Global Biogeochemical Cycles

Response of central Siberian Scots pine to soil water deficit and long-term trends in atmospheric CO2 concentration

Authors


Abstract

[1] Twenty tree ring 13C / 12C ratio chronologies from Pinus sylvestris (Scots pine) trees were determined from five locations sampled along the Yenisei River, spaced over a total distance of ∼1000 km between the cities of Turuhansk (66°N) and Krasnoyarsk (56°N). The transect covered the major part of the natural distribution of Scots pine in the region with median growing season temperatures and precipitation varying from 12.2°C and 218 mm to 14.0°C and 278 mm for Turuhansk and Krasnoyarsk, respectively. A key focus of the study was to investigate the effects of variations in temperature, precipitation, and atmospheric CO2 concentration on long- and short-term variation in photosynthetic 13C discrimination during photosynthesis and the marginal cost of tree water use, as reflected in the differences in the historical records of the 13C / 12C ratio in wood cellulose compared to that of the atmosphere (Δ13Cc). In 17 of the 20 samples, trees Δ13Cc has declined during the last 150 years, particularly so during the second half of the twentieth century. Using a model of stomatal behaviour combined with a process-based photosynthesis model, we deduce that this trend indicates a long-term decrease in canopy stomatal conductance, probably in response to increasing atmospheric CO2 concentrations. This response being observed for most trees along the transect is suggestive of widespread decreases in Δ13Cc and increased water use efficiency for Scots pine in central Siberia over the last century. Overlying short-term variations in Δ13Cc were also accounted for by the model and were related to variations in growing season soil water deficit and atmospheric humidity.

1. Introduction

[2] Plant responses to increases in atmospheric CO2 concentration, [CO2], vary considerably. In controlled experiments, more often than not, some form of “acclimation” develops. This may involve, for example, changes in stomatal conductance or reduced photosynthetic capacities or changes in plant allocation patterns [e.g., Bowes, 1993; Drake et al., 1997; Idso and Idso, 1994; Lee and Jarvis, 1995; Lloyd and Farquhar, 1996; Long et al., 1996]. It remains uncertain, however, if the results derived from controlled experiments are sufficient to predict successfully plant responses in their “natural” environment characterized by a gradually increasing [CO2].

[3] Stable carbon isotope ratios (δ13C) in tree rings reflect the balance between plant photosynthesis and water use and accurately record physiological responses in the natural growth environment over the lifetime of a tree, thus providing an alternative approach for an understanding of plant physiological responses to changing climates and [CO2] on decadal to centennial timescales. Still, it is only recently that cautious advances have been undertaken to analyze the physiological processes underlying interseasonal and interannual variations in the δ13C in wood [Berninger et al., 2000; Feng, 1999; Marshall and Monserud, 1996; Walcroft et al., 1997]. These studies expand on numerous previous analyses where δ13C from a wide range of species and environments have been used successfully as climate proxies or to deduce atmospheric δ13C [Bert et al., 1997; Francey and Farquhar, 1982; Freyer and Belacy, 1983; Krishnamurty, 1996; Leavitt, 1992, 1993; Leavitt and Long, 1983; Lipp et al., 1991; Stuiver and Braziunas, 1987; Wilson and Grinsted, 1977]; for review, see Switsur and Waterhouse [1998]. However, it is now clear that it is not possible to evaluate plant responses to increasing [CO2] using a solely statistical approach. This is because photosynthesis and transpiration vary in response to atmospheric CO2 concentration as well as with temperature, radiation, humidity, and plant water supply. However, when combined with an appropriate biochemical model and when expressed as discrimination against 13C during photosynthesis, Δ13C [Farquhar et al., 1988], stable carbon isotope ratios ultimately provide a powerful tool from which to draw conclusions concerning plant responses to short- and long-term changes in the natural environment.

[4] The objectives of the current study were to analyze Δ13C measured in α-cellulose from successive growth rings (Δ13Cc) of central Siberian Scots pine trees (Pinus sylvestris L.) to investigate photosynthetic responses to the long-term increase in [CO2] over the last 150 years or so. As is typically the case for P. sylvestris, the trees were found growing on sandy soils, in a continental climate characterized by hot and dry summers (Table 1). A second focus was therefore to use Δ13Cc to determine the extent to which short- and long-term variations in temperature, precipitation, and humidity affect the photosynthetic physiology of these trees. Samples were collected along a N-S transect following the Yenisei River in central Siberia, a natural highway that extends from south of Krasnoyarsk (56°N) to the Arctic Ocean (72°N). Climate along this gradient changes significantly, but as Scots pine is usually found on free-draining, sandy soils with low water storage capacity, soil moisture (Θ) deficit could be an important factor influencing tree carbon assimilation and transpiration during the summer period [Kelliher et al., 1998].

Table 1. Annual and Growing Season Temperatures and Precipitation From Four Meteorological Stations (Coordinates and WMO Station Name) Along the Sample Gradienta
NameGeographic Location of the Weather StationAnnual Tair, °CGrowing Season Tair, °CGrowing Season Precipitation, mmLength of Growing Season, days
  • a

    Data are medians (air temperature Tair) and sums (precipitation) derived from daily data and cover the period 1960–1989 when data were available for all four stations. The growing season is defined as the number of days with an average temperature >5°C.

Turuhansk65.8°N, 87.9°E−5.312.2218108
Bor61.6°N, 73.5°E−1.513.4260126
Eniseisk58.5°N, 92.1°E0.413.7252142
Krasnoyarsk56.0°N, 92.9°E214278154

[5] Our analysis was based on the idea that in a water-limited environment, increasing [CO2] will lead to physiological adjustments that should optimize the use of water by the trees. To test the plausibility of this hypothesis, we simulated plant Δ13Cc using a process-based model and comparing different CO2 response scenarios.

[6] 1. Stomatal conductance and the marginal evaporative cost of plant carbon gain were assumed insensitive to increasing [CO2]. The impact of climate (e.g., temperature and air saturation deficit) was investigated by running the model with (1) identical climatic input for each year and increasing CO2 concentration and (2) measured, transient climatic input for each year and increasing CO2 concentration.

