The Cretaceous physiological adaptation of angiosperms to a declining pCO 2 : a modeling approach emulating paleo-traits

. The Cretaceous evolution of angiosperm leaves towards higher vein densities enables unprecedented leaf stomatal conductance. Still, simulating and quantifying the impact of such change on plant productivity and transpiration in the speciﬁc environmental conditions of the Cretaceous remains challenging. Here, we address this issue by combining a paleo proxy-based model with a full atmosphere-vegetation model that couples stomatal conductance to carbon assimilation. Based on the fossil record, we build and evaluate three consistent proto-angiosperm vegetation parameterizations under two end-members 5 scenarios of pCO 2 (280 ppm and 1120 ppm) for the mid-Cretaceous : a reduction of leaf hydraulic or photosynthetic capacity and a combination of both, supported by a likely coevolution of stomatal conductance and photosynthetic biochemistry. Our results suggest that decreasing leaf hydraulic or/and photosynthetic capacities always generate a reduction of transpiration that is predominantly the result of plant productivity variations modulated by light, water availability in the soil, atmospheric evaporative demand and pCO 2 . The high pCO 2 acts as a fertilizer on plant productivity that strengthens plant transpiration and 10 water-use efﬁciency. However, we show that proto-angiosperm physiology does not allow vegetation to grow under low pCO 2 because of a positive feedback between leaf stomatal conductance and leaf area index. Our modeling approach stresses the need to better represent paleovegetation physiological traits. It also conﬁrms the hypothesis of a likely evolution of angiosperms from a state of low leaf hydraulic and photosynthetic capacities at high pCO 2 to a state of high leaf hydraulic and photosynthetic capacities linked to leaves more and more veins together with a more efﬁcient biochemistry at low to evaluate the combined environmental and physiological-induced response of Cretaceous vegetation to a low D v representative of proto-angiosperms. Our experimental design considers both reducing leaf hydraulic or/and capacities and applying different pCO 2 scenarios in a coupled model. The modeled g allows account for long-term plant adaptation to their changing environment. We evaluate the physiological response of the vegetation to low hydraulic and capacities at leaf scale, then at global scale. It leads us to evaluate and between


Introduction
Vegetation plays a pivotal role in the climate system as it controls water and energy fluxes at the interface between land surfaces and the atmosphere through albedo (Charney et al., 1975;Port et al., 2016;Brovkin et al., 2009), roughness length, and evapotranspiration capabilities (e.g., Bathiany et al., 2010;Betts et al., 1997;Kleidon et al., 2000;Gibbard et al., 2005). Specifically, evapotranspiration, i.e. the sum of soil evaporation, vegetation evaporation and vegetation transpiration fluxes, is a key term 20 in the continental hydrological cycle, and has been shown to control moisture convergence and convection, thereby precipitation patterns in the tropics either in present-day-like configurations (Sun and Barros, 2015;Fraedrich et al., 1999) or for past climates (e.g., Braconnot et al., 1999;Brovkin et al., 2006). It is therefore mandatory that earth system models designed to simulate climate evolution, either in the past or in the future, account for plant traits that alter these fluxes, despite the numerous evolution in the specific environmental conditions of the Cretaceous (different paleogeography and high atmospheric CO 2 concentrations). Land surface models used in climate modeling do not explicitly represent traits such as vein densities but rather directly simulate the operational stomatal conductance at the leaf scale (g s ) (Farquhar et al., 1980;Ball et al., 1987;Krinner et al., 2005;Yin and Struik, 2009). However, based on relationship obtained from extant and fossil leaves that links maximal anatomic stomatal conductance to H 2 O (g max anat ) to D v (Brodribb et al., 2007;Brodribb and Feild, 2010), they can be parameterized to represent the changes in stomatal conductance that occured with the angiosperm radiation. Depending on their design, these parameterizations can reflect changes on leaf traits controlling leaf hydraulic capacity, leaf photosynthetic capacity or both. The latter is supported by our knowledge of modern plant processes and in particular of a coevolution between stomatal conductance and V cmax (Franks and Beerling, 2009a;De Boer et al., 2012). Although there is no fossil evidence of changes in leaf photosynthetic capacity, we assume that this covariation is a time-invariant property and that such relationship was the same in the past. Moreover, atmospheric CO 2 concentration (pCO 2 ) is known to be one of the drivers that control the stomatal 70 opening and closing at very short-term (Jarvis et al., 1999). Several studies have reported that the g s /g max anat ratio is relatively stable around 20-40 %, an optimal range within which changes in guard cell turgor pressure are most efficient in controlling quick stomatal conductance changes (Dow et al., 2014;Dow and Bergmann, 2014;McElwain et al., 2016). Therefore, it is likely that pCO 2 also drives changes in g max anat on long time scales. Paleo-CO 2 reconstructions based on proxies (plant fossils and isotopes) and geochemical models show that, despite a large spread, pCO 2 was high throughout the Cretaceous, ranging 75 from 500 ppm to 2000 ppm. Values peaked during the mid-Cretaceous (Cenomanian-Turonian, ca. 95 Ma), then steadily declined towards the Cretaceous-Paleogene boundary (Fletcher et al., 2008;Wang et al., 2014). It is thus important to consider this varying pCO 2 when exploring the mechanisms of the angiosperm physiological changes.
Finally, the goal of our study is to evaluate the combined environmental and physiological-induced response of Cretaceous vegetation to a low D v representative of proto-angiosperms. Our experimental design considers both reducing leaf hydraulic 80 or/and photosynthetic capacities and applying different pCO 2 scenarios in a coupled atmosphere-vegetation model. The modeled g s allows to account for long-term plant adaptation to their changing environment. We evaluate the physiological response of the vegetation to low hydraulic and photosynthetic capacities at the leaf scale, then at the global scale. It leads us to evaluate water and carbon exchanges between the land-surface and the atmosphere. In this study, we use LMDZ and ORCHIDEE versions embedded in the IPSL-CM5A2 model (Sepulchre et al., 2020). LMDZ is the atmosphere general circulation model of the Institut Pierre-Simon Laplace earth system model. LMDZ couples a dynamical core solving the primitive equations of conservation with a set of physical parameterizations of the radiative and convective schemes. The spatial resolution of LMDZ is the regular 3.75 x 1.875°longitude-latitude grid. LMDZ has 39 vertical levels 90 to describe vertical processes (Hourdin et al., 2013). ORCHIDEE is the land surface component of the IPSL earth system model (Krinner et al., 2005). When coupled with LMDZ, it has the same horizontal resolution, and a global 4-meter soil water depth represented with a two-layer bucket model, that is adequate to simulate evapotranspiration fluxes consistent with the data (Guimberteau et al., 2014). ORCHIDEE simulates water and energy exchanges (Rosnay and Polcher, 1998;Ducoudré et al., 1993), and the key processes of the terrestrial carbon cycle such as photosynthesis, carbon allocation, or soil organic matter 95 decomposition for thirteen Plant Functional Types (PFTs). Each PFT gathers a set of plant species with similar functional characteristics that will be translated into a common set of parameter values. Photosynthesis, respiration, water and energy exchanges are computed every 30 minutes, while carbon-related slow processes (leaf phenology and carbon allocation) are computed on a daily basis (Krinner et al., 2005;Sitch et al., 2003).

