Prediction of soil organic carbon (SOC) is a crucial proxy to manage and conserve natural resources, monitoring CO 2 and preventing soil erosion strategies at the landscape, regional, and global scales. The objectives of this study were (i) to evaluate capability of regression models to predict SOC using terrain attributes and remote sensing data (ii) to evaluate capability of artificial neural network models to predict SOC using terrain attributes and remote sensing data and determining the most important factors which could explain the variability of SOC in the hilly regions and (iii) to evaluate spatial estimation of SOC by Ok, IDW, RBF and Cok using auxiliary data including the remotely sensed data and terrain attributes. A study area of 24 km 2 in hilly regions of Zargham Abad in south of Semirom, central Iran, under natural range land use located at 51 ? 39 ? E longitude and 31? 18? N latitude was selected. The average elevation of the study area is 2500 m a. s. l. The mean annual temperature and average annual precipitation are 10.6ْ C and 350 mm respectively. 125 soil samples from 0-10 cm depth were collected. Soil organic carbon was measured for the collected soil samples. The elevation data were used to create a 3m digital elevation model (DEM) using ILWIS. Then, primary and secondary topographical indices were generated from the DEM using ILWIS, DIGEM and TAS softwares. Remote sensing data used to develop the models included Landsat ETM. Image geocoding was performed using ground control points obtained through 1/25000 topographic maps with UTM coordinates, with 0.21 pixel accuracy. Finally, regression and ANN models were developed for SOC estimation in the study area and then the developed models were validated by additional samples (25 % of total data set). In the four developes models, different groups of inputes were included. In model 1 and 2 terrain attributes and remote sensing data were considered as predictors, respectively. In model 3 inclusion of terrain attributes and remote sensing data were evaluated and soil texture in addition to model 3 were examined in model 4. The results showed that the regression models explained 60%, 54%, 71% and 83% and ANN models explained 89%, 84%, 94% and 95% of the total variability of SOC in the study area using models 1, 2, 3 and 4, respectively.