[7] 2. Stomatal conductance and the marginal cost of plant carbon gain have decreased with increasing [CO2] (CO2 sensitive).

[8] 3. Stomatal conductance and the marginal cost of plant carbon gain have decreased with increasing [CO2] and also decrease with increasing soil moisture deficit (CO2/Θ sensitive).

2. Material and Methods

2.1. Sample Locations

[9] In central Siberia, Pinus sylvestris forms monospecific forests known as “Bor” in Russian. In fact, it appears etymologically likely that the boreal biome received its name from the Slavonic word Bor (U. Steltner, personal communication, 2000). Growing on sandy soils mostly to the west of the Yenisei River, Scots pine dominates the forested area between 58° and 64°N [Kalaschnikov, 1994]. North, south, and east of the Bor proper, species diversity increases, and Scots pine stands are found among a mosaic of forest types dominated by Pinus siberica, Abies siberica, Picea obovata, Larix siberica, Betula pubescence, and Populus tremula. Scots pine is also found growing along ridges in Sphagnum bogs over the entire range of its distribution in Siberia.

[10] During July and August 1998, stem discs were collected from five different Scots pine forests, spaced along a north-south gradient that extends over a large part of the species' latitudinal range (Figure 1). The site names used here refer to the closest village or city. Following the Yenisei River, the northernmost site (“Turuhansk,” 66°N) was located at the subarctic limit of Scots pine growth in central Siberia. The southernmost site (56°N), close to the city of Krasnoyarsk, was ∼300 km north of its southern limit which is in the forest-steppe transition zone in the Sajan mountains [Kalaschnikov, 1994]. Three sites interspersed along the transect (Verkhne-Imbatskoye, Zotino, and Nazimovo) represented the northern, middle, and southern area of the literal Bor, respectively.

Figure 1.

Map of the sample transect along the Yenisei River in central Siberia. The sample locations are marked with letters corresponding to names in Table 1.

[11] For four of the sampled sites, long-term weather station data are available. From this data the north-south extent of ∼1000 km translates into a sharp mean annual temperature gradient from −5.3° to 2°C along the transect (Table 1; source is http://www.ncdc.noaa.gov). During the growing season, i.e., when daily average temperatures exceed 5°C, this gradient is greatly diminished, but air temperatures in Krasnoyarsk are still close to 2° above air temperatures in Turuhansk. Growing season length also increases north to south. There is no pronounced growing season rainfall gradient and on a daily average; Eniseisk and Krasnoyarsk are only insignificantly drier (1.8 mm d−1) than the village of Bor and Turuhansk (2.1 and 2.0 mm d−1).

[12] With exception of the northernmost site (Turuhansk), samples were taken from monospecific stands growing on sandy soils (Table 2). Four emergent trees were selected from each location on the basis of stem diameter and tree height distributions in mapped plots of 100–500 m2, each plot containing between 20 and 40 trees. Stem discs for stable carbon isotope analysis were cut from 1.3 m above ground. At the northernmost site, Scots pine was growing on a south facing dolomite slope aside Larix siberica. A fire had killed an estimated 70–80% of the forest ∼30 years ago, leaving only patches of trees and from these four large specimens were selected.

Table 2. Stand Characteristics of the Sample Locationsa
NameCoordinates of the Sample PlotLive Trees Per HectareHeight, mDiameter, mGrowth Rings Per Sample DiscSoil TextureΔ13Cc
  • a

    Height and diameter data are averages ± standard deviation from trees growing within mapped plots. Data for the Zotino site are from Wirth et al. [1999]. Averages are based on 27, 239, 41, and 16 trees from Verkhne-Imbatskoye to Krasnoyarsk, respectively. Stand characteristics were not measured at the Turuhansk location because of the severe fire damage to the site. Δ13Cc denotes averages ±1 standard deviation from 1927 to 1998, when data were available from four trees at each sample location.

Turuhansk65°58′N, 88°45′E118, 88, 313, 260shallow silty sand on dolomite17.35 ± 0.43
Verkhne-Imbatskoye63°04′N, 87°32′E189431.1 ± 5.30.14 ± 0.0589, 89, 237, 235sand17.13 ± 0.42
Zotino60°45′N, 89°23′E47817.8 ± 3.50.27 ±0.07200, 200, 119, 200sand17.48 ± 0.38
Nazimovo59°18′N, 90°42′E57918.4 ± 4.70.25 ± 0.1299, 120,179,187sand17.64 ± 0.61
Krasnoyarsk56°22′N, 92°57E53623.5 ± 2.90.33 ± 0.07100, 96, 98, 100silty sand16.41 ± 0.34

2.2. Sample Preparation and Stable Carbon Isotope Analysis

[13] The stem discs were air dried, finely sanded, and dendrochronologically dated (Lintab & TSAP, Frank Rinn Distribution, Heidelberg, Germany) along two radii of each disc. Along these radii, 1 cm × 1 cm blocks were cut from the discs to extract wood for stable isotope analysis. Rings were often <0.3 mm wide, and even under a binocular lens, it was not feasible to attempt to separate annual rings without risking contamination from the previous and following years in the wood sample. Thus thin slices (20–30 μm) were cut radially, using a razorblade, each integrating a 3 year period. The number of tree rings at the sample height ranged from 89 to 313. The time period of the chronologies sampled was typically between 1849 and 1998 and was less when the trees were younger.

[14] Alpha-cellulose was extracted from wood samples (∼80 mg) that were sealed in acid- and solute-resistant bags (Laboratoires Humeau, La Chapelle sure Erdre, France) and numbered with glass plates. Resins were removed by a 24 hour Soxhlet extraction in a 2:1 toluene:ethanol mixture. The samples were subsequently dried for 24 hours, and adopting the method of Loader et al. [1997], lignin was then removed by oxidation for 6 hours in beakers containing a sodium chlorite-acetic acid mixture. The final step involved removal of hemicelluloses by 2 hour sodium hydroxide hydrolysis. The beakers were placed in an ultrasonic bath to facilitate rapid reaction. Batch processing of 50 samples was possible. During each run a bag containing α-cellulose (Sigma Aldrich GmbH, Seelze, Germany) was added, which was later analyzed for its 13C/12C ratio to ensure the absence of contamination by the chemicals used in the procedure.

