Soil temperature vegetation drought index (TVDI) is a commonly used indicator to characterize soil drought state. Land use/cover (LULC) types and nonlinear properties of dry edge and wet edge in LST-NDVI feature space are issues that need to be considered in TVDI remote sensing inversion process. However, most of the previous studies and applications of TVDI ignored these conditions, thus causing part of the error. In order to improve the inversion accuracy of TVDI with long time and wide space, this study took the Chinese character “Ji” shaped bay of the Yellow River basin, a typical resource-driven region, including five cites of Inner Mongolia Autonomous Region, namely, Hohhot, Ordos, Bayannur, Baotou and Wuhai, as a study area. A long time series Landsat data composed of 816 scences of image, during the local vegetation growing season, in eight periods, during 1986 to 2020, as well as contemporaneous 30 m LULC rasters, were used in this study. After excluding construction land and water body, the LST-NDVI feature spaces were rebuilt by seamless mosaicing and mean value synthesis of NDVI and LST during vegetation growth season under the four LULC types. Then their dry edges and wet edges were fitted through polynomial of higher power. Finally, eight 100 m TVDI grids, proven to be accurate, of soil surface in Inner Mongolia section of the Yellow River were obtained. The results of this work could provide theoretical reference and technical guidance for regional soil drought prediction.