{"type": "FeatureCollection", "features": [{"id": "10.5061/dryad.0k6djhb5k", "type": "Feature", "geometry": null, "properties": {"updated": "2026-06-23T16:22:10Z", "type": "Dataset", "created": "2023-08-29", "title": "Empirical data and model simulations of the effect of repeated hurricanes on soil carbon dynamics in a humid tropical forest", "description": "unspecifiedSite description Soils were sampled from the Bisley Experimental Watershed of the LEF, Puerto Rico (18.3157 deg. N, 65.7487 deg W), a Long-Term Ecological Research and Critical Zone Observatory and Network site (https://luq.lter.network). The mean maximum daily temperature at Bisley was 27 \u00baC between 1993 and 2010 (Gonzales, 2020), with little seasonality. The mean annual precipitation at Bisley was 3883 (\u00b1 864 s.d.) mm y-1 from 1988 through 2014 (Gonz\u00e1lez, 2017; Murphy et al., 2017). Rainfall occurs all year, though January through April experience slightly less precipitation than other months (Heartsill-Scalley et al., 2007). The site is a humid tropical forest with a diverse tree community of approximately 170 species > 4 cm diameter at breast height (Weaver & Murphy, 1990), and dominated by tabonuco (Dacryodes excelsa Vahl). Elevation of Bisley spans from 261 m a.s.l. at the base to 450 m a.s.l. on the ridges (Scatena, 1989). Soils in Bisley are derived from volcaniclastic sediments of andesitic parent material (Scatena, 1989).\u00a0 Ridge soils are classified as Ultisols (Typic Haplohumults), while slope soils are classified as Oxisols (inceptic and Aquic Hapludox), and valley soils are classified as Inceptisols (Typic Epiaquaepts) (Hall et al., 2015; McDowell et al., 2012; Scatena, 1989). Detailed site descriptions can be found in Scatena (1989), Heartsill-Scalley et al (2010), and McDowell et al (2012). Here we refer to soil organic C (SOC) and soil C interchangeably because there is no detectable inorganic C in these soils. Hurricane occurrence\u00a0 Figure 1: Timeline of major hurricanes that have affected Luquillo Experimental Forest between sampling dates. Nine major hurricanes (category 3 or higher) have impacted Puerto Rico between 1851 and 2019 (L\u00f3pez-Marrero et al., 2019), and five of these hurricanes have impacted the LEF. Until 1998, hurricanes had historically directly impacted the LEF approximately every 60 years (Scatena & Larsen, 1991). Before the initial sampling campaign of this study, Hurricane San Cipri\u00e1n in 1932 was the most recent storm to cause major disturbance to the LEF (Scatena & Larsen, 1991).\u00a0 However, since sampling in 1988, four major hurricanes have impacted the forest (Figure 1). Hurricane Hugo (Category 3-4) in 1989, Hurricane Georges (Category 3) in 1998, and Hurricanes Irma and Maria (Categories 5 and 4, respectively) within two weeks in 2017. The trajectory and windspeeds of all these hurricanes caused widespread defoliation. Litterfall historically takes over five years to return to pre-hurricane levels (Scatena et al., 1996).\u00a0 Sampling Sample collection occurred in 1988 and again in 2018. In both years, samples were collected from three depths: 0\u201310 cm (the A horizon), 10\u201335 cm (all of the B1 horizon and part of B2), and 35\u201360 cm (B2 to C) using an 8 cm diameter soil auger. Soils in this study were sampled at three separate sites at least 40 m from one another for each of three topographic locations, ridge, slope, and upland valley. Two separate cores were taken from a fourth topographic location in the riparian valley, that characterized a smaller proportion of the area of these watersheds (Scatena & Lugo, 1995). Riparian valley sites were ephemeral streambeds with a high boulder presence that limited sampling to less than 25 cm depth in one case. Sampling sites from 1988 were marked with flags, and samples from 2018 were collected from within 15 m of the same locations as the replicates from 1988, for consistency. Samples collected in 1988 were analyzed for bulk density, pH, soil moisture, and a suite of soil chemical properties (see Silver et al. 1994). Samples were then air-dried and stored in closed Ziploc bags within paper bags in a storage facility in Richmond, CA, USA before density fractionation in 2018. Fresh samples collected in 2018 were also characterized for pH, soil moisture, and soil chemistry. Approximately 3 g subsamples from each fresh sample in 2018 were immediately