Philosophical Transactions of the Royal Society B: Biological Sciences
You have accessResearch articles

Non-stomatal processes reduce gross primary productivity in temperate forest ecosystems during severe edaphic drought

Louis Gourlez de la Motte

Louis Gourlez de la Motte

Terra Teaching and Research Center, University of Liège – Gembloux Agro-Bio Tech, 5030 Gembloux, Belgium

[email protected]

Google Scholar

Find this author on PubMed

,
Quentin Beauclaire

Quentin Beauclaire

Terra Teaching and Research Center, University of Liège – Gembloux Agro-Bio Tech, 5030 Gembloux, Belgium

Google Scholar

Find this author on PubMed

,
Bernard Heinesch

Bernard Heinesch

Terra Teaching and Research Center, University of Liège – Gembloux Agro-Bio Tech, 5030 Gembloux, Belgium

Google Scholar

Find this author on PubMed

,
Mathias Cuntz

Mathias Cuntz

Université de Lorraine, AgroParisTech, INRA, UMR Silva, 54000 Nancy, France

Google Scholar

Find this author on PubMed

,
Lenka Foltýnová

Lenka Foltýnová

Global Change Research Institute CAS, Brno, Czech Republic

Google Scholar

Find this author on PubMed

, , ,
Giovanni Manca

Giovanni Manca

European Commission, Joint Research Centre (JRC), Ispra, Italy

Google Scholar

Find this author on PubMed

,
Ignacio Goded Ballarin

Ignacio Goded Ballarin

European Commission, Joint Research Centre (JRC), Ispra, Italy

Google Scholar

Find this author on PubMed

,
Caroline Vincke

Caroline Vincke

Earth and Life Institute, UCLouvain, Louvain-la-Neuve, Belgium

Google Scholar

Find this author on PubMed

,
Marilyn Roland

Marilyn Roland

Plants and Ecosystems, University of Antwerp, Universiteitsplein 1, 2610 Wilrijk, Belgium

Google Scholar

Find this author on PubMed

,
Andreas Ibrom

Andreas Ibrom

Department of Environmental Engineering, Technical University of Denmark (DTU), Bygningstorvet 115, 2800 Lyngby, Denmark

Google Scholar

Find this author on PubMed

, ,
Lukas Siebicke

Lukas Siebicke

Bioclimatology, University of Goettingen, Büsgenweg 2, 37077 Goettingen, Germany

Google Scholar

Find this author on PubMed

,
Johan Neiryink

Johan Neiryink

Institute for Nature and Forest Research, INBO, Havenlaan 88 Box 73, 1000 Brussels, Belgium

Google Scholar

Find this author on PubMed

and
Bernard Longdoz

Bernard Longdoz

Terra Teaching and Research Center, University of Liège – Gembloux Agro-Bio Tech, 5030 Gembloux, Belgium

Google Scholar

Find this author on PubMed

Published:https://doi.org/10.1098/rstb.2019.0527

    Abstract

    Severe drought events are known to cause important reductions of gross primary productivity (GPP) in forest ecosystems. However, it is still unclear whether this reduction originates from stomatal closure (Stomatal Origin Limitation) and/or non-stomatal limitations (Non-SOL). In this study, we investigated the impact of edaphic drought in 2018 on GPP and its origin (SOL, NSOL) using a dataset of 10 European forest ecosystem flux towers. In all stations where GPP reductions were observed during the drought, these were largely explained by declines in the maximum apparent canopy scale carboxylation rate VCMAX,APP (NSOL) when the soil relative extractable water content dropped below around 0.4. Concurrently, we found that the stomatal slope parameter (G1, related to SOL) of the Medlyn et al. unified optimization model linking vegetation conductance and GPP remained relatively constant. These results strengthen the increasing evidence that NSOL should be included in stomatal conductance/photosynthesis models to faithfully simulate both GPP and water fluxes in forest ecosystems during severe drought.

    This article is part of the theme issue ‘Impacts of the 2018 severe drought and heatwave in Europe: from site to continental scale’.

    1. Introduction

    With global climate change, droughts are likely to be more intense [1,2]. In 2018, a severe drought event occurred in Northern and central Europe causing forest fires and crop yield losses [3]. Europe experienced a major reduction of gross primary productivity (GPP) and transpiration (E) similarly to previous extreme events such as the 2003 Europe drought-heatwave [4] mostly because of soil water limitation [5,6]. Continuous measurements of ecosystem CO2 and water fluxes captured throughout Europe at eddy covariance (EC) flux tower stations thus provide a great large-scale ‘natural experiment’ to study the impact of drought on GPP and E [7].