100
Stomatal conductance is computed per canopy layer, with layer depth increasing exponentially from top to bottom of the canopy. This feature enables to account for the response to light radiation extinction through the canopy. Leaf stomatal conductance is coupled to photosynthesis based on original formulations from Ball et al. (1987) and Leuning et al. (1995). Within the model used in ORCHIDEE (Yin and Struik, 2009) s −1 ), the intercellular CO 2 partial pressure (C i in bar), the C i -based CO 2 compensation point in the absence of R d (C * i in bar) and the residual stomatal conductance to H 2 O (g 0 in mol m −2 [leaf] s −1 ), to account for a non-zero conductance when the carbon assimilation is zero (Farquhar et al., 1980;Ball et al., 1987;Yin and Struik, 2009). Finally, the leaf operational stomatal conductance to H 2 O is modulated by a factor fcpl, describing the strength of the coupling between A and g s , which is function of the leaf-to-air vapor pressure deficit (kPa) and that we will further name "leaf hydraulic capacity" : The intercellular CO 2 partial pressure is dependant of the leaf boundary layer conductance and the ambient air CO 2 partial pressure. The coupling factor relies on two empirical constants A1 (unitless) and B1 (kPa −1 ) : The photosynthesis scheme proposed by Yin and Struik (2009) solves analytically the net assimilation, the stomatal conductance and the leaf intercellular CO 2 concentration. The semi-empirical formalism of g s allows to account for both the structural conductance, linked to the morphologic and physiologic traits of leaves (e.g. D v , D s and S) developed on the long-term plant evolution history, and the dynamical conductance, related to the short-term stomatal opening and closing depending on the environment. A is limited by either the Rubisco enzyme (RuBP) activity (V c ) or the electron transport (V j ): V cmax is parameterized for each PFT as a function of leaf age (Krinner et al., 2005). V j (molCO 2 m −2 [leaf] s −1 ) represents 125 the regeneration rate of the RuBP. The latter depends on the radiation and the electron transport : is the actual rate of electron transport (Yin and Struik, 2009), dependent on irradiance and V jmax , the maximum rate of electron transport at saturating light (mol e − m −2 [leaf] s −1 ). V jmax is directly proportional to V cmax with the ratio V cmax /V jmax varying with surface monthly temperature (Kattge and Knorr, 2007). Then, when 130 discussing the changes in V cmax , we consider the proportional changes in V jmax , in the following. A increases with V cmax and pCO 2 , until to reach a plateau where light and temperature become the limiting factors. Both V c and V j are a function of the soil moisture stress for transpiration w lim soil , that ultimately controls photosynthesis. This factor depends on both the soil moisture and the root profile given per PFT, trees having deeper roots than grasses. w lim soil is unitless and ranges between 0 for fully moist soil and 1 for maximal water stress. The transpiration rate Tr (mm d −1 ) is a function of the potential evapotranspiration, that 135 is, the atmospheric evaporative demand, modulated by a serie of resistances : where r a and r c are the aerodynamic and canopy resistances respectively (s m −1 ), ρ is the air density (kg m −3 ), q * s is the saturation specific humidity of the evaporating surface (kg kg −1 ), function of the temperature, and q is the specific humidity of the surrounding air (kg kg −1 ). Calculation of the transpiration requires the integration of g s at the canopy level which is g c 140 (molH 2 O m −2 [ground] s −1 ), that is over the leaf area index (LAI) and expressed as : 2.1.2 Summary of dynamical coupling between stomatal conductance and photosynthesis in a given environment  At the leaf level, the structural resistance ( Fig. 1, in orange) represents traits (veins and stomata) that evolve at long time scales.
The structural resistance drives the maximum theoretical conductance of water (g max anat ). The dynamical resistance (Fig. 1, in 145 dark blue) controls the stomatal closure on a very short time scale, depending to the external forcing : pCO 2 , light, vapor pressure deficit and soil moisture stress for transpiration ( Fig. 1, in light blue). However, the structural resistance defines the upper boundary condition that constrains the dynamical resistance. The coupling factor (fcpl) controls the sensitivity of stomatal conductance to the surrounding environment : a low fcpl implies that g s is less sensitive to C i , A and the vapor pressure deficit than a high fcpl. Plants can also slightly adapt their V cmax and V jmax , in absence of nitrogen stress, on seasonal timescale and 150 in a greater range on evolutionary long time scales. The ORCHIDEE version used in this study does not include the nitrogen cycle, as we do not have good constraints regarding the C:N ratio of Cretaceous vegetation and soils. The end limit of the assimilation process is the available energy from light that controls the quantity of RuBP that can be regenerated (V j in Eq.
(5)). At high pCO 2 , C i is maintained at a sufficient level to feed the carboxylation rate (V c ) and then, for a sufficient V cmax , light becomes the main limiting factor even for a small conductance (Eq. (3) to (5)). So, the structural resistance is not a limiting 155 factor and plants will close the stomata to limit the water loss. Likewise, V c rate increases, with both C i and V cmax , and the level of V cmax for which V j becomes the main limiting factor decreases (Eq. (3)). Because nitrogen acquisition and RuBP protein maintenance consume a lot of energy, plants tend to optimize the V cmax and V jmax to reach the point where V c and V j are colimiting (Maire et al., 2012;Stocker et al., 2020). On the contrary, at low pCO 2 structural conductance (that drives the maximum conductance) can become a limiting factor since, even with full stomata opening, it can limit C i under the level 160 to feed the carboxylation rate. V cmax and V jmax should also be increased to maintain a sufficient level of carboxylation (V c ) (lower efficiency of the oxidation/carboxylation ratio).
At the canopy level ( Fig. 1), the canopy conductance g c depends on both leaf level conductance g s and LAI. As LAI is a function of plant productivity, there is an additional positive feedback: a change in g s will impact GPP and then LAI in the same direction, which will amplify the initial effect of g s on g c . Another external forcing is the soil water availability (w lim soil ).

165
In the model used, water stress acts directly on V cmax and V jmax , as supported by previous studies (Keenan et al., 2009(Keenan et al., , 2010Egea et al., 2011). A low soil water content will induce a water stress limiting the V cmax and V jmax that will indirectly also reduce g s . Hence, arid ecosystems will be less sensitive to structural change in g s than ecosystems without a large hydric stress.
On the contrary, they will be affected in the same way for direct changes on V cmax and V jmax .

