نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی آب و خاک، دانشکده کشاورزی، دانشگاه ایلام
2 گروه مهندسی آب و خاک، دانشکده کشاورزی، دانشگاه ایلام، ایلام، ایران
3 موسسه تحقیقات خاک و آب کشور، سازمان تحقیقات، آموزش و ترویج کشاورزی، کرج، ایران
4 دانش آموحته دکتری مدیریت منابع خاک ، گروه علوم و مهندسی خاک، دانشگاه تهران، کرج ، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Soil quality assessment is a key tool for evaluating the sustainability of agricultural and natural resource systems. This study aimed to model and digitally map the soil quality index (SQI) using machine learning algorithms and spatial interpolation methods in Ilam Province, western Iran. A total of 76 surface soil samples (0-20 cm) were collected using the conditioned Latin hypercube sampling (cLHS) approach, and 10 physicochemical properties were measured. Principal component analysis (PCA) was applied to identify the minimum data set (MDS), from which five variables were selected. Soil quality was assessed using the Integrated Quality Index (IQI) and Nemoro Quality Index (NQI) for both the total data set (TDS) and minimum data set (TDS) and MDS under linear and nonlinear scoring methods. Also, correlation coefficient(r) between TDS and MDS achieved 0.65 and 0.40 for IQI-LS and IQI-NON-LS, respectively. While theses value for NQI were 045 and 0.19, so these results represented that it might be acceptability to calculate SQI based on MDS instead TDS in the IQI-LS approach., in contrast, we cannot applied MDS instead TDS with acceptable confidence for calculating SQI in the other approaches (i.e. IQI-NON-LS, LS and NON-LS NQI. Spatial prediction of SQI was performed using ordinary kriging (MDS) under linear and nonlinear scoring functions. Spatial prediction of SQI was performed using ordinary kriging (OK) and inverse distance weighting (IDW), as well as two machine-learning models: random forest (RF) and k-Nearest Neighbors (k-NN). Results indicated that the soils of the study area fall within a low quality class due to low organic carbon and available phosphorus, along with high pH, sand content, and bulk density. Among interpolation methods, OK outperformed IDW, exhibiting higher accuracy and lower prediction error. RF showed the highest performance for most SQI computations, whereas k-NN performed best for NQITDS. Variable importance analysis revealed that topographic factors include Topographic position index, Analitical hillshading, Channel network distance, Modified catchment area, Aspect, Standardized height, Mass balance index, LS-factor, Covcexity, Diffuse insolation, Normalized Height and Wind effect were the most influential predictors of soil quality. Overall, combining OK with the MDS, particularly when samples are selected using cLHS, provides more accurate predictions of soil quality indices.
کلیدواژهها [English]
Soil quality assessment is essential for guiding sustainable agricultural production and land-use decisions, particularly in semi-arid regions where soil degradation, nutrient depletion, and spatial heterogeneity pose major challenges for land managers. Soil quality is not directly measurable; instead, it is inferred through a set of physicochemical indicators that collectively reflect soil functioning, productivity potential, and ecological stability. The integration of these indicators into a soil quality index (SQI) has become a widely accepted approach for evaluating soil health at field to regional scales. In recent years, DSM has emerged as an efficient method for producing spatially explicit soil information by linking point-based observations with environmental covariates derived from terrain models, remote sensing data, and climatic attributes. Despite considerable progress in DSM research, limited studies have simultaneously compared different SQI scoring functions, data-reduction approaches, interpolation techniques, and machine-learning algorithms within a unified framework. Moreover, little attention has been given to the role of minimum dataset (MDS) selection in improving SQI prediction accuracy and reducing measurement costs. The present study was conducted to address these knowledge gaps by modeling and mapping the SQI in the Valiasr–Badreh region of Ilam Province, western Iran. The specific objectives were to: (i) evaluate linear and nonlinear scoring functions for soil quality indexing; (ii) compare two geostatistical interpolation methods, ordinary kriging (OK) and inverse distance weighting (IDW), as well as two machine-learning algorithms, random forest (RF) and k-nearest neighbor (k-NN), for predicting the spatial variability of SQI; (iii) identify the most influential environmental covariates controlling soil quality distribution; and (iv) assess whether the MDS provides comparable or superior performance relative to the total dataset (TDS).