    There is increasing evidence that GPP reductions due to droughts could originate from both changes in stomatal behaviour (stomatal origin limitation, SOL) and non-stomatal traits (non-stomatal origin limitation, NSOL) [812]. Proposed NSOL mechanisms are reduced Rubisco activity (carboxylation rate) and/or electron transport activity [13], reduced active leaf area index (LAI) [13], reduced mesophyll conductance (including the intercellular airspace, cell walls, plasma membranes, cytoplasm and the chloroplast envelopes [14], gm) [15] or a combination of those [11,16]. The cause of GPP reduction is still subject to debate [17,18] and the modelling of SOL and NSOL and their (de)coupling is still poorly constrained by data. As a result, there is a strong need to examine whether different mechanisms are relevant and if models could be improved by developing more evidence-based functions for the impact of drought stress [19,20].

    In leaf/canopy photosynthesis models, gross primary assimilation (A) is very often modelled using the Farquhar et al. [21] photosynthesis model for C3 species [9,10,13]. In this model, Rubisco-limited photosynthesis (usually close to light saturation) is a function of the maximum carboxylation rate (Vcmax) and the internal CO2 leaf concentration (Ci) which implicitly considers that Ci is equal to the CO2 concentration in the chloroplasts (Cc). As Ci cannot be measured directly, it is usually approximated by employing Fick's diffusion law through the stomata using a stomatal conductance (gs). This representation requires the determination of stomatal conductance by modelling. In this study, following Zhou et al. [10], we use the concept of apparent Vcmax (Vcmax,app) recognizing that variations in Vcmax,app can result either from changes in the actual maximum rate of carboxylation or from changes in gm which are not explicitly represented in this diffusion model. Consequently, when drought occurs, it impacts directly stomatal behaviour (closure) and then photosynthesis by limiting the diffusion of CO2 into the leaf which results in reduced Ci (SOL) or/and it impacts non-stomatal mechanisms (NSOL) which result in decreases of Vcmax,app [10,20].

    A long standing stomatal conductance model from Cowan & Farquhar [22] states that stomata should act to maximize carbon gains while minimizing water losses (transpiration, E), that is to maximize the integrated sum of A–E where λ (mol C mol−1 H2O) is the carbon cost of water gain A/E or marginal water-use efficiency [23] (note that we inverted the original expression). Medlyn et al. [24] proposed a reconciliation of the optimal stomatal behaviour theory [22] with empirical stomatal models linking gs and A. Their work resulted in a unified stomatal optimization model (USO) with a form similar to former empirical expressions [25,26] (see equation (2.3)) where the slope between gs and A*f (g1) is a key parameter (called the stomatal slope parameter). g1 is directly interpretable as inversely related to λ and to intrinsic water-use efficiency (iWUE, A/gs) normalized by vapour pressure deficit (VPD) and CO2 air concentration (Ca) [27].

    The USO model has been used both at the leaf level using leaf gas exchange data [28] and at the ecosystem level using EC flux observations [29] during non-water limited periods. During water limited periods, various responses of g1 (leaf level, SOL) to soil moisture were found [10] for a large range of species while a more consistent pattern of decreasing Vcmax,app was found. In a recent work, a good correlation between leaf scale and ecosystem scale g1 (or G1) response to soil moisture was found in a woodland dominated by Acacia trees thereby demonstrating the ability of both leaf and ecosystem scale approaches to quantify drought effect [30].

    In this study, we used the USO model combined with the Farquhar C3 model (considering that Ci = Cc) to study the origin of edaphic drought impacts on GPP (SOL and/or NSOL) in forest ecosystems using EC flux measurements by replacing leaf-level variables by their ecosystems analogues using a big leaf framework [27,31]. The surface conductance (Gs analogous to gs) was estimated by inverting the Penman–Monteith equation [32]. We then inferred the bulk stomatal slope parameter (G1 analogous to g1) and the maximum apparent carboxylation rate of the ecosystem (VCMAX,APP) [33] at a daily time step for each ecosystem. The study was restricted to the growing period excluding any autumn senescence or spring leaf emergence influence on the variation of VCMAX,APP.

    In addition, drought intensity was quantified using the relative extractable water (REW) as proposed by Granier et al. [5], which is a normalized index of soil water deficit varying from 0 to 1 that allows for edaphic status inter-site comparisons. This index was used in previous studies [5,34] and, based on their results, we hypothesize that both E and GPP reductions will occur when REW falls below ≈0.4.

    The objective of this work is to examine the response of G1, as a measure for SOL, and VCMAX,APP, as a measure for NSOL, to soil water deficit using EC data collected in forests during the 2018 European drought. More specifically, we intend to answer the following questions: (1) how was REW impacted by the drought in forest sites in 2018? (2) Can we confirm the REW threshold of ≈0.4 for GPP reductions found in previous studies [5,34]? (3) To what degree did SOL and NSOL impact GPP during the drought? (4) What were G1 and VCMAX,APP responses to REW function shapes and how did these responses vary across sites?