extracted with 45 mL of 0.2 M sodium citrate/0.5 M ascorbate solution, shaken for 16 hours, then centrifuged and the supernatant decanted to measure concentrations of poorly crystalline iron (Fe) oxides. Within two days of being double-bagged in Ziploc bags, fresh samples were further subsampled and analyzed for pH in a 1:1 soil-to-water slurry (Thomas, 1996) and for gravimetric soil moisture by oven-drying ~10 g subsamples at 105 \u00baC until a constant weight. Soil samples were air-dried before further processing and analysis. Air-dried soils from both sampling years were sieved to 2 mm and large roots were sorted out. Soil Density fractionation Soil was fractionated by density following the method of Swanston et al. (2005), as modified by Marin-Spiotta et al., (2009). Approximately 20 g of air-dried soil was added to centrifuge tubes. Sodium polytungstate (SPT, Na6 [H2W12O40] TC-Tungsten Compounds, Bavaria, Germany) in solution of density 1.85 g cm-3 was added to centrifuge tubes and agitated before centrifuging. The density of the SPT followed previous studies from this and nearby sites to allow direct comparison (Guti\u00e9rrez del Arroyo & Silver, 2018; Hall et al., 2015). Particulate organic matter floating at the surface after centrifugation, the free light fraction (FLF), was aspirated and then rinsed with 100 ml of deionized water 5 times on a 0.8 \u00b5m pore polycarbonate filter (Whatman Nuclepore Track Etch Membrane, Darmstadt, Germany). Rinsed FLF was oven-dried at 65 \u00baC until weight had stabilized. The remainder of the sample was combined with 70 ml of additional SPT and mixed using an electric benchtop mixer (G3U05R, Lightning, New York, NY, USA) at 1700 rpm for 1 min and sonicated in an ice bath for 3 min at 70% pulse (Branson 450 Sonifier, Danbury, CT, USA). Sonication is intended to disrupt soil structure and liberate organic matter that has been occluded in aggregates. The sonicated slurry was centrifuged again, and the light fraction at the surface, the occluded light fraction (OLF), was aspirated, rinsed, and dried using the same method as for the FLF. The remaining soil pellet was considered the heavy fraction (HF), or mineral-associated organic matter fraction. The HF was rinsed by thoroughly mixing with 150 ml of deionized water in the centrifuge tube, centrifuging, and removing the supernatant repeatedly until the fraction had been rinsed 5 times. The rinsed HF was oven-dried at 105 \u00baC until weight stabilized. The average mass recovery was 98%. Soil C and N and \u03b413C Dried bulk and HF soils were homogenized separately using a Spex Ball mill (SPEX Sample Prep Mixer Mill 8000D, Metuchen, NJ). The FLF and OLF were homogenized separately by hand using a mortar and pestle. All homogenized samples were then analyzed at U. C. Berkeley for C and N concentrations on the CE Elantech elemental analyzer (Lakewood, NJ) and for \u03b413C in the Stable Isotope Laboratory at UC Berkeley, using a CHNOS Elemental Analyzer interfaced to an IsoPrime 100 mass spectrometer (Cheadle Hulme, UK), with a long-term external precision of 0.10 %. \u00a0Soil C stocks were calculated by multiplying the C concentrations (%) by the oven-dry mass of bulk soil (< 2 mm) and dividing by depth and the bulk density as measured in 1988 (Silver et al., 1994; Throop et al., 2012). Radiocarbon Homogenized soil samples were combusted to CO2 in sealed glass tubes along with silver (Ag) and copper oxide (CuO) at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Lab. The CO2 was then graphitized on Fe powder under pressurized hydrogen gas (Vogel et al., 1984). Graphite was pressed into aluminum targets and run on the Compact Accelerator Mass Spectrometer for radiocarbon analysis (Broek et al., 2021). Radiocarbon is reported in \u039414C, following Stuiver & Polach (1977), and calculated based on the fraction of modern isotope composition, corrected for the year of sampling, and corrected for mass-dependent fractionation with observed \u03b413C values of the sample. The compact AMS had an average \u039414C precision of 3.2 %. We report the corrected \u039414C value and \u0394\u039414C, which is calculated as \u039414C of the sample minus \u039414C of the atmosphere, to account for rapidly changing atmospheric \u039414C during the study period. Atmospheric radiocarbon has been decaying nonlinearly since the peak of weapons testing in the 1950s. Radiocarbon