Spatial Modeling and Prediction of the Soil Fertility Index in Two Arid and Semi-Arid Regions of Ilam Province, Iran | ||
| پژوهش های خاک | ||
| Articles in Press, Accepted Manuscript, Available Online from 15 April 2026 | ||
| Document Type: Research Paper | ||
| DOI: 10.22092/ijsr.2026.371527.803 | ||
| Authors | ||
| Asghar Rahmani1; Mahmood Rostaminya* 2; Nasibeh Sayedi3; Seyed Roohollah Mousavi4 | ||
| 1Soil and Water Research Institute, Agricultural Research, Education and Extension Organization, Karaj, Alborz, Iran. | ||
| 2Water and Soil Department, Agriculture Faculty, Ilam University, Ilam, Iran. | ||
| 3MS.c. in Water and Soil Department, Agriculture Faculty, Ilam University, Ilam, Iran. | ||
| 4Ph.D. in Soil Resources Management, Department of Soil Science, University of Tehran, Karaj, Iran. | ||
| Abstract | ||
| Extended Abstract Introduction Soil fertility is one of the most critical factors determining the sustainability of agricultural ecosystems and ensuring food security. Declining soil fertility directly affects crop productivity, and consequently food insecurity. Understanding the spatial variability of fertility distribution patterns are essential for efficient soil management. In recent years, the integration of advanced machine learning algorithms with geostatistical methods has provided powerful tools for modeling and predicting soil fertility indicators such as the Soil Fertility Index (SFI). In arid and semi-arid regions, water scarcity, soil salinity, and climatic variability are major challenges for sustainable agricultural production. Therefore, spatial modeling of soil fertility and identification of its driving factors can serve as a scientific basis for regional land-use planning and resource management. The present study aimed to model and predict the spatial distribution of SFI in two arid and semi-arid regions, Miameh–Dehloran and Valiasr–Badreh (Ilam Province, western Iran), using Random Forest (RF) and Cubist (CB) machine learning algorithms, and to compare their performance with the conventional Ordinary Kriging (OK) method. Ultimately, this research seeks to develop a region-based spatial model for SFI prediction in the agricultural lands of the Zagros region in western Iran. Materials and Methods Soil sampling was conducted in the Miameh–Dehloran and Valiasr–Badreh areas using the conditioned Latin hypercube sampling (cLHS) method. A total of 133 and 71 surface soil samples were collected from the respective regions. The samples were analyzed for physical, chemical, and biological properties to calculate the Soil Fertility Index (SFI). Auxiliary environmental variables, including topographic parameters derived from the Digital Elevation Model (DEM) and remote sensing (RS) indices, were used as predictors. The most relevant variables were selected using the Variance Inflation Factor (VIF) and Boruta algorithms, resulting in 9 and 12 selected predictors for the Miameh–Dehloran and Valiasr–Badreh sites, respectively. Additionally, two climatic variables—mean annual precipitation (MAP) and mean annual temperature (MAT)—were included based on expert judgment. The RF, CB, and OK models were trained and validated, and their predictive performances were evaluated using the coefficient of determination (R²) and root mean square error (RMSE). Spatial prediction maps of SFI were generated in ArcGIS based on the best-performing model. Results and Discussion The results indicated that the Random Forest model outperformed both the Cubist and Ordinary Kriging models in predicting SFI values. The R² of RF was 0.79 for Miameh–Dehloran and 0.60 for Valiasr–Badreh, while the RMSE values were 0.64 and 0.69, respectively. These results demonstrate the superior ability of RF in capturing nonlinear relationships between soil fertility and environmental covariates. According to the spatial distribution maps, approximately 74.14% of the Miameh–Dehloran and 77.33% of the Valiasr–Badreh areas fell within the “very high” (F1) and “high” (F2) fertility classes, indicating considerable potential for agricultural productivity. The climatic variables MAT and MAP were identified as the most influential predictors of SFI. In Miameh–Dehloran, remote sensing indices—especially vegetation and spectral reflectance indicators—played a major role, whereas in Valiasr–Badreh, topographic parameters such as elevation, slope, and aspect were more dominant. These spatial differences reflect the contrasting climatic and geomorphological conditions of the two regions. Combining RF with RS and topographic data significantly improved prediction accuracy and enabled the generation of high-resolution fertility maps suitable for precision agriculture applications. Conclusion The findings highlight the potential of the Random Forest algorithm as a robust and reliable approach for spatial modeling of soil fertility in arid and semi-arid environments. RF effectively captured complex interactions among climatic, topographic, and spectral variables, leading to accurate and detailed SFI prediction maps. The study confirmed that climatic variables, particularly temperature and precipitation, play a decisive role in determining the spatial variability of soil fertility. The resulting maps can serve as valuable tools for agricultural planning, selection of suitable crop types, and sustainable management of soil and water resources. Overall, the proposed modeling framework provides an efficient strategy for optimizing land potential, improving crop yields—especially for wheat—and contributing to food security and sustainable agricultural development in the drylands of western Iran. | ||
| Keywords | ||
| Digital soil mapping; Soil fertility index; Auxiliary variables; Machine learning; spatial interpolation | ||
|
Statistics Article View: 31 |
||
Journal management system. designed by sinaweb