    2. Material and methods

    (a) Site and data description

    Data have been processed by the Ecosystem Thematic Centre of the Integrated Carbon Observation System (ICOS) and form the 2018 drought ICOS/Fluxnet dataset [35], which is a compilation of EC fluxes, meteorological and edaphic data during the 2018 European drought at half-hourly resolution. Only sites with a sufficiently resolved vertical profile of soil water content sensors were selected. The main site characteristics are summarized in table 1. Flux data followed the standard FLUXNET processing [46], including friction velocity (u*) filtering [47] and GPP determination by night-time flux partitioning [48]. Only data marked with highest quality flags were used for this study. Latent heat fluxes were not corrected for energy balance closure.

    Table 1. Main characteristics of the flux tower sites included in this study. The LAI corresponds to the maximum LAI typically observed at the sites.

    site ID country latitude longitude dominating species LAI m2 m−2 soil texture rooting depth m ref
    BE-Bra Belgium 51.308 4.52 Pinus syvlestris 3 sand 1 [36]
    BE-Vie Belgium 50.305 5.998 Fagus sylvatica/Pseudotsuga menziesii 5 silty clay loam 1.4 [37]
    CZ-Lnz Czechia 48.682 16.946 Quercus robur/Fraxinus angustifolia/Carpinus betulus/Tilia cordata 6.5 sandy loam 1.2 [38]
    CZ-Raj Czechia 49.444 16.697 Picea abies 5 sandy loam 0.7 [39]
    CZ-Stn Czechia 49.036 17.97 Fagus sylvatica 5.5 sandy loam 0.7 [40]
    DE-Hai Germany 51.0792 10.453 Fagus sylvatica 6 clay loam 0.7 [41,42]
    DK-Sor Denmark 55.486 11.645 Fagus sylvatica 5 sandy clay loam 1 [43]
    FR-Bil France 44.494 −0.956 Pinus pinaster 2.5 sand 1.1 [44]
    FR-Hes France 48.674 7.065 Fagus sylvatica 6.5 silty clay loam 1.6 [45]
    IT-Sr2 Italy 43.732 10.291 Pinus pinea 2.5 sand 1.2

    (b) Quantification of drought

    The intensity of edaphic drought was quantified by computing the REW content [5] at each time step and for the entire root depth using:

    REW=i((SWCiSWCWP,i)/(SWCFC,iSWCWP,i))Δhihmax,2.1
    where i is the index of each soil layer over the rooting depth, SWCi is the actual soil water content, SWCWP is the soil water content at the wilting point, SWCFC is the soil water content at field capacity, Δh is the thickness of each layer and hmax is the maximum rooting depth. Each soil horizon was divided into soil layers corresponding to the number of sensors installed in the horizon. The layer boundaries were the horizon limits or the midway point between two sensors. Soils related data are summarized in the supplements (electronic supplementary material, table S1). For each layer, SWCWP and SWCFC were estimated using soil retention curves based on either measurements (by research teams) or modelling (based on soil textures) and checked for consistency with SWC 48 h after a rain event for SWCFC and with minimum SWC values observed at the site for SWCWP to avoid negative REW values. When not available, the maximum depth was defined as the bedrock depth [5]. When data were available (BE-Vie and BE-Bra), REW was corrected for the coarse fraction by applying a correction factor for each layer. According to Granier et al. [5,34] it is expected that both GPP and Gs start to decrease when REW drops below ≈0.4. The evolution of REW at each site in 2018 is presented in the electronic supplementary material, figure S1.

    (c) Canopy surface variables

    Detailed computation procedures for canopy surface variables are fully described in Knauer et al. [31]. First the aerodynamic conductance to water transfer (Gaw) was computed as a combination of an aerodynamic conductance to momentum (first term) and a boundary layer conductance (second term) as follows:

    Gaw=(u2u(z)+(6.2u0.67)1)1,2.2
    where u* is the friction velocity (m s−1), and u(z) the wind speed at measurement height (z). Canopy surface conductance for water (Gs, m s−1) was computed by inverting the Penman–Monteith equation [32]:
    Gs=LEGawγs(RnGS)+ρCpGawVPDaλE(s+γ),2.3
    where LE is the latent heat flux (W m−2), γ is the psychrometric constant (Pa K−1), s is the slope of the saturation vapour pressure curve at air temperature (Pa K−1), Rn is the net radiation (W m−2), G is the ground heat flux (W m−2), S is the sum of all storage terms (W m−2), Cp is the heat capacity of dry air (1005 J kg−1 K−1) and VPDa is the VPD of ambient air (Pa). G was considered negligible when not available while S was not available and was set to 0 at all sites.