170
Vein density as well as stomatal size and density are both used to reconstruct past variations of g max anat (Franks and Beerling, 2009a, b;Brodribb et al., 2007;Brodribb and Feild, 2010;De Boer et al., 2012;Franks and Farquhar, 2001). Here, we have chosen to account for D v changes rather than D s and S. Indeed, using D v is a good proxy to constrain g max anat since D v and D s are correlated and that observed relationship between D v and g max anat gives the highest correlation coefficient (Feild et al., 2011b). Vein densities from angiosperm fossils published in Feild et al. (2011b) record a 2 to 5-fold increase in angiosperm D v during the Cretaceous (Fig. 2a), compared to early angiosperms and non-angiosperms (Table S2).
D v allows to reconstruct the maximal water that can flow through the stomata g max anat (mol m −2 [leaf] s −1 ) using the relationship developed by Brodribb et al. (2007) and Brodribb and Feild (2010) : Where VPD is the leaf-to-air vapor pressure deficit (MPa), ∆Ψ leaf is the leaf water potential gradient (MPa), D v is the vein 180 density (mm mm −2 [leaf]) and y is the distance from vein terminals to epidermis (µm). The past VPD and ∆Ψ leaf are set to 0.002 MPa and 0.4 MPa respectively, values that are typical for temperate-tropical environments (Brodribb and Feild, 2010).
An estimation of y is 70-130 µm, producing a span of predicted values encompassing the likely morphological variability of leaves (Brodribb and Feild, 2010). A lower value for y indicates a lower stomata-to-vein distance and a higher g max anat . The highest increase in g max anat is obtained 185 for the largest variation of D v over time combined with the smallest y. Conversely, the lowest increase in g max anat is obtained for the smallest variation of the highest D v values over time, combined with the highest y (Table S2). From the early to the late Cretaceous, the 2 to 5-fold increase in angiosperm D v corresponds to a 3 to 5-fold increase in g max anat ( Fig. 2b and Table S2). Thus, fossil D v provides an estimation of the increase in g max anat over time. Our land surface model does not explicitly represent vegetation traits nor g max anat but only g s , the operational stomatal conductance. However, based on the strong relationship between g max anat and g s (Dow et al., 2014;Dow and Bergmann, 2014), one assumes that variations of g s due to the long-term evolution of leaf hydraulic and photosynthetic capacities together with that of environmental factors such as pCO 2 would reflect into proportional changes on g max anat .

Experimental setup
To assess the impact of angiosperm leaf evolution on transpiration and photosynthesis, we performed climate model simulations 195 while varying leaf hydraulic or photosynthetic capacity, or both, for two contrasted pCO 2 levels ( Table 2). This is not a direct functional trait-based modeling approach, but we mimic paleo-traits through constraints on the maximum operating stomatal conductance g s with different factors. The model continental boundary conditions include mid-Cretaceous (115 Ma) paleogeography reconstructions from Sewall et al. (2007). Global vegetation distribution is set by establishing a first-order estimate of correspondence between biomes, inferred from paleobotanical data (Sewall et al., 2007), and PFTs (Table S1).
200 Figure 3 shows only the dominant PFT, but several PFTs can be present in a pixel. For instance, regions indicated as C3 grass are savannas with a fraction of trees (Table S1). We prescribe PFT distribution rather than activate the dynamical vegetation model to constrain our interpretations of the sensitivity experiments and avoid additional feedback linked to vegetation dynamics.
This configuration allows the carbon cycle to be activated and the LAI to be prognostically calculated as a function of dynamic carbon allocation (Krinner et al., 2005).  Although scattered paleo soil texture data do exist for the Cretaceous, sensitivity studies carried on present-day have shown a rather weak sensitivity of the terrestrial water budget to soil texture in the ORCHIDEE model (Tafasca et al., 2020). Thus, we set this parameter worldwide to an averaged value corresponding to loam in Zobler classification (Zobler, 1986) : 0.39 % silt, 0.43 % sand and 0.18 % clay. Soil color determines the bare soil albedo and is set globally at 0.16 (Wilson and Henderson-Sellers, 1985). We set the solar radiation at 99 % of the current value (Gough, 1981) and the orbital parameters as today 210 (Laskar et al., 2004). The ocean is prescribed through sea-surface temperatures obtained from an unpublished Aptian fullycoupled simulation obtained with the ESM IPSL-CM5A2 (Sepulchre et al., 2020). To assess the impact of angiosperm trait evolution on transpiration and photosynthesis, we consider (i) two different factors applied to angiosperm-PFTs only, and (ii) one external forcing (pCO 2 ), that alter the stomatal conductance in our land-surface model (Table 2). Change in leaf hydraulic capacity is considered via a change in the coupling factor (fcpl) that describes the slope of the relationship between g s and the dynamical driving parameters (namely A, pCO 2 and vapor pressure deficit). Because fcpl is related to the structural conductance, the change in g max anat can then be empirically represented by a factor α, modulating fcpl in the g s formulation (Eq. (1)). Changes in leaf photosynthetic capacity are taken into account via changes in V cmax (Eq. (4)). To this end, we use a factor β that is applied to both V cmax and V jmax (i.e V cmax /V jmax ratio kept constant, Eq. (3) to (5)). We carry out three different factorial experiments using the upper value of changes in g max anat induced from D v proxy as depicted earlier.

220
(i) α is set to 1 /5 while β is 1, accounting for a direct decrease of angiosperm leaf hydraulic capacity by a factor of 5 (NOAN-GIOh, Table 2). (ii) β is set to 1 /5 while α is 1, corresponding to a direct decrease of angiosperm leaf photosynthetic capacity by a factor of 5 (NOANGIOp, Table 2). This sensitivity experiment is supported by the coupling between stomatal conductance and assimilation in the model and by several studies suggesting to mimic proto-angiosperm capacities by decreasing modern V cmax by a factor of 5 Lee and Boyce, 2010). Since leaf hydraulic and photosynthetic capacities are 225 likely to have co-evolved over the Cretaceous (Franks and Beerling, 2009a;De Boer et al., 2012), (iii) we set α and β to 1 /5, i.e. with half the forcing simultaneously applied to the leaf hydraulic and photosynthetic capacities (NOANGIOhp, Table 2).
These three sensitivity tests represent three likely Cretaceous worlds with proto-angiosperm types. We perform the reference control experiment where the factors α and β are set to 1 in order to keep the modern leaf hydraulic and photosynthetic capacity of extant angiosperms as the reference ones (ANGIO, Table 2). Since pCO 2 simultaneously impacts leaf hydraulic and pho-230 tosynthetic traits evolution on the long-term (Franks and Beerling, 2009a), we repeat this set of experiments for two different extreme pCO 2 forcings. At 1120 ppm, we refer to the mid-Cretaceous (115 Ma) estimates, contemporary with the beginning of the angiosperm radiation. At 280 ppm, we make a sensitivity test to assess the response of proto-angiosperm vegetation to a preindustrial-like pCO 2 . Regarding the average g s value over a long time period, we assume that the g s /g max anat ratio is a constant at long-time scales because of plants structural adaptation to their environment (Dow et al., 2014;Dow and Bergmann, 2014). 235 We consider that all the changes in the model operational g s , arising from the above-mentioned proto-angiosperm parameterizations, reflect that of the maximal anatomic stomatal conductance. The experiments were run for 60 years, a sufficient time to balance gross primary productivity and evapotranspiration. We analyse and show the averaged last 10 years of simulations.

