Fine-root mortality rates in a temperate forest: estimates using radiocarbon data and numerical modeling

Authors


Author for correspondence:
W. J. Riley
Tel: +1 510 486 5036
Email: wjriley@lbl.gov

Summary

  • • We used an inadvertent whole-ecosystem 14C label at a temperate forest in Oak Ridge, Tennessee, USA to develop a model (Radix1.0) of fine-root dynamics. Radix simulates two live-root pools, two dead-root pools, non-normally distributed root mortality turnover times, a stored carbon (C) pool, and seasonal growth and respiration patterns.
  • • We applied Radix to analyze measurements from two root size classes (< 0.5 and 0.5–2.0 mm diameter) and three soil-depth increments (O horizon, 0–15 cm and 30–60 cm).
  • • Predicted live-root turnover times were < 1 yr and ∼10 yr for short- and long-lived pools, respectively. Dead-root pools had decomposition turnover times of ∼2 yr and ∼10 yr. Realistic characterization of C flows through fine roots requires a model with two live fine-root populations, two dead fine-root pools, and root respiration. These are the first fine-root turnover time estimates that take into account respiration, storage, seasonal growth patterns, and non-normal turnover time distributions.
  • • The presence of a root population with decadal turnover times implies a lower amount of belowground net primary production used to grow fine-root tissue than is currently predicted by models with a single annual turnover pool.

Introduction

In a typical year, terrestrial plants assimilate about 20 times as much CO2 as is emitted by fossil fuel combustion (Houghton et al., 2001). Of the assimilated carbon (C), some is rapidly respired back to the atmosphere (Bowling et al., 2001), but a substantial fraction is used to build plant tissues. In forest ecosystems, the production of fine roots is an important component of the overall forest C balance. Roots supply C to microorganisms and soil organic matter (SOM) through root mortality, sloughing, support of mycorrhizal fungi, and exudates. Over time, root-derived SOM is returned to the atmosphere via mineralization by soil microorganisms.

To characterize tree growth, models need to include representations of fine-root mortality turnover times, decomposition turnover times, and turnover times associated with other belowground C-cycle processes, such as respiration and exudation. The need to separately include these processes arises because their C fluxes depend differently on environmental factors, life histories, soil properties, and nutrient conditions. These basic components of the root C cycle remain uncertain (Trumbore & Gaudinski, 2003; Johnston et al., 2004; Majdi et al., 2005) and poorly characterized in models.

Recent studies using isotopic approaches have shown that root lifespans are very heterogeneous and range from months to more than a decade (Gaudinski et al., 2001; Luo, 2003; Matamala et al., 2003; Tierney et al., 2003; Joslin et al., 2006; Keel et al., 2006). Fine roots have a positively skewed population age distribution, with young roots much more likely to die than older roots (Wells & Eissenstat, 2001; Tierney & Fahey, 2002). There is also a growing body of evidence that mortality turnover time depends on nitrogen (N) content and mycorrhizal association (Pregitzer et al., 1997; Bidartondo et al., 2001; Wells & Eissenstat, 2001; King et al., 2002; Pregitzer, 2002; Guo et al., 2004, 2008).

Most methods for calculating fine-root turnover have assumed uniform or normal, rather than positively skewed, age distributions. Tierney & Fahey (2002) showed that using a normal age distribution underestimated mean root ages in minirhizotron applications and overestimated ages in isotopic applications. Guo et al. (2007) used a statistical model of fine-root populations that included root order and mortality probability distribution (e.g., lognormal or normal) to investigate differences between minirhizotron- and 14C-based inferences of root turnover. They concluded that mortality estimates did not depend strongly on the turnover time distribution. Additionally, they concluded that the two main reasons for differences between minirhizotron and isotope-derived turnover time estimates were overemphasis of fast cycling roots by the root-number-based (minirhizotron) method, and under-emphasis of fast cycling roots by the root-mass-based (isotope) method.

Root respiration is one of the largest C fluxes through roots, and may play a large role in controlling the isotopic composition of root tissue. Nevertheless, to our knowledge it has not been explicitly included in any root turnover models that use isotopes as constraints (Caldwell & Camp, 1974; Milchunas et al., 1985; Gaudinski et al., 2001; Luo, 2003; Matamala et al., 2003; Trumbore et al., 2006), at least partly because it is difficult to measure and therefore uncertain in magnitude and temporal variability.

Our goals in this study were: to estimate mortality and decomposition turnover times of live and dead roots, respectively; to estimate C fluxes out of the dead-root pool; and to characterize the sensitivity of predicted fine-root Δ14C values to assumptions about root mortality turnover distributions, fine-root pool structure, and respiration. We developed a new model of fine-root C dynamics (Radix1.0), which accounts for: (1) short-lived and long-lived roots, each with right-skewed age populations; (2) stored-C and 14C inputs to root growth and respiration; (3) seasonal variation in root respiration and growth rates; (4) structural versus nonstructural C in long-lived fine roots; (5) two dead-root pools; and (6) uncertainty in forcing variables and model parameters. We tested Radix using published 14C data from live and dead roots from a mature deciduous forest (Joslin et al., 2006) that was labeled with 14CO2 in 1999 (Trumbore et al., 2002). The site, on the Oak Ridge Reservation (ORR), Oak Ridge, Tennessee (USA), is part of the Enriched Background Isotope Study (EBIS; Joslin et al., 2006) and provides a unique opportunity to quantify C cycling rates through mature trees on timescales ranging from months to decades.