    The CO2 concentration at the canopy surface (Cs), needed in the USO and diffusion equations, was computed as:

    Cs=Ca+NEE(Gaw/1.32),2.4
    where Ca (μmol CO2 mol−1) is the CO2 air concentration at the measurement height, NEE (μmol CO2 m−2) is the net CO2 ecosystem exchange and the factor 1.32 is the ratio of diffusivities of CO2 and water vapour in the boundary layer. The VPD at the canopy surface (VPDs, Pa) was also computed (see [31] for more details). Gs is a good predictor of bulk stomatal conductance only when evaporation is small compared to transpiration; data collected during a period of 48 h following a rain event were discarded. Secondly, the analysis was restricted to the growing season, avoiding senescence and leaf emergence periods. We defined this period as the days when the daily GPP (averaged over all the available years) smoothed with a 15 days moving average window was higher than 70% of the 95th percentile of the daily GPP distribution. Gs data were also filtered excluding half hour with LE < 0 or Rn < 0. Negative Gs values were filtered and Gs outliers were also discarded by removing data when Gs were higher than the 98th percentile of the Gs distribution.

    (d) Stomatal origin limitations

    Similarly to previous work [10,49], reductions of GPP originating from SOL were assessed by analysing dependence on REW of the G1 parameter used in the USO model developed by Medlyn et al. [24] but adapted to the ecosystem scale using bulk ecosystem parameters [27]:

    Gs=G0+1.6(1+G1VPDs)GPPhighCs,2.5
    where, GPPhigh is the GPP at high radiation (Rg > 500 W m−2) and replaces net assimilation in the original leaf scale expression of the model, Gs replaces stomatal conductance and leaf surface variables were replaced by their corresponding canopy surface values [50] (Cs the air CO2, VPDs). In this expression, the nocturnal stomatal conductance G0 was set to 0 as its magnitude can be considered negligible when compared to the other terms at saturating daylight conditions [27]. G1, the slope parameter, is a physiologically meaningful parameter as it was shown to be inversely related to λ [24] and to iWUE [27].

    G1 was obtained by inverting equation (2.5) using half-hourly measurements. Because leaf respiration was neglected in equation (2.5), the equation was inverted only for high radiation data so that GPPhigh should be much higher than leaf respiration. Negative G1 values were filtered and outliers were also discarded by removing data when absolute G1 was two times higher than the average absolute deviation from the median. Finally, daily G1 averages were then computed for days with at least five valid half-hourly values.

    The response of G1 to REW was fitted with a segmented linear response curve (two segments) in order to test the presence of an REW threshold (break point) above which G1 is constant (no effect range) and under which stomatal regulation occurs (G1 decreases or increases) [51]. After a first fit, outliers were removed by exclusion of the G1 value having absolute residuals more than 2.2 times the standard deviation of the residuals distribution. A second fit was then done. Parameters obtained from a second fit were G1*, i.e. the average G1 value within the ‘no effect range’, the break point REW value (REWB,G1 I) and the slope/intercept of the G1 decrease/increase. The presence of the break point was further tested by comparing the residuals of the model to those of a simple linear regression model using an F-test.

    (e) Non-stomatal origin limitations

    Reductions of GPP originating from NSOL were studied by assessing the effect of water stress on apparent bulk Vcmax hereafter called VCMAX,APP. It was obtained by inverting the expression of Rubisco-limited photosynthesis during high radiation conditions [21]:

    VCMAX,APP=GPPhigh(Ci+Km)(CiΓ),2.6
    where VCMAX,APP is expressed per m2 of soil and not of leaf as usual. Km is the effective Michaelis Menten coefficient kinetics and Γ* is the CO2 compensation point, which were both computed using temperature responses following Bernacchi et al. [52], GPPhigh is the GPP at Rg > 500 W m−2 while Ci was computed using Fick's diffusion law [53]:
    Ci=CsGPPhigh(Gs/1.6).2.7

    Note than in equations (2.5)–(2.7), leaf respiration was neglected which should have a small effect on the results as, at high radiation, leaf respiration should be much smaller than GPP. Half-hourly values of VCMAX,APP were then normalized for temperature to 25°C using an Arrhenius equation [31] fitted for each decile of REW as VCMAX,APP response to temperature was found to decrease under drought conditions. Finally, VCMAX,APP was averaged on a daily basis and days with less than 5 half-hourly values were discarded. Considering the way we estimated VCMAX,APP, a decrease in VCMAX,APP indicates NSOL of GPP including either changes in mesophyll conductance or in actual VCMAX or other processes limiting GPP (apart from stomatal closure). The response of VCMAX,APP to REW was assessed using the same segmented linear regression model as explained in the previous section. VCMAX,APP* and REWB,VCMAX were defined as the average of VCMAX,APP values (normalized at 25°C) within the no effect range and the break point REW value, respectively.

