Using Multivariate Regression and Artificial Neural Network in Predicting and Zoning Organic Carbon of Soil Under the Influence of Topographic Variables1

Considering the role of organic carbon in soil management and sustainability, the prediction of the organic carbon of soil at regional, national and global scale is of great importance. Along with the prediction of organic carbon, zoning is also effective as a way to select the appropriate strategy for controlling this element and make appropriate decisions inmanaging and preventing environmental disasters.The purpose of this research is to predict and zonethe organic carbon of soil based on primary and secondary topographic variables using artificial neural network. For this purpose, 127 circular sample plots,with an area of 1000 square meters,were harvested from the educational and research forest of Natural Resources Faculty of TarbiatModarres University following a systematically randomized procedure. At the center of each sample plot, a soil sample was taken from a depth of 0-10 cm. In addition to measuring the organic carbon from thecollected samples, the primary and secondary topographic variables of the area were calculated on the basis of the digital elevation model and finally, using artificial neural network and multivariate regression, the storage model of the organic carbon was created. The aforementioned models were evaluated using R2 and RMSE criteria. The results showed that the artificial neural network model with a capacity to justify 76% of the changes in the organic carbon of soil has a better performance in comparison to the multivariate regression model with a capacity to justify 26% of the amount of changes. In addition, plan curvature, curvature and slope indices have the greatest impact on the changes in the organic carbon of soil and together justify 90% of the changesin organic carbon.