A comparative study of dissolved organic carbon transport and stabilization in California forest and grassland soils

Abstract

For soil carbon to be effectively sequestered beyond a timescale of a few decades, this carbon must become incorporated into passive reservoirs or greater depths, yet the actual mechanisms by which this occurs is at best poorly known. In this study, we quantified the magnitude of dissolved organic carbon (DOC) leaching and subsequent retention in soils of a coniferous forest and a coastal prairie ecosystem. Despite small annual losses of DOC relative to respiratory losses, DOC leaching plays a significant role in transporting C from surface horizons and stabilizing it within the mineral soil. We found that DOC movement into the mineral soil constitutes 22% of the annual C inputs below 40 cm in a coniferous forest, whereas only 2% of the C inputs below 20 cm in a prairie soil could be accounted for by this process. In line with these C input estimates, we calculated advective transport velocities of 1.05 and 0.45 mm year⁻¹ for the forested and prairie sites, respectively. Radiocarbon measurements of field-collected DOC interpreted with a basic transport-turnover model indicated that DOC which was transported and subsequently absorbed had a mean residence time of 90–150 years. Given these residence times, the process of DOC movement and retention is responsible for 20% of the total mineral soil C stock to 1 m in the forest soil and 9% in the prairie soil. These results provide quantitative data confirming differences in C cycles in forests and grasslands, and suggest the need for incorporating a better mechanistic understanding of soil C transport, storage and turnover processes into both local and regional C cycle models.

Appendix: soil water flux models

Water balance model

A 6-layer soil water balance model, similar to models employed by Lohse and Matson (2005) and Spittlehouse and Black (1981), was solved for drainage (D, mm h⁻¹):

$$ D_{t,z} = D_{t,z - 1} - ET_{t,z} - \Delta S_{t,z} , $$

where ET is evapotranspiration (mm h⁻¹), ΔS is change in storage over time (mm h⁻¹), t = time (h) and z = model layer (D _z=0 is rainfall). Hourly rainfall or throughfall (for forested site) was recorded with a series of four recording tipping buckets (Spectrum Technologies, Plainfield, IL) at each site. We used the hourly ET record scaled to site vegetation from the nearest California Irrigation Management Information System (CIMIS 2007) station which uses a modified version of the Penman-Monteith equation (Allen et al. 1998). ET was divided among soil layers using an exponential drop with depth. Storage capacity or the maximum water holding capacity (WHC, mm) for each layer was calculated as the product of porosity (mm³ mm⁻³) and layer thickness (mm). We adopted the following simplifying rules: (1) water only drains down; and (2) there is no drainage from a layer until θ _z  > WHC _z and then excess water drains into the next layer. The model was run using the hourly rainfall and ET records from the first rain of the each rainy season until drainage ceased following the last storm of the season.

Soil hydrology model

We also utilized soil water content and tension measurements, and applied Richard’s equation to calculate vertical fluxes (Cassel and Klute 1986; Harr 1977):

$$ J_w = K(h)\frac{{\partial H}}{{\partial z}}, $$

where J _w is water flux (mm h⁻¹), K(h) is the unsaturated hydraulic conductivity (mm h⁻¹) as a function of pressure head (h), and dH/dz is the change in total head (gravity + pressure head) over the measurement interval. K(h) was estimated from the SPAW hydrologic model (Saxton and Rawls 2006; Saxton et al. 1986) using the volumetric water content (θ) record and measured soil physical properties, and dH/dz was calculated directly from soil water tension sensor data (Spectrum Watermark sensors, Spectrum Technologies, Plainfield, IL).

Goodness of fit was assessed by calculating the root mean square error (RMSE) and bias between the two models:

$$ RMSE = \sqrt {\frac{1}{n}\sum\limits_{i = 1}^n {d_1^2 } } $$

and

$$ bias = \frac{1}{n}\sum\limits_{i = 1}^n {d_i } , $$

where d _i is the difference between model outputs for time interval i.

Model evaluation

Overall, we found a very good agreement between seasonal totals for the two methods. At Tennessee Valley, the water balance model estimated 870 and 590 mm of drainage at 10 and 50 cm, respectively, while the hydrometric model estimated 930 and 640 mm at the same depths. At Caspar Creek, for the period of overlapping measurements drainage at 20 cm was estimated to be 660 and 700 mm by the water balance and hydrometric models, respectively.

The calculated daily water fluxes were a bit more variable between methods and between sites (Fig. 9). For depths where we can compare the two methods, the mean relative standard deviation of daily flux at Tennessee Valley is 12.4 ± 5.5% and at Caspar Creek, the mean RSD is 8.7 ± 5.7%. At Caspar Creek, where soils saturate and drain very rapidly, the two methods agreed very well (at 20 cm, RMSE = 5.02 and bias = −0.38 (towards hydrometric model)). However, at Tennessee Valley, the water balance approach was likely moving water through the soils at too rapid of a rate (RMSE = 11.75 and 9.35, and bias = −0.26 and −0.20 (towards hydrometric model), for 10 and 50 cm depths, respectively). The hydrometric results here (Fig. 9b) show slower but more sustained drainage pattern which was in line with field data and observations that significant saturated flow continued for a few days following large storms. This pattern was especially pronounced at greater depths (data not shown).

Fig. 9

Modeled vertical soil water fluxes at (a) 20 cm at Caspar Creek and (b) 10 cm at Tennessee Valley. Hydrometric data is missing prior to December 29, 2004 for Caspar Creek