    (f) Degree of stomatal and non-stomatal origin limitation

    To illustrate the degree of SOL and NSOL due to edaphic drought, following Zhou et al. [10], a sensitivity analysis was performed to explore the impact of drought-induced changes in G1 and VCMAX,APP on GPP. The impact of NSOL was assessed by comparing measured GPP to a theoretical non-affected value corresponding to a ‘modelled’ GPP computed using inverted equation (2.5) with VCMAX,APP* instead of VCMAX,APP and a Ci obtained from equation (2.6) using observed Gs values (equation (2.2)) and measured GPP values. Similarly, the impact of SOL was assessed by comparing measured GPP to ‘modelled’ unaffected GPP computed using constant G1* values and observed VCMAX,APP values. The degree of limitation (DoL) was computed as the ratio of modelled ‘unaffected’ GPP against measured GPP and represents the factor by which GPP was divided because of SOL or NSOL.

    3. Results

    (a) Gs and GPPhigh

    GPPhigh and Gs (for Rg > 500) normalized by their respective maximum values in relation with REW are presented in figure 1. At all sites, we can observe that both GPPhigh and Gs behave similarly. High values of both GPPhigh and Gs were observed for high REW while both variables decreased simultaneously with REW. There was an exception at BE-Vie where such a pattern was not observed although both variables still behaved similarly. The lowest GPPhigh and Gs values were observed at sites such as CZ-Raj, FR-Bil and DE-Hai where very low REW values (lower than 0.15) were reached. At IT-Sr2, Gs and GPPhigh were still quite high (around half of maximum values) even for very low REW values most probably because, in this sandy soil, rooting depth was probably deeper than the deepest available SWC sensor (1.2 m, see electronic supplementary material, table S1) which caused an underestimation of REW.

    Figure 1.

    Figure 1. Dependence of GPPhigh (blue, left axis) and Gs at high radiation (Rg > 500, red, right axis) normalized by their maximum value on REW for each site. (Online version in colour.)

    (b) Response of G1 to edaphic drought

    G1 was found to be constant at all sites apart from DE-Hai even for sites where REW values lower than 0.4 were observed (figure 2). In BE-Bra G1 seems to be enhanced at low REW but the segmented model did not perform significantly better than the linear one. In DE-Hai, G1 was found to increase when REW dropped below a very low value of 0.2, which is quite close to the wilting point. Such low REW were also observed during the growing season at CZ-Raj and IT-Sr2 but no similar behaviour was observed. The lowest G1 * (1.5 kPa0.5 at CZ-Raj) was three times lower than the highest value (4.5 kPa0.5 at FR-Bil, table 2).

    Figure 2.

    Figure 2. Dependence of G1 on REW for each site. REWB,G1 are marked by a vertical solid line with 95% confidence intervals (dashed vertical lines). Regression lines are only shown when F-test p-values (comparison with linear regression) were smaller than 0.05.

    Table 2. Maximum extractable water (EW), minimum observed REW in 2018 during the growing season, REWb,VCMAX and REWb,G1 (REW break points for VCMAX,APP and G1, respectively) given with 95% confidence intervals. VCMAX,APP* and G1* (VCMAX,APP and G1 values in unstressed conditions) given with 95% confidence intervals. P-values are given for the F-test comparing the segmented mode (three parameters) to the linear model (two parameters).

    site ID EW min REW VCMAX,APP* REWb,VCMAX p-value G1* REWb,G1 p-value
    mm μmol m−2 s−1 kPa−0.5
    BE-Bra 133 0.30 58 ± 3 0.57 ± 0.19 <0.05 2.9 ± 0.2 0.28
    BE-Vie 215 0.19 81 ± 5 1 1.5 ± 0.1 1
    CZ-Lnz 241 0.49 116 ± 10 1 2.0 ± 0.2 1
    CZ-Raj 92 0.07 80 ± 7 0.36 ± 0.11 <0.001 1.2 ± 0.8 1
    CZ-Stn 236 0.46 120 ± 7 1 2.3 ± 0.3 1
    DE-Hai 143 0.00 86 ± 17 0.36 ± 0.07 <0.001 1.3 ± 0.2 0.2 ± 0.05 <0.001
    DK-Sor 176 0.25 104 ± 3 0.80 ± 0.09 <0.001 2.2 ± 0.2 1
    FR-Bil 159 0.08 76 ± 7 0.45 ± 0.15 <0.05 4.5 ± 0.4 1
    FR-Hes 338 0.33 121 ± 13 0.44 ± 0.08 <0.05 2.0 ± 0.5 1
    IT-Sr2 108 0.08 89 ± 6 1 2.1 ± 0.2 1