Materials and Methods

In this section we describe the ORR site, 14C data for fine-root biomass and respiration, and Radix model structure and parameter definitions; a series of sensitivity analyses used to improve our understanding of C cycling through fine roots is given in the supporting material. The definition of ‘fine roots’ varies in the literature, but for this paper we define fine roots as those < 2 mm in diameter. We define root mortality turnover time to be the annually averaged stock of C in the root pool divided by the annual C flux leaving the pool via mortality once the system has come to a steady annual biomass cycle. Analogously, the decomposition turnover time is defined to be the annually averaged stock of C in the dead-root pool divided by the C flux leaving the pool via decomposition (at a steady annual cycle). In this paper, unless otherwise noted, ‘turnover time’ refers to the turnover time associated with mortality for live roots and decomposition for dead roots. The turnover times of live-root pools estimated here are not equivalent to their mean residence time or age because we imposed a right-skewed turnover time distribution (Wells & Eissenstat, 2001; Tierney & Fahey, 2002).

The atmosphere near ORR was highly enriched in 14CO2 sometime between 12 June and 22 August 1999, presumably from a hazardous waste incinerator near West ORR. In this analysis, we used previously published data on root biomass and 14C content from before, during, and after this period in four upland oak forest sites on and near the ORR (Joslin et al., 2006). Mean annual precipitation on ORR is 1358 mm and mean annual temperature is 14.1°C (Johnson & Van Hook, 1989). For further information on the site, 14C measurements, and estimated local atmospheric 14C content, see Supporting Information Methods S1 and Gaudinski et al. (2009).

Model description

We designed Radix to: represent processes and ecosystem characteristics important in root growth and function; interpret 14C measurements in the context of fine-root C cycling rates; and have a sufficiently general structure that the model can be applied at other sites. Radix is a departure from previous fine-root models in that it explicitly includes two live- and two dead-root pools, each with their own turnover time distributions. To estimate turnover times and C fluxes, we run the model with root data sorted into depth intervals and two size classes (diameter < 0.5 and ≥ 0.5–2). Each modelled size class therefore has fast and slow cycling roots.

The development of models (like Radix) requires a balance between the desire to include all mechanisms hypothesized to be important and restrictions based on (1) uncertain parameter characterization; (2) uncertainty in boundary and initial conditions; (3) uncertainty in assumed system structure; (4) limited availability of measurements to test model predictions; and (5) computational resources. We attempted to balance these factors in the model development; however, we expect the model structure and parameterizations will improve as more information becomes available.

Model structure

Radix represents C flows through fine roots with the following pools (Fig. 1): storage (S), live roots with fast turnover (L1), live roots with slow turnover (divided into nonstructural (L2) and structural (L3) components), dead roots from the fast-turnover pool and nonstructural C in dead roots from the slow-turnover pool (D1), and structural C from dead roots with slow turnover (D2). A fraction (fs) of recently fixed photosynthate is stored while the remainder (1 −fs) is used immediately by roots (Fig. 1). The model conceptualizes storage as well-mixed carbohydrate pools of equal turnover times in one or more locations within the tree. While stored C is used in both aboveground and belowground growth, we assumed that the isotopic composition of the storage pool used to grow roots is controlled by C transfers to roots. However, in this forest the distinction is not critical because the turnover time of storage used to grow leaf buds, expanding leaves, and fine roots is similar (0.7 yr) (Gaudinski et al., 2009). Carbon from recent photosynthate and storage is directed to live roots using the parameters f1 and f2 (Fig. 1); the effect of uncertainty in these parameters is explored in the sensitivity analyses described in the Results section.

Figure 1.

 Schematic structure of the Radix root model. Carbon (C) enters the root system and is allocated to storage (S) and live-root pools L1 (short-lived) and L2 (total nonstructural C (TNC) in longer lived roots). L2 and L3 (structural C) make up a single root but are considered as separate pools to distinguish nonstructural and structural C. C moves between L2 and L3 with turnover time τts and all live pools respire and experience mortality. C flows into the fast-cycling dead roots from the L1 and L2 pools and into the slow-cycling dead roots from the L3 pool.

In the model, C can exit the live-root pools via mortality, transfer to another live pool, and respiration. Carbon can exit the dead root pools via decomposition. Because they are extremely difficult to quantify (Hogberg & Read, 2006), we did not explicitly represent fluxes associated with exudation or mycorrhizal fungi. Mortality and decomposition losses are characterized in the model using turnover times. We assumed that turnover times for each of the pools (τL1, τL23, τD1, and τD2) are lognormally distributed (Tierney & Fahey, 2002) with geometric standard deviations (GSDs) of 2, thereby generating a right-skewed distribution. The turnover time distributions are limited to be within a factor of 3 of the geometric mean (GM). We imposed uncertainty on the mean and GSD of this distribution. As described in the sensitivity analysis, we also explored the effects of assuming normal turnover time distributions. Respiratory fluxes are notated with ‘R’ (g C m−2 s−1). Live roots grow from stored C, newly fixed C, or a mixture, depending on the season.

