Predicting soil organic carbon density using auxiliary environmental variables in northern Iran

In the present study, artificial neural networks (ANNs) were employed to develop models to predict soil organic carbon density (SOCD) at different depths of soil layers. Selected environmental variables such as vegetation indices, soil particle size distribution, land use type, besides primary and secondary terrain attributes were considered as the input variables. According to the results, the ANN models explained 77% and 72% of the variability in SOCD at soil layer depths of 0–20 cm and 20–40 cm, respectively, at the site studied. Sensitivity analyses showed that the most considerable positive contribution of variables for predicting SOCD included the land use type, normalized difference vegetation index (NDVI) > normalized difference water index (NDWI) > silt > clay > elevation in the 0–20 cm soil layer. On the other hand, for the 20–40 cm soil layer, the land use type following NDVI > NDWI > clay > silt were identified as the most powerful predictive factors. In the Deylaman region, in both soil layers, sand had a considerable negative effect on SOCD and most of the terrain attributes had no significant impact on the SOCD prediction. Therefore, these results provide valuable information for sustainable management and decision-making on a landscape scale for governors and other users.