spatial modeling of soil saturation percentage using machine learning methods in sistan plain

Soil maps are an urgent need for different land users and decision-makers. in recent years, attention to digital soil mapping has greatly increased, but most studies have focused on surface soil, even though land users are faced with the three-dimensional (3d) structure of soil. saturation percentage (sp) is one of the physical attributes of soil moisture, which can be considered in land management, especially in the direction of soil water retention in connection with other attributes, in arid areas. therefore, the present study was conducted with the aim of digital mapping of sp in 3d using some machine learning methods in the sistan plain, which is located on the hirmand river delta in a hyper-arid region. to carry out this research, the information from 576 soil profiles located in the sistan plain was used and the percentage of saturated soil moisture was measured using the standard method at depths of 0-15, 15-30, 30-60, and 60-100 cm using the weighted average method. random forest (rf), quantile regression forest (qrf), and cubist methods were used for spatial modeling. the results showed that the variables derived from remote sensing showed a significant correlation with the sp parameter only at the first and second depths, which were close to the ground surface, but the variables derived from dem had a significant correlation at all depths. these variables were mainly related to alluvial activities, which had the greatest effect on soil changes in the studied area. among the models, the rf method showed the best performance for spatial modeling of sp in all depths. the 3d modeling of the percentage of sp showed that the value of sp is the lowest in the south and medium in the middle of the area, and the highest in the north of the sistan plain at the edge of the hamoun wetlands. sp value is repeated with the same spatial trend, but the average value of sp increases from the surface to the depth. it seems that the changes in this attribute are in line with the 3d changes in the soil texture components in the region. based on the results of the three-dimensional zoning of sp, it could be recommended that in the northern areas of the sistan plain, irrigation should be done with a longer time interval than in the southern regions for the same agricultural products. in the fields of natural resources, to manage vegetation and especially to deal with wind erosion, plants with shallow and deep roots in the northern regions, and trees and plants with deep roots in the southern regions can be considered. machine learning methods, especially rf, can be effective in preparing digital and 3d maps of soil characteristics and can help different land users manage their land better.