Spatial sampling, landscape modeling, and interpretation of soil organic carbon on zero-order watersheds

High-resolution (2 m) raster spatial models of soil organic carbon were created for zero-order watersheds at the North Appalachian Experimental Watershed in Coshocton, Ohio. Soil samples, management practice, and topographic parameters obtained from a digital elevation model were used to create spatial models through multivariate regression. Soil organic carbon was studied because of its importance in the global CO₂ cycle and soil quality. Four small watersheds (5000 m2 to 10,000 m2) were studied, representing continuous no-till corn, conventional tillage, hay, and pasture management practices. Topographic information (slope, wetness index) was derived from a 2 m digital elevation model (DEM). Nearly 500 new soil samples were collected (30 cm depth). Sample locations were chosen using a rational scheme. Legacy and initial sample sets (61 legacy, 250 initial samples) were used to obtain preliminary information on the structure and predictors of spatial variability. Such information was used to design a final sample set (185 samples) that was compatible with statistical (ANOVA and least squares regression), and geostatistical (kriging) modeling techniques.; The results from statistical modeling were used to identify the major challenges to carbon mapping at regional scales for inventory purposes. Soil carbon exhibited a large amount of variability (4 kg C m−2) over small distances (101 meters). Carbon values assigned to county soil survey mapping units are based on few laboratory measurements that cannot properly account for such large gradients. Geostatistical autocorrelation ranged from 30 to 100 meters. Accordingly, the maximum sample spacing permitted for kriging was <30 meters. Kriging was excluded as a viable technique for county-scale carbon inventory because of the required sample intensity. Regression-based techniques were favored because of decreased sample requirements. Management (no-till vs. the others) was the most important predictor of soil carbon (from comparison of management model (ANOVA) to a coupled management-topographic model (multiple regression). Long-term no-till fields must be identified for accurate inventory at broad scales. Finally, local rises and pits (2–10 meters in wavelength) were the most predictive topographic features for soil carbon. High-resolution (<10 m) DEMs are needed to capture this relationship.