Comparison of Observed and Modeled Ae Estimates

Good agreement between measured and modeled Ae values is found for the boreal zone, the humid tropics as well as for ecosystems in the Southern Hemisphere (Fig. 4). However, measured values for Ae are higher than modeled estimates in the temperate latitudes. In the mid-latitudes, there are several reasons modeled results might differ from observations by up to 5%o. We will discuss several potential explanations in the paragraphs below. The main issue is probably a result of measurement sampling strategy (see below) and the wide integration over bioclimatic space that the model makes in a 0.5° latitudinal band.

The observed global mean S,3CER ratios and Ae estimates were surprisingly similar, independent whether free tropospheric C02 or canopy air C02 were measured (Table 3). Due to higher measurement precision and the use of only one analytical laboratory,

South North


FIGURE 4 Comparison of observed and modeled ecosystem discrimination estimates for natural vegetation and crop systems. Observed Ae values were calculated using Eq. 3 (see Appendix); modeled Ae values were estimated using BIOME3.5.

South North


FIGURE 4 Comparison of observed and modeled ecosystem discrimination estimates for natural vegetation and crop systems. Observed Ae values were calculated using Eq. 3 (see Appendix); modeled Ae values were estimated using BIOME3.5.

TABLE 3 Global Means for "C Signatures of Ecosystem Respiration 8n Ci R and Biospheric Discrimination A Estimates


S"C||( (%o)


Observations, Trolier ct nl. (1996 i

-25.9 ±0.4

—17.8± 0.4a

Obscrvatios, Bakwin cl «/.(1998)

— 24.7± 0.8

—16.8± 0.8b

Observation, this study

— 25.3 ± 2.2

—18.0± 2.3c

Model Input, Keeling ct al. (1989)


Model input, Tans ct til. (1993)


Model input, Enting et al (1995)


Model, Lloyd and Farquhar (1994)


Model, SiB-GCM, Ciais ct al. (1995)

17— 18d

Model, SiB2-GCM, Fung cl al. (1997)


Model, BIOME3.5, this study


Means and standard deviations are given a Estimate for Point Barrow, Alaska, b Kstimates for near continental source/sink regions, c Estimates averaged from Fig. 2.

d Estimates of photosynthetic discrimination of the canopy, no soil compartment conisdered. e Estimates for potential natural vegetation.

Means and standard deviations are given a Estimate for Point Barrow, Alaska, b Kstimates for near continental source/sink regions, c Estimates averaged from Fig. 2.

d Estimates of photosynthetic discrimination of the canopy, no soil compartment conisdered. e Estimates for potential natural vegetation.

— 24.7 and — 25.9%o, with a global mean 5"C,:r value of — 25.3 ± 0.6 (SD) %o, close to the model input values for 8' 1CER by Keeling etal. (1989) and Hnting et al. (1995). Biospheric l3C fractionation during the CO, exchange between the terrestrial biosphere and the atmosphere (Ae) was found to be between 16.8 and l8%o, with a global mean for observations of 17.5%o, independent of the experimental method used.

However, modeled estimates differed by a maximum of 2.7%o from this observed mean value for Ae (l7.5%o). Closest agreement was found with the model estimate by Ciais and colleagues (l7-l8%o). Differences might arise from the fact that Ae values from actual measurements did naturally include the soil compartment (see above) whereas most of the models estimated only the canopy l3C discrimination, excluding the soil compartment and its associated isotopic effects. However, although BIOME3.5 did include the "long-term memory" effect of the soil compartment, its Ae estimate (l5.6%o) was l.9%o lower than the mean Ae estimate based on field measurements. Furthermore, this difference between observed and modeled Ae estimates cannot be explained by the continuous decrease of tropospheric SI3C values by about l.3%o since 1744 (Friedli et al, 1987, Trolier et al, 1996). Accounting for this effect (and potential interactions with low turnover rates) would even further increase the observed At, estimates and therefore the difference. In the following sections, we will discuss potential reasons for this difference.

6.1 Differences Due to Vegetation and PFT Distribution

Differences between observed and modeled Ae values could arise due to differences between the modeled and the real world in vegetation distribution or PFT distribution. Particularly, the distribution of C4 plants is assumed to be one of the major factors contributing to differences between Ae values based on tropospheric air measurements and modeled Ae values (Bakwin et al, 1998). Unlike other analyses, BIOME3.5 predicts the dominant natural vegetation type for a given region (grid-cell), including the natural distribution of C4 vegetation. While satellite data continue to improve, there is a lack of consistent data on actual vegetation type, seasonal variability, and distribution. BIOME3.5 circumvents the need for a predefined vegetation map, but includes a few other caveats.

Assumptions must be made about the predominant vegetation in several areas. One important limitation of BIOME3.5 is its inability to simulate biogeochemically the coexistence of different PFTs within a grid-cell and the variation due to habitats. For an analysis of Ae values, this limitation is not important in most cases because the physiological parameters of the PFTs are similar enough that under localized environmental conditions the PFTs behave similarly. Thus, flux rates and correspondingly the magnitude of isotopic fractionation are similar. However, in tropical savannas and warm-temperate grasslands, C3 and C4 plants may often be codominants in the same grid-cell. In this situation, the model takes an empirical approach of arbitrarily assigning a percentage of the grid-cell's NPP and consequent Ae to the grass and woody PFTs based on each PFTs NPP relative to the other. The NPP of any given PFT is calculated as if it was growing alone in the grid-cell. In the case of mixed C3-C4 biomes, NPP and Ae are scaled to reflect a mixed ecosystem. Since the main differences between observed and modeled Ae values (BIOME3.5) were observed in the mid-latitudes, natural vegetation and PFT distribution could contribute to this observed difference (see below).

