Spatial analysis of soil quality in agricultural land using machine learning and environmental covariates: A case study of Khuzestan Province

The Soil Quality Index (SQI) serves as a comprehensive assessment tool, encompassing various soil properties and providing a holistic measure of soil health and productivity. This study aimed to analyze the spatial variation of SQI at the regional level in the agricultural areas of Khuzestan province, employing a random forest (RF) machine learning (ML) algorithm along with environmental covariates. A total of 811 soil composite samples were collected from depths of 0–25 cm, and the physical and chemical soil properties including total nitrogen (TN), available phosphorus (P av ), exchangeable potassium (K ex ), soil acidity (pH), electrical conductivity (EC), soil organic carbon (SOC), cation exchange capacity (CEC), calcium carbonate equivalent (CCE), exchangeable sodium percentage (ESP), silt, sand, and clay were analyzed in the laboratory. Additionally, remote sensing (RS) data, topographic attributes and climatic factors were used as environmental covariates. Two approaches, the total data set (TDS) and the minimum data set (MDS) were applied, along with linear (L) and non-linear (NL) scoring functions, to assess SQI, resulting in four SQI-IQI outputs (MDS L , MDS NL , TDS L and TDS NL ) and two Nemero Quality Index (NQI) (MDS L , TDS L ). The results demonstrated that the RF algorithm, in conjunction with selected environmental covariates, accurately predicted the SQI map, achieving an R 2 of 0.70 (IQI TDSNL ) and 0.79 (IQI MDSNL ) with low uncertainty. Furthermore, the relative importance emphasizes the significant role of climatic factors in SQI prediction, followed by RS indices. The developed mapping approach for SQ provides a valuable tool for sustainable agricultural development , contributing to food security and facilitating agricultural assessments. • The Soil Quality Index provides a comprehensive assessment of soil properties essential for agricultural applications. • Non-linear scoring method had higher correlation (r = 0.88) compared to linear method. • Climatic factors were major contributors to SQI distribution, while topography had less influence. • RF algorithm with environmental covariates accurately predicted SQI (R 2 =0.70–0.79). • SQI prediction map shows most of the area falls into the high restriction class (IV).