Examining the large-scale convergence of photosynthesis-weighted tree leaf temperatures through stable oxygen isotope analysis of multiple data sets

Authors


Author for correspondence:
Brent R. Helliker
Tel: +1 215 746 6217
Email: helliker@sas.upenn.edu

Summary

  • The idea that photosynthesis-weighted tree canopy leaf temperature (Tcanδ) can be resolved through analysis of oxygen isotope composition in tree wood cellulose (δ18Owc) has led to the observation of boreal-to-subtropical convergence of Tcanδ to c. 20°C.
  • To further assess the validity of the large-scale convergence of Tcanδ, we used the isotope approach to perform calculation of Tcanδ for independent δ18Owc data sets that have broad coverage of climates.
  • For the boreal-to-subtropical data sets, we found that the deviation of Tcanδ from the growing season temperature systemically increases with the decreasing mean annual temperature. Across the whole data sets we calculated a mean Tcanδ of 19.48°C and an SD of 2.05°C, while for the tropical data set, the mean Tcanδ was 26.40 ± 1.03°C, significantly higher than the boreal-to-subtropical mean.
  • Our study thus offers independent isotopic support for the concept that boreal-to-subtropical trees display conserved Tcanδ near 20°C. The isotopic analysis cannot distinguish between the possibility that leaf temperatures are generally elevated above ambient air temperatures in cooler environments and the possibility that leaf temperature equals air temperature, whereas the leaf/air temperature at which photosynthesis occurs has a weighted average of near 20°C in cooler environments. Future work will separate these potential explanations.

Introduction

Stable oxygen isotope ratios in plant cellulose (δ18Oc) have been found to be significantly related to a suite of climatic variables, such as air temperature (Libby et al., 1976; Burk & Stuiver, 1981; Rebetez et al., 2003), precipitation (Saurer et al., 1997; Robertson et al., 2001; Treydte et al., 2006), and relative humidity (RH; Ramesh et al., 1986; Roden & Ehleringer, 1999; Porter et al., 2009). Because of significant relationships with these variables in many scenarios, δ18Oc has attracted interest among paleobiologists as a tool for past climate reconstructions. Mechanistically, these relationships are either a reflection of the strong plant source water control over δ18Oc or a result of an RH effect on leaf water evaporative enrichment and the resultant δ18Oc signal (Roden et al., 2000). In addition to the influence of these abiotic factors, δ18Oc is also known to be subject to regulation by several biotic factors, among which leaf stomatal conductance is especially noteworthy because of the important physiological roles it plays in plant carbon and water dynamics (Barbour et al., 2000a; Scheidegger et al., 2000). It was by taking advantage of the intrinsic link between stomatal conductance and δ18Oc as predicted by isotope theory that researchers were able to expand the utility of δ18Oc to new areas, such as agriculture, where δ18Oc (or, more precisely, Δ18Οc, which denotes isotope enrichment above source water) has been proposed as an effective indicator of genotypic differences in gas exchange correlated traits such as crop yield potential (Barbour et al., 2000a; Cabrera-Bosquet et al., 2009, 2011), and ecology, where δ18Oc has the potential to be of use in evaluating the intensity of competitive interactions occurring in natural plant communities (Ramirez et al., 2009).

Traditional reconstruction studies based on δ18Oc have largely relied on empirical relationships established between δ18Oc and the target variables of reconstructive interest. Such an empirical approach is appealing because of its simplicity and minimal requirement for understanding the underlying processes driving the focal relationship (Sternberg, 2009). Nevertheless, there are clearly some limitations inherent in this approach, which may prevent the realization of the reconstructive potential of δ18Oc (McCarroll & Loader, 2004). For example, to generate an empirical relationship one needs to obtain a sufficient amount of empirical data for both δ18Oc and the target variable, which in many cases would be a challenging task as a result of either logistic constraints or the lack of instrumental measurement in the past. In these cases, the mechanistic modeling approach, which takes advantage of the establishment of a process-based plant δ18Oc model, becomes an alternative option. By allowing for direct calculation of the target variable through rearranging a mechanistic isotope model, the mechanistic modeling approach could circumvent the need for data to generate empirical relationships. The power of this approach was exemplified by Jahren & Sternberg (2003), who made use of the tree wood δ18Oc18Owc) model to reconstruct the amount of water vapor contained in the middle Eocene air in northern Canada from δ18Owc (in combination with hydrogen isotope composition in tree wood) excavated from a fossil forest; by Brooks & Coulombe (2009), who inverted the model to calculate the degree of changes in stomatal conductance in Douglas-fir (Pseudotsuga menziesii) trees in response to a long-term nitrogen treatment; and, more recently, by Kahmen et al. (2011) in their exploratory study that led to successful identification of leaf-to-air vapor pressure difference (VPD) as the single, most predominant environmental control on δ18Owc in tropical ecosystems.

Helliker & Richter (2008; hereafter H&R) recently developed a new application of the ‘mechanistic modeling approach’ by using δ18Owc to estimate tree canopy temperatures during the period of active photosynthesis. The basic principle of this application is related to the fact that leaf temperature exerts direct control over two components of the tree δ18Owc model, namely, the leaf saturation vapor pressure (ei) and the equilibrium fractionation factor (ε+). So as long as other components in the model are properly quantified (either by measurement or by modeling), photosynthesis-weighted leaf temperature (Tcanδ) can then be calculated from δ18Owc. Continuing this rationale, H&R applied this method to a data set of tree ring δ18Oc collected from 39 tree species at 25 different North American sites (Richter et al., 2008). Their calculation revealed a mean Tcanδ of 21.4 ± 2.2°C across the 50° of latitude that the entire data set encompasses, from subtropical to boreal biomes. This boreal-to-subtropical convergence of Tcanδ is an intriguing, and perhaps controversial pattern and it begs for both confirmation and an extension to larger data sets.

