Comparison of Different Data Mining Methods in Predicting Soil Organic Carbon Storage in Some Lands of Behbahan City

Soil organic carbon is an important factor in determining the global carbon cycle and global climate regulation. Soil is also the input/output source of carbon to the atmosphere which is depended on the land use. For this purpose, the objective of this study was to compare different methods of data mining in predicting soil organic carbon storage in irrigated, mixed cultivation (irrigated and rainfed), pasture and palm trees lands in some parts of Behbahan city in southwestern of Iran. Soil sampling from depths of 0-30 and 30-60 cm was carried out using conditional Latin hypercube square method. Organic carbon content of the soil samples was determined by Walky-Black method. Bulk density of the soils was determined using paraffin method. The auxiliary parameters used in this study included territory components, OLI sensor image data from landsat 8 and land use map. The results showed that the SAVI, NDVI, NDSI, salinity, carbonate, gypsum and clay indices have the highest correlation with the soil organic carbon stock values. The results also showed that the random forest (RF) (R2= 0.983, RMSE=2.32) was the best model to predict soil organic carbon storage followed by artificial neural network model (R2= 0.887, RMSE= 4.257) and Support Vector Regression Machine model (SVR) (R2 = 0.707, RMSE=7.344).