signatures in the soil are strongly influenced by the atmospheric D14C signature, making them useful for modeling soil C age and transit time, especially since the 1950s. To compare the contribution of modern C between 1988 and 2018, it is useful to take the difference between soil and atmospheric D14C values, or DD14C, because atmospheric D14C declined between 1988 (98 %) and 2018 (4.4 %) in Northern Hemisphere Zone 2 (Hua et al., 2013). We note that the decline in atmospheric D14C is nonlinear, and thus the DD14C in 2018 soil will be less sensitive to short-term shifts in D14C inputs than the samples from 1988. Carbon age and transit time modeling Transit times and ages of C were modeled with the package \u201cSoilR\u201d (Sierra et al., 2012, 2014) in R, version 4.0.2. The change in C density fractions over time, termed C flow, was modeled using a 3-pool structure with a series flow matrix, under the simplifying assumption that C flows from the litter pool to the FLF, where it is sequentially transferred into the OLF and HF pools (Figure 2). The model structure is depicted in basic form in equation 1, \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 (1)\u00a0 dC(t)/dt = Inputs - k*C \u00a0in matrix form with explicit pools in equation 2, \u00a0 \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 \u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 (2)\u00a0 dC(t)/dt = [Litter Inputs; 0; 0] + [-kFLF, 0, 0 ; a21,\u00a0-kOLF, 0; 0, a32, -kHRF] * [CFLF; COLF; CHF] where k is the first-order decay constant for each pool, a is the C transfer rate between pools (i.e. a21 is the transfer from FLF (pool 1) to OLF (pool 2) and a32 is the transfer from OLF (pool 2) to HF (pool 3)), and C is the C stock of each pool. The transitTime and systemAge functions within the \u201csoilR\u201d package use this model structure to solve for the distribution of ages (time since entry) of each pool, and the distribution of transit times (times between entry and exit from the bulk soil) (Sierra et al 2016). Distributions of age and transit time were time-independent and did not assume a specific distribution (Sierra et al., 2014, 2017). Figure 2: Hypothesized flow of C in soils. Free light fraction (FLF) C (pink) is either decomposed (at cycling rate -kFLF * FLF) or transferred to the occluded light fraction pool (OLF, blue) with the transfer proportion defined by a21. Carbon transfer between the OLF and heavy fraction (HF, purple) is defined by transfer coefficient a32, and is respired from these pools at cycling rates -kOLF* OLF and -kHF* HF, respectively. Figure adapted from Sierra et al. (2012). Soil D14C and C stock mean and standard deviations from each time point, depth, and fraction were used to constrain the matrix model describing the movement of C through three soil pools and losses of C from each pool. Topography was not a strong predictor of patterns in D14C, C stocks, or C fractions, so samples from all topographies were aggregated for model simulations. The model used mean observed C content in each pool for each depth in 1988 as initial conditions for SOC stocks. Above and belowground litter inputs at 0\u201310 cm were assumed to be 900 g C m-2 in non-hurricane or hurricane recovery years, based on observations from the same site (Liu et al., 2018; Scatena et al., 1996; Silver et al., 1996; Vogt et al., 1996). Inputs to the 10\u201335 cm and 35\u201360 cm depths were estimated using observations of live fine roots on the surface and typical root distribution in the forest (Silver & Vogt, 1993). Total root input is approximately threefold the input of fine roots alone (McCormack et al., 2015; Yaffar & Norby, 2020), and live fine roots in the 0\u201310 cm depth had a mean biomass of 80 - 250 g C m-2 \u00a0(Hall et al., 2015), suggesting that total root C inputs of approximately 450 g C m-2 to the surface would be well within the expected range. Root inputs below 0\u201310 cm were estimated assuming that inputs follow the typical distribution of root biomass in Puerto Rican tropical forests, with 60\u201370% of root biomass in 0\u201310 cm, an additional 20-30% of biomass in 10\u201335 cm (~135 g C m-2\u00ad), and 5\u20138% of biomass is in the 35\u201360 cm depth (~40 g C m-2\u00ad) (Silver & Vogt, 1993; Yaffar & Norby, 2020). The model was parameterized under two scenarios for each depth: 1) constant inputs, assuming a steady-state undisturbed