In the present study, we attempted to generally confirm the isotopic approach of H&R and to extend the overall analysis to high latitudes and the tropics. We used two tree δ18Owc data sets to assess if and how well the boreal-to-subtropical Tcanδ convergence pattern was supported by independent data sets that have even broader coverage of mean annual air temperatures (MAT) than those of H&R. We extended the analysis into the tropical biome, which was not represented in H&R, using a combination of new isotopic analysis and published data. As the calculation of Tcanδ via δ18Owc is reliant upon a mechanistic cellulose isotope model, in the calculation we addressed two aspects of the uncertainties that could potentially compromise the model accuracy, namely, the ‘source water’ uncertainty and the ‘Péclet effect’ uncertainty. In brief, for the ‘source water’ uncertainty we first evaluated the relevance of two well-established source water isotope models to our cellulose data sets, and then selected a model that has better predictive power; whereas to address the ‘Péclet effect’ uncertainty, we performed Tcanδ calculations under several hypothetical scenarios of different Péclet numbers, and then we assessed how error in this poorly constrained parameter would influence our calculation results.

Materials and Methods

Data set description

The majority of the results presented in this study are based on two data sets of δ18Owc from previously published papers. The first data set (hereafter referred to as the ‘Barbour data set’; Barbour et al., 2001) was a collection of 42 tree stem/twig samples from sites across the world, spanning a range in latitudes from 35.18°S to 69.67°N (see Fig. 1) and MATs from −6.2°C to 22.2°C. The second data set (hereafter referred to as the ‘Saurer data set’; Saurer et al., 2002) was a collection of 26 tree-ring δ18Owc samples from boreal conifers grown in Eurasia high-latitudinal regions (mostly between 60°N and 70°N). This data set covers a broad range in longitudes from 10.88°E to 167.67°E (see Fig. 1) and a range in MATs from −12.5°C to 5.1°C. The two data sets combined include 23 tree species from four tree genera: Quercus, Pinus, Larix, and Picea, and span boreal to subtropical biomes. The δ18Owc values in the Barbour and Saurer data sets are both multi-year averaged values; each δ18Owc in the Saurer data set is an integration of 30 yr of isotope signal during the time period of 1961–1990, and the δ18Owc values in the Barbour data set usually represent integrations of 3–10 yr in the 1990s. A full description of the data sets, together with the methodology for determining δ18Owc of the wood samples, can be found in Barbour et al. (2001) and Saurer et al. (2002).

Figure 1.

Distribution of the tree oxygen isotope composition in tree wood cellulose (δ18Owc) sample collection sites. Closed circles, open circles and triangles represent the sites of the Barbour, Saurer and tropical data sets, respectively.

As a separate assessment of tropical tree leaf temperatures, we compiled a relatively small, tropical tree δ18Owc data set. There are 10 samples (species–site combinations) in this data set, located in either Costa Rica or Southeast Asia (see Fig. 1 and Table 1 for more details); four of them were donated by a colleague and were collected from a blowdown site in the Area de Conservacion, Guanacaste, Costa Rica, and the data of the other six samples were obtained from the literature. For the samples donated by a colleague, cellulose was extracted from well-homogenized tree cores (one core per tree; three trees per species) following the Brendel procedure modified by the addition of a 17% NAOH step (wash) to remove hemicellulose (Brendel et al., 2000; Gaudinski et al., 2005). Cellulose samples of 90–100 μg were weighed into silver capsules and pyrolyzed at 1100°C in a Costech Elemental analyser (Costech Analytical Technologies, Valencia, CA, USA). The isotopic composition of the evolved CO gas was determined on a Thermo-Finnigan Delta Plus isotope ratio mass spectrometer (Thermo-Finnigan, Bremen, Germany), which had a measurement precision of < 0.23‰ on a standard reference cellulose powder. All samples were run in triplicate and data were reported on the Standard Mean Ocean Water (SMOW) scale.

Table 1.   Details of the tropical data set reported in this study
SpeciesCountryLat.Long.Elev. (m)Data source
Acrocromia aculeataCosta Rica10.84°N85.62°W295Colleague donation
Enterolobium cyclocarpumCosta Rica10.84°N85.62°W295Colleague donation
Gliricidia sepiumCosta Rica10.84°N85.62°W295Colleague donation
Guazuma ulmifoliaCosta Rica10.84°N85.62°W295Colleague donation
Cordia sp.Costa Rica10°N85°W300Evans & Schrag (2004)
Hyeronima alchorneoidesCosta Rica10.4°N84°W40Evans & Schrag (2004)
Tectona grandisIndonesia7.27°S111.33°E50Poussart et al. (2004)
Quercus kerriiThailand19.57°N98.27°E1100Poussart & Schrag (2005)
Podocarpus neriifoliusThailand18.3°N98.3°E700Poussart et al. (2004)
Podocarpus neriifoliusThailand18.3°N98.3°E750Poussart et al. (2004)

Because of the lack of visible ring structures in the tropical wood, we were unable to determine the exact time span over which each tropical sample is integrated. Some rough estimates based either on average tree wood increment rate or on high-resolution (sub-annual) isotopic measurement-derived annual cycles indicate that these samples represent integrations of 20–40 yr of cellulose formed mostly in the second half of the 20th century (one exception is the sample for Acrocromia aculeate, which lacks secondary xylem growth).

