• TAZ unit is applied to analyzing the relationships between urban features and LST. • SQR model explores heterogeneity of urban features on conditional low- or high-LST. • Spatial dependence bears a downward trend across the whole quantiles of LST. • SQR model is a promising method on estimation of LST. To seek mitigation strategies of urban heat island (UHI), many statistical methods have been applied to quantitatively explore the impact of driving factors on UHI. Nonetheless, the commonly used statistical models, such as ordinary least-squares (OLS) model and spatial autoregression model (SAM), were limited to explore the homogeneity relationship of driving factors on UHI. Although the spatial dependence of spatial data was considered, SAM failed to address the heterogeneity of relationships. In this study, a spatial quantile regression (SQR) model is innovatively introduced to investigate the relationships between driving factors and land surface temperature (LST) at different quantiles (heterogeneity) while considering spatial dependence. Substantial variations in SQR are found, compared with OLS and the spatial lag model (SLM). The coefficients of all driving factors in SQR are not constant and the spatial dependence shows an obvious downward trend throughout the whole quantiles of LST. The results of root mean square deviation (RMSD), coefficient of determination (R 2 ) and concordance correlation coefficient (CCC) show that SQR has better performance than SLM in LST prediction. These indicate that the characteristics of both spatial dependence and heterogeneity are concurrence. Therefore, joint prevention strategies with regional governance according to local conditions are suggested to mitigate LST.