[15] Between 2 and 2.5 mg of α-cellulose per sample were sealed into tin cups, and the carbon isotope ratios of the samples were measured in a Finigan Delta Plus XL mass spectrometer (Finnigan MAT, Bremen, Germany) using a ConFlo III interface [Werner et al., 1999]. The δ13C values are expressed in per mil versus Vienna pee dee belemnite (VPDB) standard. The standard deviation for δ13C values of a laboratory standard was 0.09‰, and the standard deviation of δ13C measured in 15 samples from the same 3 year period was also 0.09‰. Matching the decline in atmospheric δ13C (subscript “a”) due to burning of isotopically depleted fossil fuel the δ13C of plant material (subscript “p,” in this case taken to be that of tree ring α-cellulose) has also declined steadily in the recent past. For analysis of time trends the isotopic abundances were therefore converted into discriminations (Δ13C [Farquhar et al., 1989a]), namely,

equation image

where Δ13Cc is the plant isotopic discrimination as reflected by concurrent measurements of δ13Ca and δ13C of the tree ring cellulose. Values for δ13Ca between 1849 and 1993 were taken from an ice core record from Antarctica [Francey et al., 1999]. Values were extrapolated to 1998 by assuming a continuing annual decline of 0.02‰. There is an annual gradient between δ13Ca between the hemispheres, but the differences in δ13Ca measured at oceanic locations at the South Pole and at Point Barrow during the months May to September, essentially the growing season in Siberia, were <0.02‰ in 1993–1997 (source is ftp://cmdl.noaa.gov). Consequently, the δ13Ca data were not adjusted further, bearing in mind the potentially large uncertainties likely due to a lack of atmospheric data from inland locations [Berninger et al., 2000]. Similarly, the possible influence of a changing hemispherical δ13Ca gradient is likely to be 1 order of magnitude smaller than the changes in Δ13Cc analyzed here [Berninger et al., 2000].

[16] In many cases, when compared with average discrimination over the lifetime of trees, time series of δ13C in wood and cellulose during the early stages of tree growth show sometimes spuriously low and/or steadily increasing values (i.e., high and decreasing Δ13Cc). This “juvenile effect” has been attributed to distinct microenvironments during seedling establishment, uptake of depleted respired CO2 from the canopy floor, or age-related physiological factors [Bert et al., 1997; Francey, 1981; Freyer, 1979; Marshall and Monserud, 1996]. Regardless of the causes for the juvenile effect, where it occurs, it overrides trends attributable to climatic variation or to the influence of CO2], which are the main concerns of this analysis. On the basis of a change point identification [Lanzante, 1996; Siegel and Castellan, 1988], juvenile data were thus excluded from further analysis.

[17] An agreement between the absolute values of observed δ13C in tree growth ring cellulose and photosynthetic δ13C should not necessarily be expected. This is because carbon of cellulose generally is enriched 3–4‰ compared to bulk leaf carbon [Livingston and Spittlehouse, 1996]. Thus a model estimate of photosynthetic δ13C based on leaf and canopy gas exchange characteristics (see below) should be larger than the corresponding Δ13Cc [Berninger et al., 2000].

2.3. Model Simulations

[18] For the Zotino sampling location a process-based model analysis of Δ13C was possible because all the necessary information required to simulate canopy photosynthesis and transpiration was available. The model analysis is based on the notion that stomata serve to minimize the marginal evaporative cost of plant carbon gain [Cowan, 1977]. As is described in Appendix A, this has been achieved by the derivation of analytical solutions to a combination of equations incorporating the stomatal response predictions of Cowan [1977] and the model of plant photosynthesis developed by Farquhar et al. [1980]. Briefly, this requires the concept that the rate of carbon assimilation A in the leaf cell is either limited by the activity of ribulose-1,5 bisphosphatase-carboxylase (Av) or by the rate of ribulose1,5 bisphosphate regeneration (AJ)

equation image

to be combined with the requirement of Cowan [1977] that

equation image

where λ is a Lagrange multiplier that represents the marginal water cost of plant carbon gain, E is transpiration, and gs is stomatal conductance. An analytical solution of the combined model provides an expression for calculating the substomatal CO2 concentration Ci and is given in Appendix A for the Av and the AJ cases ((A19) and (A20)).

[19] Using hourly climatic input data, the modeled Ci were used to calculate

equation image

where Ca is ambient CO2 concentration, a = 4.4‰, and b = 27‰ and a and b are the discrimination due to diffusion in air and discrimination due to Rubisco, respectively [Farquhar et al., 1989a]. The calculated Ci were also used to compute instantaneous A (t = 1 hour) ((A15) and (A16)). For comparison with growth ring cellulose data, photosynthesis-weighed annual Δ13C (Δ13Cyear) were subsequently calculated [Berninger et al., 2000] from

equation image

The assimilation rates Av and AJ are largely dependent on the maximum rate of Rubisco activity (Vmax) or electron transport (Jmax); both vary with temperature. The model thus requires determination of λ, Vmax, Jmax , and Rd (respiration rate) and of the energies of activation and deactivation that describe the temperature response of Jmax and Vmax (Table A1). To estimate the values of these parameters we utilized data from eddy flux measurements of whole forest and ground CO2 and H2O fluxes that have been conducted since 1998 at Zotino (see Table 2). Canopy values for the input parameters could be derived by fitting the combined conductance-assimilation model to data of forest photosynthesis [Arneth et al., 1999; Lloyd et al., 1995] that were obtained from the difference between measured whole forest-atmosphere CO2 exchange minus ground CO2 flux (Table A1). Data from 5 days collected between May and July 1999 were chosen to fit the model on the basis that these days were mostly sunny, a large temperature range was covered (0.2°–33°C), and the data showed that no reduction in carbon or water fluxes due to soil water stress. Computations of the temperature dependence of Jmax and Vcmax were similar to those described by Arneth et al. [1999]. All other parameter values required in the modeling exercise were similar to elsewhere (e.g., Table A2) [Harley et al., 1992; Leuning, 1990; Lloyd et al., 1995].