6.2 Differences Due to Vegetation Change

Only a few of the new generation of computationally expensive dynamic vegetation models can accurately simulate a situation that must incorporate the transient effects of competition, disturbance, and mortality. The potential "memory effect" that decaying soil carbon may have when the dominant vegetation is in a successional or transition stage (Houghton, 1995; Neill et al, 1996; Buchmann and Ehleringer, 1998) cannot be simulated by an equilibrium vegetation model used to date (this paper; Lloyd and Farquhar, 1994; Fung et al., 1997). These effects, while exacerbated by anthropogenic land-use change, may also be present in natural ecosystems. Low-frequency but catastrophic disturbance regimes such as those in the arid subtropics may effect a long-term shift in Ae. Simultaneously, seasonal variability can cause a shift from C3-grass or shrub-dominated ecosystems to C4-grass-dominated ones, with expected lags in the response of §I3CER. Shifts in vegetation distribution due to climate change may also be a source of incongruity in the signature of £>'3CER and therefore Ae.

Because BIOME3.5 cannot simulate the dynamics of changing carbon pools and is not supplied with information on the 8I3C of atmospheric CO,, it is impossible to make an estimate of isotopic disequilibrium. However, global isotopic disequilibrium of l3C is estimated to be less than 0.3-0.5%o, thus within the range of uncertainty of the observed Ae as well as the modeled Ae (Enting et al, 1995; Fung et al, 1997). This analysis further suggests that the information from inverse modeling techniques about isotopic disequilibrium is limited by the wide spatial, and possibly temporal heterogeneity in Ae. The estimates presented here along with estimates of global biosphere Ac presented by others (Table 3) differ substantially. These differences illustrate the uncertainty in prescribing a mean global value of Ac, which has often been the case when Ac was used for constraining deconvolution analyses (Tans et al, 1993).

While the current state of land use has been incorporated into some modeling studies (Fung et al, 1997; Lloyd and Farquhar, 1994), no model to date has performed a sensitivity analysis on the importance of transient effects of land-use change. Transient land-use changes are especially important for addressing the question of isotopic disequilibrium. When the dominant vegetation changes to that of a different photosynthetic pathway, i.e., with the conversion of forest to C4 cropland, the SI3CER ratios would be expected to respond, albeit with a time-lag that would vary among ecosystems (Tans et al, 1993). However, recent widespread conversion to C4 crops is seen only in a few areas in the tropics (J. Lloyd, personal communication). There, a significant disequilibrium may exist between the isotopic signatures of soil-respired C02 and the carbon in the standing biomass.

In temperate regions of the Northern Hemisphere, widespread maize production may cause an increase in the isotopic content of the regional carbon stock (i.e., a shift to heavier l3C signatures). However, this signal would result in a higher global mean 8I3CER and a lower Ac; thus it cannot explain the observed 5%o difference in Ac at mid-latitudes between field observations and BIOME3.5 estimates.

6.3 Differences Due to the Water Regime

The BIOME3.5 model may underestimate ecosystem discrimination in places where water stress or environmental limitations on plant productivity are present during prolonged periods of the growing season (e.g., in winter-rain areas or deserts). Other modeling studies (Lloyd and Farquhar, 1994; Fung et al, 1997) take an even more empirical approach to vegetation distribution and physiology and also have difficulty to simulate Ae properly in dry places. Discrepancies between observed and modeled Ae estimates were expected for latitudes where agricultural C3 crops replaced the natural vegetation. Thus, higher observed than modeled Ae could arise because crop species are mainly bred for productivity, and only to a minor extent for low water use. In addition, great efforts are generally taken to ensure high water availability to agricultural fields, thus lowering the need to conserve water through stomatal regulation of photosynthesis. For the similar reason, Ae estimates for C4 crops were expected to be lower than modeled Ae values. These factors could contribute to both the observed differences, though the result may be confounding.

Other limitations of the model include assumptions made based on the driving data, soil hydrology model, and physiology of PFTs. BIOME3.5, using only monthly means, does not simulate the nonuniform nature of weather events. Physical parameters regarding soil structure, depth, and water-holding capacity are poorly constrained. We do not model plants' access to deep groundwater and other aquifers. Finally, various unknowns in the physiological parameters of certain PFTs, such as photosynthetic response to low temperatures, are coarsely parameterized.

6.4 Differences Due to Selection of Field Sites

Further discrepancies between observed and modeled Ae estimates might arise from biased site selection. Most of the study sites were located in the higher latitudes between 30° and 60° (Fig. 2). Terrestrial ecosystems within certain latitudinal bands such as 40°-20° S, 20°-30° N, or >70° N have not been studied at all (Fig. 4). Thus, the representation of global vegetation is still rather poor despite the 50 different study sites used for this comparison. This lack of field observations in deserts, C3 or C4 grasslands, savannas, or shrublands skews the distribution of observed Ae estimates. This could result in overestimation of Ae from field measurements. Spatial heterogeneity within a biome is often better known than differences among biomes or vegetation types (Flanagan et al, 1996; Buchmann et al, 1997; 1998).

In general, only limited information is available for ecosystems under naturally or anthropogenically disturbed conditions (e.g., her-bivory, fire, wind-fall, logging, clear-cut, severe air pollution). However, these conditions do not cause disagreements with modeled Ac estimates since they are not considered in BIOME3.5 or the other models either. Thus, the lack of predominantly C4 sites could contribute to the observed pattern of lower Ae values from the models.

Was this article helpful?

0 0

Post a comment