Do changes in tree‐ring δ¹⁸O indicate changes in stomatal conductance?

Abstract

Plant scientists have long been fascinated by the balance between photosynthetic carbon gain and transpirational water loss in terrestrial plants, because such a balance can have wide-ranging implications with respect to plant productivity and survival, ecosystem service and function, and even global-scale hydrology and climate forcing (Cowan & Farquhar, 1977; Bonan, 2008; Keenan et al., 2013). Intrinsic water use efficiency (iWUE, defined as the ratio of photosynthetic rate, A, to stomatal conductance, gs) is a frequently used trait to assess plant carbon–water balance. Due to the well-established, theoretical link between iWUE and photosynthetic carbon isotope discrimination (Farquhar et al., 1982), analysis of plant carbon isotope composition (δ13Cplant) has emerged as a popular method of estimating time-integrated iWUE during plant growth. However, the carbon isotope technique alone does not allow partitioning of iWUE between its component gas-exchange parameters (i.e. A and gs), and so is deemed to be insufficiently informative for a mechanistic understanding of iWUE responses to abiotic/biotic factors (Scheidegger et al., 2000; Farquhar et al., 2007). To address this limitation, researchers have suggested adding plant oxygen isotope composition (δ18O) into the isotope toolbox to help tease apart the influences of A vs gs on δ13Cplant-derived iWUE (Scheidegger et al., 2000; Grams et al., 2007). This suggestion stemmed from the theoretical prediction that, for plants exposed to similar environmental conditions, gs should exhibit a negative relationship with transpirative enrichment of leaf lamina water (Δ18Olw), and hence with the isotope composition of cellulose (δ18Ocel), which is known to preserve much of the Δ18Olw signal (Barbour et al., 2000; Scheidegger et al., 2000; Farquhar et al., 2007). In this case ‘similar environmental conditions’ refers primarily to similar δ18O of source water (δ18Osw) and ambient relative humidity (RH), as these are two important controlling factors of δ18Ocel. The predicted negative association of gs with δ18Ocel has gained support from the results of several glasshouse and field tests on different plant species (Barbour et al., 2000; Barbour & Farquhar, 2000; Scheidegger et al., 2000; Sullivan & Welker, 2006; Grams et al., 2007; Moreno-Gutiérrez et al., 2011), and consequently has resulted in the use of δ18Ocel as a proxy for gs to aid interpretation of physiological underpinnings of iWUE in numerous investigations (Siegwolf et al., 2021). While the early glasshouse/field examinations were conducted in ways to satisfy the ‘similar growth environment’ prerequisite of the theoretical negative gs–δ18Ocel relationship, an increasing number of 18O applications have instead focused on tree-ring cellulose in a temporal context, to gain retrospective insights into tree canopy gs responses to global change factors (Nock et al., 2011; Barnard et al., 2012; Fajardo et al., 2019; Guerrieri et al., 2019; Savard et al., 2020; Mathias & Thomas, 2021). Field-grown trees commonly experience pronounced interannual variations in δ18Osw and/or RH, so tree-ring cellulose δ18O (δ18Otr-cel)-based gs reconstructions do not necessarily conform to the ‘similar growth environment’ prerequisite. Several authors have thus urged caution over this type of temporal-context application (Roden & Siegwolf, 2012; Gessler et al., 2018), with Roden & Siegwolf (2012) explicitly highlighting several caveats/complexities that require attention, and recommending that one should rigorously account for the confounding isotopic and environmental variations either through measurement or modeling, if the goal is to assess the physiological responses that influence the measured δ18Otr-cel chronology. In this context, it is important to note that Guerrieri et al. (2019) recently developed a method to take care of the confounding variations in their attempt to infer temporal changes in gs from δ18Otr-cel. The method involves converting δ18Otr-cel to Δ18Olw following a three-step process: first, by assuming the local, current-year precipitation as the dominant source for tree source water, the δ18Osw of each year is simulated from climatic factors of the same year using a globally derived empirical model of precipitation δ18O (δ18Oprep; Barbour et al., 2001); second, tree-ring cellulose 18O enrichment above the source water (Δ18Otr-cel) is then calculated from the simulated δ18Oprep and the measured δ18Otr-cel; and third, the resultant Δ18Otr-cel chronology is subsequently used to derive a Δ18Olw chronology by inverting the well-established mechanistic model of Δ18Otr-cel (Sternberg, 2009), with potential climatic sensitivity of the model parameters (i.e. carbonyl oxygen exchange rate or pex) explicitly considered. Specifically, Guerrieri et al. (2019) suggested that one should check for interannual variation in RH, a potential driver of temporal variation in Δ18Olw. In cases where RH does not display an apparent trend during the time span encompassed by the δ18Otr-cel chronology, a negative