Spatial Estimation of Coefficient of Variation of Soil Organic Carbon Stocks Using Spectral and Digital Data

Soil entities depends on its physical, chemical, and biological characteristics. Soil organic carbon (SOC) stocks is one of the main factors whose variations affect all these parameters. So, this study was performed for mapping the coefficient of variations (CV) of SOC stocks in some parts of the Simineh Roud watershed. Soil sampling performed using the Latin Hypercube method (cLHm) at 210 points from 0 to 30 cm of the soil surface, and the organic carbon was measured, then SOC stocks was calculated. In the next step, using Random Forest (RF) model the effective parameters were calculated (in this step, to model the SOC stocks, standardized spectral reflectance index and extracted data from digital elevation model were used). Finally, RF model was performed 100 times, as well as mapping of the values of upper (95th %), lower (5th %), and average SOC stocks for each pixel with a spatial resolution of 30 × 30 m was obtained. To obtain the CV with a confidence coefficient of 90%, the percentile of 95% and 5% were subtracted. The CV was obtained by dividing it by the mean. The results showed that the accuracy coefficient (R2) for modeling SOC stocks by the RF model was 0.81 and the mean accuracy coefficients including RMSE and MAE were 0.44 and 0.34 (kg/m2), respectively. Also, the results of CV mapping for the amount of SOC stocks in the study area showed that the amount of variation of this parameter varies between 3.9- 55%. Based on the results of the CV mapping of the study area, the most and lowest variations were observed in dry farming and grasslands, respectively. Probably, continuous cultivation and low return of organic matter in dry farming have increased the CV of SOC stocks in dry farming use.