Pyrogenic carbon distribution in mineral topsoils of the northeastern United States

Publication date: 15 June 2017
Source:Geoderma, Volume 296
Author(s): Verena Jauss, Patrick J. Sullivan, Jonathan Sanderman, David B. Smith, Johannes Lehmann
Due to its slow turnover rates in soil, pyrogenic carbon (PyC) is considered an important C pool and relevant to climate change processes. Therefore, the amounts of soil PyC were compared to environmental covariates over an area of 327,757km² in the northeastern United States in order to understand the controls on PyC distribution over large areas. Topsoil (defined as the soil A horizon, after removal of any organic horizons) samples were collected at 165 field sites in a generalised random tessellation stratified design that corresponded to approximately 1 site per 1600km² and PyC was estimated from diffuse reflectance mid-infrared spectroscopy measurements using a partial least-squares regression analysis in conjunction with a large database of PyC measurements based on a solid-state ¹³C nuclear magnetic resonance spectroscopy technique. Three spatial models were applied to the data in order to relate critical environmental covariates to the changes in spatial density of PyC over the landscape. Regional mean density estimates of PyC were 11.0gkg⁻¹ (0.84Ggkm⁻²) for Ordinary Kriging, 25.8gkg⁻¹ (12.2Ggkm⁻²) for Multivariate Linear Regression, and 26.1gkg⁻¹ (12.4Ggkm⁻²) for Bayesian Regression Kriging. Akaike Information Criterion (AIC) indicated that the Multivariate Linear Regression model performed best (AIC=842.6; n=165) compared to Ordinary Kriging (AIC=982.4) and Bayesian Regression Kriging (AIC=979.2). Soil PyC concentrations correlated well with total soil sulphur (P<0.001; n=165), plant tissue lignin (P=0.003), and drainage class (P=0.008). This suggests the opportunity of including related environmental parameters in the spatial assessment of PyC in soils. Better estimates of the contribution of PyC to the global carbon cycle will thus also require more accurate assessments of these covariates.