[20] While the stomatal optimality concept requires instantaneous λ to be constant over periods of 1 day or so, there is no reason why this should be the case for longer time periods [Cowan, 1977]. Indeed, Hall and Schulze [1980] found a reduction in λ with increasing soil water deficits, and such a reduction is also anticipated on theoretical grounds [Cowan, 1982; Mäkelä et al., 1996]. Other things being equal, a reduction in λ represents a reduction in gs (equation (A11)) and an associated reduction in Ci/Ca (Figure 5). Testing the hypotheses that stomatal closure in response to increasing CO2 and/or water stress will be reflected in the measured cellulose Δ13Cc, we compared the measured data with simulated values and ran the simulations (1) assuming no effect of [CO2] or soil water (Θ) deficit on λ (“insensitive”), (2) assuming a linear decline of gs by 7% with increasing [CO2] over the simulation period (“CO2 sensitive”), and (3) assuming a linear decline of gs with increasing [CO2] (case 2) and with decreasing available soil water deficit below a threshold of 0.4 of field capacity (“CO2/Θ sensitive”).

[21] The rationale for the form and magnitude of the stomatal response to soil water deficit in scenario 3 reflects the observation that in coniferous forests, maximum conductances decline with increasing soil water deficit below a threshold level of soil moisture. In an old-growth P. sylvestris forest in central Siberia a linear decline in surface conductance has been measured when soil volumetric moisture content Θ was <0.05 [Kelliher et al., 1998]. A linear decline of gs translates in our model into a quadratic decline of λ [see Lloyd and Farquhar, 1994, equation (3)] of the form

equation image

where λmax = λ, the fitted canopy value (Table A1) when gs is assumed insensitive to Θ. Here gs and gs0 are (canopy) stomatal conductances with and without the prescribed effect of H2O.

[22] In the CO2/Θ-sensitive case, λ varies between λmax and zero, namely, gs/gs0 = 1 above the Θ threshold (0.05 m3 m−3). Below this threshold, gs declines linearly with declining Θ, such that in the hypothetical case, when Θ = 0, gs/gs0 = 0. In the CO2-sensitive case we have assumed a linear decline of gs by 7% in the simulation period (1884–1998), above and beyond any change predicted by the stomatal model. This is one half of an observed reduction in gs for Scots pine grown at elevated [CO2] [Wang, 1996] and corresponds to λ/λmax = 0.88 in 1998.

2.4. Soil Water Balance Subroutine

[23] Case 3 requires the calculation of an ecosystem soil water balance. Daily changes in soil water storage were calculated using a simple single-bucket simulation as

equation image

where dWt/dt is daily change in soil water storage (volumetric water content Θ multiplied by the depth of soil, in units of mm), P is daily precipitation, reduced by 30% to account for interception by the canopy [Whitehead and Kelliher, 1991], and Et is daily forest (canopy plus ground) evaporation.

[24] For the sandy soils at Zotino, Θ was allowed to vary between 0.13 and 0.02 m3 m−3, which corresponds to a suction of 20–1500 kPa [Kelliher et al., 1998]. From observations in soil pits the main rooting depth for water extraction was ∼0.3 m, although some roots were found down to 2 m depth. Daily evaporation from the canopy Ec was calculated from hourly values as Ec = Σ(gsD (t)), where gs is the canopy conductance for water vapor transfer, D is air saturation deficit (mol mol−1), and t is hour of day. As

equation image

Ec is in molar units but can be easily transferred into units of mm d−1 (required in (7)). Total forest evaporation was taken as twice the tree canopy value as soil and lichen evaporation in Siberian pine forests typically accounts for 50% of the total [Kelliher et al., 1998].

[25] Running the model in a forward mode requires input of hourly quantum flux Q (μmol m−2s−1), air temperature T (°C), and air saturation deficit D (mol mol−1). A continuous record of daily minimum and maximum temperatures (Tmin and Tmax) and precipitation was available from 1936 to 1998 for the township of Bor, located along the Yenisei River, ∼150 km north of the eddy flux site. The data were assembled from records at the National Oceanic and Atmospheric Administration (NOAA) National Climate Data Center (NCDC) archive (source is from http://www.ncdc.noaa.gov), in combination with data covering the years 1990–1992, which were obtained directly at Bor village airport. An extended record of daily temperature and precipitation from 1884 to 1989 was also available from the city of Eniseisk, located ∼300 km south of Zotino. Linear regressions between Bor village and Eniseisk daily average temperatures (TavBor = 0.97 TavEn − 0.96, r2 = 0.85, when Tav > 0) and between Bor village Tav and Tmin and Tmax (TminBor = 0.99 TavBor − 5.09, r2 = 0.97; TmaxBor = 1.01 TavBor + 5.41, r2 = 0.97) were established, which allowed the reconstruction of the growing season temperature record for Bor village between 1884 and 1936. Although the above r2 values are satisfactory for our purposes, a possible bias in the Eniseisk average daily temperatures (and hence in the corrected Bor village values) before 1936 cannot be excluded because the recording procedure did not account correctly for nighttime temperatures (see http://www.ncdc.noaa.gov). A reconstruction of the precipitation record prior to 1936 was not possible, as there was no significant relationship between daily precipitation at Bor-village and at Eniseisk.