L 1 can lose C via respiration (R1) and mortality (τL1). Because these roots are short-lived, we assumed that there is no significant isotopic difference among nonstructural, structural, and respired C. Pools L2 and L3 collectively comprise living long-lived fine roots. L2 represents total nonstructural carbohydrates (TNC; starch and sugar) while L3 represents the structural (e.g., cellulose) portion. L2 receives stored and new photosynthate and loses C via respiration (R23), transfer of carbon to L3 (characterized by the turnover time τts), and mortality (τL23). We chose a value of τts (0.5 yr) that produced average annual L2 values that were within the range of published values for nonstructural carbohydrate concentrations for white oak roots (< 10 mm in diameter) growing in the Walker Branch Watershed (McLaughlin et al., 1980). L3 receives C from L2, loses C via root mortality (τL23), and C associated with L3 respiration is removed from L2. The mortality turnover times for L2 and L3 are equivalent because when a root dies both TNC and structural pools are simultaneously lost. Pool D1, comprised of the fast cycling component of dead roots, receives inputs from L1 and TNC from L2 and loses C via decomposition. Pool D2 receives only structural C (from L3) and loses C via decomposition.

Model parameter determination

Values for model parameters (i.e., τL1, τL23, τD1, τD2) were estimated using a minimization of the squared differences between model predictions and observations, weighted by measurement uncertainty (Press et al., 1989). The model was run from 1905 so that the inter-annual C pool size variations are steady by the time the elevated atmospheric 14C event occurs in 1999.

To estimate the turnover times (τL1, τL23, τD1, and τD2) we used the live- and dead-root Δ14C measurements from East and West ORR. The parameter fitting procedure sampled the following ranges of mortality turnover times: τL1 = [0.1, 4] yr; τL23 = [4, 19] yr; τD1 = [0.1, 4] yr; and τD2 = [4, 17] yr; and compared predictions averaged over a 30-d period surrounding the measurement time.

Storage

The size of the modeled storage pool is controlled by the fraction (fs) of belowground gross primary productivity (BGPP) input to the storage pool during May−October and losses throughout the entire year. Values for fs (0.55) and τs (0.7 yr) were estimated using measurements of new roots grown on the East ORR (Gaudinski et al., 2009).

Root respiration

Fine-root respiration comes predominantly from recently assimilated C (Horwath et al., 1994; Hogberg et al., 2001; Keel et al., 2006). Further, measured Δ14C of root respiration at ORR had values similar to atmospheric Δ14C (Trumbore et al., 2002). Therefore, in Radix, respiration for short-lived roots (R1) comes from the L1 pool and respiration from long-lived roots (R23) comes only from nonstructural C in the L2 pool. The L1 and L2 pools are supplied by recently fixed and stored C (Fig. 1), with the relative amounts depending on the season. Evidence for autotrophic respiration containing some stored C, particularly in winter, has been found in some temperate and boreal forests (Gaudinski et al., 2000; Schuur & Trumbore, 2006; Carbone et al., 2007; Czimczik & Trumbore, 2007). As the mean age of stored C is young in this study (∼0.7 yr; Gaudinski et al., 2009), predicted Δ14C values of L1 and L2 are always relatively close to the atmospheric value.

Total respiratory rates were estimated from field measurements of rhizosphere respiration (for roots < 1 mm (and primarily < 0.5 mm)) from four similar forests studied by Burton & Pregitzer (2002). In that study the roots were brushed, but not washed, so that some of the measured CO2 emission may have included heterotrophic sources using labile C on the root surface. We used the average specific respiration for a mixed Quercus forest in Georgia, a Quercus–Carya and mixed hardwood forest in North Carolina, an Acer saccharum (sugar maple) forest in Michigan (adjusted to 18°C; 0.05 µg C g−1 s−1), and an average temperature sensitivity factor (Q10) of 2.7 for the same four sites (Burton et al., 2002) to calculate specific respiration rates for the four seasonal periods: November−March, April, May−July, and August−October. Mean 2000 and 2001 TDE soil temperatures at 10 cm depth for the four time periods were used for the Q10 conversions (14.0, 19.1, 20.7, and 8.6°C; Paul Hanson, unpublished data). With this method, estimated respiration rates for the four periods were 0.020, 0.033, 0.055, and 0.064 µg C g−1 root s−1, respectively.

We note that these specific respiration rates are higher than measured rates for roots in some other forests (e.g., Majdi & Andersson, 2005; E. A. Davidson & K. Savage, unpublished data for Harvard Forest). However, root respiration rates varied by over a factor of 3 as a function of N content and diameter (Pregitzer et al., 1998) in two sugar maple forests in Michigan. Use of these unscaled specific respiration rates led to unrealistically high predictions of R23 (i.e., larger than the proportion of C entering L2). Therefore, we decreased the respiration rate for the long-lived roots (R23) by a factor of 3 after consultation with A. J. Burton (pers. comm.) and comparison with other studies. We did not change R1 because the L1 root population is more representative of the types of roots measured by Burton & Pregitzer (2002). Finally, we investigated uncertainty and sensitivity of our predictions to respiration by varying the respiration rate via an adjustable scale factor (fr).