Leaf operating stomatal conductance 240
The leaf stomatal conductance to H 2 O (g s , mol m −2 [leaf] s −1 ) is the global mean of annual daylight leaf stomatal conductance of the top of canopy leaves (where exposure to sunlight is maximal). Figure 4 shows that leaf stomatal conductance decreases as we sequentially reduce leaf hydraulic (NOANGIOh) or photosynthetic (NOANGIOp) capacities, or both (NOANGIOhp), to mimic the proto-angiosperm world. Reducing the high leaf hydraulic and/or photosynthetic capacities from modern angiosperm leads to a 3-fold drop of leaf stomatal conductance on average, a factor in the lowest bracket of the range expected from g max anat 245 inferred from the fossil record (Fig. 2b). Within every PFT group, g s decrease is similar amongst sensitivity experiments and validates our chosen combinations of values for α and β. The maximal decrease in g s is simulated for the tropical broad-leaved evergreen and the C3 grass PFTs which also depict the largest stomatal conductance in the reference ANGIO simulations. Leaf hydraulic and photosynthetic capacities have not been altered for the temperate needleleaf evergreen PFT, as they correspond to gymnosperms. Thus, gymnosperm g s was expected to be constant for each of the pCO 2 group of experiments. However, g s 250 slightly decreases in the sensitivity experiments for this PFT, suggesting a feedback from the atmosphere-vegetation coupling that ultimately altered soil moisture stress and stomatal conductance. Indeed, soil moisture stress for gymnosperm transpiration increases for all the simulations with the proto-angioperm vegetation prescribed (Fig. S2)   Simulations forced with "Cretaceous-like" pCO 2 (1120 ppm) show systematically lower g s values than simulations with pCO 2 set at preindustrial (280 ppm). This illustrates that the model well reproduces the expected closure of stomata as a response to high pCO 2 , in direct link with Eq. (4) and (5). This decrease in g s with increasing pCO 2 is the highest (52 % on average) for experiments with the modern-like leaf hydraulic and photosynthetic capacities (ANGIO(1120) versus ANGIO (280)).
For these experiments, tropical broad-leaved evergreen PFT reacts to the high pCO 2 with g s values dropping from 0.14 mol 260 m −2 [leaf] s −1 at 280 ppm to 0.06 mol m −2 [leaf] s −1 at 1120 ppm, while that of C3 grasses drops from 0.09 to 0.04 mol m −2 [leaf] s −1 . At 1120 ppm, as the leaf stomatal conductance of ANGIO runs is already strongly reduced compared to that at 280 ppm, the relative decrease in g s when modern leaf hydraulic and photosynthetic capacities are reduced is also less important : -66 % for tropical broad-leaved evergreen at 1120 ppm to be compared to -73 % for the same PFT at 280 ppm (Fig. 4).  for regions dominated by tropical broad-leaved evergreen, that are set in mid-to-high latitudes in the specific Cretaceous configuration (Fig. 3). LAI is also slightly higher at 1120 ppm than at 280 ppm, as a response to CO 2 fertilization effect on photosynthesis that increases biomass production (Fig. 1). The vegetation response to the prescribed perturbations of fcpl and V cmax strongly differs depending on the pCO 2 . At high pCO 2 , LAI is unchanged with the sole perturbation of the leaf hydraulic capacity (fcpl* 1 /5, NOANGIOh, Fig. 5c) meaning that even with a reduced g s at the leaf level, the high pCO 2 allows for the 275 fertilization effect to drive biomass growing (C i in Eq. (1)). Conversely, V cmax perturbation (NOANGIOp, Fig. 5e) induces a strong decrease in the vegetation cover of all angiosperm PFTs, that is replaced by bare soil (Fig. S4) (1120). Lastly, when both fcpl and V cmax perturbations are combined (fcpl* 1 /5 and V cmax * 1 /5, NOANGIOhp, Fig. 5g), LAI is mostly unchanged, suggesting that in a high pCO 2 context, slightly reduced leaf hydraulic and photosynthetic 280 capacities does not affect actual assimilation. At high pCO 2 , the impact of reducing V cmax on photosynthesis is limited because the limiting factor is likely V j (and then light) rather than V c (and C i ) (Eq. (3) except in the very southern latitudes, where no significant change is simulated (Fig. 5d). The strongest signal occurs with the 285 reduction of leaf photosynthetic capacity through V cmax (NOANGIOp, Fig. 5f). Apart from the unchanged needleleaf PFT, the modified vegetation cannot grow in NOANGIOp experiment at 280 ppm, as shown by LAI values close to zero for every angiosperm PFTs. Simulated LAI in NOANGIOhp shows that at a lower V cmax associated with a lower fcpl, tropical PFTs also globally collapse, although vegetation is maintained at high latitude ( Fig. 5h and S4h). At 280 ppm, intercellular CO 2 partial pressure (Eq. (1) to (5)) becomes more limiting than the solar energy whatever the vegetation prescribed. The latitudinal 290 gradient of LAI reduction testifies to another limiting factor to be integrated, which plays a role at low latitudes but not at high latitudes. Figure S5 shows that the soil moisture stress for transpiration (Fig. 1, Eq. (4) and (5)) is larger in the tropics to mid-latitudes and near zero at high latitudes. As expected, water in the soil available for transpiration is thus a limiting factor of plant growth in the tropics and mid-latitudes at low pCO 2 .