[26] Hourly climatic variables were estimated from the daily temperature record using the following procedure: (1) Daily global radiation at the surface was estimated from extraterrestrial global radiation using an atmospheric transmissivity calculated from the daily temperature amplitude [Bristow and Cambell, 1984; Spitters et al., 1986]. The required parameter values in the empirical relationship were determined by fitting the model to minimum and maximum temperatures and daily totals of global radiation measured between June and September 1998 at the Zotino eddy flux site. Additionally, a linear relation was used to calculate Q from total global radiation (Q = 1.98 total global radiation + 1.34, r2 = 0.99, where Q is in mol m−2 d−1 and total global radiation is in MJ m−2 d−1). (2) The diurnal course of Q was estimated from modeled daily Q and solar geometry [Ephrat et al., 1996], and Q absorbed by the canopy was calculated from Beers law and solar geometry [Campbell, 1981] assuming a constant leaf area index of 1.0 [Wirth et al., 1999]. (3) The diurnal course of air temperature was estimated from solar geometry and daily minimum and maximum temperature data [Ephrat et al., 1996]. (4) The diurnal course of air saturation deficit was established assuming that daily Tmin represented the dew point temperature, taken to be constant over each day. The predicted data were tested against meteorological data that were measured at the Zotino eddy flux site during the 1998 growing season. Linear regressions between measured and modeled data delivered r2 between 0.71 and 0.87.

3. Results and Discussion

[27] Overall, observed Δ13Cc in the 20 trees varied from 15.2 to 19.7‰ (Figure 2). These values were similar to values observed in α-cellulose of Scots pine growing in open, maritime forests in northern Finland [Berninger et al., 2000] and somewhat higher than reported for Scots pine parklands in central England [Hemming et al., 1998]. The observed variation included up to 1‰ high-frequency fluctuations between single 3 year periods which were of a similar magnitude to the differences observed between single trees. Differences between trees in Δ13Cc at any one location were generally nonsystematic and <0.5‰, except for one tree at Nazimovo which, for nearly the entire 150 year period investigated, had values consistently 0.5–1‰ less than the other trees at the same site (Figure 2, triangles). Juvenile Δ13Cc within the first 10–21 growth years were identified (see section 2) for all four trees at Krasnoyarsk and for two trees at Verkhne Imbatskoye; while these periods are shown in Figure 2, they were excluded from the calculation of the trend lines.

Figure 2.

Time course of average Δ13Cc ± standard deviation from 20 Pinus sylvestris trees growing on five locations along a 1000 km latitudinal transect across central Siberia. The values are from measurements of tree growth ring α-cellulose in a time resolution of 3 years. Four emergent trees per location were sampled that had between 88 and 313 growth rings at sample height (1.3 m above ground). The time period of the chronologies sampled was typically between 1849 and 1998 and was less when the sample trees were younger. For locations Turuhansk, Zotino, and Krasnoyarsk, symbols represent averages of all four trees. For location Verkhne-Imbatskoye, solid circles indicate the two young sample trees, while open circles represent averages of the four trees. At Nazimovo, solid circles are averages of three trees with one outlier tree separated (open circles). Lines are three-parameter power function fits through the data, excluding periods of possible “juvenile effects.”

[28] There was no continuous latitudinal discrimination gradient observable between Turuhansk and Krasnoyarsk, but Δ13Cc values at Krasnoyarsk were statistically significantly lower than Δ13Cc values at any of the other sites (Table 2). These results are in accordance with the theoretical considerations of Lloyd and Farquhar [1994], who emphasized that on a global scale, only a minor latitudinal gradient of decreasing Δ13C can be expected for the Northern Hemisphere. However, their analysis aggregated data from all C-3 land cover classes without accentuating additional regional constraints due to nutrient or water supply. Along the Siberian gradient, available Θ was anticipated to decrease from northern to southern latitudes, as growing temperatures and evaporative demand increase while average daily precipitation remains constant (Table 1); without additional constraints, such a gradient should be reflected in declining Δ13Cc. There are, however, a variety of factors that might override a simple soil water deficit-Δ13Cc gradient: First, the sampling locations were located between 10 and 40 km inland from the Yenisei, but the climatic data compiled in Table 1 originate from meteorological stations located directly at the Yenisei River. The extent to which the recorded surface weather close to the river is modulated by the river's water body is uncertain. Second, precipitation is intercepted by canopies, and the amount of precipitation reaching the soil thus depends on canopy structure. Third, soil water storage capacity as well as soil C:N varied between sampling locations, and there was no systematic trend along the transect (M. Bird, unpublished data, 1999). The differences in Δ13Cc between single trees suggest that detecting (or rejecting) a potential water stress-induced discrimination gradient would require a larger sample size per latitude. Bearing in mind the remoteness of the area as well as the effort involved with cellulose extraction and stable isotope analysis, this was not possible with the resources available for the current study.

[29] Nonetheless, when the average time course of Δ13Cc at each location was examined, a long-term trend of declining discrimination was discernable at four sites (Figure 2), but in the case of Nazimovo, only when the outlier tree (Figure 2, open circles) was removed from the average. This decline in Δ13Cc appears to have intensified in the second half of this century. The decline has been steepest at Krasnoyarsk, where the long-term Δ13Cc have decreased by >1‰ since 1924, thus equaling in magnitude the superimposed short-term fluctuations. In contrast, however, at Verkhne-Imbatskoye a reversed trend to increasing discriminations has been observed, largely attributable to increasing Δ13Cc values in the two old trees at this location (Figure 2, open circles). When restricted to the two younger trees (solid circles), average Δ13Cc declines at this location as well, albeit at a lesser rate than seen at the other locations. The data thus suggest large-scale environmental modulation of Δ13C this century, observed along the entire transect. The trend is small but genuine, and both the residuals of the power function fit shown in Figure 2 as well as their autocorrelation coefficients had a mean value of zero (not shown [Chatfield, 1989]).

[30] Tree age strongly affects net primary productivity and may also alter carbon allocation patterns [Schweingruber, 1989]. In Siberia, stand density and net primary productivity are regulated by age and by the frequency and severity of fires [Wirth et al., 1999]. It is unlikely, however, that the overall decline in discrimination resulted from age-related physiological changes or from changes in stand structure related to regrowth after fire. The observed trend in central Siberia was found in trees that were (in 1998) ∼100–320 years old, based on the number of rings grown at a height of 1.3 m. Furthermore, as pointed out before, at 63°N, only the two younger trees showed the declining trend, while Δ13Cc of the two old trees from this location increased with time. At Turuhansk, no step change in Δ13Cc was observed after the severe fire that had killed ∼80% of the stand ∼30 years ago. From reviewing other studies we have to conclude that there is no general pattern of Δ13Cc and hence Ci/Ca; during the recent past, values either increased, decreased, or remained constant even for similar species and when climatic trends were accounted for [Bert et al., 1997; Duquesnay et al., 1998; Feng, 1999, 1998; Leavitt and Lara, 1994; Marshall and Monserud, 1996]. For example, Δ13Cc of Pinus sylvestris growing in the maritime climate of northern Finland clearly increased in the second part of the twentieth century [Berninger et al., 2000], while Δ13Cc from Pinus sylvestris in England declined between 1930 and 1975 but then increased again until the mid-1990s [Hemming et al., 1998]. How can these seemingly contradictory observations be explained?