Belowground biomass and productivity

Measured biomass values by live and dead status (live and dead categories were based on tensile strength, integrity, and color of the vascular tissue; Vogt & Persson, 1991), diameter size class, and depth interval are shown in Table 2 (see also Joslin et al., 2006). Monthly, total soil column BGPP has been estimated for these sites (Hanson et al., 2003b), but we were unable to directly apply these values because there was no method to partition BGPP by depth without first assuming turnover times (we used the Hanson et al. (2003b) estimate to partition BGPP over the year, as described two paragraphs hence). We therefore estimated BGPP by depth using measured live biomass and the best-fit turnover times as constraints.

Table 2.  Predicted belowground gross primary productivity (BGPP), predicted belowground net primary productivity of new fine-root biomass (BNPP*), and measured biomass for two root size classes at three depth intervals
Diameter (mm)Predicted BGPP (g C m−2 yr−1)Predicted BNPP* (g C m−2 yr−1)Measured live biomass (g C m−2)Measured dead biomass (g C m−2)
< 0.50.5–2< 0.50.5–2< 0.50.5–2< 0.50.5–2
  1. Measured values are from data collected at all four enriched background isotope study (EBIS) sites over 3 yr of sampling.

Horizon
O20 12 7 510 ± 16 ± 14 ± 12 ± 0
0–15 cm66134213534 ± 371 ± 544 ± 443 ± 5
15–30 cm    13 ± 316 ± 218 ± 413 ± 2
30–60 cm14 38 5127 ± 020 ± 112 ± 217 ± 3
60–90 cm    8 ± 210 ± 411 ± 410 ± 3
0–90 cm    71 ± 6123 ± 489 ± 785 ± 13
Total        

We estimated annual BNPP* (belowground net primary productivity of new fine-root biomass) by subtracting predicted total annual respiration from predicted BGPP. We used the term BNPP* to distinguish it from total BNPP, which would also include production of exudates, fine-root hairs, and C export to mycorrhizal fungi (Clark et al., 2001; Hanson et al., 2003a). If root biomass is in an annual steady cycle (i.e., does not change year to year), then annual production is equal to annual mortality, and BNPP* is equivalent to the annual fine-root mortality-derived C inputs to soil. Estimates of BGPP and BNPP* were made for each fine-root size class and depth interval.

As long as a pool is not completely depleted of C, the predicted 14C content of the roots does not depend on BGPP. Root 14C content does, however, depend on seasonal BGPP partitioning because of its dependence on the timing of growth relative to the changing atmospheric 14C content (in our case, primarily associated with the early summer 1999 14C pulse). We assumed that BGPP was zero between November and April, when the leaves have senesced or dropped. At Walker Branch, leaf expansion occurs between 10 April (5% completion) and 11 May (95% completion; Joslin et al., 2001; Hanson et al., 2003c). 14CO2 labeling of mature white oaks (Quercus alba) on Walker Branch showed that leaves translocate C out of leaves in May through October (Edwards et al., 1989). Therefore, we assumed that photosynthate produced in April is used for aboveground growth. The periods May−July and August−October receive 72 and 28% of annual BGPP, respectively, based on minirhizotron observations at ORR of < 2 mm diameter root-length growth during November−March (5%), April (10%), May−July (65%), and August−October (20%) (Joslin et al., 2001). We divided the 15% of observed root-length growth occurring in November through April evenly among the 6 months between May and October. The resulting BGPP partitioning for May through October was: 0.24, 0.24, 0.24, 0.09, 0.09, and 0.09.

Uncertainty analysis

Each parameter and forcing variable used in the model is uncertain to some extent. We applied a Monte Carlo technique (Press et al., 1989) to characterize the effect of these uncertainties on model predictions. For this analysis we assumed limited normal distributions for the following parameters: fs, fr, τts, f1, and f2. The distributions were limited in that we enforced a limit of two standard deviations (SDs), thereby ensuring that unrealistic parameter values were excluded. We did not include the effects of parameter covariation in this analysis. For the model turnover times (τL1, τL23, τD1, τD2), the uncertainty in GM was normally distributed. For values that vary seasonally (e.g., BGPP), the annual value of the parameter changed between Monte Carlo simulations, but the relative monthly proportion did not. Because of the large uncertainty in atmospheric Δ14C (ΔARPC), a scaling factor with a GM of 1 and GSD of 1.3 (East ORR) or 1.7 (West ORR) was applied to ΔARPC in each Monte Carlo simulation (Gaudinski et al., 2009).

The Monte Carlo technique involved performing 300 simulations, each with a different set of parameters and boundary conditions based on the probability distributions described above. Mean and uncertainty ranges for the predicted quantity of interest (e.g., the Δ14C value of pool L1) were then computed from the ensemble simulation results.

