Accurate mapping of the spatial distribution of soil properties is essential for soil resource management and environmental protection. This study used 34 environmental variables derived from Landsat 8 images and a digital elevation model, alongside 96 surface soil samples (0-20 cm), to compare the performance of six modeling algorithms: generalized linear model (GLM), generalized additive model (GAM), random forests (RF), support vector machine (SVM), classification and regression trees (CART), and an ensemble model. The aim was to create accurate maps of soil properties, including electrical conductivity (EC), pH, pOH, carbonate, bicarbonate, Na, and Cl in an arid area located in the central plateau of Iran. Key environmental covariates included Principal Component Analysis (PCA) and Tasseled Cap (TC) transformations of Landsat images, Land Surface Temperature (LST), Temperature Vegetation Dryness Index (TVDI), vegetation indices, and salinity indices. Pearson correlation analysis was used to study relationships between soil variables and these covariates. Model performance was evaluated using mean absolute error (MAE), root-mean-square error (RMSE), and coefficient of determination (R2). The ensemble and random forest models provided the most reliable spatial distribution of soil properties. In contrast, GLM and GAM algorithms produced relatively poor estimates. Important predictors identified included PCA1, PCA2, Tasseled Cap Wetness (TCW), B5, B7, and elevation. This study demonstrates that machine learning algorithms, combined with freely available remote sensing data, can effectively model and map soil properties in arid and desert regions. The methodology offers a cost-effective approach for improving soil information at local and regional scales, particularly in data-poor areas like Iran. This strategy supports enhanced soil management and environmental protection efforts.