[31] The response of plants to rising [CO2] is varied and will depend on the plant species and on its growth environment [e.g., Beerling et al., 1996; Lee and Jarvis, 1995]. Evidence from studies that were carried out over several months to years points to abundant physiological acclimation enabling plants to respond plastically to the changing CO2 environment. Acclimation is thought to result in an increasing resource use efficiency [Lloyd, 1999; Long et al., 1996], and while adaptative responses have not been found unanimously, they may be particularly advantageous where other environmental parameters (e.g., nutrients and available water) are limiting.

[32] Of relevance here is that Pinus sylvestris tress grown over several years at elevated CO2 show only minor modulation of photosynthetic capacities. In the absence of water stress both observed and modeled photosynthetic rates increase with additional effects of increased temperature on the photosynthetic response being negligible [Wang, 1996; Wang and Kellomaki, 1997]. Similar increased photosynthetic rates were also found for Scots pine growing at an elevated CO2 and temperature site in southern Norway [Beerling, 1997]. In contrast to more or less unaffected photosynthetic capacities, stomatal conductances of P. sylvestris have been observed to be reduced in high CO2 and/or temperature environments [Wang and Kellomaki, 1997]. If ambient temperature and D are unchanging, stomatal closure reduces transpiration, but the reduced gs in itself may exert a relatively small influence on photosynthesis [Farquhar and Sharkey, 1982]. As a result, the ratio of assimilation to transpiration, the instantaneous water use efficiency, should rise [Farquhar et al., 1988; Farquhar et al., 1989b; Korol et al., 1999].

[33] To see if the observed responses of P. sylvestris to increased CO2 concentrations and climate change this century are consistent with the above, the results of the four simulations (“unsensitive plus identical climate,” “unsensitive plus transient climate,” “CO2 sensitive,” and “CO2/Θ sensitive”) are shown in Figure 3. Figure 3a shows the measured triannual variations in growing season temperature and precipitation for Bor village (Figure 1 and Table 1). It is apparent that the temperature record, extended back with Eniseisk data, shows continuously increasing temperatures since the late 1930s. This trend was preceded by a period of comparatively low temperatures during the 1920s and high temperatures during the turn of the century, respectively. Increasing values of D, which exert a significant influence upon gs (equation (A11)), are associated with increasing temperatures. Figures 3b–3e summarize model calculations of canopy conductance for water vapor transfer (Figures 3b and 3d) and Δ13Cc (Figures 3c and 3e) for the variety of scenarios.

Figure 3.

(a) Air temperature T (solid triangles) and precipitation P (open triangles) at the town of Bor. The data are growing season medians (T) and sums (P) calculated for 3 year periods from daily T and P data. Data were available for 1936–1998, but in case of the temperature record it was extended using a record of daily T at the city of Eniseisk and linear regressions between the Bor and the Eniseisk data. (b) Simulated canopy stomatal conductance gs using the combined conductance-assimilation model described in Appendix A. Simulations were run with keeping λ constant (“insensitive” scenario) and using (1) identical climatic data for every year and increasing atmospheric [CO2] (dashed line) and (2) transient climatic record from Bor and increasing atmospheric [CO2] (squares). Conductances are given as average daily values during the growing season for each 3 year period when Δ13Cc was measured. (c) Simulated Δ13C for the scenarios shown in Figure 3b. (d) Simulated gs using the CO2-sensitive (solid circles) and CO2/Θ-sensitive (open circles) scenarios. (e) Simulated Δ13C for the scenario shown in Figure 3d.

[34] While keeping λ, Jmax, and Vmax constant, the combined biochemistry-conductance model predicts stomatal aperture to be relatively constant and Ci/Ca to increase over a wide range of [CO2] (Figures 5 and 6). Accordingly, when climatic influences were excluded, i.e., when the simulation was run with identical climatic input for each year of the calculation, over the range of [CO2] in this study (∼290–360 ppm), gs showed only a minute decline with time (Figure 3b, dashed line), while A, as expected, increased but at a proportionally slower rate than the increasing [CO2] (not shown). Consequently, Δ13C increases in this scenario (Figure 3c, dashed line). In the insensitive plus transient climate simulation scenario, when both [CO2] and meteorological input variables are allowed to vary but λ remains constant, gs remains more or less constant until 1938, succeeded by declining values during the second half of the twentieth century (Figure 3b, squares). Using this scenario, this decline shows the combined result of increasing [CO2] and high D during periods with high temperature. This influence is enough to remove the trend of increasing Δ13C that was predicted in response solely to increasing [CO2] (Figure 3c). In Alaskan White spruce, decreasing Δ were attributed to temperature-induced drought stress on the basis of a statistical analysis of Δ and growth ring data [Barber et al., 2000]. However, while the observed increased temperatures and increased D this century are considered here to have contributed to a reduced long-term gs and hence the observed long-term reduction in Δ13Cc, this effect is not enough on its own to explain the magnitude of the observed decline in Δ13Cc in central Siberia.

[35] Scots pine growing in central Siberia may experience regular shortages in water availability, a result from the combination of low precipitation, high temperatures, and low soil water storage capacity [Kelliher et al., 1998]. Testing our original assumption that in this environment, increasing [CO2] will lead to physiological adaptations that result in a more efficient use of water, in a third simulation run, gs was assumed to decline linearly as [CO2] increased, which translates into a quadratic decline of λ. The linear relation between gs and CO2 is a necessary simplification because we are not aware of any study that aimed at defining a functional relationship between stomatal acclimation and increasing [CO2]. The assumed 7% decline in gs is about half of what has been observed in a doubled CO2 environment [Wang, 1996].