Canopy operating stomatal conductance
Albeit the fossil record and empirical models can help to infer the foliar maximal stomatal conductance to H 2 O, the land surface model is required to explore the stomatal conductance to H 2 O at the canopy level that is given by Eq. (7) : g c is the integral of g s over the entire canopy and is inversely proportional to the canopy resistance r c . The canopy stomatal conductance is the global mean of annual daylight canopy stomatal conductance and is weighted by surface unit of ground of a grid-cell where each PFT is found (i.e. written ground P F T ). g c is the combined result of changes in g s arising from the forcing factors 300 α and β, and changes in vegetation leaf area, i.e., both have an influence that can either be reinforcing or counterbalancing. As expected, figures 4 and 6 show significant differences between g s and g c in respect to the pCO 2 prescribed: while g s decreases by about a factor of 3 for each proto-angiosperm sensitivity experiment compared to the reference (Fig. 4), the decrease in g c is not equal through experiments (Fig. 6). The dramatic drop in g c when reducing the high leaf photosynthetic capacity (NOANGIOp) is well explained by the significant LAI decrease (Fig. 5e and f). Besides, the effect of pCO 2 is more complex 305 at the canopy scale. Indeed, g c is still reduced in the perturbed experiments, but is not systematically lower at 1120 ppm than at 280 ppm ( Fig. 6) as opposed to g s (Fig. 4).  NOANGIOh simulations do not show a change in g c between 280 ppm and 1120 ppm. A decrease in fcpl by a factor of 5 implies a reduction of the LAI at 280 ppm (Fig. 5d) because plants cannot assimilate enough carbon to sustain their biomass (C i in Eq. (1) to (5)) whereas it is not the case at 1120 ppm (Fig. 5c). Indeed, at the canopy scale, at low pCO 2 , higher g s just 310 makes up the LAI loss. Conversely, at high pCO 2 , lower g s is compensated by the higher LAI, ultimately making angiosperm g c being almost identical for the two pCO 2 scenarios. In the NOANGIOp and NOANGIOhp experiments, g c is higher at 1120 ppm than at 280 ppm. By increasing the carbon assimilation during photosynthesis, high pCO 2 allows for higher vegetation cover ( Fig. 5e and g) and higher canopy stomatal conductance whereas low pCO 2 together with low leaf photosynthetic capacity imply a collapse of the vegetation (Fig. 5f and h).

315
LAI changes described earlier modulate the relationship between g s and g c (Fig. 1). At low pCO 2 , the LAI decrease between the ANGIO and any of the three perturbed experiments (Fig. 5d, f and h) strengthens the initial decrease in g s (positive feedback, Fig. 4b and 6b). Mean g s decreases by 69 %, 73 % and 69 % while mean g c decreases by 72 %, 97 % and 91 % respectively for NOANGIOh, NOANGIOp and NOANGIOhp compared to ANGIO (Fig. 4b and 6b). Indeed, decreasing V cmax under low pCO 2 implies a reduction of the plant capacity to assimilate carbon (Eq. (4)) and directly impacts the GPP ( Fig. S6f and h) 320 and then the LAI at the canopy scale ( Fig. 5f and h). However, decreasing fcpl and thus g s reduce the CO 2 concentration at the chloroplast level (C i in Eq. (1)) and have only an indirect effect on GPP (Fig. S6d) and thus on LAI (Fig. 5d). At high pCO 2 , the LAI is almost sustained for NOANGIOh and NOANGIOhp compared to ANGIO (Fig. 5a, c and g) because the assimilation remains high when C i is not the limiting factor (Fig. S6a, c and g). The latter lessens the initial decrease in g s (Fig. 6a) on g c (Fig. 4a). Nevertheless, NOANGIOp(1120) experiment shows a much lower g c than the two previous experiments because of 325 the direct impact of decreasing V cmax on the LAI (Fig. 5e). Therefore, comparing g c of perturbed simulations to that of the reference allows us to account for the structural conductance linked to plant trait evolution at the canopy level.

Transpiration rate
As shown in Eq. (6), transpiration rate is controlled by (i) the atmospheric evaporative demand (Fig. S7), that depends on air 330 vapor pressure deficit, the surface temperature (Fig. S8) as well as the aerodynamic resistance, and (ii) the capacity of plants to transpire, driven by the canopy conductance (Fig. S9). For both pCO 2 cases, experiments with modern-like vegetation depict the highest transpiration rates in the tropics and the mid-latitudes, where they reach up to 2.5 and 3 mm day −1 at 1120 and 280 ppm, respectively ( Fig. 7a and b). These regions have optimal conditions in water availability in the soil (Fig. S5) and light, resulting in higher canopy stomatal conductance ( Fig. S9a and b) together with a high atmospheric evaporative demand (Fig. 335 S7a and b) as a response to high temperatures ( Fig. S8a and b). Transpiration is also slightly strengthened (+ 0.2 mm day −1 ) at low pCO 2 compared to high pCO 2 ( Fig. 7a and b) as a consequence of higher g c (Fig. 4 and 6). Parameterizing the vegetation without the modern angiosperm leaf hydraulic and photosynthetic capacities systematically leads to lower transpiration rates (Fig. 7). Overall, in line with changes depicted for LAI and g c , transpiration rates react stronger in a 280 ppm world than in a 1120 ppm world to decreasing leaf hydraulic and photosynthetic capacities. At 280 ppm, NOANGIOh shows a decrease of 0.6 mm d −1 (-44 %) in transpiration compared to ANGIO, especially over equatorial Gondwana and paleo Southeast Asia, whereas the decrease is limited to 0.3 mm d −1 (-24 %) at 1120 ppm ( Fig. 7c and d) as a response to g c changes described earlier (Fig. 6). Transpiration also significantly drops when leaf photosynthetic capacity alone is reduced (NOANGIOp, Fig. 7e and   345 f). At 1120 ppm, transpiration drops by 0.5 mm d −1 (-53 %) (Fig. 7e). The signal is stronger at 280 ppm, where a complete collapse of transpiration is simulated (Fig. 7f). This latter result is a direct consequence of the LAI collapse described earlier ( Fig. 5e and f). Combining reduction in leaf photosynthetic and hydraulic capacities (NOANGIOhp) leads to little decreases in transpiration rate at 1120 ppm (Fig. 7g), comparable to NOANGIOh (Fig. 7c), because the high pCO 2 prevents the decrease in carbon assimilation (Fig. 5g) and canopy stomatal conductance (Fig. 6). Conversely, at low pCO 2 , the limitation of C i (Eq.
Transpiration anomalies result from variations in the atmospheric evaporative demand (Fig. S7) combined with that of the canopy stomatal conductance (Fig. S9), arising from g s and LAI changes which are modulated by the soil moisture stress for transpiration and light. Transpiration anomalies are located over the paleo-tropics ( Fig. 7c and d) for NOANGIOh compared 355 to ANGIO experiments because decreasing leaf hydraulic capacity by a factor of 5 acts mainly on wet regions where plants efficiently transpire in the ANGIO world ( Fig. 1 and Eq. (1) to (5)), while arid belt regions are less sensitive to any change in g c . In theses regions, transpiration is already constrained by water shortage, so the upper stomatal conductance limit is not the limiting factor. However, mid-latitude do not show transpiration anomalies because the decrease in g c (Fig. 6, S9c and d) is compensated by the increase in atmospheric evaporative demand ( Fig. S7c and d) that strengthens transpiration rate.