Theoretical basis for δ18Owc-based Tcanδ calculation

The tree wood δ18Owc can be described by the following general model (Barbour & Farquhar, 2000; Roden et al., 2000):

image( Eqn 1)

18Osw and δ18Olw, oxygen isotope ratios of source and leaf water, respectively.) The term pex represents the fraction of oxygen in the cellulose molecule that exchanges with water at the site of cellulose synthesis. In tree species, pex has generally been shown to have a nearly constant value ranging from 0.38 to 0.42, with a mean of 0.40 (Roden et al., 2000; Cernusak et al., 2005). One exception is the study of Gessler et al. (2009), who showed that pex in field-grown Scots pines (Pinus sylvestris) may display some variation at the intra-annual scale. However, as the present study deals with spatial variation in multi-year integrated δ18Οwc values, consideration of temporal variation in pex is neither warranted nor even realistic. The variable px represents the proportion of unenriched xylem water at the site of cellulose synthesis, which is very close to 1 for the stems of mature trees (Roden et al., 2000; Cernusak et al., 2005). εο is the biochemical fractionation factor associated with the exchange of oxygen atoms between carbonyl group and the tissue water. εο is treated as a constant of 27‰, regardless of plant type and growth temperature (Sternberg & Deniro, 1983; Sternberg, 1989; Yakir & Deniro, 1990; Roden & Ehleringer, 2000; Helliker & Ehleringer, 2002).

Leaf mesophyll water enrichment is determined by a balance of enriched water at the evaporative site and unenriched vein water in the leaf. This balance can be described by a Péclet effect (inline image) that accounts for the ratio of advection of unenriched vein water via transpiration stream to back-diffusion of the enriched water from the evaporative site (Farquhar & Lloyd, 1993):

image( Eqn 2)

18Οlw and Δ18Οes, enrichment of leaf mesophyll water above source water (approximated by δ18Οlw − δ18Οsw), and enrichment of water at the evaporative site above source water (approximated by δ18Oes − δ18Osw), respectively.)

image( Eqn 3)

(E, the transpiration rate (mol s−1 m−2); L, the scaled effective path length (m); C, the concentration of water (5.55 × 104 mol m−3); D, the diffusivity of heavy water in water (D = 119 × 10−9 exp(−637/(136.15 + T  )) m2 s−1); T, temperature (°C) (Cuntz et al., 2007)).

Δ18Οes is conventionally described by the steady-state Craig–Gordon model (Craig & Gordon, 1965; Flanagan et al., 1991; Farquhar & Lloyd, 1993)

image( Eqn 4)

18Οv, atmospheric water vapor relative to source water (approximated by δ18Ov − δ18Osw); ε+ and εΚ, the temperature-dependent equilibrium fractionation factor for the water evaporation and the cumulative kinetic fractionation factor of water vapor diffusing out of the leaf, respectively; ea/ei, the ratio of ambient water vapor pressure (ea) to the pressure of water vapor saturated at leaf temperature (ei).) For studies that focus on long-term average isotope values over broad spatial scales, it is reasonable to assume that the water vapor of the air is in isotopic equilibrium with source water (or Δ18Οv = − ε+) (West et al., 2008; Ogee et al., 2009), such that Eqn 4 can be rewritten as:

image( Eqn 5)

which implies that Δ18Οes is linearly dependent on inline image

Saturation water vapor pressure ei can be resolved through combining and rearranging Eqns 1, 2 and 5, such that,

image( Eqn 6)

Leaf temperature TL (°C) can then be obtained from ei (kPa), by rearranging a well-quantified saturated vapor pressure–temperature relationship (Buck, 1981):

image( Eqn 7)

Data sources and model parameterization

Monthly climate data required for model parameterization were obtained from high-resolution global climatological data sets. Monthly precipitation and mean, minimum and maximum air temperatures were obtained from WorldClim (http://www.worldclim.org) (Hijmans et al., 2005); monthly RH data were obtained from CRU CL 2.0 (New et al., 2002). Worldclim and CRU CL 2.0 are climate surfaces developed on the basis of world-wide weather station records of the 1950–2000 and 1961–1990 periods, respectively (New et al., 2002; Hijmans et al., 2005). We recognize that temporal variations are usually present in the δ18O values of tree rings formed in different years, presumably driven by temporal variations in local climates. The best way to avoid temporal asynchrony between δ18Owc and climate is to parameterize the isotope model with ‘customized’ climate data that have exactly the same temporal coverage as the corresponding sample δ18Owc. However, this was practically impossible to achieve in the current study, and therefore we assume that the multi-year averaged values characteristic of most of our δ18Owc samples are primarily subjected to the influence of the sample sites’ mean climatic conditions, which can in turn be well described by both Worldclim and CRU CL 2.0.

Monthly ea values were calculated as a product of monthly RH and saturated water vapor pressure at ambient air temperature (esat), where esat is determined from monthly mean air temperature using Eqn 7. As photosynthesis occurs during the day, daytime conditions are most relevant to isotope signals in tree cellulose, so we also derived monthly values of daytime air temperature (Tday) and RH (RHday). Tday was given by Tday = 0.67(max air T ) + 0.33(min air T ) (Barbour et al., 2002; Ferrio & Voltas, 2005); RHday is determined as the ratio of monthly ea over the monthly average of daytime esat, where esat is calculated using Eqn 7 parameterized with monthly Tday values.