[36] Notwithstanding the simplification associated with the assumed linear response of gs, when λ is allowed to decrease with CO2, this model is a much improved predictor of the observed long-term trend in Δ13Cc (Figure 3e, solid circles). Modeled Δ13C decrease from the late 1930s onward, and a linear regression through modeled and measured Δ13C explained 41% of the variation. However, while the CO2-sensitive model run predicts the long-term trend in observed Δ13Cc, measured high-frequency variation in measured data exceeds modeled high-frequency variation significantly (Figure 2).

[37] In coniferous forests, maximum and average canopy and surface conductances are greatly reduced when available soil water drops below a threshold of ∼0.5 field capacity; in Siberian pine forests this decline was linear when Θ < 0.05 m3 m−3 [Arneth et al., 1999; Kelliher et al., 1998]. Thus, while the increasing [CO2] may explain much of the long-term trend, interannual differences in precipitation and soil water supply should be reflected in overlying short-term variation of Δ13Cc. Investigations of the possible magnitude of this effect using the fourth scenario, the CO2/Θ-sensitive model, was possible only for the years from 1936 to 1998 for which reliable precipitation data are available and soil water deficit could be added as additional constraint to forest-atmosphere gas exchange. However, when this effect was incorporated for these years at Zotino, high-frequency variations in simulated Δ13C were amplified, and the fit between measured and modeled data was further improved compared to the CO2 sensitive model alone (Figure 3e, open circles). This combined CO2 and Θ sensitive model explained 62% of the observed variation in the data (Figure 4).

Figure 4.

Measured versus modeled Δ13C at Zotino. Modeled values are taken from the CO2/Θ-sensitive scenario (Figure 3e). The line represents a linear regression through the data.

[38] This modeling result supports the theoretically proposed decline in λ [Cowan, 1977] under water-limiting conditions, which has been demonstrated from leaf gas exchange measurements for cowpeas (Vigna unguiculata) after several days of drought [Hall and Schulze, 1980]. It has also been shown over a period of 15 summer days in an eastern Siberian Larix gmeliniicanopy by Hollinger et al. [1998], who used the optimal stomatal regulation model of Mäkelä et al. [1996] and Berninger et al. [1996]. In those models a parameter closely related to λ varies with time since last rainfall. This general decline of λ under water-limiting conditions appears independent from the absolute value of λ itself. Nevertheless, for both the cowpeas and for the Siberian larch trees, λ was ∼1000, more than twice the value that was found in this study for Scots pine.

[39] While agreement between modeled and measured Δ13Cc was good, one remaining important source of variation between measured and modeled Δ values will have been introduced by the treatment of the climatic input data. The diurnal courses that were calculated from daily weather information are particularly unreliable on cloudy days because the simulation of Q is based on a simple sine function. Likewise, our model did not include possible increased canopy photosynthesis on cloudy days, e.g., when the contribution of the diffuse part of Q might be noticeable.

[40] Accepting these uncertainties, we conclude that the model analysis here, based on plant physiological responses, can explain much of the long-term and short-term trends observed in Δ13Cc. The results of our analysis strongly suggest that these are indicative for long-term stomatal closure in response to increasing CO2 and D. The combination of reduced gs while long-term A remained unaffected points toward a more efficient use of water as evaporation rates declined. The observed long-term trend in Δ13Cc was accompanied by pronounced short-term fluctuations that reflected annual variations in precipitation, soil water availability, and air saturation deficit. Although the simulations could only be run for one of the five sampling locations, the similarity in the observed triannual variation in Δ13Cc over a 1000 km gradient suggests a common and plastic response of stomatal conductance, λ, and water use efficiency to changes in climate and atmospheric CO2 concentrations this century.

Appendix A:: Combined Cowan-Farquhar Model of Stomatal Conductance and Photosynthesis

[41] Cowan [1977] suggested that stomata should act to minimize the marginal water cost of plant carbon gain. In formal mathematical terms this requires

equation image

where E is the transpiration rate, A is the CO2 assimilation rate, and λ is a Lagrange multiplier that represents the marginal water cost of plant carbon gain. One can also write

equation image

where gs is the stomatal conductance. For individual leaves, (A2) provides a ready means of determination of λ as ∂E/∂gs and ∂A/∂gs can be determined from gas exchange measurements [Hall and Schulze, 1980; Lloyd, 1991]. The full equations are [Cowan, 1977]

equation image

where rs is the stomatal resistance (equal to 1/gs), rb is the boundary layer resistance, and

equation image
equation image

where Tl is the leaf temperature, L is the latent heat of vaporization of water, CP is the heat capacity of air, ∂w′/∂Tl is the slope of the curve relating the saturation vapor pressure of water to temperature, σ is the Stefan-Boltzmann constant, and Ci is the intercellular partial pressure of CO2 within the leaf.

[42] The rb and rb* terms in (A3)(A5) are necessary to account for the fact that stomatal conductance itself can influence leaf temperature. However, for aerodynamically rough needle leaf canopies, it can be expected that rb and rb* are relatively low. Making the assumption that rb and rb* can be ignored, as is effectively done in the simplified models of Lloyd [1991], Berninger and Hari [1993], and Lloyd et al. [1995], allows considerable simplification. Equation (A3) then becomes

equation image

where D is the leaf-to-air vapor pressure difference, (A4) is not required, and (A5) becomes

equation image

Combining (A2), (A6), and (A7) then gives

equation image

One can also write [e.g., Farquhar and Sharkey, 1982]

equation image

Combining (A8) and (A9) then gives

equation image

Rearranging

equation image

and substituting again with (A9) and with gs = 1/rs,

equation image

Equation (A12) can also be expressed as a quadratic equation with respect to Ci, i.e.,

equation image

where α = λ, β = 1.6 D − 2 Caλ, and γ = λ Ca2 − 1.6 DCa − 1.6 • DA/(∂A/∂Ci), which can be solved in the normal way.