gs–Δ18Olw relationship can be safely used to infer historical changes of gs from the derived Δ18Olw chronology. That is, when ‘no systematic RH trend’ is detected, a significant correlation between the derived Δ18Olw and time suggests a significant change in gs over time (the direction of which is opposite in sign from the calculated correlation coefficient or regression slope), while an insignificant correlation suggests gs remained relatively constant. By applying the above method to 12 tree species from eight forested sites in North America, Guerrieri et al. (2019) found that reductions in gs over the last 30 years were restricted to species from xeric sites. This result, combined with the tree-ring δ13C data showing a consistent increase in iWUE in response to rising atmospheric CO2 across all the sites, prompted the authors to reject the hypothesis that rising iWUE in forests ‘is primarily caused by the widespread, CO2-induced reductions in gs’ (Guerrieri et al., 2019). More recently, this method was also used by Mathias & Thomas (2021) in a meta-analysis of tree-ring isotope chronologies from 34 species across 84 sites globally. The results indicated no significant historical changes in gs over the past century in a majority (77.9%) of the tree–site combinations analyzed, again ‘challenging the idea of widespread reductions in gs as the major driver of increasing tree iWUE’ in response to past atmospheric CO2 increases (Mathias & Thomas, 2021). However, is the method proposed by Guerrieri et al. (2019) robust enough to guarantee a reliable inference of temporal directionality of changes in gs from δ18Otr-cel? Here, by examining this method in detail we highlight three aspects of insufficiency, each of which in our opinion is severe enough to render this method to be ineffective, as far as a δ18Otr-cel-based gs reconstruction is concerned. First, in Guerrieri et al. (2019) and Mathias & Thomas (2021), temporal changes in δ18Osw at each site were simulated using a spatial δ18Oprep model developed empirically by Barbour et al. (2001; hereafter the Barbour model). Such a ‘space-for-time’ substitution assumed that the environmental influences on δ18Oprep were the same across space and through time. However, it has long been recognized that the drivers of temporal variability in δ18Oprep can vary from site to site (Dansgaard, 1964; Gat et al., 2001). With this concern in mind, we tested the ability of the Barbour model for predicting the δ18Oprep variability at the site level using the δ18Oprep datasets archived in the Global Network of Isotopes in Precipitation (GNIP; https://www.iaea.org/services/networks/gnip). It turned out that only one (Ottawa, Canada) out of the eight selected sites displayed a significantly positive relationship between the simulated and the observed interannual variations in δ18Oprep, and the slope coefficients for the regression of the simulated δ18Oprep on the observed δ18Oprep were all far from unity (Fig. 1). This analysis (see Fig. 1 legend for more details) revealed that the Barbour model, for the most part, performed poorly in capturing the site-specific, temporal dynamics of δ18Oprep. As accurate prediction of δ18Oprep is critical in determining Δ18Otr-cell and/or Δ18Olw from δ18Otr-cell, the use of an unconfirmed spatial model is not appropriate for producing trustworthy time series of Δ18Otr-cel and/or Δ18Olw at each site. Interestingly, this point is evident in Guerrieri et al.'s (2019) own data (fig. S9 from Guerrieri et al., 2019), showing that the δ18Otr-cel-derived Δ18Olw could be quite different (i.e. by up to 5‰) from the actual measurement of Δ18Olw. While predicting large-scale spatiotemporal δ18Oprep is challenging (Terzer-Wassmuth et al., 2021), better options such as IsoMAP (Bowen et al., 2014) and IsoGSM (Yoshimura et al., 2008) do exist. Notably, Nelson et al. (2021) recently developed a machine learning approach to simulate temporal (intra and interannual resolution) δ18Oprep variations in Europe, and their simulations were substantially better than the existing models. We expect this promising modeling framework to be readily extended to other regions and deliver reliable estimates of δ18Oprep for retrospective research in tree-ring stable isotopes. Further, due to the complex mechanisms regulating soil water available for plant uptake (Allen et al., 2019; Goldsmith et al., 2022), the aforementioned assumption that local, current-year precipitation is the major source of tree source water might not always be met. If this is the case, isotope-enabled hydrological models can be helpful in capturing species-specific dynamics of δ18Osw at high temporal resolutions (Ogée et al., 2009; Brinkmann et al., 2018). Second, in their attempts to infer the directional changes in gs from the δ18Otr-cel-derived Δ18Olw chronology, both Guerrieri et al. (2019) and Mathias & Thomas (2021) considered the negative gs–Δ18Olw relationship to be universally applicable to all species included in their studies. The negative gs–Δ18Olw relationship was derived from the theoretical analyses of how gs modifies several isotopic/physiological