We then averaged monthly data for the above-mentioned climatic variables into growing season values. In the context of this paper, growing season ambient RH (RHgs) and growing season ambient air temperature (Tgs) refer to daytime averages of ambient RH and air temperature across the growing season months, respectively. For non-Mediterranean-type sites in our data sets, we define the growing season months as those having monthly minimum air temperatures > 6°C. In the case of some cold sites where no single month has its minimum temperatures exceeding 6°C, July and August were taken as the growing season months. For Mediterranean-type sites, we used monthly precipitation as the filter criterion, and defined growing season months as the months that have precipitation amount > 15 mm. It is certainly possible that tree growth would occur outside the defined growing months. However, we emphasize that the defined growing season months are where the majority of tree growth takes place and therefore should dominate the isotope signature in tree cellulose.

Another essential step for leaf temperature calculation was to estimate the δ18O of the amount-weighted annual precipitation (δ18Oppt, a surrogate for δ18Osw). The assumption that δ18Osw is close to the long-term average of precipitation water δ18O has been shown to be generally valid in a number of previous studies via either direct δ18Osw − δ18Oppt comparison (Sternberg et al., 2007), or strong correlations found between δ18Owc and annual δ18Oppt (or its proxy, mean annual temperature), on regional or continental scales (Gray & Thompson, 1976; Burk & Stuiver, 1981; Richter et al., 2008). To estimate annual δ18Oppt, we initially tried out two δ18Oppt models that rely on different sets of independent prediction variables: the Bowen δ18Oppt model and the Barbour δ18Oppt model. The Bowen model is a grid-interpolated, isoscape-type model with its computing algorithm dependent upon geographic coordinates (latitude and altitude) as the prediction variables (Bowen & Wilkinson, 2002; Bowen & Revenaugh, 2003). We obtained the Bowen modeled δ18Oppt values from the online isotope calculator hosted at http://www.waterisotopes.org. The Barbour δ18Oppt model, however, bases its prediction more on climatic variables such as MAT and precipitation amount (Barbour et al., 2001). For the sites with annual precipitation < 1.4 m, the Barbour model gives the following prediction equation: inline image (Pa, the annual precipitation amount of the site.) Both of these two models have been shown to explain > 75% of the spatial isotopic variation at the global scale. Nevertheless, in the present study, we chose to use the Barbour modeled δ18Oppt values to parameterize the cellulose isotope model, for the reasons given in the discussions on δ18Oppt model selection.

Given the current lack of understanding of the species-dependent nature of L (and E to some extent), we did not specifically parameterize the Péclet effect for each of the sampled species. While there is evidence showing that the Péclet effect may be of limited importance in determining and hence modeling δ18Owc (Roden & Ehleringer, 1999; Ogee et al., 2009; B. R. Helliker et al., unpublished), several studies do include the offset to extract information from δ18Owc (Barbour et al., 2002; Evans, 2007; Brooks & Coulombe, 2009; Kahmen et al., 2011). Therefore, in our calculation we assumed two separate but constant Péclet effects as well as considering a ‘no Péclet effect’ scenario by allowing Δ18Οlw to be equal to Δ18Οes. For the constant Péclet effect, we performed leaf temperature calculations under two hypothetical scenarios where the proportional discrepancy between Δ18Οlw and Δ18Οes (hereafter referred as f, which is equal to inline image) according to Eqn 2) is assumed to be either 0.05 or 0.1, over the entire data sets. It should be noted that studies performed on crop species at the single leaf level often found f values to be above 0.1, and this was primarily attributed to the high transpiration rates of these species (Flanagan et al., 1991, 1994; Barbour et al., 2000b; Ripullone et al., 2008; also see discussion in Ferrio et al., 2009 and B. R. Helliker et al., unpublished). However, as far as tree cellulose isotope modeling is concerned, there are at least two reasons to expect f to be lower: tree leaves are known to have much lower inherent E than crop leaves (Schulze et al., 1994); and more relevant to the cellulose isotope signal is the canopy-scale E, which is expected to be lower than the commonly reported leaf-level E measured on sunlit leaves. Based on our estimate, a realistic allowance of canopy E to vary in the range of 0–1 mmol m−2 s−1 and L in the range of 0–30 mm led the f value to be within the small range of 0–0.12. This lends support to the treatment of f as a small value in models interpreting the δ18O of tree cellulose.

During the calculation, we first used growing season Tday to estimate ε+ following the well-established ε+−temperature relationship (Bottinga & Craig, 1969):

image( Eqn 8)

After solving the isotope model for ei and TL with this initial ε.+, we used the solved TL to determine a second value of ε.+, which was subsequently used for the second iterative determination of ei and TL. To arrive at final the TL we repeated this iteration four times (see H&R for more details). The final calculated TL was taken as Tcanδ in our study.

Results

The Barbour and Saurer data sets

Across all the Barbour sample sites, the Bowen modeled δ18Oppt was found to agree with the Barbour modeled δ18Oppt well, that is, the slope of the relationship between the Bowen and Barbour modeled δ18Oppt values was not significantly different from 1 and the intercept was not different from 0 (= 0.11 and 0.21, respectively; R2 = 0.89). Accordingly, when the measured δ18Owc in the Barbour data set was regressed against the Bowen and Barbour modeled δ18Oppt values, respectively, we found no difference between both slopes and intercepts of these two regression lines (= 0.32 and 0.83, respectively; Fig. 2a,c). For the Saurer data set, however, the slope of the regression line of the Bowen vs Barbour modeled δ18Oppt was significantly different from 1 and the intercept was significantly different from 0 (< 0.001 and < 0.001, respectively). Accordingly, for the Saurer data set the slopes and intercepts of the linear regression of δ18Owc against the Barbour modeled δ18Oppt significantly differed from those of δ18Owc against the Bowen modeled δ18Oppt (< 0.001 and < 0.001, respectively; Fig. 2b,d). Further, in the Saurer data set, the correlation coefficient for the δ18Owc−Barbour modeled δ18Oppt relationship was 0.83 (Fig. 2b), much higher than 0.41, the correlation coefficient for the δ18Owc−Bowen modeled δ18Oppt relationship (Fig. 2d). In the remainder of this article, ‘δ18Oppt’ without specification refers to the Barbour modeled δ18Oppt values, because we selected them against the Bowen modeled δ18Oppt values to parameterize the tree cellulose isotope model, for the reasons specified in the discussions of ‘δ18Oppt model selection’.

