Predicting Spatial Distribution of Organic Carbon Pool In Soils of Ohio Using Four Statistical Approaches.

Assessment of spatial variability of soil organic carbon (SOC) pool at state and regional scale is constrained by extensive soil sampling. A wide range of methods and tools are thus used to estimate the SOC pool at these scales. Since, SOC is influenced by different environmental and anthropogenic variables, and the interaction among them. A better estimation of SOC pool can be made by considering the relationship among these variables and by using geographical information systems (GIS) modeling which can enhance the understanding of the global C cycle. Thus, the specific objective of the study was to estimate the SOC density (C stock per unit area) to 1-m depth for soils of Ohio. Four predictive GIS models used were: ordinary kriging (OK), multiple linear regression (MLR), regression kriging (RK) and geographically weighted regression (GWR). Analytical data for a total of 1424 soil profiles were extracted, of which 80% were used for calibration and 20% for validation. A total of 20 variables (except in OK) including terrain attributes, climate data, bedrock geology, and land use data were used for estimating the SOC density. Four approaches were compared and observed that the GWR provided better predictions with lowest (3.81 kg m^-2) root mean square error (RMSE) followed by RK (3.85 kg m^-2), MLR (3.90 kg m^-2), and OK (4.70 kg m^-2). The relative improvement (RI) values showed that predictions with GWR, RK and MLR approaches were 79, 53, and 49% higher compared to those with OK. Total estimated SOC pool for soils in Ohio ranged from 727 to 742 Tg. The Data show that GWR approach, a local spatial statistical technique, enhances the precision for predicting the SOC pool compared to other global spatial techniques (e.g. MLR and RK).