    (c) Response of VCMAX,APP to edaphic drought

    The effect of NSOL caused by drought was studied by analysing the dependence of the temperature normalized VCMAX,APP values on REW (figure 3). At all sites that experienced low REW conditions (below ≈ 0.4) apart from BE-Vie and IT-Sr2, constant VCMAX,APP were observed for large REW values followed by a decrease when REW declined below an REWB,VCMAX threshold. The REWB,VCMAX were not significantly different (according to the confidence intervals) than the value of 0.4 which was found in previous studies, with an exception at DK-Sor where REWB,VCMAX was higher (REWB,VCMAX = 0.8 ± 0.09, table 2). The high REWB,VCMAX observed at DK-Sor might result from an overestimation of REW as the shallowest available SWC probe was at 15 cm depth (see electronic supplementary material, table S1) and was not able to catch the beginning of the progressive drying of the upper layers that contain a large amount of roots. The most impacted site was DE-Hai where REW almost reached the wilting point (REW ≈ 0) with very low VCMAX,APP values (≈15 µmol m−2 s−1) probably because of shallow soil and rooting depth (0.6 m).

    Figure 3.

    Figure 3. Dependence of VCMAX,APP normalized at 25°C on REW for each site. REWB,VCMAX are marked by a vertical solid line with 95% confidence intervals (dashed vertical lines). Regression lines are only shown when F-test p-values (comparison with linear regression) were smaller than 0.05.

    (d) Degree of stomatal and non-stomatal limitation

    DoL reached values of 5 for NSOL at DE-Hai while it remained close to 1 for SOL at all sites (figure 4). This analysis therefore confirms that, at all sites, NSOL as the dominant mechanism. As a result, reducing VCMAX,APP while maintaining G1 constant could capture the variations of both GPPhigh and Gs with drought (identical conclusions are obtained if we focus the analysis on Gs instead of GPP, data not shown). It is also worthwhile noticing that the increasing G1 observed at DE-Hai (and at BE-Bra to a lesser extent) did not lead to important changes in GPP.

    Figure 4.

    Figure 4. Degree of limitation (DoL) by both SOL and NSOL. The degree of limitation was computed as the ratio of modelled GPP on measured GPP at high radiation. Modelled GPP was computed by using either fixed VCMAX,APP* (red points, NSOL) and varying G1 or fixed G1* and observed VCMAX,APP (blue points, SOL). (Online version in colour.)

    4. Discussion

    (a) Methodological limitations

    Although the responses of VCMAX,APP and G1 to REW were relatively consistent, some sites showed unexpected behaviours. For example, REWB,VCMAX at DK-Sor was much higher than expected (0.85) because some SWC sensors experienced failures during the drought. At IT-Sr2, no limitation of GPP was found although very low REW values were estimated, probably because the SWC sensor profile was not deep enough to capture the whole rooting depth. Multiple and deeper sensor profiles (with matching wilting points and field capacities) would certainly help to reduce these uncertainties. Complementary measurements such as predawn leaf water potential and soil matric potential [17,54], when REW approaches values close to 0.4 at the site, would also be useful.

    The big leaf approach used in this study also has several limitations [31] which could be critical when comparing leaf scale-derived parameters to big leaf canopy scale estimates [27,29,31] or when attributing a behaviour to a specific species. First, the approach is only able to derive bulk parameters and is unable to distinguish the vertical and horizontal distribution of the properties. Horizontal heterogeneity is especially crucial at mixed forest sites where different species could show different responses to drought (and different root depth and therefore REW) which would blur their respective responses in the measured signal. This is especially critical at BE-Vie where the two most frequent wind directions (southwest and northeast) correspond to different stands (coniferous and beech stands) with possibly different root depth, REW and weather conditions [37]. Separating the data between each sector did not however improve the relation because of the lack of data (data no shown).

    At sites with dense canopy and high LAI, vertical gradients of the parameters (Gs, G1 and VCMAX,APP) could result from vertical gradients within plants of the same species or from physiological differences across species [27]. Sun leaves, developed under high irradiance, usually exhibit higher WUE (lower G1) than those developed under shady conditions [55], primarily because of higher photosynthetic capacities. However, more critical for this study, little is known about to what degree these vertical gradients (within and across species) could affect the response of G1 and VCMAX,APP to drought. To our knowledge, in most Earth system models, the same reduction functions of photosynthesis during edaphic drought (either NSOL or SOL) are used for sun and shade leaves [17]. More complex multiple layers and/or sun-shade models as well as additional data gathered at multiple canopy layers would be needed to assess this question more closely.

    Moreover, soil and vegetation components cannot be distinguished so that critical variables such as Gs (and variables depending on it, such as G1 and Ci) will inevitably contain some signal from the soil. This signal can be reduced by filtering the data after rain events [27] and should be small for dense canopies with LAI higher than 2–3 [56] (which is the case in all sites) and even smaller when the upper soil layer dries.