parameters imbedded in the widely used Δ18Olw model that incorporates a Péclet effect (Sullivan & Welker, 2006; Grams et al., 2007; Roden & Siegwolf, 2012). Here, note that for plant cellulose-related isotopic signatures, the term Péclet effect implicitly refers to leaf lamina-associated radial Péclet effect, which by definition represents the relative balance between advection of unenriched leaf vein water via the transpiration stream and back diffusion of the enriched water from the evaporative sites in the leaf lamina (Farquhar & Lloyd, 1993). It is worth emphasizing that at the time when the gs–Δ18Olw negative relationship was theoretically inferred, it was conveniently assumed that: (1) the Péclet effect is pervasively present in different species, and (2) a key component governing the Péclet effect − known as the effective pathlength of water movement inside the leaf lamina or L − is a species-specific constant (Barbour & Farquhar, 2004; Kahmen et al., 2009). Under these assumptions, it can be predicted that Péclet-mediated isotope dynamics is a viable mechanism driving the negative gs–Δ18Olw association, as, for example, an increase in gs and/or transpiration rate (E) is predicted to cause a concomitant increase in the advective flux of unenriched xylem water to the leaf lamina, consequently leading to a decrease in Δ18Olw (Guerrieri et al., 2019). However, more recent studies have yielded experimental evidence that is inconsistent with either of the above two Péclet-related assumptions. For example, in their experimental survey of 27 species, Barbour et al. (2021) found that the Péclet effect was present in only 10 (37%) species; this indicates that the Péclet effect should not be taken for granted for all plant species (Song et al., 2015; Holloway-Phillips et al., 2016), thereby invalidating assumption 1. Furthermore, even for species in which the Péclet effect is operational, evidence has been mounting that L within a given species is variable and related to leaf transpiration rate (Ferrio et al., 2012; Song et al., 2013; Roden et al., 2015; but see Kahmen et al., 2009 for an exception), meaning that the ‘species-specific L’ assumption (assumption 2) is not necessarily valid. To examine the dynamics of Δ18Olw in response to gs with the updated knowledge related to the two assumptions, we performed a simple modeling exercise under three scenarios: no Péclet effect, a conventional Péclet effect (hereafter ‘Péclet effect with a constant L’) and a Péclet effect with a variable L based on an empirically determined L–E negative power relationship (Song et al., 2013). Given the annual resolution and integration of whole-canopy activities of tree-ring isotopic signatures, an isotopic steady state was assumed during our Δ18Olw modeling, similar to Roden & Siegwolf (2012). For demonstration purposes, we considered typical growing season environmental conditions for temperate trees (as detailed in the legend to Fig. 2) and simulated leaf temperature (Tl) at a given gs following Campbell & Norman (1998) before simulation of Δ18Olw (refer to Supporting Information Methods S1 for simulation details). The modeling results revealed a much-dampened sensitivity of Δ18Olw to gs under the ‘no Péclet effect’ scenario (gray dashed line in Fig. 2) than under the ‘Péclet effect with a constant L’ scenario (light blue line in Fig. 2). Concerning the ‘no Péclet effect’ scenario in particular, it can be seen that for every 0.1 mol m−2 s−1 change in gs, there is on average a change in Δ18Olw of −0.5‰, which can be translated to a −0.3‰ change in Δ18Otr-cel if the standard assumption is made that 60% of the Δ18Olw signal gets imprinted onto Δ18Otr-cel (Roden et al., 2000). Such a result − when viewed in the context of cellulose 18O measurement precisions typically reported in the literature (±0.3‰; Boettger et al., 2007) − leads us to suggest that an unambiguous detection of the gs signal in Δ18Otr-cel may not always be possible when the Péclet effect is absent. In this regard, it should be noted that a recent study combining isotope measurements with leaf hydraulic design considerations has identified coniferous needles as being more likely to fall into the ‘no Péclet effect’ category than broadleaf trees (Barbour et al., 2021). If this is the case (but see Roden et al., 2015), we would expect greater difficulty in detecting a true gs signal from isotope chronologies of conifers than from broadleaf trees, assuming everything else is equal. Such interspecies differences may at least partially explain why proportionally fewer conifers than broadleaf trees exhibited a significant gs changing trend in the metadataset of Mathias & Thomas (2021). The simulation we performed under the ‘Péclet effect with a variable L’ scenario yielded a pattern (dark blue line in Fig. 2) showing that the gs–Δ18Olw relationship was no longer negative, but nonmonotonic. That is, Δ18Olw increased with gs at low gs but decreased at high gs. As such, we conclude from our modeling results that the strength and direction of the gs–Δ18Olw relationship can be variable, depending on the specific parameter