360
In contrast, decreasing leaf photosynthetic capacity (NOANGIOp) acts more globally on terrestrial plants and explains the widespread transpiration anomalies (Fig. 7e and f), always modulated by the atmospheric evaporative demand (Fig. S7e and f).
At 1120 ppm, NOANGIOhp shows the same pattern as NOANGIOh ( Fig. 7c and g) because plants are more sensitive to reduction in fcpl than to reduction in V cmax (C i is not limiting) while the decrease in transpiration is more extended at 280 ppm ( Fig.   7h) because it arises mainly from the decrease of V cmax at low pCO 2 that drives carbon assimilation to decline (Eq. (3) and (4)).

Water use efficiency
The annual mean water use efficiency (WUE) is the ratio of annual mean GPP to annual mean transpiration over each grid point (gC kgH 2 O −1 ). It pictures the capacity of vegetation to maximise carbon uptake while minimizing water loss and thus is a good indicator of plant adaptation to their environment. For the modern-like vegetation (ANGIO), maximal values of WUE 370 are found in the mid to high-latitudes where optimal conditions of water and light ( Fig. 8a and b, Fig. S5) are available for the vegetation. WUE increases with pCO 2 : while it barely exceeds 10 gC kgH 2 O −1 at 280 ppm, it reaches more than 15 gC kgH 2 O −1 at 1120 ppm. This is explained by the CO 2 fertilization effect generated by the high pCO 2 on C i (Eq. (3) to (5)) that strengthens plant productivity. At 1120 ppm, the highest value of WUE is found for NOANGIOh, with a 1.7 gC kgH 2 O −1 (+ 30 %, Fig. S10c) increase compared to ANGIO (Fig. 8a and c). Indeed, reducing the leaf hydraulic capacity does not 375 imply a change in GPP which remains similar to that of ANGIO(1120) (Fig. S6a and c) but rather generates a slight decrease in transpiration rate (Fig. 7c), explaining the increase of WUE. Therefore, plants with lower leaf hydraulic capacity than today are better adapted to the high pCO 2 environment. For NOANGIOhp(1120), the WUE is partly reduced compared to ANGIO(1120) but stays at a high level ( Fig. 8a and g). It results from the larger decrease in GPP than in transpiration rate ( Fig. S6g and 7g), as GPP is more sensitive to the V cmax reduction (Eq. (4)) than transpiration is to the fcpl reduction (Eq. (1) and (6)). In 380 contrast, WUE is largely degraded in NOANGIOp(1120) compared to ANGIO(1120) (Fig. 8a and e). Both GPP (Fig. S6a and e) and transpiration rate (Fig. 7a and e) significantly drop when reducing the leaf photosynthetic capacity (NOANGIOp) at high pCO 2 . However, V cmax acts directly on the carbon assimilation (Eq. (3) and (4)) while it is more indirect on the stomatal conductance, thus implying that the reduction of V cmax has a larger effect on GPP than on transpiration rate. In contrast to experiments at high pCO 2 , ANGIO(280) gives the highest WUE at low pCO 2 . NOANGIOh(280) depicts a lower WUE than 385 ANGIO(280) (Fig. 8b and d) which demonstrates that plants with lower leaf hydraulic capacity than today are less adapted to the low pCO 2 environment. At low pCO 2 , the low C i decreases GPP (Eq. (4), Fig. S6b and d) while it increases g s (Eq. (1), Fig.   4), modulating the transpiration decrease (Fig. 7d). However, at low pCO 2 , WUE collapses to very low values for NOANGIOp and NOANGIOhp ( Fig. 8f and h), that is driven by the large decrease in GPP when combining the low C i to the reduction of V cmax (Fig. S6f and h). Once the leaf photosynthetic capacity is decreased, changes in GPP are the main contributor to the 390 changes in WUE whatever the pCO 2 level prescribed.

