Overview of the U.S. Rapid Carbon Assessment Project: Sampling Design, Initial Summary and Uncertainty Estimates

Abstract

The Rapid Carbon Assessment (RaCA) project was undertaken by the Soil Science Division of the Natural Resource Conservation Service (NRCS) to capture baseline soil carbon stocks across the conterminous US (CONUS). A multi-level hierarchical design was used to ensure that samples were distributed across regions, soils and land use/land cover classes (LULC). Within those strata, sites were selected at random locations where five pedons were described and sampled at 0-5 cm and by genetic horizon from 5 to 100 cm. A total of 6,148 sites, 32,084 pedons and 144,833 samples were described. Bulk density was calculated for samples from the upper 50 cm and predicted for deeper samples using pedon and horizon information in a regression tree developed with random forests. Soil organic carbon (SOC) concentration was predicted for each sample using processed Visible-Near Infrared spectra and a random forest model. Pedon SOC stocks were calculated by fixed depth to 100 cm. Expected variance was introduced into the stock calculations using analytical and modeling prediction errors (e. g., SOC concentration and bulk density measurements) and the stratified sampling design was partitioned using a hierarchical Bayesian modeling approach. Pedons were averaged by site. The mean of all RaCA site SOC stocks to 100 cm was 321.1, with a median of 173.3 and range of 2 to over 5,000 Mg ha(-1). Geometric means of soil groups and LULC classes were used to extrapolate results to all assessed areas. Further work is needed to properly weight averages by areal extent and assess the cause of higher than expected site SOC stock values.