Sensitivity analyses

Once the best-fit values for τL1, τL23, τD1, and τD2 had been determined, we investigated model sensitivity to various model structures and parameters. We performed a series of six analyses (focusing on roots < 0.5 mm in the 0–15 cm depth interval) on the sensitivity of model predictions to: (1) live fine-root mortality turnover times (τL1 and τL23); (2) the assumption of lognormally distributed turnover times; (3) the use of a simpler, one-pool model construct; (4) the distinction between structural and nonstructural C in live-root pools; (5) variations in seven critical model parameters; and (6) separating the East and West ORR 14C observations (see Methods S1 for a more detailed description of these analyses).

Results

Predicted turnover times

Predicted and measured fine root Δ14C values from East and West ORR were well above the atmospheric background during the entire 3-yr sampling period (2001–2003), demonstrating the substantial influence of the local 14C release (Figs 2, 3). Best-fit turnover times for the three depth intervals (O horizon, 0–15 cm, and 30–60 cm), and two size classes of roots (< 0.5 and 0.5–2.0 mm diameter) were 0.1–0.3 yr for the fast-turnover root pool (τL1), 7–9 yr for the slower turnover live pool (τL23), 2 yr for the fast-turnover dead-root pool (τD1), and 9–10 yr for the slow-turnover dead-root pool (τD2) (Table 1). An example, best-fit turnover time distribution for τL1 is shown in Fig. S1. We predicted no substantial differences in turnover time with depth or size class.

Figure 2.

 Measured and predicted East Oak Ridge Reservation (ORR) Δ14C values for roots from 0–15 cm depth and of < 0.5 mm diameter in (a) live and (b) dead roots.

Figure 3.

 Measured and predicted West Oak Ridge Reservation (ORR) Δ14C values for roots from 0–15 cm depth and of < 0.5 mm diameter in (a) live and (b) dead roots.

Table 1.  Predicted turnover times for two root size classes at three depth intervals using 3 yr of root 14C data from the Oak Ridge Reservation sites
Diameter (mm) τL1 τL23 τD1 τD2
< 0.50.5–2< 0.50.5–2< 0.50.5–2< 0.50.5–2
Horizon
O0.20.17822109
0–15 cm0.20.38922109
30–60 cm0.20.29922 99

Comparison of measured and predicted Δ14C values

Predicted mean Δ14C values of live roots < 0.5 mm from the 0–15 cm depth interval were slightly higher than those measured on the East ORR (Fig. 2a) and somewhat lower than those measured on the West ORR (Fig. 3a). All the live-root measurements fell within the 1 SD uncertainty bounds of the predicted Δ14C values. The peak in live-root Δ14C at the beginning of 2000, and all subsequent peaks, resulted from seasonal increases in use of storage C.

Predicted dead-root Δ14C values on the East ORR matched measurements relatively well (Fig. 2b). Most predicted West ORR (Fig. 3b) dead-root Δ14C values were lower than measurements, although again within the 1 SD uncertainty bounds. Differences between measured and predicted Δ14C values were similar to those shown in Fig. 3 for the other depth intervals and size classes.

The uncertainty ranges in predicted Δ14C values for the live pools were large and dominated by uncertainty in local atmospheric 14CO2ARPC). The uncertainty bounds were largest when ΔARPC was largest, and declined after the peak values in 1999. To illustrate the effect of uncertainty in ΔARPC on live-pool Δ14C values, we performed simulations that eliminated uncertainty in ΔARPC. In these simulations, the 1 SD uncertainty bounds in live-root 14C content were reduced by about one-third compared with simulations including ΔARPC uncertainty.

Belowground biomass and productivity

We predicted large seasonal variability in the biomass of short-lived roots (L1): a factor of about 3 between minimum (early spring) and maximum (mid-summer) values (Fig. S2a). Because L2 + L3 are long-lived, the overall L2 + L3 biomass was less variable than L1 biomass (Fig. S2a). Predicted variation in root biomass between summer and winter was similar to that observed in other temperate hardwood forests for which monthly or bimonthly sampling of live- and dead-root biomass has been performed (McClaugherty et al., 1982; Aber et al., 1985).

For the combined O horizon, 0–15 cm, and 30–60 cm depth intervals, 35 and 65% of predicted BGPP were associated with roots < 0.5 and 0.5–2 mm, respectively (Table 2). Total mortality-derived C input to soils (BNPP*) for the three depth intervals combined was 30% of BGPP, with 40 and 60% of that derived from roots < 0.5 and 0.5–2 mm, respectively. To estimate BGPP and BNPP* in the 15–30 cm and 60–90 cm intervals (which were not simulated because 14C data were unavailable, although biomass data were available), we assumed that the ratio of production to biomass was the same in these depth intervals as in the 30–60 cm interval. Including the values estimated in this way, BGPP and BNPP* to 90 cm depth were 360 and 110 g C m−2 yr−1, respectively.

Sensitivity analyses

The sensitivity analyses were designed to probe important aspects of the model structure and parameterization, to aid in understanding which system components most strongly affect fine-root C exchanges, and inform future experimental and observational work. Results for the first sensitivity analysis (varying the live fine-root short- and long-lived mortality turnover times; τL1 and τL23) showed that, as τL1 increased, the L1 pool size increased and its response rate to the input 14C pulse decreased, as did the rate of subsequent 14C loss (Fig. 4a). Predicted Δ14C values for L1 using τL1 = 0.2 and 2 yr differed by > 100‰ and c. 50‰ immediately following the atmospheric pulse and 1 yr later, respectively. The effect of these differences in τL1 on the total live fine-root pool (L1 + L2 + L3) Δ14C value was c. 50‰ in spring 2000, very small in spring 2001, and c. 20‰ thereafter (Fig. 4c).