forest, and 2) hurricane inputs, which simulated the input fluxes from defoliation during the three major hurricanes, followed by a subsequent reduction in litter inputs and then litterfall increasing linearly to pre-hurricane inputs over 6 years (Scatena et al., 1996; Silver et al., 1996; Vogt et al., 1996). Hurricane inputs were imposed as an additional pulse of litter inputs to each depth interval, declining with depth. \u00a0The 0\u201310 cm interval received 100% of the surface input pulse, the 10\u201335 cm depth received a pulse of root inputs equivalent to 30% of the surface input pulse, and the 35\u201360 cm depth received root inputs equal to 10% of the surface input pulse. Surface litter pulses under hurricanes were specified according to measured litterfall values and were 42.5 g C m-2\u00ad to the surface in 1989 (Hurricane Hugo) and 1998 (Hurricane Georges) (Scatena et al., 1993; Silver et al., 1996) and 1611 g C m-2 \u00a0in 2017 (Hurricanes Irma and Maria) (Liu et al. 2018a). The same soil D14C and C stock observations were used to constrain the model under each scenario, with only the input regime varying. Parameters of the transfer matrix (-k\u00ad\u00adFLF, -k\u00ad\u00adOLF, -k\u00ad\u00adHF, a21, a32) were constrained using a cost function to accept or reject potential parameter sets over 1000 iterations, based on observed D14C and C stock means and standard errors from both time points (1988 and 2018). A Markov chain Monte Carlo (MCMC) simulation initialized with cost-optimized parameters was run to assimilate observed data and optimize parameter choices to the observations using function modMCMC() from R package \u201cFME\u201d (Sierra et al., 2014; Soetaert & Petzoldt, 2010). The MCMC was iterated over at least 20,000 simulations or until parameter solutions converged according to the trace, which was over 100,000 iterations at the 35\u201360 cm depth. The first half of the iterations was considered the burn-in period before the chain started to converge near an equilibrium, and these iterations were discarded in calculations of optimal parameters. The model output for the surface soils of the HF pool was validated using published radiocarbon values from the mineral-associated fraction (the only fraction analyzed) of samples from the site taken in 2012 (Hall et al., 2015).\u00a0 Bulk and pool soil C age and transit time density distributions and mean values were calculated using the systemAge() and transitTime() functions from the \u201cSoilR\u201d package. Mean density distributions were calculated using the mean parameter set given from the MCMC analysis. Standard deviation from the mean was calculated using the systemAge() and transitTime() functions on 200 sets of five parameters selected randomly within one standard deviation of the mean of each parameter given as output from the MCMC. Lower and upper limits of SOC ages and transit times were calculated using the upper and lower ranges of these iterations. Statistics Statistics were run in R, version 4.0.2 (R Core Team, 2020). The statistical model selection followed the recommendations of Zuur et al (2009). Statistical models were chosen using a linear mixed effects model in package \u201clme4\u201d, with random slopes accounting for the influence each core, or sampling site, had on the response variable values as they varied with depth. This random effect of the core site on the depth effect was evaluated using a restricted maximum likelihood approach and was included in the initial evaluation of all model comparisons. Linear mixed effect models included year, topographic position, depth, and interactions as fixed factors, and the depth effect of each core as a random factor for each of the response variables: C concentration, N concentration, d13C, DD14C. In evaluations of some response variables with AIC and BIC criteria, the random effect no longer enhanced the model, and model comparison proceeded using ANOVAs of linear models without random effects. Topographic effects on C concentrations are discussed in the supplemental information. Model assumptions were evaluated using the check_model function in R package \u201cperformance\u201d, to check for multicollinearity, normality of residuals, homoscedasticity, homogeneity of variance, influential observations, and normality of random effects. In the cases when random effects were significant (bulk soil d13C