Soil Carbon Storage Estimation in a Forested Watershed using Quantitative Soil-Landscape Modeling

Proxy information is used to stratify larger areas, and then measurements within each of these strata are agCarbon storage in soils is important to forest ecosystems. Moreover, gregated and multiplied by the area of each stratum forest soils may serve as important C sinks for ameliorating excess (Schimel and Potter, 1995). Soil survey maps and laboraatmospheric CO2. Spatial estimates of soil organic C (SOC) storage tory characterization data are the primary resources for have traditionally relied upon soil survey maps and laboratory characterization data. This approach does not account for inherent variability estimating the amount of SOC stored in soils using this within map units, and often relies on incomplete, unrepresentative, approach (e.g., Homann et al., 1998; Kern et al., 1998; or biased data. Our objective was to develop soil-landscape models Galbraith et al., 2003; Tan et al., 2004). There are numerthat quantify relationships between SOC and topographic variables ous benefits to this approach (Arnold, 1995), but there derived from digital elevation models. Within a 1500-ha watershed also are several limitations. There may be significant in eastern Kentucky, the amount of SOC stored in the soil to a depth variability of SOC content within map units due to natuof 0.3 m was estimated using triplicate cores at each node of a 380-m ral soil variability and unmapped inclusions of higher grid. We stratified the data into four aspect classes and used robust or lower C soils (Eswaran et al., 1995). Galbraith et al. linear regression to generate empirical models. Despite low coeffi(2003) attributed the greatest source of uncertainty in cients of correlation between measured SOC and individual terrain their SOC maps to the high variation among SOC data attributes, we developed and validated models that explain up to 71% of SOC variability using three to five terrain attributes. Mean SOC from replicate samples from the same soil series. Also, content in the upper 30 cm, as predicted from our models, is 5.3 kg the soil characterization data that are commonly used m 2, compared with an estimate of 2.9 kg m 2 from soil survey data. to establish SOC levels within a soil map unit were not Total SOC storage in the upper 30 cm within the entire watershed is originally collected for examining SOC content, and 82.0 Gg, compared with an estimate of 44.8 Gg from soil survey data. therefore may not include all of the necessary data for A soil-landscape modeling approach may prove useful for future SOC calculating SOC storage (Amichev and Galbraith, 2004). spatial modeling because it incorporates the continuous variability of These data sets also may be biased toward different soil SOC across landscapes and may be transportable to similar landscapes. types or landscape settings, and may not adequately represent true range in variability of SOC (Tan et al., 2004). A important component in understanding the role An alternative to the measure and multiply approach of soils in the global C cycle is developing reliable is referred to as “paint by numbers” (Schimel and Potestimates of the amounts of C stored in the soil and ter, 1995). This approach incorporates information on other terrestrial C pools. Estimates of SOC storage have multiple environmental factors within geographic areas been made at global (Post et al., 1990; Akin, 1991; Esthat are used as input variables to models, which then waran et al., 1995), continental (Bajtes, 2000), national are used to make predictions that can be multiplied by (Kern, 1994), state (Bliss et al., 1995; Kern et al., 1998; the areal extent of given combinations of each of these Amichev and Galbraith, 2004; Tan et al., 2004), regional factors. This approach is akin to soil-landscape modeling (Homann et al., 1998; Galbraith et al., 2003), and land(McSweeney et al., 1994), in which the variability of soils scape (Bell et al., 2000; Arrouays et al., 1995, 1998; is analyzed with respect to changes in environmental Chaplot et al., 2001; Terra et al., 2004) scales. These variables known to influence soil property variability, studies have used a range of techniques by which point such as topography, hydrology, or geology. measurements of SOC are extrapolated to larger scale Soil-landscape modeling has been successfully appredictions of C storage. plied to predict soil variability at the site or hillslope These various techniques can be divided into two scale, focusing almost exclusively on small-scale landgeneral methods of spatial extrapolation. The most scapes of 100 ha, with some as small as 2 ha (Moore prominent method of producing coarse predictions of et al., 1993; Thompson et al., 1997, 2001; Chaplot et al., SOC storage at regional to global scales is often referred 2000; Gessler et al., 2000; Park et al., 2001; Florinsky et to as “measure and multiply” (Schimel and Potter, 1995). al., 2002). These studies have demonstrated that combinations of one to five terrain attributes derived from a digital elevation model (DEM) can explain 20 to 88% of J.A. Thompson, Division of Plant and Soil Sciences, West Virginia Univ., Morgantown, WV 26506-6108; R.K. Kolka, USDA Forest Serthe variability of selected soil properties. The empirical vice-North Central Research Station, Grand Rapids, MN 55744-3399. relationships between soil properties and terrain attriReceived 30 Sept. 2004. *Corresponding author (james.thompson@ butes are unique to each soil property and each soilmail.wvu.edu). forming environment. Modeling examples at the waterPublished in Soil Sci. Soc. Am. J. 69:1086–1093 (2005). shed scale (and coarser) are more limited and require Pedology doi:10.2136/sssaj2004.0322 © Soil Science Society of America Abbreviations: CFI, continuous forest inventory; DEM, digital elevation model; SOC, soil organic carbon. 677 S. Segoe Rd., Madison, WI 53711 USA 1086 Published online June 2, 2005