Figure 2.

The relationship between oxygen isotope composition in tree wood cellulose (δ18Owc) and the modeled oxygen isotope composition of the amount-weighted annual precipitation (δ18Oppt) for the Barbour data set (a, c) and the Saurer data set (b, d). δ18Oppt values in the top two panels are modeled using the Barbour δ18Oppt model, and δ18Oppt values in the bottom two panels are modeled using the Bowen δ18Oppt model. In (a), δ18Owc = 33.04 + 0.55δ18Oppt, R2 = 0.61, < 0.001. In (b), δ18Owc = 32.97 + 0.57δ18Oppt, R2 = 0.83, < 0.001. In (c), δ18Owc = 33.2 + 0.52δ18Oppt, R2 = 0.75, < 0.001. In (d), δ18Owc = 35.37 + 0.73δ18Oppt, R2 = 0.41, < 0.001.

Of the climatic factors that are relevant to tree cellulose isotope signals, MAT was significantly positively related to δ18Owc across the entire Barbour and Saurer data sets: an increase in δ18Owc of 0.30‰ corresponded to 1°C of increase in MAT (Fig. 3a). However, the correlation with MAT became negative when the impact of source water (or δ18Oppt) was removed by expressing oxygen isotope values of wood cellulose as Δ18Οwc (Fig. 3b; R2 = 0.73, < 0.001). Across the entire MAT range from −15.17 to 22.20°C, Δ18Οwc fell in the range of 33.7–44.9‰.

Figure 3.

The relationship between (a) the oxygen isotope composition in tree wood cellulose (δ18Owc) and mean annual temperature (MAT); and (b) Δ18Οwc and MAT. The Saurer samples are represented by open circles and the Barbour samples are represented by closed circles. In (a), δ18Owc = 25.53 + 0.30MAT, R2 = 0.83, < 0.0001. In (b), δ = 38.69 − 0.23MAT, R2 = 0.73, < 0.001.

For growing season ambient RH (RHgs), we found a negative relationship with δ18Owc (= 0.04), but a positive relationship with Δ18Οwc (= 0.04). After resolving leaf-based RH (RHleaf, or the term ‘ea/ei’ in Eqn 5) from Δ18Οwc following Eqns 1–5, we plotted both the resolved RHleaf and RHgs against MAT, only to find that RHgs values substantially deviated from the RHleaf values in many of the sample sites in our data sets (Fig. 4). Specifically, RHleaf strongly positively trended with the MAT gradient, with its value being as low as 31% at the low-MAT end and as high as 72.75% at the high-MAT end. By contrast, RHgs negatively trended with MAT, with its values being as high as above c. 70% in many of the low-MAT sites (Fig. 4).

Figure 4.

The patterns of relative humidity (RH) resolved through isotopes (RHleaf; circles) and of the growing season ambient RH (RHgs; triangles) following the mean annual temperature (MAT) gradient. The Saurer samples are represented by open symbols and the Barbour samples are represented by closed symbols. For RHgs: RHgs = 64.58 − 0.25MAT, R2 = 0.08, = 0.01; for RHleaf: RHleaf = 49.98 + 0.95MAT, R2 = 0.73, < 0.0001.

Based on the resolved RHleaf values, we further calculated Tcanδ. When Tcanδ (calculated under the ‘= 0.05’ senario) was expressed as canopy-growing season temperature deviation (Tcanδ minus Tgs, or Tdiff), it became clear that Tdiff significantly and negatively trended with MAT, regardless of whether the circled outliers were included or not for data analysis (for all data points, Tdiff = 4.57 – 0.41MAT; R² = 0.66 and < 0.0001; with the outliers excluded in the analysis, Tdiff = 4.75 – 0.41MAT; R² = 0.75, < 0.0001; Fig. 5a). A significant negative relationship was also found between Tdiff and growing season temperature; including the outliers led to Tdiff = 18.27 – 1.05Tgs; R2 = 0.50, < 0.0001, while excluding the outliers led to Tdiff = 20.08 – 1.16Tgs; R2 = 0.75, < 0.0001 (Fig. 5b). These relationships reveal that, toward the cold extreme of the Barbour and Saurer sample sites, tree canopy temperature can be well above the growing season daytime temperatures by > 10°C, whereas toward warmer climates the tree canopy temperature switched from being above to being slightly below growing season daytime temperatures.

Figure 5.

The relationship between photosynthesis-weighted tree canopy leaf temperature (Tcanδ) minus growing season ambient air temperature (Tgs) (or Tdiff) and (a) mean annual temperature (MAT) and (b) Tgs. The Saurer samples are represented by open circles and the Barbour samples are represented by closed circles. In (a) including the circled outliers leads to Tdiff = 4.57 − 0.41MAT, R2 = 0.66, < 0.0001; excluding the circled outliers leads to Tdiff = 4.75 − 0.41MAT as shown by the regression line, R2 = 0.75, < 0.0001. In (b) including the circled outliers leads to Tdiff = 18.27 − 1.05Tgs, R2 = 0.50, < 0.0001; excluding the circled outliers leads to Tdiff = 20.08 − 1.16Tgs as shown by the regression line, R2 = 0.75, < 0.0001. Note that f is assumed to be 0.05 in the calculation.

