Taking the urban landscape of Shijiazhuang as the research object, the Landsat remote sensing image data of 1987, 2004 and 2019 are selected, and the supervised classification method is used to distinguish the studied area into four types of landscapes: green land, water body, impervious surface, and unused land. The window algorithm and split window algorithm are used to invert land surface temperature (LST). From the perspective of landscape ecology, Fragstats4.2 is used to calculate the four types of landscape pattern indexes, and explore and analyze the landscape granularity and mobile window scale selection, using the ArcGIS spatial analysis method and statistical analysis method to analyze the four types of landscape correlation between pattern index and LST. The results show that, from 1987 to 2019, the green patch type area (CA), maximum patch area index (LPI) and aggregation index (AI) gradually decreased, and the CA, LPI and AI of impervious surface gradually increased. With the urbanization process, the area of green land gradually reduced and cracked, the advantage of green landscape was declining, the surface area of impervious land was gradually increasing and converging, and the advantage of impervious surface landscape was constantly strengthening, gradually forming an advantageous landscape. The Plaque percentage index (PLAND), LPI, AI and LST show a consistent and extremely significant correlation; green land and water are negatively correlated; impervious surface and unused land were positively correlated. The SPLIT index was the opposite, green lands and water bodies are positively correlated, and impervious surface and unused land are negatively correlated. The correlation coefficient of LST with PLAND and LPI is significantly higher than that with AI and SPLIT, indicating that the effect of a dominant landscape on the surface temperature is significantly greater than that of several more scattered or broken landscapes.