Implications of Using USDA-NCSS Bulk Density to Estimate Carbon Stocks in Forest Soils Across the Southeastern United States

Thursday, 17 December 2015
Poster Hall (Moscone South)
Allan R Bacon1, Katherine Akers2, Josh Cucinella1, Sabine Grunwald3, Eric J Jokela1, Daniel Markewitz2, Marshall A. Laviner4, Jason G Vogel5, Timothy Martin1, Thomas D. Fox4, Michael Kane2, Gary F Peter1, John M Davis1 and C. Wade Ross6, (1)University of Florida, School of Forest Resources and Conservation, Gainesville, FL, United States, (2)University of Georgia, Warnell School of Forest Resources, Athens, GA, United States, (3)University of Florida, Soil and Water Science, Gainesville, FL, United States, (4)Virginia Polytechnic Institute and State University, Department of Forest Resources and Environmental Conservation, Blacksburg, VA, United States, (5)Texas A & M University College Station, College Station, TX, United States, (6)University of Florida, Soil and Water Science, Ft Walton Beach, FL, United States
Estimates of soil bulk density (Db) are critical for accurate estimates of soil carbon (C) stocks, and thus, greatly influence the balance and interpretation of soil C budgets at plot, regional, and national scales. Large scale soil C investigations in the United States (US) almost always utilize a compilation of more than 20,000 Db observations across the US within the USDA-NRCS National Cooperative Soil Survey (NCSS) database. NCSS observations can be manually extracted as point data and then stratified or modeled by a variety of soil taxonomic, geographic, and environmental factors to estimate Db across large scales. NCSS observations also underpin the popular Soil Survey Geographic (SSURGO) database which provides continuous Db estimates across most of the US. Here, for the first time, we evaluate the precision and accuracy with which NCSS data can estimate forest soil Db across the southeastern United States and explore how using these observations impacts soil C budgets in forests across the region.

We analyze and compare nearly 3,000 Db observations from the NCSS database to nearly 1,500 Db observations from the PINEMAP Tier II Network (325 experimental forest plots) across the southeastern US. We model all NCSS observations and 70% of the PINEMAP Tier II observations (a calibration dataset) separately with Random Forest algorithms to create a variety of Db predictive models at 0-10, 10-20, 20-50, and 50-100 cm depths. We validate all models against 30% of the PINEMAP Tier II observations (a validation dataset). As indexed by the mean prediction error (MPE), NCSS observations tend to over predict forest soil Db across the validation dataset by an average of 0.20 g/cc. Incorporating this positive bias of NCSS Db predictions into C stock estimates in the top 100 cm of soil across the PINEMAP Tier II network inflates C stock estimates by an average of 13 Mg/ha. Our findings identify significant potential for NCSS observations to over predict soil Db, and thus soil C stocks, in forests across the southeastern US. We suggest that similar results might be found for forest soils elsewhere in the US due to an inherent agricultural bias of NCSS data, and conclude that the potential for artificially high stock estimates warrants consideration when forest ecosystems and NCSS Db observations are included in regional soil C analyses.