A plot of Tcanδ against MAT showed a significant relationship between these two variables, with the regression slope being negative (for all data points, slope = −0.17, R2 = 0.22, < 0.001; in the case of outliers being excluded, slope = −0.11, R2 = 0.23, < 0.05; Fig. 6a). However, whether or not the outliers were considered, we found no significant relationship between Tcanδ and Tgs. The mean Tcanδ from the Barbour and Saurer data sets was 19.35 ± 3.28°C, or 19.48 ± 2.05°C after the outliers as marked in Fig. 6 were removed. Introducing a larger Péclet effect into the model led to a slightly higher mean Tcanδ and a slightly larger SD, that is, setting = 0.1 raised the calculated mean Tcanδ by 0.96°C and SD by 0.38°C (Table 2); by contrast, a ‘no Péclet effect’ case slightly decreased the mean and SD, by 0.80°C and 0.19, respectively. For comparison purposes, we also applied the same calculation procedure to the H&R data set; the results show that mean Tcanδ calculated from H&R was c. 1°C higher than that calculated from our data sets (Table 2).

Figure 6.

Photosynthesis-weighted tree canopy leaf temperature (Tcanδ) over (a) the mean annual temperature (MAT) gradient and (b) the growing season air temperature (Tgs) gradient. The Saurer samples are represented by open circles and the Barbour samples are represented by closed circles. In (a) including the circled outliers leads to Tcanδ = 19.93 − 0.17MAT, R= 0.22, < 0.001; excluding the circled outliers leads to Tcanδ = 19.90 − 0.11MAT, R2 = 0.23, < 0.05. In (b), including the circled outliers leads to leaf T = 19.22 + 0.1Tgs, R2 < 0.001, = 0.95; excluding the circled outliers leads to leaf T = 20.91 − 0.10Tgs, R2 < 0.01, = 0.26. The dashed line enclosed region indicates the mean and SD of the leaf temperature of the data points after excluding the circled outliers (19.48 ± 2.05°C). Note that f is assumed to be 0.05 in the calculation.

Table 2.   Mean and SD of photosynthesis-weighted tree canopy leaf temperature (Tcanδ) for all species–site combinations after excluding the outliers (as circled in Fig. 5), under three scenarios of the Péclet effect as described by f, the proportional discrepancy between the enrichment of leaf mesophyll water above source water (Δ18Οlw) and the enrichment of water at the evaporative site above source water (Δ18Οes)
Data sourceTcanδ ± 1 SD (°C)
= 0= 0.05= 0.1
Barbour and Saurer data sets18.68 ± 1.8619.48 ± 2.0520.44 ± 2.43
Helliker and Richter (H&R) data set19.53 ± 1.8120.44 ± 1.9521.51 ± 2.52

The tropical data set

Given the fact that the Barbour δ18Oppt model is only applicable to the sites with mean annual precipitation amount < 1400 cm (Barbour et al., 2001), but that all of our tropical sample sites have average annual precipitation amounts > 1400 cm, we therefore used the Bowen δ18Oppt model to parameterize δ18Osw in our tropical Tcanδ calculation. Tcanδ values calculated from the tropical cellulose isotope data set were plotted against mean growing season temperatures (Fig. 7). When ‘= 0.05’ was assumed in the calculation, mean Tcanδ of all data points in this biome was 26.40 ± 1.03°C, significantly higher than the calculated mean of the Barbour and Saurer data sets (< 0.0001).

Figure 7.

Photosynthesis-weighted tree canopy leaf temperature (Tcanδ) calculated from the oxygen isotope composition (δ18O) of tree wood cellulose of tropical tree species collected from Costa Rica, Thailand, and Indonesia. f is assumed to be 0.05 in the calculation. The gray region represents the ‘19.48 ± 2.05°C’ area derived from the Barbour and Saurer data sets.

Discussion

δ18Oppt model selection

We found that the Bowen and Barbour δ18Oppt models gave very similar δ18Oppt values for the Barbour sample sites; however, for the Saurer sample sites, δ18Oppt values calculated by these two models did not agree. This highlights the necessity of selecting a model that has more predictive power for the Saurer sample sites.

As is well known, air temperature is the primary driver of the spatial δ18Oppt variations in mid- and high-latitude regions, through its influence on Rayleigh distillation (Dansgaard, 1964). Yet, without incorporating air temperature as a prediction variable, the Bowen δ18Oppt model inherently takes latitude (together with altitude) as an approximation to air temperature. However, for the Saurer data set in our study, it was longitude and not latitude that accounted for most of the across-site variation in temperature (data not shown). This is apparently inconsistent with the underlying assumption of the Bowen model. Several of the Sauer sample sites are geographically close to Siberia Network of Isotopes in Precipitation (SNIP) stations (Kurita et al., 2004), allowing direct comparisons to be made between the models and SNIP observations. With this approach, we found that the Bowen model had a tendency to overestimate annual δ18Oppt, by up to c. 5‰ at some sites, while the Barbour model much more closely reflected the SNIP observations (the model–observation difference was less than c. 1‰ at all of these sites). Further, in the Saurer data set, the correlation coefficient for the relationship between δ18Owc and the Bowen modeled δ18Oppt was only 0.41, much less than either that of the relationship between δ18Owc and the Barbour modeled δ18Oppt (0.83), or what has been reported in most other studies for the δ18Owc−δ18Oppt relationships at the geographic scale (Fig. 2b,d; Burk & Stuiver, 1981; Sternberg et al., 2007; Richter et al., 2008). Given the fact that δ18Oppt has been demonstrated to exert a strong control over δ18Owc, the weak δ18Owc−Bowen modeled δ18Oppt correlation may serve as indirect evidence that the Bowen model is less of a reflection of the real δ18Oppt values of the Saurer sites than the Barbour δ18Oppt model. For these reasons, the Barbour δ18Oppt model should be of better predictive power for our data sets, and hence was chosen to parameterize the cellulose isotope model in our calculation. Our analysis may also suggest that multi-year analysis of δ18Owc across broad spatial scales could be, in turn, a way to constrain model predictions of δ18Oppt.