    Finally, systematic errors (energy balance non-closure [57]) in EC fluxes are also major sources of uncertainties that affect G1 and VCMAX,APP magnitudes. However, it was found that, at multiple flux tower sites, the surface energy balance was not modified during the 2018 drought [58]. This source of error is therefore unlikely to affect G1 and VCMAX,APP responses to REW.

    Nevertheless, despite all the limitations of the big leaf approach detailed above, this framework was very suitable for this multi-site study as it relied on very few ancillary data [31]. If the comparison of G1 and VCMAX,APP (which inherently has a different meaning than leaf-level Vcmax,app) is not straightforward, analysing the dynamics of these parameters inferred from in situ EC data during drought provides very useful information about how forest ecosystems reacted to these events.

    (b) Implications of non-stomatal origin limitation for the modelling of gross primary productivity and transpiration

    In this study, similarly to Granier et al. [5], we found that GPP and Gs reductions can be expected when REW drops below ≈0.4. To account for these reductions, using empirical reduction factors (ranging from 1 to 0) when soil water content falls below a given threshold is a widely used approach [17]. However, it is questionable whether the reduction factors should be applied to SOL and/or to NSOL. This was previously investigated in Mediterranean ecosystems [8,13,59] and it was found that, calibrating the model on either GPP or transpiration (E) (not both) by considering only SOL during edaphic drought conditions systematically led to overestimates of WUE which did not allow the correct simulation of both fluxes. Surprisingly, it was found that applying NSOL only was sufficient to correctly simulate both GPP and E.

    In this study, we found that reducing VCMAX,APP when REW dropped below ≈0.4 and using a constant G1 parameter (from the USO model [24]) allowed the capture of both GPP and Gs reductions at European forest sites. Similar conclusions were also found by Chen et al. [60] in four different ecosystems (temperate grassland, tropical savannah, boreal and one temperate forest). More specifically, relatively consistent behaviour was observed at the three beech (Fagus sylvatica) forest sites (FR-Hes, DK-Sor and DE-Hai) where NSOLs were the main source of photosynthesis reductions with relatively constant G1. Similarly, in a study carried out on adult beech using leaf-level measurements, the Ball–Berry slope was found to be almost insensitive to soil water potential [61]. Our results are in agreement with Granier et al. [5] who observed constant WUE even for very low REW during the 2003 drought and with Hentschel et al. [62] who also found unchanged annual iWUE derived from tree ring carbon isotopic composition.

    Studies were also performed at the leaf scale to study the impact of drought on NSOL and SOL. In their meta-analysis, Zhou et al. [10] found highly variable responses of g1 (leaf level) for woody species ranging from rather constant to severely decreasing g1 with drought. Decreasing Vcmax,app (NSOL) were however found for all species. It was also highlighted that NSOL was the main factor limiting photosynthesis under severe stress in 10 Mediterranean herbs and shrubs species [11,63] and, more importantly for this study, for four tree species [20]. Unfortunately, a direct comparison with our results cannot be carried out as the studied species were different. Such direct comparisons have been carried out in a woodland dominated by Acacia trees by Tarin et al. [30] who found a close agreement between G1 and g1 estimated from ecosystem (EC big leaf) and leaf-level approaches, respectively.

    According to Keenan et al. [64], NSOL could be caused by the variations of a finite mesophyll conductance with soil water availability [65] which, if not taken into account, leads to wrong estimates of actual Vcmax [66] (which was implicitly taken into account by using VCMAX,APP). In addition, the hypothesis that, under severe droughts, GPP can be directly impacted by biochemical limitations which cause the reduction of actual VCMAX should not be discarded [67]. Separating NSOL between these two mechanisms (mesophyll conductance and actual VCMAX) was not done in this study. Currently, without additional leaf-level data to better understand the mechanisms underlying mesophyll conductance changes during droughts, we use an apparent VCMAX,APP [15,17,20].

    (c) Optimal stomatal behaviour during drought and intrinsic water-use efficiency

    We did not find a general pattern of systematically decreasing G1 during drought or, in other words, an increasing iWUE for stomatal closure (increasing λ) across ecosystems as theoretically predicted [23,68]. To the contrary, we found constant G1 (and therefore λ) values at most sites, and even increasing values at DE-Hai (and BE-Bra to a lesser extent). This result (a constant G1) is rather surprising as it would suggest that changes in stomatal conductance responses were not needed to model Gs under long-term water stress events [59] and that λ does not increase with drought. We argue that this is caused by the fact that NSOLs were not considered by Manzoni et al. [23] and Mäkelä et al. [68] in their analyses as stomatal closure (reduced Gs) is known to regulate leaf water flows in response to soil water availability [69]; without such mechanisms, leaves would be quickly dehydrated. However, one should consider that stomatal closure in response to drought does not necessarily lead to a decrease in G1 as, in USO, any reduction of VCMAX,APP lead to a reduction in stomatal conductance [10]. At very low REW values, previous studies showed that Ci could even increase because of NSOL [70], which would explain the increase of G1 we observed at DE-Hai.