and DD14C, FLF DD14C and HF C and N concentrations), fixed effects were chosen using ANOVA of subsequent models using maximum likelihood estimation, with the random effects held constant. Once fixed effects were established, the model was re-fitted using a restricted maximum likelihood approach to report model estimates, and an ANOVA was run to determine the significance of the response variable. In all cases, P-values were estimated using Tukey\u2019s honest significant post-hoc test to assess significant differences between variables, in the package \u201cagricolae\u201d in R, and contrasts and standard errors of contrasts were estimated using lsmeans() function in package \u201clsmeans\u201d in R. Values of\u00a0P < 0.10 were reported as significant unless otherwise specified. The topographic position was not a significant predictor for most variables, so results are reported as means aggregated across positions.", "keywords": ["soil organic carbon", "Transit time", "Tropical forest soil", "FOS: Earth and related environmental sciences", "Soil R", "density fractions", "Radiocarbon"], "contacts": [{"organization": "Mayer, Allegra", "roles": ["creator"]}]}, "links": [{"href": "https://doi.org/10.5061/dryad.0k6djhb5k"}, {"rel": "self", "type": "application/geo+json", "title": "10.5061/dryad.0k6djhb5k", "name": "item", "description": "10.5061/dryad.0k6djhb5k", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main/items/10.5061/dryad.0k6djhb5k"}, {"rel": "collection", "type": "application/json", "title": "Collection", "name": "collection", "description": "Collection", "href": "https://repository.soilwise-he.eu/cat/collections/metadata:main"}], "time": {"date": "2024-04-01T00:00:00Z"}}, {"id": "10.1007/s10533-021-00838-z", "type": "Feature", "geometry": null, "properties": {"updated": "2026-06-23T16:15:20Z", "type": "Journal Article", "created": "2021-08-27", "title": "Soil organic matter turnover rates increase to match increased inputs in grazed grasslands", "description": "Abstract

Managed grasslands have the potential to store carbon (C) and partially mitigate climate change. However, it remains difficult to predict potential C storage under a given soil or management practice. To study C storage dynamics due to long-term (1952\uffe2\uff80\uff932009) phosphorus (P) fertilizer and irrigation treatments in New Zealand grasslands, we measured radiocarbon (14C) in archived soil along with observed changes in C stocks to constrain a compartmental soil model. Productivity increases from P application and irrigation in these trials resulted in very similar C accumulation rates between 1959 and 2009. The \uffe2\uff88\uff8614C changes over the same time period were similar in plots that were both irrigated and fertilized, and only differed in a non-irrigated fertilized plot. Model results indicated that decomposition rates of fast cycling C (0.1 to 0.2\uffc2\uffa0year\uffe2\uff88\uff921) increased to nearly offset increases in inputs. With increasing P fertilization, decomposition rates also increased in the slow pool (0.005 to 0.008\uffc2\uffa0year\uffe2\uff88\uff921). Our findings show sustained, significant (i.e. greater than 4 per mille) increases in C stocks regardless of treatment or inputs. As the majority of fresh inputs remain in the soil for less than 10\uffc2\uffa0years, these long term increases reflect dynamics of the slow pool. Additionally, frequent irrigation was associated with reduced stocks and increased decomposition of fresh plant material. Rates of C gain and decay highlight trade-offs between productivity, nutrient availability, and soil C sequestration as a climate change mitigation strategy.Soils store large quantities of carbon in the subsoil (below 0.2\uffe2\uff80\uff89m depth) that is generally old and believed to be stabilized over centuries to millennia, which suggests that subsoil carbon sequestration (CS) can be used as a strategy for climate change mitigation. In this article, we review the main biophysical processes that contribute to carbon storage in subsoil and the main mathematical models used to represent these processes. Our guiding objective is to review whether a process understanding of soil carbon movement in the vertical profile can help us to assess carbon storage and persistence at timescales relevant for climate change mitigation. Bioturbation, liquid phase transport, belowground carbon inputs, mineral