The relative humidity record in Δ18Οwc

The Barbour and Saurer data sets showed strong positive relationship between δ18Owc and MAT, which is consistent with the common expectation that MAT has a strong control on δ18Owc at the spatial scale (Burk & Stuiver, 1981; Saurer et al., 2002; Richter et al., 2008), and thus provides empirical evidence indicating that using δ18Owc to reconstruct air temperature is reasonably well grounded. In addition to MAT, we found that the correlation of δ18Owc with RHgs was also significant, but negative. This observation has two possible explanations. One possibility stems from the fact that across the entire data sets RHgs co-varies with MAT in a negative way (i.e. colder climates tend to have ambient water vapor pressures that are closer to saturation; see Fig. 4), such that the observed negative δ18Owc−RHgs correlation may well be an artefact of the observed positive δ18Owc−MAT relationship. The second possible explanation is that growing season RH signal is directly recorded in δ18Owc, as a result of the significant role it plays in controlling the evaporative enrichment of leaf water. However, if this explanation holds true, it is to be expected from the isotope theory that Δ18Οlw, an expression that highlights the enrichment process by removing δ18Owc and the associated MAT variations, must be negatively trending with RHgs in a more clear-cut manner. However, contrary to this expectation, we found a positive trend between Δ18Οlw and RHgs, thereby ruling out this second explanation. More importantly, this inconsistency with the isotope theory suggests that RHgs may not be the actual control over season-long integrated isotopic enrichment of leaf water (or wood cellulose). This suggestion was supported when the δ18Owc values were used to solve the tree cellulose oxygen isotope model for RH. As is revealed in Fig. 4, the resolved RHleaf (the actual RH signal controlling growing season integrated isotope enrichment) followed along the MAT gradient a very distinct pattern when compared with RHgs. Notably, at the low extreme of the MAT range, the deviation (or more precisely, decrease) of RHleaf from the RHgs could be as large as > 40%. As was shown by H&R – and further shown by Kahmen et al. (2011) via leaf-to-atmosphere vapor pressure deficit – these substantial RH deviations therefore suggest that caution is warranted regarding simple interpretation of ambient RH signals from isotope analysis of tree wood cellulose, especially in cold climates.

Large-scale Tcanδ patterns and possible mechanisms

Tcanδ deviated considerably from Tgs in many of the Barbour and Saurer sample sites. The negative correlation of Tcanδ minus Tgs with MAT (Fig. 5a) is similar to the results of H&R. More specifically, our results resemble those of H&R in that Tcanδ was > 10°C elevated from the Tgs in boreal forest sites where MAT is extremely low (below −10°C), and that Tcanδ values approached their growing season counterparts in the temperate biome where MAT is c. 10 to 15°C. As MAT increased toward values more characteristic of hot temperature or subtropical biomes, Tcanδ was lower than Tgs in the current study. This result is slightly different from that of H&R, where subtropical canopy and air temperatures were relatively equal.

Across the Barbour and Saurer data sets, Tcanδ remained within a narrow range irrespective of the Péclet uncertainty (Table 2). As such, this study offers independent isotopic support for the concept that trees from boreal to subtropical biomes display conserved Tcanδ near 20°C. The small offset (c. 1°C) seen between the mean Tcanδ of the Barbour and Saurer data sets and the H&R data set indicates (Table 2) that inter-laboratory variations may be present in the methods employed in determining δ18Owc. Regarding the several outliers circled in Fig. 6, it should be noted that the two outlier sites (Mexico and Idaho sites) that have low calculated Tcanδ values are both located in mountainous regions, hinting at the possibility of common causes. First, it is certainly possible that the Tcanδ values are correct and most photosynthesis occurs at much cooler leaf temperatures at these sites. Second, it is possible that our calculated Tcanδ values underestimate the real Tcanδ, presumably as a result of a violation of our assumption that δ18Osw is equal to δ18Oppt in these mountainous regions, where isotopically depleted water (i.e. snow-melt water) from higher altitudes may mix with local precipitation water. We are currently working at a montane site in Colorado, USA to clarify these discrepancies.

