Accounting for two-billion tons of stabilized soil carbon

Highlights

• Environmental heterogeneity is a large source of soil carbon variability.

• Strategic feature selection is used to identify critical, regional-scale relationships.

• Precipitation, nitrogen, and soil moisture have the strongest association with topsoil C.

• Parent material and elevation have the strongest association with subsoil C.

• Approximately 2.6 Pg C are stored in the upper 1 m of production forestland soil.

Abstract

The pedosphere is the largest terrestrial reservoir of organic carbon, yet soil-carbon variability and its representation in Earth system models is a large source of uncertainty for carbon-cycle science and climate projections. Much of this uncertainty is attributed to local and regional-scale variability, and predicting this variation can be challenging if variable selection is based solely on a priori assumptions due to the scale-dependent nature of environmental determinants. Data mining can optimize predictive modeling by allowing machine-learning algorithms to learn from and discover complex patterns in large datasets that may have otherwise gone unnoticed, thus increasing the potential for knowledge discovery. In this analysis, we identify important, regional-scale determinants for top- and subsoil-carbon stabilization in production forestland across the southeastern US. Specifically, we apply recursive feature elimination to a large suite of socio-environmental data to strategically select a parsimonious, yet highly predictive covariate set. This is achieved by recursively considering smaller and smaller covariate sets—or features—by first training the estimator on the full set to obtain feature importance. The least important features are pruned, and the procedure is recursively repeated until a desired number of covariates is identified. We show that although carbon ranges from 0.3 to 8.2 kg m⁻² in the topsoil (0 to 20 cm), and from 0.4 to 17.6 kg m⁻² in the subsoil (20 to 100 cm), this variability is predictably distributed with precipitation, soil moisture, nitrogen and sand content, gamma ray emissions, mean annual minimum temperature, and elevation. From our spatial predictions, we estimate that 2.6 Pg of soil carbon is currently stabilized in the upper 100 cm of production forestland, which covers 34.7 million ha in the southeastern US.

Introduction

Soil is a critical component of the global climate system, serving as both a sink and a source of CO₂ by actively exchanging carbon with the atmosphere (Davidson, 2016, Luo et al., 2015). Resource management, carbon cycle projections, and policy all rely on accurate representations of soil carbon, yet global estimates remain highly uncertain despite comprehensive efforts to quantify this large reservoir (Gianelle et al., 2010). Depending on the depth modeled, published estimates range from 863 to over 3800 Pg C (Batjes, 2014, Eglin et al., 2010, Sanderman et al., 2017, Watson et al., 2000). Much of the uncertainty is attributed to the vertical and horizontal variability of soil, the scale of analysis, and the subsequent loss of information that occurs during spatial aggregation (Jobbágy and Jackson, 2000, Ross et al., 2013, Xiong et al., 2015), which is particularly problematic when aggregating non-linear data over large areas (Easterling, 1997).

Because soil carbon stabilization is governed by a multitude of non-linear relationships, the strength of relationships can vary with the scale of analysis (Miller et al., 2015, Xiong et al., 2016). Across larger spatial scales, for example, carbon sequestration varies with plant productivity, which in turn is affected by atmospheric CO₂ (Roy et al., 2016), growing season length (Hilton et al., 2017), and resource availability (Eskelinen and Harrison, 2015). However, soil carbon responds to socio-environmental conditions that can vary dramatically at different temporal scales and across regional and sub-regional scales. Factors affecting soil carbon persistence include temperature, precipitation, and acidity (Chen et al., 2018, Schmidt et al., 2011), as well as management (Noormets et al., 2015), and disturbance from land use change (Ross et al., 2016, Xiong et al., 2014b), fire (Godwin et al., 2017), and erosion (Pimentel, 2006). Characterizing these factors at regional scales may be required to upscale soil carbon to global estimates and to refine our understanding of soil carbon stabilization (Mulder et al., 2016).

A recent US Department of Agriculture funded Coordinated Agricultural Project referred to as PINEMAP (Pine Integrated Network: Education, Mitigation, and Adaptation Project) addressed this issue by establishing a monitoring network across the southeastern US to refine our understanding of carbon storage and dynamics in managed forests at the regional-scale (Will et al., 2015). Forests cover 99 million hectares of land in the southeast and account for almost one third of all forested lands in the conterminous US (Oswalt et al., 2014). Not only are these forests an economically-important resource—providing approximately 60% and 16% of the US and global industrial wood supply by volume (Oswalt et al., 2014)—but are ecologically important as well, and sequestered enough aboveground carbon (176 Tg C yr⁻¹) to mitigate 42% of the regions CO₂ emissions between 2000 and 2005 (Lu et al., 2015). About one third of the regions forests are pinelands, of which 19% are comprised of managed pine plantations (Wear and Greis, 2013). The most dominant species—loblolly pine (Pinus taeda L.)—accounts for more than two thirds of all planted tree species in the region (Wear and Greis, 2013).

Intense silvicultural production cycles in this region are a large source of land-cover change, which subsequently affects the region’s carbon cycle. An accurate estimate of soil-carbon distribution in southeastern production forestland is therefore a critical step towards further resolving carbon-cycle science in this region, and to identify factors potentially affecting soil carbon at the global scale. By identifying important regional-scale associations, we hypothesize that our models will provide improved estimates of soil carbon stock when compared with those derived from global models. In this analysis, we develop a data-driven approach to model topsoil (0 to 20 cm) and subsoil (20 to 100 cm) carbon, which is based on a regional compilation (N = 2,564) of soil samples collected from PINEMAP research sites. Variable selection is performed by applying recursive feature elimination to a comprehensive set (N = 73) of environmental predictors to identify parsimonious covariate sets (N = 5) for each depth interval, which are used with the random forest (RF) algorithm to produce soil carbon prediction maps for top- and subsoil depth intervals.