association, and microbial activity are the main processes contributing to the formation of soil carbon profiles, and these processes are represented in models using the diffusion\uffe2\uff80\uff93advection\uffe2\uff80\uff93reaction paradigm. Based on simulation examples and measurements from carbon and radiocarbon profiles across biomes, we found that advective and diffusive transport may only play a secondary role in the formation of soil carbon profiles. The difference between vertical root inputs and decomposition seems to play a primary role in determining the shape of carbon change with depth. Using the transit time of carbon to assess the timescales of carbon storage of new inputs, we show that only small quantities of new carbon inputs travel through the profile and can be stabilized for time horizons longer than 50\uffe2\uff80\uff89years, implying that activities that promote CS in the subsoil must take into consideration the very small quantities that can be stabilized in the long term.Soils store large quantities of carbon in the subsoil (below 0.2\uffe2\uff80\uff89m depth) that is generally old and believed to be stabilized over centuries to millennia, which suggests that subsoil carbon sequestration (CS) can be used as a strategy for climate change mitigation. In this article, we review the main biophysical processes that contribute to carbon storage in subsoil and the main mathematical models used to represent these processes. Our guiding objective is to review whether a process understanding of soil carbon movement in the vertical profile can help us to assess carbon storage and persistence at timescales relevant for climate change mitigation. Bioturbation, liquid phase transport, belowground carbon inputs, mineral association, and microbial activity are the main processes contributing to the formation of soil carbon profiles, and these processes are represented in models using the diffusion\uffe2\uff80\uff93advection\uffe2\uff80\uff93reaction paradigm. Based on simulation examples and measurements from carbon and radiocarbon profiles across biomes, we found that advective and diffusive transport may only play a secondary role in the formation of soil carbon profiles. The difference between vertical root inputs and decomposition seems to play a primary role in determining the shape of carbon change with depth. Using the transit time of carbon to assess the timescales of carbon storage of new inputs, we show that only small quantities of new carbon inputs travel through the profile and can be stabilized for time horizons longer than 50\uffe2\uff80\uff89years, implying that activities that promote CS in the subsoil must take into consideration the very small quantities that can be stabilized in the long term.Managed grasslands have the potential to store carbon (C) and partially mitigate climate change. However, it remains difficult to predict potential C storage under a given soil or management practice. To study C storage dynamics due to long-term (1952\uffe2\uff80\uff932009) phosphorus (P) fertilizer and irrigation treatments in New Zealand grasslands, we measured radiocarbon (14C) in archived soil along with observed changes in C stocks to constrain a compartmental soil model. Productivity increases from P application and irrigation in these trials resulted in very similar C accumulation rates between 1959 and 2009. The \uffe2\uff88\uff8614C changes over the same time period were similar in plots that were both irrigated and fertilized, and only differed in a non-irrigated fertilized plot. Model results indicated that decomposition rates of fast cycling C (0.1 to 0.2\uffc2\uffa0year\uffe2\uff88\uff921) increased to nearly offset increases in inputs. With increasing P fertilization, decomposition rates also increased in the slow pool (0.005 to 0.008\uffc2\uffa0year\uffe2\uff88\uff921). Our findings show sustained, significant (i.e. greater than 4 per mille) increases in C stocks regardless of treatment or inputs. As the majority of fresh inputs remain in the soil for less than 10\uffc2\uffa0years, these long term increases reflect dynamics of the slow pool. Additionally, frequent irrigation was associated with reduced stocks and increased decomposition of fresh plant material. Rates of C gain and decay highlight trade-offs between productivity, nutrient availability, and soil C sequestration as a climate change mitigation strategy.