How to better account for the proto-angiosperm conductance traits in land surface models?
Fossil maximal anatomic stomatal conductance has been widely used to estimate the maximum of water flow through the stomata before and after the angiosperm radiation. Still, determining how a 5-time increase in maximal anatomic stomatal 395 conductance translates into actual flux at the top of the canopy is challenging. We show that the complete response of the vegetation to evolving physiological and morphological traits is modulated by environmental factors such as pCO 2 , light, vapor pressure deficit and water availability in the soil (Fig. 1).
The simplest representation of proto-angiosperm vegetation is to account for the lower maximal anatomic stomatal conductance by a factor of 5, consistent with the fossil records, directly by applying this factor to fcpl in the calculation of the leaf stomatal 400 conductance (Eq. (1)). Our results show that a lower fcpl of angiosperms at high pCO 2 does not change plant photosynthesis ( Fig. 5c and S6c) but slightly decreases transpiration (Fig. 7c), driving WUE to increase (Fig. 8c) compared to the modern angiosperm prescription. Hence, a lower maximal stomatal conductance at high pCO 2 appears as an advantage compared to modern angiosperms because of a better optimization of carbon uptake over water loss. At low pCO 2 , both transpiration ( Fig. 7d) and photosynthesis are decreased because a reduced LAI entails a reduction of canopy stomatal conductance, which 405 strengthens the initial reduction of leaf stomatal conductance compared to the modern vegetation. Despite the absence of a direct proxy for fossil plant maximum V cmax , several studies have suggested to mimic the proto-angiosperm capacities by decreasing modern V cmax Lee and Boyce, 2010), rather than fcpl, by a factor of 5. This approach is supported by our knowledge of modern plant processes that stomatal conductance interacts with assimilation in order to optimize carbon gain against water loss (Bonan, 2015) and is made possible by the coupling between leaf stomatal conductance 410 and leaf photosynthetic capacity in our land-surface model (Farquhar et al., 1980;Ball et al., 1987;Krinner et al., 2005;Yin and Struik, 2009). However, applying this method leads vegetation cover to decrease at high pCO 2 (Fig. 5e and S6e) or even collapse at low pCO 2 (Fig. 5f and S6f), which is not recorded in the fossil record. As a consequence, transpiration rates ( Fig.   7e and f) and WUE (Fig. 8e and f) significantly decrease at high pCO 2 and are zero at low pCO 2 compared to the modern vegetation. Hence, taking into account stomatal conductance reduction only through V cmax reduction does not appear to be 415 adequate. However, several studies suggested a decrease of V cmax with increasing pCO 2 (Ainsworth and Rogers, 2007) driven by (i) a coevolution between stomatal conductance and V cmax through time (Franks and Beerling, 2009a;De Boer et al., 2012) and (ii) the photosynthesis coordination theory that states that plants optimize V cmax to be near the co-limitation between carboxylation rate and RuBP regeneration (Maire et al., 2012;Stocker et al., 2020). This theory has been recently improved considering also the cost related to stomatal conductance (Stocker et al., 2020). This V cmax limitation is related to the high 420 energetic (and then respiration) cost needed to maintain a high level of Rubisco (acquisition of nitrogen). Rather than the two extreme cases that decrease leaf hydraulic or photosynthetic capacity of angiosperms by a factor of 5, we consider a covariation of fcpl and V cmax , by applying half the forcing given by the fossil records (i.e. 1 /5) to fcpl directly and the other half to V cmax . Experimental studies on extant plant types (Lin et al., 2015) have shown differences in water-use strategy between modern angiosperm trees and gymnosperm trees. They argue that modern angiosperm trees have 2 times higher stomatal 425 conductance sensitivity response to driving factors than gymnosperm trees, showing that our choice for fcpl* 1 /5 together with V cmax * 1 /5 seems the most realistic. When applying this factor jointly to fcpl and V cmax at a high pCO 2 , our results suggest that vegetation is barely impacted, with a slight reduction of GPP (Fig. S6g), that remains sufficient to sustain LAI values close to those of the control scenario (Fig. 5g), and a relatively high WUE (Fig. 8g) compared to the modern vegetation ( Fig. 5a and 8a). Conversely at low pCO 2 , LAI collapses (Fig. 5h) and GPP, transpiration and WUE reach zero (Fig. S6h, 7h   430 and 8h) as a response to carbon assimilation drop.
At 280 ppm, sensitivity tests with a low leaf photosynthetic capacity or together with a low leaf hydraulic capacity can be considered as extrema where plants cannot grow ( Fig. 5f and h, S6f and h) whereas experiment with modern trait vegetation corresponds to a maximum in plant productivity and transpiration (Fig. 5b, 7b and 8b). At 1120 ppm, two sets of forcing factors simulate a sustainable paleovegetation productivity : lowered leaf hydraulic capacity only or lowered leaf hydraulic 435 and photosynthetic capacities together ( Fig. 5c and g, S6c and g). However, lowering only the leaf hydraulic capacity while keeping the high V cmax as in the modern vegetation is probably not realistic. Indeed, under high pCO 2 , because of increasing limitation of Rubisco regeneration, the actual photosynthesis can be maintained with a smaller V cmax . Sustaining a given level of V cmax is associated with a respiration cost, for nitrogen acquisition and protein maintenance. So, to optimize its carbon gain, we expect the plant to also have a lower V cmax . Although our model does not represent the nitrogen cycle, we infer that this 440 supplementary cost would lead to a decrease in productivity when simulating the proto-angiosperm vegetation with a modern leaf photosynthetic capacity together with a low leaf hydraulic capacity. Taking into account the differentiated response of vegetation to a declining pCO 2 , we suggest that simultaneously decreasing fcpl and V cmax by a factor of 1 /5 is the most realistic representation of proto-angiosperm vegetation. The latter is consistent with a previous study of Franks and Beerling (2009a) suggesting that past fluctuations in pCO 2 acted as a forcing on both V cmax and stomatal conductance. 445 Finally, our study confirms the hypothesis that paleovegetation has evolved from a likely state with a relatively low stomatal conductance (Fig. 4a) and V cmax under high pCO 2 that allowed nonetheless to sustain a high productivity (Fig. 5g, 7g and 8g) to a state of high stomatal conductance (Fig. 4b) and V cmax as found in the modern vegetation, in order to maintain high productivity under low pCO 2 (Fig. 5b, 7b and 8b). The study also shows that paleovegetation characteristics can be better represented by parameterizing models fully describing the coupling between stomatal conductance and plant productivity 450 from leaf to the canopy scale. Furthermore, we show that decreasing leaf hydraulic and/or photosynthetic capacities does not coincide with a decrease of the leaf operational stomatal conductance to the same extent. Indeed, accounting for a decrease by a factor of 5, given by the maximal bound of the range expected from the maximal anatomic stomatal conductance, leaf stomatal conductance is only 3-fold lower than the reference. This difference is due to the fact that plants never operate at the level of the maximal anatomic stomatal conductance because of light and water limitation effects on photosynthesis. While some previous 455 studies directly decrease the transpiration rate by the factor of vein density changes from the fossil record (Boyce et al., 2009;Boyce and Lee, 2017;Boyce et al., 2010), we suggest to explicitly represent changes in leaf hydraulic and photosynthetic capacities. Nevertheless, when embedding parallel changes on pCO 2 , leaf stomatal conductance of plants with reduced leaf hydraulic and photosynthesis capacities at high pCO 2 is nearly 6-time lower than the reference one at low pCO 2 . It confirms that accounting for pCO 2 changes is mandatory to model the evolution of angiosperm leaf traits. Our work relies upon the assumption that stomata aperture maximizes carbon gain while minimizing water loss (Bonan, 2015).
Both photosynthesis and stomatal conductance to H 2 O are sensitive to environmental variables such as light, pCO 2 and water availability in the soil (Fig. 1). As pCO 2 has varied a lot through the Cretaceous (Fletcher et al., 2008;Wang et al., 2014), 465 plants had to adjust their stomatal conductance. Our study suggests that having a high stomatal conductance, which means a large vein density and a high V cmax , provides little advantage compared to a low stomatal conductance under high pCO 2 . In contrast, having a high stomatal conductance under low pCO 2 may confer a competitive advantage over plants with limited stomatal conductance to assimilate carbon. But this higher stomatal conductance should be linked to an increase in V cmax to sustain growth. In that sense, fossil records provide evidence of plant traits evolution during the angiosperm radiation. Fossils 470 of early angiosperms show that vein density was as low as that of other plant types (Feild et al., 2011b) because having a high vein density under high pCO 2 was not necessary for the plants to grow. Then, the pCO 2 likely declined (Fletcher et al., 2008;Wang et al., 2014). For this time, our results support the hypothesis that angiosperms evolved towards leaves with increasing vein density combined with a more efficient biochemistry that allowed them to have an increasing stomatal conductance and photosynthetic capacity to counteract the effect of pCO 2 decrease on carbon assimilation. Among other factors, this 475 evolution of physiological leaf traits has given a competitive advantage to angiosperms compared to gymnosperms dominating the vegetation of the period that enabled them to colonize almost all the terrestrial ecosystems. Our results are consistent with those of Franks and Beerling (2009a), which have shown that WUE co-evolves positively with variations in pCO 2 over the Phanerozoic : periods with high pCO 2 enhanced GPP while simultaneously allowing a reduction of transpirational water losses due to reduced stomatal conductance. They also show that even after the evolution of angiosperm leaf morphology and 480 biochemistry, WUE is estimated to have been at its lowest level since the Carboniferous. Our model consistently represents the range of WUE deducted by Franks and Beerling (2009a) under different pCO 2 : between 5 and 9 gC kgH 2 O −1 (Fig. 8a and b). Moreover, there is a likely co-adaptation of stomatal traits and leaf venation which implies a better optimisation of carbon gain against water loss (De Boer et al., 2012). Progressively, under decreasing pCO 2 , angiosperms with high stomatal density and low stomatal size (Franks and Beerling, 2009a, b) likely invested increasingly more energy in building more and 485 more veins to sustain the higher stomatal conductance and then carbon assimilation, while other plants did not. The innovation of angiosperms in dense water transport networks could have become a necessity to support higher stomatal conductance and prevent plant desiccation (De Boer et al., 2012).