Figure 4.

 Sensitivity analysis of live root Δ14C values to mortality turnover times of short- and long-lived roots (τL1 and τL23). Nominal values for τL1 and τL23 are 0.2 and 8 yr. (a) L1Δ14C value for the nominal case and τL1 = 1 and 2 yr. (b) L2 + L3Δ14C values for the nominal case and τL23 = 4 and 16 yr. (c) L1 + L2 + L3Δ14C value for the nominal case and τL1 = 1 and 2 yr. (d) L1 + L2 + L3Δ14C value for the nominal case and τL23 = 4 and 16 yr.

The response of the long-lived fine-root (L2 + L3) pool to changing τL23 was complicated by the fact that the predicted Δ14C value of the pool at the beginning of the atmospheric pulse depends on τL23 (Fig. 4b). This difference is not seen for the effect of τL1 on the Δ14C of L1 because τL1 (∼0.2 yr) is small relative to the characteristic time of variability in background atmospheric Δ14C. Analogous to the response of L1 to changes in τL1, the Δ14C of the long-lived pool responded most rapidly to the pulse when τL23 was smallest (∼100‰ and ∼40‰ changes when τL23 = 4 and 16 yr, respectively), reflecting the relative rates at which the atmospheric 14C pulse was assimilated into L2 + L3. The 14C content of the entire live-root pool (L1 + L2 + L3) was sensitive to variations in both τL1 and τL23 (Fig. 4c,d).

The second sensitivity analysis assumed normal distributions for the mortality turnover times (τL1 and τL23) instead of lognormal distributions. The effect on the mean predictions over time was between 5 and 20‰, with the normal turnover time distributions resulting in more enriched values than the lognormal distributions. This result is consistent with the lognormal distributions resulting in higher flux-weighted turnover times than the normal distributions when using the same values for GMs and means. Therefore, the live pools with the lognormal distribution acquired relatively less of the 14C pulse in 1999, but had a relatively smaller decline over time. Given the uncertainty ranges in the data, these differences were not significant enough to distinguish which turnover time distribution type was more appropriate for this system.

Our third sensitivity analysis tested whether using only one live pool, one dead pool, and a storage pool (compared to the nominal structure of two live and two dead populations with different turnover times) changed the model's ability to match the observations. For this scenario, the best-fit turnover times were 2 and 1 yr for the live and dead pools, respectively. The fit to the data was substantially worse for both live and dead roots (Fig. 5), and the amounts of BGPP and BNPP* both increased substantially (factors of 4 and 10 for BGPP and BNPP*, respectively). The fits to the biomass data, however, were about the same for both cases. This sensitivity analysis demonstrates that conceptualizing live and dead fine roots as single pools can lead to substantial errors in C transfers from roots to soil.

Figure 5.

 East Oak Ridge Reservation (ORR) measured Δ14C for (a) live and (b) dead roots from 0–15 cm depth and of < 0.5 mm diameter and model predictions using the nominal model construct (three live- and two dead-root pools) and two simplified model constructs.

The fourth sensitivity analysis investigated the need for both the L2 and L3 pools (in addition to L1). In the scenario that excluded L3, the best-fit turnover times changed only for τL23, which was reduced from 8 to 6 yr. Additionally, there was a substantially worse match with the observations (Fig. 5). In the nominal scenario (which included L2 and L3), most of the 14C variability in the long-lived roots occurred in L2, which made up a relatively smaller fraction of the biomass. Therefore, the total live pool 14C content in the scenario that did not include L3 responded more strongly to the atmospheric 14C pulse than in the case where L3 was included. This sensitivity analysis demonstrates that, for pulse label experiments, separating TNC and structural C in the long-lived roots is critical for accurate prediction of fine-root 14C content.

As a general sensitivity analysis for eight important model parameters, we imposed variations of ± 50% on the parameters and evaluated changes during 1998, the year before the ORR 14C pulse (Table 3). The largest effects on live biomass were from perturbations to respiration (fr) and the partitioning of C leaving the storage pool (f1). The largest effects on dead biomass were from fr, the slow pool mortality turnover time (τL23), and the decomposition turnover times of the two dead pools (τD1 and τD2). Overall, the largest changes in live plus dead fine-root biomass occurred with perturbations in the magnitude of respiration (fr). The largest effects on live Δ14C values were from τL23, and the largest effects on dead Δ14C values were from fr, f1, τL23, τD1, and τD2.