    Another more complex approach to stomatal conductance modelling is to model stomatal conductance in the function of leaf water potential which is expected to regulate Gs [71]. This approach requires a complete model of water flow from the soil through the plant to the atmosphere [72]. This kind of model was tested by reducing the stomatal slope of the Ball–Berry–Leuning model [26] with leaf water potential, but the model did not account for NSOL [54]. Recently, Dewar et al. [73] proposed a new optimization model in which stomatal behaviour maximizes photosynthesis and where the costs of stomatal closure arise from NSOL (mesophyll conductance and/or carboxylation rate) and/or loss of hydraulic conductance [74]. This results in a parameter, equivalent to G1, which is expressed as a function of measurable variables such as hydraulic conductivity, leaf water potential and Vcmax. This model has been successfully tested on saplings for different plant functional types [75] and fitted well sub-daily leaf scale observations; however, this still needs to be tested for longer term in situ ecosystem droughts. This could not be done in this study as leaf and soil water potentials were lacking. It does, however, highlight a promising research path for the future.

    5. Conclusion

    In this study, we used a big leaf framework to investigate the origin of edaphic drought impacts on GPP (stomatal origin limitation and non-stomatal origin limitation) in European forest ecosystems during the 2018 drought. In agreement with Granier et al. [5], we found that GPP and Gs were both greatly affected by soil moisture depletion at many sites. We went a step further by showing that these reductions could be faithfully modelled by decreasing VCMAX,APP (NSOL) when the REW dropped below around 0.4 while keeping the G1 (SOL) parameter from the USO model [24] constant. These results were rather unexpected as it would suggest that stomatal closure was not responsible for GPP reductions with drought. We argue that this was caused by the fact that G1 was not representative of stomatal behaviour during drought because GPP was not only regulated by stomatal closure but also by NSOL. Nevertheless, these results strengthen the increasing evidence that NSOL should be included in stomatal conductance/photosynthesis models to faithfully simulate both GPP and water fluxes in forest ecosystems.

    Data accessibility

    Most of the flux data used in this study are available on the ICOS carbon portal at: https://doi.org/10.18160/PZDK-EF78. Ancillary information about the soil is provided in the electronic supplementary material.

    Authors' contributions

    L.G.dlM. was the leader of the data analysis and writing of the paper. Q.B. and B.H. participated in the design of the study, gave intellectual inputs and feedbacks to the paper. B.L. designed the study and followed the data analyses and writing process closely. All the other co-authors were involved in the data acquisition as principal investigators for each site and actively contributed to the paper by commenting on intermediate versions.

    Competing interests

    We declare we have no competing interests

    Funding

    The financial support of the Research Foundation-Flanders (FWO) to the ICOS infrastructure is acknowledged. The BE-Vie was funded by the Service Public de Wallonie (Convention 1217769) which also gave us the opportunity for research at that site. The Czech sites were funded by the project ‘SustES - Adaptation strategies for sustainable ecosystem services and food security under adverse environmental conditions' (grant no. CZ.02.1.01/0.0/0.0/16_019/0000797). We would like to thank the University of Goettingen, Deutsche Forschungsgemeinschaft and German Federal Ministry of Education and Research for funding the operation of DE-Hai. The work was funded by the Technical University of Denmark (DTU), the Danish Research Council (DFF - 1323-00182), the Danish Ministry of Higher Education and Science (5072-00008B) and the EU research infrastructure projects RINGO and ICOS. The data collected from the Salles ICOS station (FR-Bil) funded by INRA were obtained by C. Chipeaux (site manager), S. Lafont and D. Loustau (INRA, UMR ISPA).

    Acknowledgements

    The study has been performed thanks to the European Fluxes database cluster. The authors would like to thank all the persons that contributed to this study. We would also like to thank all the financial support provided for sites. We thank the administration of the Hainich National Park for the opportunity for research in the National Park. Data from the Sorø beech forest site (DK-Sor) have been measured, evaluated and provided by Kim Pilegaard and Andreas Ibrom and the station team. We acknowledge the support of successive European projects, by European regional development programs with the Region Lorraine, by GIP Ecofor and SOERE F-ORE-T, by ADEME and by the INRA Department of Forest, Grassland and Freshwater Ecology (EFPA). Finally, we would also like to thank Carsten Gruening for his technical assistance.

    Footnotes

    One contribution of 16 to a theme issue ‘Impacts of the 2018 severe drought and heatwave in Europe: from site to continental scale’.

    Electronic supplementary material is available online at https://doi.org/10.6084/m9.figshare.c.5077590.

    Published by the Royal Society. All rights reserved.

    References