space confined by the species and its environment. As aforementioned, this modeling exercise was based on a commonly adopted steady-state assumption, and so did not consider nonsteady-state isotope dynamics under which more complicated processes involving a role of δ18Ov through vapor one-way flux through stomata (Lehmann et al., 2018, 2020; Farquhar et al., 2021) as well as its potential interaction with the Péclet-caused within-leaf enrichment gradients may also serve to mediate the gs–Δ18Olw relationship. Notwithstanding this, our modeling purpose was to demonstrate a diverse gs–Δ18Olw relationship by incorporating recent advances in 18O modeling and was not to provide an exhaustive survey on all possible patterns. Therefore, given the uncertainties in the validity of the Péclet-related assumptions, we consider it imprudent to assume a priori all species under different environmental/physiological conditions to consistently display a negative gs–Δ18Olw correlation that is strong enough for gs reconstruction. Third, as discussed above, interannual environmental fluctuations can be pronounced in long-term field studies such as Guerrieri et al.'s (2019), making the ‘similar growth environment’ prerequisite a challenge to satisfy. Among the multiple environmental variables, RH is usually the most important control for Δ18Olw (see Simonin et al., 2013 for a demonstration of their sensitivity analysis). Presumably for this reason, Guerrieri et al. (2019) proposed to relax the ‘similar RH’ requirement to a less stringent ‘no systematic RH trend’ one for applying the negative gs–Δ18Olw relationship. From a statistical perspective, increased data variability increases the likelihood of mistakenly accepting the null hypothesis (i.e. the Type II error, see Ott & Longnecker, 2001) if sample size remains the same. Thus, allowing RH as well as other environmental variables such as atmospheric water vapor δ18O (Gat et al., 2001; Munksgaard et al., 2020) to fluctuate from year to year (in spite of no systematic trend) could introduce additional variability to Δ18Olw, in turn lowering the probability of detecting a significant correlation between the derived Δ18Olw and time, if it actually exists. That is, the tree–site combinations with a true directional change in gs over time may be mistakenly judged as those without, due to the background ‘noise’ in Δ18Olw produced by environmental fluctuations. As a result, even with the assumptions that δ18Osw was adequately simulated and a negative gs–Δ18Olw relationship is robust for use in Mathias & Thomas (2021), the insignificant correlation between the derived Δ18Olw and time in 77.9% of their tree–site combinations may not be interpreted with certainty that gs remained relatively constant in these cases. Our analyses have highlighted three outstanding issues related to the method proposed by Guerrieri et al. (2019) for retrospectively inferring gs changes from tree-ring cellulose δ18O chronologies: the proposed use of a δ18Oprep spatial model does not guarantee accurate estimation of interannual variation in δ18Osw, making the practice of converting δ18Otr-cell to Δ18Olw prone to error; due to the complications related to the Péclet effect, no universal gs–Δ18Olw relationship can be applied to interpret variations in Δ18Olw; and even in a case where a negative gs–Δ18Olw/Δ18Otr-cel relationship is robust, the varied interannual environmental influences on Δ18Olw/Δ18Otr-cel may produce a strong background an effective detection of the gs in highlighted uncertainties it is also important to note that the of δ18Otr-cel to Δ18Olw in Guerrieri et al.'s (2019) method the mechanistic model of the mechanisms driving interannual variability in model parameters pex) are far from being (Song et al., 2015; & et al., which could the of this these were not adequately by Guerrieri et al. (2019) and Mathias & Thomas (2021), the gs from δ18Otr-cell in these studies (i.e. gs remained relatively in a majority of their should be viewed with In spite of the in this we that the isotope approach for our understanding of the physiological responses of plants to the (Siegwolf et al., 2021). it is to note that experimental of this approach have focused on in the only experimental performed so Roden & Farquhar (2012) that the isotope framework was not studies to use the isotope approach at the tree-ring level should to where the gs signal can be from the tree-ring and/or rigorously and the uncertainties in this by examining species and environmental conditions on a so as to in the gs from the δ18Otr-cel time This was by the of to and to and to this study based on simulation data and and the and that the The data that support the of this study are available in Global Network of Isotopes in Precipitation and and The data that support the of this study are available in Global Network of Isotopes in Precipitation and and Methods S1 The estimation of leaf temperature and leaf water 18O enrichment above source are not for the or of Supporting Information by the than should be to the The is not for the or of by the than should be to the for the

Topics & Concepts