[43] We need now expressions for A and ∂A/∂Ci . According to the model of Farquhar et al. [1980], the observed photosynthesis rate can be written as

equation image

where AV is the rate of CO2 assimilation limited by the potential rate of Rubisco carboxylation,

equation image

where Vmax is the maximum activity of the enzyme ribulose-1,5 bisphosphate carboxlase/oxygenase (Rubisco), Cc is the partial pressure of CO2 at the sites of carboxlation in the chloroplast, Γ* is the CO2 compensation point in the light, Kc is the Michaelis-Menten constant of Rubisco for CO2, O is the partial pressure of oxygen in the atmosphere, and Ko is the Michaelis-Menten constant of Rubisco for O2. In (A14), AJ is the rate of CO2 assimilation limited by the potential rate of ribulose-1,5 bisphosphate regeneration, usually considered to be limited by the rate of electron transport J, according to

equation image

The derivates of (A15) and (A16) with respect to Cc are

equation image
equation image

At this point, some discussion of use of Ci in (A5)(A13) versus Cc in (A15)(A18) is appropriate. This difference arises due to a finite resistance for the diffusion of CO2 from the substomatal cavity to the sites of carboxylation within the chloroplast. For woody species this resistance can be of a considerable magnitude [Lloyd et al., 1992] and is attributable to a diffusion resistance both through the intercellular airspaces and across the cell wall to the Rubisco catalytic sites [Syvertsen et al., 1995]. Cc may be as much as 8 Pa less than Ci with the magnitude of this dependence increasing with increasing with photosynthetic rate which has serious implications for modeled values of Δ13C (equation (4)).

[44] However, as in the scenarios here, the differences in modeled A and hence in CiCc are quite small over the last 150 years. Moreover, the Rubicso kinetic constants used here, originally from Harley et al. [1986], are usually applied for the (unstated) assumption that Ci = Cc. If an appreciable internal resistance is assumed, then the Rubisco kinetic constants must be modified accordingly [von Caemmerer et al., 1994]. Especially as the study here is concerned with long-term trends and short-term fluctuations in Δ13C rather than the absolute values (the latter being confounded in any case by the large but uncertain difference between the δ13C of cellulose versus that of bulk leaf carbon discussed in section 3.2), then especially as the absolute magnitude of the internal resistance in pine leaves is uncertain, the usual simple simplification of Ci = CC allows the substitution of either (A17) or (A18) into (A13)), giving for the Vmax case

equation image

where k′ = Kc (1 + O/Ko), or for the Jmax case,

equation image

respectively.

[45] It is noteworthy that for a given Ci, ∂A/∂Ci will differ in the Jmax and the Vmax cases, which for similar λ, results in dissimilar Ci/Ca (Figure 5). The model also predicts an increase in Ci/Ca at higher Ca, particularly so at high values of D. While over the range of Ca investigated in this study (e.g., ∼300–365 ppm), gs remains fairly constant, values decrease at low Ca (Figure 6). This pattern does not reflect the observed stomatal opening at low Ca even in the dark and is a general weakness of combined assimilation-conductance models where A is used to calculated gs (e.g., (8)). If in the model scenario, assimilation is held constant, the simulated gs displays the expected increase at low values of Ca.

Figure 5.

The dependence of modeled Ci on air saturation deficit D for the Jmax and the Vmax cases ((A19) and (A20)). Calculations were done for two atmospheric CO2 concentrations Ca with T = 20°C and Q = 1500 μmol m−2 s−1.

Figure 6.

The relationship of modeled stomatal conductance gs and CO2 concentration Ca. Calculations were done (1) with assimilation rate A and Ca varying (solid line) and (2) with A being held constant and only Ca varying (dotted line). In both cases, T = 20°C, Q = 1500 μmol m−2 s−1, and D = 5 mmol mol−1.

Table A1. Parameters of the Photosynthesis/Conductance Modela
ParameterDescriptionValue
  • a

    Parameter values were obtained from fitting the model to 5 fine days of canopy photosynthesis measurements that were made between May and July 1999 at Zotino. Where appropriate, rates are based on ground area. The remaining parameter values were derived from the literature (Table A2).

Vcmax20maximum rate of Rubisco at 20°C46.2 × 10−6 mol m−2 s−1
Jmax20light-saturated rate of electron transport at 20°C99.2 × 10−6 mol m−2 s−1
Rd20canopy respiration rate at 20°C0.7 × 10−6 mol m−2 s−1
λmarginal cost of plant carbon assimilation445.2 mol mol−1
Ejactivation energy for the Jmax temperature response55,990 J mol−1
Hjdeactivation energy for Jmax temperature response157,900 J mol−1
Sjentropy term in the Jmax temperature response525.7 J mol−1
Evactivation energy for Vcmax50,250 J mol−1
Table A2. Literature-Derived Parameter Values of the Photosynthesis/Conductance Modela
ParameterDescriptionValue
Γ*CO2 compensation point in the absence of dark respiration (at 20°C)34.6 × 10−6 mol mol−1
Γ0empirical parameter to describe the temperature response of Γ*0.0451
Γ1empirical parameter to describe the temperature response of Γ*0.000347
Kc20Michaelis Menten constant for carboxylation (at 20°C)302 × 10−6 mol mol−1
Ko20Michaelis Menten constant for oxygenation (at 20°C)256 × 10−3 mol mol−1
Oambient O2 concentration209 × 10−3 mol mol−1
Ecactivation energy for Kc59,430 J mol−1
Eoactivation energy for Ko36,000 J mol−1

Acknowledgments

[46] A. Arneth acknowledges support from a postdoctoral fellowship from the New Zealand Foundation for Research, Science, and Technology. Access to the sampling sites was granted with a boat provided by the Krasnoyarsk Forest Institute. R. Werner, M. Saurer, and C. Cook commented on cellulose extraction techniques. Isotopic abundances of ∼1500 cellulose samples and standards were meticulously analyzed by Beate Bruch and Stephan Bräunlich from the ISOLAB of MPI-BGC. We are thankful to Roger Francey for making the ice core data available prior to publication.

Ancillary