relating to the urban heat island (UHI) phenomenon has been significantly increasing in recent years, there is still a lack of a continuous and clear recognition of the potential gradient effect on the UHI—landscape relationship within large urbanized regions. In this study, we chose the Beijing-Tianjin-Hebei (BTH) region, which is a large scaled urban agglomeration in China, as the case study area. We examined the causal relationship between the LST variation and underlying surface characteristics using multi-temporal land cover and summer average land surface temperature (LST) data as the analyzed variables. This study then further discussed the modeling performance when quantifying their relationship from a spatial gradient perspective (the grid size ranged from 6 to 24 km), by comparing the ordinary least squares (OLS) and geographically weighted regression (GWR) methods. The results indicate that: (1) both the OLS and GWR analysis confirmed that the composition of built-up land contributes as an essential factor that is responsible for the UHI phenomenon in a large urban agglomeration region; (2) for the OLS, the modeled relationship between the LST and its drive factor showed a significant spatial gradient effect, changing with different spatial analysis grids; and, (3) in contrast, using the GWR model revealed a considerably robust and better performance for accommodating the spatial non-stationarity with a lower scale dependence than that of the OLS model. This study highlights the significant spatial heterogeneity that is related to the UHI effect in large-extent urban agglomeration areas, and it suggests that the potential gradient effect and uncertainty induced by different spatial scale and methodology usage should be considered when modeling the UHI effect with urbanization. This would supplement current UHI study and be beneficial for deepening the cognition and enlightenment of landscape planning for UHI regulation.
Land-surface temperature measurement from space: physical principles and inverse modeling
