Increasing pressure on soil resources due to population growth, development of urban and industrial areas has led to a decrease in agricultural land area, therefore, the optimal and sustainable use of soil and land resources seems necessary. So, the present study aims to digital land suitability evaluation for irrigated horticultural products (grapes, pistachios, and olives) and crop (wheat, barley, and saffron) in the vicinity of Badreh city by using machine learning (ML) algorithms in an area with 1840 ha. A total of 80 soil profiles location were designed by conditional Latin hypercube (clhs) method, dug, describe, and sampled. After that, all of the soil samples were transferred to the laboratory to measure soil physio-chemical properties. Based on laboratory analysis, the average of soil properties was determined by considering the depth weighted coefficient up to 100 and 150 centimeters for annual and perennial crops, respectively. Then, the soil characteristics of each profile were matched with the land criteria requirement table and the climatic characteristics required with the climatic requirement tables of different crops. Then, final qualitative land suitability classes were determined using the square root- parametric method. In order to spatially model and predict the classes and sub-classes of land suitability evaluation, two ML models of boosted regression tree (BRT) and classification decision tree (DTc) were used. The variance inflation factor (VIF) was used to feature selection. The applied models were calibrated with 80% of data (n=64 soil profile) for training and 20% of data (n=16 soil profile) for validating. Overall accuracy (OA) and Kappa index were used for evaluating model performance. The feature selection results obtained that, 12 covariates ie., included slope length, midslope position, geomorphons, modified catchment area, vertical distance to channel network, convergence index, clay index, wind effect, analytical Hillshading, normalized difference vegetation index, and multi-resolution valley bottom flattens index (MRVBF) as a representative of topography and RS factors were selected by VIF method among a total of 53 environmental variables. The results of spatial modeling indicated that the BRT method for the class and subclass land suitability of the selected crops based on OA statistical index for wheat, barley, and saffron (85%, 75%, 75%) and (66%) (59%, 70%) and for horticultural crops of grapes, olives, and pistachios (85%, 91%, 89%) and (74%, 67%, 60%), respectively, and provided higher accuracy than the DTc method. In general, based on both models, the OA and kappa index values had a decreasing trend from class to subclass level. Also, among the environmental variables predicting irrigated crops horticultural land suitability classes and subclasses are wind effect indices, MRVBF, analytical Hillshading, vertical distance to distance channel network, slope length, and carbonate index, and for crops had the highest relative importance in comparison to other variables. Generally, the use of the BRT ML algorithm in relation to environmental variables, is well and able to predict class and subclass land suitability map of strategic crops included wheat, barley, saffron, grapes, pistachios, and olives in the study area with high accuracy.
