Abstract: The soil salinization results in the loss of a large area of land resources, and is a direct threat to the ecological sustainable development, and also damages to the natural environment for human survival seriously. So it has a very important practical significance to use remote sensing technology in soil salinization monitoring. Ground is covered by different vegetation, so the retrieval of soil moisture of traditional NDVI-T_s space increases the error especially in large-scale. In this paper, taking corn farmland and grassland of saline as cases, firstly make use of landsat8 data for the retrieval of the land surface temperature T_s (Land surface temperature) by split window algorithm combined with the ground test station meteorological data. And then establish the NDVI (Normalized Difference Vegetation Index)-T_s feature space under two types of land use. Use the NDVI-T_s space to calculate the TVDI (Temperature Vegetation Dryness Index). Combined with the actual measured surface soil moisture data to calculate the correlation coefficients between actual soil moisture and TVDI:the whole area is 0.16, farmland is 0.22, grassland is 0.35;and the correlation coefficients got through the 0.01, 0.1, 0.05 significance level respectively. Lastly analyze the soil moisture retrieval error by different NDVI-T_s feature space. From the above conclusion:Separating different types of land use increase the soil moisture retrieval accuracy.