Table 3.  Sensitivity of annual averaged live (L1 + L2 + L3) and dead (D1 + D2) biomass and Δ14C values (before the large 1999 pulse) to 50% increases and decreases in model parameters
Parameter L 1 + L2 + L3 (%) D 1 + D2 (%)Δ14C of L1 + L2 + L3 (‰)Δ14C of D1 + D2 (‰)
ReduceIncreaseReduceIncreaseReduceIncreaseReduceIncrease
  1. Values shown are either per cent or per mil (‰) changes from the nominal case. Parameters causing a > 20% change in biomass or a > 6‰ (the analytical error in the 14C measurement) change in Δ14C value are shown in bold. The largest effects on live biomass were from fr; those on dead biomass were from fr, τL23, τD1, and τD2; those on live Δ14C values were from τL23, and those on dead Δ14C values were from fr, f1, τD1, and τD2.

f r  72 −28  51 −21   2 1  10  −5
f 1  20 −20   0  2  0−4  18 −17
τts  4 −4  5 −4  1−1  3 −1
τ1 −4  0 11  −7  1−1 −5  4
τL23−16  8  32 −14 −30 13   2 −10
τD1  0  0 −23  25   0 0  16  −7
τD2  0  0 −26  26   0 0 −23   6

For our sixth sensitivity analysis we tested whether different turnover times would be predicted if separate East and West ORR analyses were performed instead of the nominal analysis, which combined observations from both sides of ORR into a single data set. Both East and West ORR best-fit turnover times were within the ranges shown in Table 1. This result implies that the predicted mortality turnover times were robust for two very different atmospheric 14C pulses and that the forests behaved similarly on the two different ORR ridges.

Discussion

We used an inadvertent whole-ecosystem 14C label at a temperate forest in Oak Ridge, Tennessee to develop, test, and apply a model (Radix1.0) of fine-root dynamics. The model simulates two live-root populations, two dead-root pools, non-normally distributed root mortality turnover times, a stored C pool, and seasonal growth and respiration patterns. After using the model to estimate turnover times for two size classes and three depths, we performed sensitivity analyses to elucidate mechanisms responsible for C exchanges through the fine-root system.

While root lifetimes undoubtedly span a continuum, we found that fine roots were well described as comprising a short-lived and a long-lived population with turnover times at ORR of < 1 yr and ∼10 yr, respectively (Fig. 5). Our results also indicated that it is important to distinguish structural from nonstructural components. Without the physiologically realistic separation of nonstructural and structural C in the long-lived root pool, the Δ14C value of root respiration is significantly different from that of atmospheric C and forces predicted root Δ14C values to be overly enriched following the atmospheric 14C pulse.

Joslin et al. (2006) reported that roots < 0.5 mm in diameter had more rapid turnover than roots 0.5–2 mm, and that roots in the O horizon had more rapid turnover than deeper roots. We did not predict similar trends in this study, although uncertainty in predicted turnover times might have obscured such differences. Although this study used measurements of the 14C content of bulk roots, better characterization of live-root turnover times could be achieved by measuring the 14C content of root cellulose and TNC separately.

The Radix model structure and predictions are consistent with a growing body of literature arguing that roots vary widely in probability of mortality and that including this variation is necessary to model root dynamics accurately (Wells & Eissenstat, 2001; Pregitzer, 2002; Tierney & Fahey, 2002; Trumbore & Gaudinski, 2003; Majdi et al., 2005; Joslin et al., 2006). Our results also support the idea that the large differences in fine-root mortality turnover times derived from minirhizotrons (3 months to < 1 yr; Hendrick & Pregitzer, 1992; Jackson et al., 1997; Fahey et al., 1999) versus isotopic techniques (1.2–18 yr; Gaudinski et al., 2001; Matamala et al., 2003; Keel et al., 2006) occur because these approaches are sensitive to different ends of the fine-root mortality turnover time spectrum. In other words, the minirhizotron results are strongly influenced by the short-lived roots, while results from the isotopic approaches are influenced predominantly by the long-lived roots, which have more biomass.

Uncertainty in our turnover time predictions was dominated by uncertainty in local atmospheric 14CO2 and the lack of live-root 14C measurements immediately following the 14C pulse enrichment. In particular, constraints on the turnover time of the short live-root pool (τL1) would have improved markedly if we had fine-root Δ14C measurements in spring 2000, because root 14C content during this period strongly reflects variations in the fast-turnover pool. These observations highlight the fact that, although isotopes are useful tracers of ecosystem C fluxes, frequent sampling in the months and years immediately after any pulse labeling is required to obtain the most useful information.

Assuming that the two depth intervals for which we did not have 14C data (15–30 and 60–90 cm) had BGPP and BNPP* that scaled with live biomass, we estimated column BGPP and BNPP* to be 360 and 110 g C m−2 yr−1, respectively, for roots < 2 mm in diameter. Previous BGPP estimates (Curtis et al., 2002; Hanson et al., 2003b; Joslin & Wolfe, 2003) from sites on the ORR using a C budget approach were c. 30–70% larger than our estimate (478–619 g C m−2 yr−1). This previous carbon budget approach may overestimate BGPP because it used: the fine-root production estimate of Joslin & Wolfe (2003); measurements of total soil respiration which included respiration by larger roots; and assumptions about the mix of heterotrophic versus autotrophic respiration. In contrast, our measured live-root biomass may be an underestimate given our root sorting protocol (see Supporting Information), which could lead to an underestimate of BGPP. Improved measurements of biomass, autotrophic respiration, and exudation, coupled with root models such as Radix, could help reduce uncertainty in predicted BGPP.

