Assessing the relationship between environmental variables, vegetation indices, and soil properties on rainfed wheat yield

Introduction
Rainfed wheat is one of the most important strategic crops in arid and semi-arid regions, where crop production is highly dependent on rainfall variability and environmental conditions. In these systems, spatial and temporal differences in soil properties, vegetation dynamics, and topographic conditions can significantly affect crop growth and final yield. Soil physical and chemical characteristics influence water retention, nutrient availability, aeration, and root development, while vegetation indices derived from satellite imagery can reflect crop vigor and seasonal growth conditions. Similarly, topographic variables such as elevation, slope, and moisture-related indices affect runoff generation, soil moisture distribution, erosion processes, and local microclimatic conditions. In semi-arid regions of Iran, including Semnan Province, identifying the dominant environmental factors controlling rainfed wheat yield is essential for improving agricultural management and reducing production risks. Previous studies have often investigated soil, vegetation, or topographic factors separately; however, fewer studies have simultaneously evaluated their combined effects and relative importance under field conditions. In addition, the interrelationships among environmental variables may complicate the interpretation of their independent effects on yield. Therefore, the present study was conducted in the Kalpoosh Plain to investigate the relationships among soil properties, NDVI, topographic indices, and rainfed wheat yield using correlation analysis and Boruta feature selection. The study also evaluated multicollinearity among environmental variables to improve the interpretation of variable importance and environmental interactions.
Materials and Methods
This study was conducted in rainfed wheat fields of the Kalpoosh Plain located in northeastern Mayami County, Semnan Province, Iran. Data were collected from 112 wheat fields during the growing season. Wheat yield was measured using 5 × 5 m plots, and soil samples were collected from the 0–30 cm layer at the center of each plot. Geographic coordinates of sampling points were recorded using GPS. Soil analyses included texture fractions (sand, silt, and clay), bulk density (BD), pH, electrical conductivity (EC), equivalent calcium carbonate (TNV), organic carbon (OC), total nitrogen (N), soluble potassium (K), sodium (Na), calcium plus magnesium (Ca+Mg), and sodium adsorption ratio (SAR). Laboratory analyses were performed using standard procedures. Monthly NDVI values from April to July were extracted from Sentinel-2 imagery with 10 m spatial resolution after atmospheric correction and cloud removal. Topographic variables including elevation, slope, topographic wetness index (TWI), LS-factor, plan curvature, profile curvature, valley depth, and relative slope position (RSP) were derived from the 30 m SRTM digital elevation model. Pearson correlation analysis was used to investigate relationships among environmental variables and wheat yield. Correlation matrices and heatmaps were generated for graphical interpretation. Because strong correlations among predictor variables can influence statistical interpretation, multicollinearity was evaluated using Variance Inflation Factor (VIF) and Tolerance (TOL) indices. Variable importance and sensitivity analysis were performed using the Boruta algorithm based on Random Forest in the R-Studio environment.
Results and Discussion
The results showed that many soil properties were strongly interrelated, whereas their direct relationships with wheat yield were generally weak. Among soil variables, TNV (r = −0.25) and soil pH (r = −0.21) showed the relatively weak to moderate negative correlations with yield. Soil texture fractions also exhibited very strong interrelationships, particularly between sand and silt (r = −0.97), while their correlations with yield were not significant. NDVI values across all months showed positive correlations with yield. The highest correlation was observed in May (r = 0.53), followed by June (r = 0.46), July (r = 0.43), and April (r = 0.42), highlighting the importance of vegetation conditions during stem elongation and reproductive growth stages. Strong correlations among monthly NDVI values also indicated strong temporal continuity in crop growth dynamics. Among topographic variables, elevation showed the strongest (though weak) negative correlation with yield (r = −0.29), whereas TWI showed a weak positive correlation (r = 0.24). The negative effect of elevation may be associated with lower temperatures and shorter growing periods at higher altitudes. Multicollinearity analysis indicated that most variables did not exhibit serious collinearity; however, sand, clay, soluble sodium, and NDVI (month 2) showed high VIF values, suggesting the presence of multicollinearity among these predictors. Therefore, these variables should be interpreted with caution in model-based inference due to potential redundancy and overlapping information. Boruta analysis identified NDVI, elevation, TNV, pH and Ca+Mg the most influential variables affecting rainfed wheat yield.
Conclusion
This study identifies four key factors affecting rainfed wheat yield: altitude as a microclimatic factor, TNV, pH, and Ca+Mg as the main soil chemical constraints, and NDVI as the optimal remote sensing indicator for the critical growth stage. The insignificant contribution of soil texture and most topographic indicators highlights the nonlinear and interactive nature of these environmental factors, which can complicate linear modeling approaches. Theoretically, these findings emphasize the improvement of the efficiency of the feature selection algorithm (Boruta). In practical terms, targeted monitoring of the four identified variables (altitude, TNV, pH, Ca+Mg and NDVI) can support sustainable management through methods such as soil amendment to reduce lime and salinity, combined with seasonal remote sensing monitoring. In general, in semi-arid regions, rainfed wheat yield depends more on a few dominant environmental factors than on a wide range of variables. Future research should integrate dynamic climate data and use advanced models (e.g., neural networks) for yield prediction. Such approaches will help reduce yield variability, enhance food security, and support climate adaptation strategies in rainfed agriculture.