Document Type : Research Paper
Authors
1 . Department of Soil Science, Faculty of Agriculture, University of Zanjan, Zanjan, Iran
2 Department of Soil Science, Faculty of Agriculture, University of Zanjan, Zanjan, Iran
3 Soil and Water Research Institute, Agricultural Research Education and Extension Organization (AREEO), Karaj, Iran,
Abstract
Keywords
Main Subjects
Soil quality, as a fundamental pillar of sustainable agricultural production systems and the maintenance of soil ecosystem functions, plays a decisive role in food security, environmental sustainability, and the sustainable management of natural resources. In recent decades, increasing anthropogenic pressures resulting from land-use change, unsustainable exploitation, overgrazing, improper tillage practices, and inefficient use of agricultural inputs have led to a gradual decline in soil quality across many regions of the country. Therefore, quantitative and spatial monitoring of soil quality and the identification of its driving factors are considered essential prerequisites for sustainable land management planning and precision agriculture. The Soil Quality Index (SQI) provides an integrated framework for the simultaneous incorporation of soil physical, chemical, and biological properties, enabling quantitative assessment of the functional health status of soils across different spatial scales. The main objective of this study was to assess soil quality and map its spatial distribution in the study area by applying the SQI and integrating field-based soil data with remote sensing information and advanced machine-learning models.
A total of 76 composite soil samples were collected from the 0–30 cm soil layer. A suite of physical, chemical, and biological properties, including soil texture, bulk density, aggregate stability, water-dispersible clay, pH, electrical conductivity, soil organic carbon, total nitrogen, available phosphorus and potassium, soluble calcium and magnesium, cation exchange capacity, calcium carbonate equivalent, microbial respiration, microbial biomass carbon, metabolic quotient (qCO₂), and microbial quotient (MQ), were measured. Soil organic carbon stock was also calculated as a key indicator of soil ecosystem functioning. To derive environmental covariates, Sentinel-2 satellite imagery was obtained from Google Earth Engine and processed to extract spectral indices including NDVI, NDWI, NDMI, and BSI, as well as land surface temperature (LST). These variables were used as auxiliary predictors in digital soil quality mapping. Principal component analysis (PCA) combined with correlation analysis was applied to reduce the number of indicators and to determine the minimum data set (MDS). The PCA results indicated that several principal components with eigenvalues >1 explained most of the data variance; accordingly, soil organic carbon, bulk density, available phosphorus, magnesium, and pH were selected as key indicators. In parallel, the total data set (TDS), comprising all measured indicators, was used to compute SQI. Indicator scoring was performed using fuzzy membership functions according to the response type of each variable (more-is-better, less-is-better, or optimum range), and indicator weights in the TDS and MDS approaches were derived based on factor analysis and the proportion of variance explained by principal components, respectively. The SQI was ultimately calculated as a weighted additive index for each sampling point. For spatial prediction of SQI, multiple linear regression (MLR) and random forest (RF) models were employed.
The results showed that the mean SQI obtained from the TDS approach was slightly higher than that derived from the MDS approach. However, the very strong correlation between SQI values obtained from the two approaches indicates that, despite the substantial reduction in the number of input variables, the MDS approach was able to reproduce the spatial pattern of soil quality with acceptable accuracy. The wider range of SQI values under the MDS approach reflects its higher sensitivity in discriminating different soil quality levels and identifying areas with more severe limitations. These findings confirm the efficiency of the MDS approach as a cost-effective and practical method for regional-scale soil quality monitoring. Model performance evaluation further demonstrated that the random forest algorithm outperformed multiple linear regression in both TDS and MDS scenarios, yielding higher coefficients of determination and lower prediction errors. This superiority highlights the greater capability of machine-learning algorithms to capture complex and nonlinear relationships between soil quality indices and environmental covariates. In contrast, the multiple linear regression model, due to its reliance on linear assumptions, showed limited ability to represent the inherent complexity of soil quality controlling processes. Accordingly, the application of machine-learning approaches, particularly random forest, is recommended for digital soil quality mapping in heterogeneous landscapes. Variable importance analysis in the random forest model revealed that moisture- and vegetation-related spectral indices, including NDWI, NDVI, NDMI, and the bare soil index (BSI), contributed most to explaining the spatial variability of soil quality. These results emphasize the key role of soil moisture conditions and vegetation cover in regulating soil biological and chemical properties and, ultimately, soil quality. Higher NDVI and NDWI values were generally associated with increased soil organic carbon content, enhanced microbial activity, and improved structural stability, whereas higher BSI values reflected more exposed and degraded soil surfaces, which were directly linked to soil quality degradation. Therefore, remote sensing indices can serve as rapid, cost-effective, and efficient tools for spatial monitoring of soil quality at regional scales.
Soil quality maps indicated that a considerable proportion of the study area falls within moderate to low soil quality classes, highlighting the need for implementing conservation-oriented management strategies at the watershed scale. In contrast, areas under sustainable land-use systems such as olive orchards, particularly in low-slope and downslope positions, exhibited higher soil quality. This pattern confirms the positive role of permanent vegetation cover, reduced tillage intensity, and increased organic matter inputs in improving soil quality. Accordingly, the adoption of conservation management practices, including residue management, enhancement of surface cover, reduction of intensive tillage, optimized irrigation management, and erosion control, is recommended as effective strategies for improving soil quality in the region.
Overall, the findings demonstrate that integrating the Soil Quality Index with remote sensing data and machine-learning algorithms provides an efficient framework for regional-scale assessment and digital mapping of soil quality. The MDS approach can be recommended as a practical, cost-effective, and reliable alternative to the TDS approach for soil quality monitoring, while the random forest algorithm, due to its superior predictive performance, represents a robust tool for supporting management decision-making in soil conservation, precision agriculture, and sustainable land-use planning. This integrated approach can serve as a scientific basis for policy-making related to soil and water resources management at local and regional scales.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
For research articles with several authors, a short paragraph specifying their individual contributions must be provided. The following statements should be used “Conceptualization, S.H., K.M, M.N and A.G.; methodology, S.H., K.M, N.N, A.G and M.A.; software, S.H, M.N.; validation, A.G., K.M, M.N. and M.A.; formal analysis, S.H.; investigation, S.H. , A.G; resources, S.H, M.A.; data curation, S.H, M.N, A.G, K.M and M.A; writing—original draft preparation, S.H.; writing—review and editing, S.H, K.M, A.G, M.N.; visualization, A.G, M.A, M.N.; supervision, K.M.; project administration, S.H, K.M, M.N.; funding acquisition, S.H,.K.M. All authors have read and agreed to the published version of the manuscript
All authors contributed equally to the conceptualization of the article and writing of the original and subsequent drafts.
The authors did not use any artificial intelligence tools in preparing this manuscript.
Data available on request from the authors.
The authors would like to thank all participants in the present study.
The authors avoided data fabrication, falsification, and plagiarism, and any form of misconduct.
The authors declare no conflict of interest.
)