While our results suggest a boreal-to-subtropical convergence of Tcanδ toward c. 20°C, there are different, yet not necessarily mutually exclusive, interpretations of what this implies: (1) Tcanδ is c. 20°C across biomes because of leaf temperature ‘control’ and true temperature homeostasis; or (2) leaf temperature on average equals air temperature, and leaf (or air) temperature at which photosynthesis occurs has a weighted average of c. 20°C. As for interpretation (1), it is a well-recognized fact that trees can regulate their leaf temperatures away from the ambient through a combination of physiological and morphological controls (Hadley & Smith, 1987; Smith & Carter, 1988; Leuzinger & Korner, 2007; Smith & Hughes, 2009). As a result of these controls, in cold climates leaf temperatures tend to be warmer than ambient air and the reverse is true in warmer climates (Grace, 1987; Smith & Carter, 1988; Gates, 2003). This is at least seemingly compatible with our observed mean climate-dependent leaf–air temperature relationship during the growing season (Fig. 5b). Alternatively, contrasting patterns of diurnal and seasonal photosynthesis across biomes could, upon being recorded in δ18Owc, look like leaf homeothermy and support interpretation (2) above. Take the cold, boreal sites as an example: marked elevations of Tcanδ from Tgs may mean that there is strong leaf temperature ‘control’ in operation, raising leaf temperature well above the ambient during photosynthesis (interpretation (1)); however, if the common assumption is made that boreal forests have aerodynamically well-coupled canopies (Jarvis & Mcnaughton, 1986), the elevation of Tcanδ from the ambient could then be an indirect result of dynamic diurnal/seasonal photosynthesis patterns as stated in our interpretation (2), that is, as a result of temperature being a major limiting factor in boreal forest, the majority of photosynthetic activities in a boreal tree may largely be confined to the warm part of the day (i.e. midday) or the warm period of the growing season, which leads to photosynthesis-weighted leaf temperature being higher than Tgs (or correspondingly photosynthesis-weighted RH being lower than RHgs). What makes this interpretation plausible is the common observation that summer midday ambient RH values of Siberia boreal forests can frequently reach values as low as 30 to 40% (Zimmermann et al., 2000; Ohta et al., 2001), in general consistence with the low δ18Owc-resolved RH values from many of our Siberia samples (Fig. 4).

The two points above are not mutually exclusive, and our isotope-based approach provides no clear evidence regarding which of these two mechanisms is more accountable for the observed boreal-to-subtropical convergence of Tcanδ. The relative importance of each of the mechanisms may well be biome- and/or species-dependent. To tease these mechanisms apart, a more direct approach would be required. By overcoming the conventional logistic constraints, such an approach must involve the continuous measurement of tree canopy leaf temperature and ambient air temperature, combined with simultaneous estimates of photosynthetic rates (i.e. using tower-based eddy covariance measurements) throughout the entire growing season. This is the direction of our future work.

Canopy Tcanδ in tropical regions

The reconstruction of Tcanδ using tropical tree cellulose shows that Tcanδ in the tropical biome is significantly elevated from 20°C, and quite close to Tgs (Fig. 7). This is in general agreement with the argument that tree canopy temperatures in hot regions should not be significantly lower than growing season temperatures, as was made by Woodward (2008) based on his observations of maximum rates of canopy transpiration and energy balance considerations. Tropical tree canopies would need to maintain temperatures 4–5°C below ambient temperatures when most carbon uptake occurs to maintain canopy temperatures near 20°C, presumably through transpirative cooling. However, tropical biomes are characterized by high ambient RH as well as high dew-point temperatures and these physical constraints would serve to prevent significant transpirative cooling.

Evaluating calculation errors associated with the steady-state assumption

An assumption inherent in our calculation procedure is that δ18Olw is at isotopic steady state (SS) during the day, when active photosynthesis occurs. We note that potential errors could be introduced by this simplistic SS assumption, because daytime δ18Olw may not be realistically close to the SS in field conditions, especially when leaf water has a slow turnover rate, or when the surrounding environment undergoes drastic change (Farquhar & Cernusak, 2005; Pendall et al., 2005; Seibt et al., 2006). Alternatively, to predict δ18Olw (and the resultant δ18Owc) more accurately a nonsteady-state (NSS) isotope model has to be applied. In order to evaluate to what extent the calculation error may inhere in our simplistic SS assumption, we performed a modeling study to estimate the difference between the SS and NSS isotope model calculated δ18Owc, across the biomes/tree types (see Supporting Information, Methods S1 for more details). The results show that for coniferous trees δ18Owc values modeled by the SS model were more enriched than δ18Owc values modeled by the NSS approach, by 0.8 and 0.6‰ for the boreal and temperate biomes, respectively; while for broad-leaved trees we did not find a significant difference between the SS and NSS model calculated δ18Owc values, across the biomes. We then used these biome/tree type-specific offsets between the SS and NSS modeled δ18Owc values (wherever applicable) to correct the sample δ18Owc values in our data sets (subtropical samples were treated as temperate samples in this case), and after this correction we recalculated a Tcanδ of 20.41 ± 2.44°C for the Barbour and Saurer data sets, which is quite comparable to the Tcanδ of 19.48 ± 2.05°C as originally calculated based on uncorrected δ18Owc values. In addition, we found no SS–NSS difference in the calculated Tcanδ for our tropical data set. These results therefore indicate that the SS assumption inherent in our Tcanδ calculation is reasonable.

Conclusions

Despite the uncertainties/simplistic assumptions associated with the calculation procedure, our isotope analysis of the multiple δ18Owc data sets has provided strong evidence that trees from the boreal to subtropical biomes display conserved Tcanδ of c. 20°C, and that the tropical biome is an exception, with Tcanδ much higher than 20°C and close to Tgs. What accounts for the observed Tcanδ convergence pattern is currently unknown, and further studies are needed to tease apart the plausible explanations.

Acknowledgements

We thank Dan Janzen for retrieving tree cross-section samples from Area de Conservación Guanacaste, Costa Rica, Yun Ouyan for contributions to Fig. 1, and Dustin Bronson for valuable discussions. We also thank the three anonymous reviewers for their constructive comments. This work was supported by the National Science Foundation under award number IOS-0950998 (to B.R.H.).

Ancillary