Soil salinity is one of the crucial factors which affects eco-environmental quality in the oasis of arid regions. Consequently, there is a great need to monitor soil salinity for prevention and mitigation of land degradation and further promote regional sustainable development. The variation in soil salinity is affected by environmental factors that occur at different scales with varying intensities. It is critical to adequately consider environmental variables under scale effects for digital soil mapping which has been minimally discussed in previous studies. Totally 58 soil samples were collected from the Qitai Oasis, Xinjiang Uygur Autonomous Region of China. In the laboratory, the soil samples were prepared for analysis of electrical conductivity (EC) when prepared into suspensions 1∶5 in soil and distilled water ratio. In addition, the corresponding Landsat-5 TM data was collected and preprocessed for up-scale transformation by raster resampling (spatial resolution were 30~990m) and neighborhood filtering (window size were 3×3，5×5, …, 31×31), and the environmental variables (vegetation index (VI), normalized difference infrared index (NDII), principal component analysis (PCA), and tasseled cap transformation (TC)) were further generated. Then, the gradient boosting decision tree (GBDT) model was employed for the estimation of surface soil salinity based on these 720 environmental variables at various spatial scales. The results showed that for individual scale mode, the neighborhood filtering method based on 30m pretreated data was generally better than those of the raster resampling modes, and the maximum analytical power reached 78.55% and 75.31%, respectively. In terms of mixed scale, the analytical effect of soil salinity was significantly improved compared with the mode of individual scale, and the analytical power could reach up to 90.66%, which suggested the effective information complementarity. Compared with the neighborhood filtering, the range of adjusts R2 of the resampling mode was broader, which indicated that the scale transformation of grid resampling was more sensitive to the characterization of the relationship between soil salinity and environmental variables. The research result was helpful for understanding specific scale-dependent relationships and had the potential to reveal the scale control of soil salinity variation in arid regions.