Kourosh Shirani; Morteza Khodagholi; Rostam Khalifehzadeh
Abstract
Awareness of organic carbon status of rangeland soil is important for erosion control and soil protection management. The aim of this study is to prioritize the effective factors, modelling and predicting organic carbon amount using Landsat 8 satellite imagery, accurate digital elevation model (DEM) ...
Read More
Awareness of organic carbon status of rangeland soil is important for erosion control and soil protection management. The aim of this study is to prioritize the effective factors, modelling and predicting organic carbon amount using Landsat 8 satellite imagery, accurate digital elevation model (DEM) related to ALOS sensor and the combined application of factor analysis and multivariate regression model in Semirom watershed located in the south of Isfahan province. For this purpose, after determining the homogeneous units and Stratified Random Sampling of 218 soil samples from these units, the amount of organic carbon, percentages of sand, silt and clay were determined in the laboratory. The development of the combined method was performed using 15 spectral and non-spectral variables and two sets of training data (70%) and test data (30%) of soil samples in order to implement and validate the model, respectively. Then, effective factor prioritization, determination of main components and spatial soil organic carbon zonation map were prepared. Finally, using error measurement criteria, the model was validated and evaluated in the training and test stages. The results showed that fifteen independent variables in the form of six principal components namely vegetation, soil particle size, surface reflectance, soil surface shape, moisture storage and chemical properties have the largest contribution in soil organic carbon storage. Based on the error evaluation metrics (RMSE) and correlation coefficients (R2), the model implementation stage (Training Phase), with respective values of 0.23 and 0.84, demonstrates higher efficiency and captures greater variability in soil organic carbon, as compared to the prediction stage (Test Phase) characterized by a higher error (0.27) and a lower correlation coefficient (0.80). Also, the soil organic carbon content classes of 0.70-0.80 and 1.20-2.35 with an area of 24% and 6% have the highest and lowest area outcrops of soils in the study area, respectively.
