Towards site-specific management of soil organic carbon: Comparing support vector machine and ordinary kriging approaches based on pedo-geomorphometric factors

Abstract

Site-specific management of soil organic carbon (SOC) is crucial for cleaner, sustainable agricultural production and climate change mitigation. The Neyshabur plain of northeastern Iran is plagued with spatially variable critically low SOC. Thus, in this study management zones (MZs) were developed for the site-specific management of SOC. Soil samples (0–0.3 m) and digital elevation model-derived geomorphometric variables were collected at 288 locations. Soil pH, clay content, available phosphorus, available potassium, and the vertical distance to channel network significantly (p < 0.05) correlated with SOC. Subsequently, support vector machine (SVM) and ordinary kriging (OK) approaches were used to map the spatial variability and develop MZs based on SOC and its above-listed pedo-geomorphometric covariates. Generally, the coefficient of determination (R2) and root mean square error (RMSE) of OK and SVM models used to interpolate pedo-geomorphometric covariates, were similar. However, the Fuzzy Performance Index (FPI) and Normalized Classification Entropy (NCE) indices suggest that the MZs based on the SVM model (FPI = 0.091, NCE = 0.113) are more accurate compared to those determined using the OK (FPI = 0.135, NCE = 0.164) method. Although the results demonstrate the potential of both OK and SVM approaches for site-specific management of SOC in the Neyshabur plain, the average difference of 0.8 g kg−1 SOC in the SVM approach compared to OK (p < 0.05) was found to be significant, especially along transitional boundaries between MZs. Thus, the SVM-based MZs developed in this study can be used for more accurate site-specific application of organic amendments, mineral fertilizers, and management practices that improve carbon sequestration and sorption of chemical contaminants in the Neyshabur plain of Iran.