A total of 76 surface soil samples (0–20 cm) were collected using the conditioned Latin hypercube sampling (cLHS) method, which ensures optimal coverage of environmental gradients while minimizing sampling redundancy. Ten physicochemical attributes were measured according to standard laboratory procedures, including soil texture fractions (sand, silt, clay), bulk density, soil organic carbon (SOC), available phosphorus (P), electrical conductivity (EC), pH, calcium carbonate equivalent, and exchangeable potassium (K). Principal component analysis (PCA) was applied to the TDS to extract the most influential indicators, resulting in the selection of five variables for the MDS. Both datasets were used to compute SQI using the Integrated Quality Index (IQI) and Nemoro Quality Index (NQI), each evaluated with linear and nonlinear scoring functions. Digital environmental covariates were derived from ALOS-PALSAR DEM and Landsat-8 imagery, including slope, elevation, terrain wetness index, topographic position index, vegetation indices, and soil brightness indices. After preprocessing and resampling to a 30-m spatial resolution, the covariates were incorporated into spatial modeling. Four approaches were used for SQI prediction: OK and IDW for geostatistical interpolation, and RF and k-NN for machine-learning modeling. Model performance was assessed based on standard accuracy metrics, and RF variable-importance analysis was used to identify the most influential predictors.
The results showed that the soils of the study area generally fall within a low soil quality class. Low SOC and available P were the most limiting factors, while high pH, high sand content, and elevated bulk density further contributed to reduced soil functionality. These findings align with expectations for semi arid ecosystems where organic matter inputs are low, soil weathering is limited, and nutrient cycling is restricted. Model comparison demonstrated clear differences in predictive performance. Ordinary kriging consistently outperformed IDW across all SQI types, indicating that soil quality exhibits spatial autocorrelation strong enough to be effectively captured by geostatistical methods. Among machine learning approaches, random forest produced the highest accuracy for most SQI variants, particularly those based on the MDS and linear scoring functions. In contrast, k-NN delivered moderate performance and excelled only in one case (NQI using TDS). Variable-importance analysis revealed that topographic factors, especially elevation, slope, and terrain wetness index, were the primary predictors of SQI. Spectral indices linked to vegetation cover and soil surface characteristics ranked next in importance. These results highlight the strong influence of landscape position on soil formation, moisture availability, erosion susceptibility, and nutrient distribution in the region’s complex topography. Comparison of the TDS and MDS showed that the MDS provided results highly comparable to the TDS, demonstrating its efficiency in reducing labor and analytical costs without compromising predictive accuracy. In several cases, MDS-based SQI even produced slightly higher accuracy when combined with OK or RF.
Overall, the study concludes that integrating MDS selection, linear scoring functions, OK, and RF models offers a robust framework for digital SQ mapping in semi-arid regions. The produced maps provide essential information for land use planning, targeted soil conservation, nutrient management, and sustainable agricultural strategies in the Valiasr Badreh region and similar environments. The study underscores the importance of combining geostatistical and machine learning methods with optimized sampling designs to enhance SQ assessment at landscape scales.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.This research was financially supported by Ilam University, Faculty of Agriculture in the form of research for the first author's student thesis and also research for other authors.
“Conceptualization, M.R. and A.R., S.R.M; methodology, M.R., N.B., A.R., S.R.M; software, N.B., S.R.M., A.R.; validation,M.R; formal analysis, N.G.; investigation, N.G., M.R.; resources, M.R.; data curation, N.G., M.R; writing—original draft preparation, N.G.; writing—review and editing, M.R., A.R., S.R.M; visualization, N.B., M.R; supervision, M.R., A.R., S.R.M; project administration, M.R.; funding acquisition, M.R.
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.
Authors state that in their manuscript have not used of generative AI and AI-assisted technologies
Data available on request from the authors.
The authors would like to thank the Ilam university for supporting the research.
The authors thank all participants in this study.
The authors would like to thank anonymous reviewers for their constructive comments.
The authors avoided data fabrication, falsification, and plagiarism, and any form of misconduct.
The author declares no conflict of interest.
)