Estimates of annual fine-root production with root length observations (via minirhizotrons), measured biomass (from soil cores), and an implicit one-pool model for live roots on the ORR (110–140 g C m−2 yr−1; Joslin & Wolfe (2003) and reported in Curtis et al. (2002)) were closer to our estimate (110 g C m−2 yr−1). We expected our BNPP* estimate to be lower than estimates based on a one-pool model because we accounted for a large portion of fine-root biomass with decadal turnover times. Our third sensitivity analysis illustrated that using one live-root pool with a fast turnover time can accurately predict the biomass but will overestimate fine-root production. We predict that root production estimates will in general decrease as models begin to account for short- and long-lived fine roots.

Root decomposition rates are a critical component of ecosystem C modeling. We predicted that two pools are required to simulate dead root decomposition: a rapidly decomposing pool with turnover time ∼2 yr and a slower pool with turnover time ∼10 yr (Fig. 5). This conclusion is consistent with previous litter bag and litter recovery studies (Bird & Torn, 2006). Similar to the live-root pools, we did not predict a consistent trend of turnover time with either depth or size class. Dead-root biomass was approximately equally divided between D1 and D2, yet their turnover times were substantially different (Table 1, SM Fig. 2). Therefore, we predict that more of the organic C entering soil organic matter is coming from the shorter decomposition turnover time pool.

Effect of root respiration on predicted ecosystem parameters

Previous studies have ignored the effects of respiration when using isotopic measurements to infer C turnover times (e.g., Luo, 2003; Gaudinski et al., 2001). To illustrate the effect of respiration on transient Δ14C values (and therefore on inferred turnover times) we performed two simulations for the East ORR using the best-fit parameters (Fig. 6). The simulations differ only in that one has respiration from all the live pools forced to zero. For the live-root pools, ignoring respiration led to a substantially lower predicted peak Δ14C value (by ∼100‰), higher subsequent values, and much lower seasonal variability. For the dead-root pools, ignoring respiration led to more enriched predictions after about 1 yr following the pulse.

Figure 6.

 Effect of ignoring respiratory CO2 fluxes on the Δ14C value of (a) live and (b) dead fine roots for Oak Ridge Reservation (ORR) following the 1999 pulse. Solid line, with respiration; dashed line, without respiration. Parameters used are the same as the nominal case discussed in the text. Predicted Δ14C values of live-root pools after mid-2000 are larger and have smaller seasonal cycles in the absence of respiration. Dead-root pools in the absence of respiration also have larger Δ14C values.

The effects of respiration can be important for studies using 14C even in the absence of a large 14C pulse like that at ORR. To illustrate these effects, we performed two simulations (one with and one without respiration) using the background atmospheric 14C record (i.e., the ‘bomb spike’) (Fig. S3). A more pronounced seasonal cycle in live-root 14C content is predicted when respiration is included. In this sensitivity analysis, ignoring respiration led to differences in Δ14C values of the total live- and dead-root pools of c. 20 and 40‰ in 2000, respectively, which would affect a 14C-derived mortality turnover time by ∼3 yr for live roots and ∼7 yr for dead roots. These analyses demonstrate that ignoring respiration when using an isotopic label to trace C exchanges to the root system can lead to errors in estimated mortality turnover times. The errors will be larger for 14CO2 pulse labeling experiments but are potentially significant when using more gradual changes in input 14C values, such as the bomb spike.

Implications

The results of this study have important implications for ecosystem models that include C transfers through tree root systems. A large portion of live fine roots live much longer (∼10 yr) than previous approaches (i.e., minirhizotron) have indicated. We demonstrated that the use of two pools to represent live roots and two pools to represent dead roots, and inclusion of root respiration are critical for accurate characterization of fine-root C fluxes. The typical ecosystem model assumption that live roots turn over annually will lead to large over-predictions of root inputs to soil organic matter. Even a model with an accurate flux-weighted turnover time, but that still treats roots as a single pool, will predict very different responses to changes (e.g., in net primary production) than would a model with two pools with distinct turnover times.

Our results highlight the need for research to understand the complexities of fine-root dynamics, including: the controls on the proportion of roots with different lifetimes; the plasticity of root growth and mortality as a function of species and environmental conditions; and the magnitude and variability of autotrophic root respiration and heterotrophic respiration of recently fixed, root-derived C. Simplifications to the Radix model structure should be investigated, including omitting seasonal variability in respiration and BGPP if the model is being used in scenarios not including an isotopic pulse label.

Acknowledgements

We thank Don Todd and Heather Cooley for assistance in the field and Jessica Westbrook, Shuhui Zheng, Deborah Williard, and John Southon for help in the lab. Funding for the EBIS project was provided by the Director, Office of Science, Office of Biological and Environmental Research, Climate Change Research Division, of the US Department of Energy under Contract No. DE-AC02-05CH11231, as a part of the Terrestrial Carbon Processes (TCP) Program. EBIS project participants appreciate access to and use of Tennessee Valley Authority (TVA) land on Chestnut Ridge near the Oak Ridge Reservation allowed under Contract No. 105906 between TVA and the Oak Ridge National Laboratory.

Ancillary