Abstract:Based on MODIS remote sensing data of Beijing-Tianjin-Hebei region, using landscape ecology and related principles of spatial econometrics, the pattern characteristics of ecological space and surface temperature in Beijing-Tianjin-Hebei region were discussed, Pearson correlation was used to explore the correlation between the two, and the spatial bivariate autocorrelation and spatial autocorrelation were used to explore the spatial correlation of the two. The results showed that the forest coverage in the central, northeast and southwest borders of Beijing-Tianjin-Hebei area was increasing, the cultivated land coverage in the northeast, southwest and east coastal areas was increasing, and some areas of Chengde City in the northwest and south, Baoding City, Shijiazhuang City, Xingtai City and Handan City were at risk of land exposure. The green space or blue space in July 2018 was extracted, and the spatial distribution of ecological space and surface temperature in various areas had significant spatial autocorrelation. Sample areas 5 and 7 were located in the north of Hebei Province. The proportion of forest landscape was higher, and the correlation and bivariate spatial autocorrelation were higher than that of other sample areas, which were related to landscape dominance and patch fragmentation. Due to the low proportion of ecological space between sample area 1 and sample area 4, the impact on LST was limited. The plant had a large impact on LST image, the proportion of sample area 7 was high, and the effect of ecological space patches and concentrated patches on surface temperature was obvious. The fitting effect of the spatial lag model and the spatial error model of sample areas 1 to 7 was much better than that of OLS. R2 of the spatial error model of each sample area was greater than that of the spatial lag model, and the spatial error model had stronger ability to interpret variables. The LIK value of the spatial error model of each sample area was larger, the value of AIC, SC and Moran’s I of the model residual were smaller, and the fitting effect of the spatial error model was better than that of the spatial lag model.