Multifractal analysis has been more commonly used than fractal analysis to give a more precise characterization and a multi-scale analysis of non-linear phenomena. Multifractal analysis can also be used to analyze processes having a greater number of parameters. The objective of this study was to use a multifractal model to quantitatively describe the texture structure, spatio-temporal features, and relationship between regional temperature and Normalized Difference Vegetation Index (NDVI) at different invariant intervals for the urban thermal environment field in the Zhengzhou metropolitan area. We used Landsat TM images for three different periods. The results were calculated for the entire area as follows: (1) the difference between the maximum value α_max and minimum value of the singularity index α_min decreased from 0.6716 to 0.6419; (2) α_max and α_min of the singularity index decreased from 1.4644 and 2.136 to 1.4304 and 2.0723, respectively; (3) the singularity value of the background α(f_max) decreased from 2.0111 to 2.0083; and (4) the fractal dimension of the maximum and minimum value f(α_min), f(α_max) increased from 0.2354 and 1.4877 to 0.3412 and 1.734, respectively. These results show that with the economic development and progression of urbanization in the past 20 years, the information contained in the texture of the thermal field in the Zhengzhou metropolitan area constantly decreased, and its structure and maximum and minimum texture values became simpler and larger, respectively. For selected locations, we observed that urban areas have fluctuated on a small scale with respect to the amplitude of texture, and the largest fractal dimensions of the maximum and minimum values of texture are well distributed at different scales with high clustering properties. Contrasting results were observed for water areas. Furthermore, there were smaller α_min and D₂ values for older urban districts than those reported for newly developed areas. The relationship between the multifractal parameters and regional temperature statistics were clearly linear, with a stronger mean value than the standard deviation, and the order of variation was as follows: water areas, agricultural land, suburbs, and urban area. In addition, the singularity index of the influence of the relationship between regional temperature and the NDVI was evaluated by obtaining a three scale-invariant segmentation field using a concentration-area model based on a 2011 image. The results showed that the regional average value of land surface temperature (LST), composed of patches of edge pixels, is sensitive to the singularity index, and the contribution of the singularity index to the background value is positive, which can provide some guidance for the scaling transformation of thermal images.