Developing a Pedotransfer Function to Estimate Soil Organic Carbon in the United States.

Soil organic carbon (SOC) is a highly influential component of soil that greatly affects soil function. Scientists have long understood that human activities destroy soil carbon storage. However, only in recent years has the significance of the depletion of soil carbon, and by association soil organic carbon (SOC), been evaluated at the ecosystem and global scales. In this study, we evaluated the relationships of SOC with other, more commonly measured soil properties using a variety of statistical methods, such as principal component regression (PCR), multiple linear regression (MLR), and genetic algorithm (GA). Establishing pedotransfer functions between SOC and other common variables has the value of saving money and time for soil samples that are already limited in the quantity that can be taken cost effectively. We analyzed data for 331 samples across 43 states within the USA, of which 100 samples (30%) were chosen for independent validation. Potential covariates evaluated for inclusion in the pedotransfer function to predict SOC were clay, fine silt, coarse silt, very fine sand, fine sand, medium sand, coarse sand, very coarse sand, potassium (K), iron (Fe), aluminum (Al), manganese (MN), calcium (Ca), magnesium (Mg), phosphorous (P), and pH. These covariates were selected because of their regular use in soil testing, particularly in the context of soil fertility and nutrient management. After applying stepwise regression, clay, Al, Ca, and fine silt were identified as the predictive variables. Using those four covariates, the GA method had the highest prediction performance (R²=0.74 and RMSE=0.038) as compared to the other methods, PCR (R²=0.70 and RMSE=0.04) and MLR (R²=0.61 and RMSE=0.044), for estimating SOC.