Abstract: In order to study the evolution of urban pattern and the change of heat island effect in Hefei in recent 20 years, land classification and land surface temperature inversion were carried out based on the Landsat satellite images of October 2005, October 2009, October 2015 and October 2020. The normalized difference between bare land and building index (NDBBI), fraction vegetation coverage (FVC), modified normalized difference water index (MNDWI) and population density were extracted for multiple regression analysis, and then a mathematical model was established to analyze the heat island effect and its influencing factors in the main urban area of Hefei. The results show that: (1) From 2005 to 2020, the strong heat island area has increased by 15.03 km². The distribution direction of standard deviation ellipse of heat island is from northeast to southwest, and the scope of the ellipse is expanding year by year. The mass center of heat island is concentrated in Shushan Economic Development Zone, and 81.90% of the strong heat island areas are high-density industrial areas, indicating a good corresponding relationship between the strong heat island areas and high-density industrial areas. (2) The analysis results of geographical detector show that the explanatory power of each influencing factor on land surface temperature from large to small is, NDBBI (0.542), MNDWI (0.409), FVC (0.379) and population density (0.018). (3) The results of multivariate linear model (R2 = 0.654) indicate that the main factors affecting land surface temperature is NDBBI, while the population density has little effect. (4) The analysis of geographical weighted regression (GWR) model shows that the R2 of each point is in the range of 0.489-0.667, and the R2 of urban construction area with dense buildings and roads is highest. The high value of NDBBI coefficient is concentrated in the economic development zone and other places, with the highest value reaching more than 0.9, the coefficient of population density is still very small, the high value areas of FVC coefficient are concentrated in areas with high vegetation coverage, while MNDWI high value areas are distributed in water areas.

Key words: surface temperature, heat island effect, geographic detector, ridge regression, multiple regression model