Prediction of Soil Organic Carbon (SOC) in Semi-Arid Rangeland Using Multivariate Statistical Analysis based on Remotely Sensing Data (Case study: Neyshabur Rangeland, Khorasan-Razavi Province, Iran)
Authors
Abstract
Prediction of Soil Organic Carbon (SOC) reservoirs is high priority in the rangelands managements in arid and semi-arid regions. This study was conducted to estimation carbon sequestration using Multivariate Linear Regression Analysis (MLR), Principal Component Analysis (PCA) and Euclidean Distance from the soil line (D) on remote sensing data in semi-arid rangelands of southwest of Neishabour, Khorasan Razavi province, Iran. The map of SOC was prepared using a total of 102 soil samples (depth 0-10 cm). Landsat 8 images of the study area were provided on 5 July 2018 and used to develop the models including (OLI, TIRS), visible, near-infrared, middle-infrared, and thermal infrared bands. Models were developed using SOC as dependent variable and spectral data of MLR, PC1 and Euclidean D soil line as independent variables. Then, the developed models were validated using additional samples (30 points). The results illustrated that the MLR, PCA, and Euclidean D soil line models explain 62, 45, and 53% of the total variability of SOC coupled with root Mean Square Error (RMSE) values 0.09, 0.21, and 0.05, respectively. Therefore, the MLR technique could provide superior predictive performance than that for PCA and Euclidean D soil line techniques. It was concluded that the SOC spatial information derived using the MLR technique had much greater spatial detail and higher quality than to that derived from the conventional soil map.