What are the limitations of our modeling choices?
As a first step toward understanding the impact of trait evolution on the leaf hydraulic and photosynthetic capacities of an-490 giosperms, we have chosen to simulate the Aptian (115 Ma) because this time period corresponds to the first step of increasing vein density found in the fossil record (Feild et al., 2011a). To get an exhaustive view of the angiosperm evolution, future studies will benefit from considering similar experiments with boundary conditions set several million years before and after the Aptian. Specifically, exploring cold and warm extremes of the Cretaceous, such as the Cenomanian-Turonian (95 Ma) and the Maastrichtian (70 Ma) would be valuable, as climate and pCO 2 have been shown to vary a lot during these periods (Ladant 495 and Donnadieu, 2016).
The vegetation map we used (Fig. 3) results from two efforts of (i) compilation and spatialization of the Aptian paleobotanical records (Sewall et al., 2007) and (ii) conversion of the fossil data into plant functional types combination (Table S1). Each of these two steps include uncertainties that can propagate into our results, but can be hardly quantified. We acknowledge that the prescribed vegetation cover, especially in the tropics, can potentially alter the radiative balance and the hydrological cycle 500 (e.g., Port et al., 2016). It is however unlikely that the Aptian vegetation cover would be very different from the one provided by Sewall et al. (2007), given the compilation effort made for this reconstruction. Further studies could still circumvent this potential issue by running a full dynamical vegetation model, i.e. by allowing PFTs to spatially settle in regions where the simulated climate is the most appropriate.
Recent studies on past vegetation transitions indicate that differences in transpiration rates also arise from the ratio of carbon 505 over nitrogen (White et al., 2020;Richey et al., 2021). As mentioned earlier, the ORCHIDEE version we used does not explicitly represent the nitrogen cycle, preventing us from considering the additional cost to maintain high leaf photosynthetic capacity with lower leaf hydraulic capacity. The recent developments of a new version of ORCHIDEE that does include the nitrogen cycle (Vuichard et al., 2019), based on previous developments , will help to account for this process, provided that good constraints can be obtained regarding the C:N ratio of Cretaceous vegetation and soils.
Through the use of the coupled LMDZ and ORCHIDEE models, our approach includes the pivotal coupling between atmosphere and vegetation. However by using fixed sea-surface temperatures, we neglect the feedbacks from the ocean-atmosphere coupling that could occur as a response to simulated changes in vegetation cover. Although sensitivity experiments with strong changes in vegetation suggested ocean feedbacks could play a significant role on the continental hydrological cycle (Davin and 515 de Noblet-Ducoudré, 2010), our choice was motivated by (i) the will to focus on first-order continental processes, (ii) the computing cost required to equilibrate fully coupled simulations that typically require more than 3000 simulated years (Sepulchre et al., 2020) and the fact that comparable studies either used fixed-SSTs Lee and Boyce, 2010) or slab oceans (White et al., 2020).
Our parameterization of stomatal conductance in ORCHIDEE from Yin and Struik (2009) (Eq. (1)) is semi-empirical. A re-520 finement of our modeling approach would be to use a stomatal conductance model based on optimisation theory, that explicitly describes the stomata functioning so as to optimise carbon gain against water loss (Medlyn et al., 2011;Buckley and Mott, 2013;Buckley, 2017). In particular, the model we use here simply links external forcing to leaf stomatal conductance by an empirical term of coupling fcpl that describe all the processes related to stomatal conductance. The α factor applied in Eq. (1) to the coupling factor (fcpl) does not fully represent the change in the maximal anatomic stomatal conductance as it should be 525 considering changes inferred from vein density. We have seen that reducing the leaf hydraulic or photosynthetic capacity (or a combination of both) by a factor of 5 leads to a 3-fold decrease in leaf stomatal conductance. It emphasizes the need, in the future, to improve the parameterization of the stomatal conductance in global models by explicitly modeling both structural and dynamical conductance. Such a parameterization already exists for individuals plants (Dow et al., 2014), growing under controlled conditions, but not for land-surface models. Moreover, moving beyond PFT-based approaches towards truly trait-530 based ones would allow a more direct accounting for changes in functional traits.
Finally, we only consider the Cretaceous evolution of leaf hydraulic and photosynthetic capacity in this paper, as leaves are the end limit of plant water transport (Brodribb et al., 2007;Brodribb and Feild, 2010;Feild et al., 2011a, b). However, root thickness and wood structure for instance have evolved towards greater hydraulic efficiency (Brundrett, 2002;Wheeler and Baas, 2019). A comprehensive approach of the evolution of the plant hydraulic system would require to consider the entire soil-535 plant-atmosphere continuum, based on coupled resistances. Future studies would benefit from implementing a full hydraulic architecture of plants in the land surface model.

Conclusions
In line with recent studies focusing on Paleozoic vegetation transitions (White et al., 2020;Richey et al., 2021), the purpose of our study is to better represent past vegetation in earth system models by emulating "paleo-traits" in the vegetation parameteri-540 zations. Our approach involves an atmosphere-vegetation model, which couples stomatal conductance and carbon assimilation, motivated by an ecophysiological model based on angiosperm fossil records. Here, it allows us to evaluate three different pale-ovegetation prescriptions under two end-member scenarios of pCO 2 for the Cretaceous. We show that the simulated vegetation cover, transpiration rate and water use efficiency are sensitive to the paleovegetation trait prescribed. Only accounting for leaf hydraulic capacity reduction provides no significant change in LAI, GPP and transpiration, while slightly increasing WUE at 545 high pCO 2 . In contrast, global transpiration decreases at low pCO 2 because of the positive feedback between LAI and stomatal conductance. On the other hand, only accounting for leaf photosynthetic capacity reduction gives a substantial decrease or even a collapse of vegetation at high or low pCO 2 respectively, which is in contradiction to the fossil record. Combining a reduction of leaf hydraulic capacity with that of photosynthetic capacity does not affect the plant productivity and LAI at high pCO 2 while vegetation collapses at low pCO 2 . All the results in combination demonstrate that under high pCO 2 the reduced 550 stomatal conductance of the proto-angiosperm vegetation is not a limiting factor on productivity. It also shows that high values of V cmax as observed in modern angiosperms do not enhance plant productivity, whereas maintaining the high V cmax likely requires higher leaf nitrogen concentration and higher energy demand. Therefore, a combination of lower-than-modern leaf hydraulic and photosynthetic capacities seems the most realistic physiological parameterization for proto-angiosperms in the specific high pCO 2 context of the Cretaceous. This is supported by evidence of coevolution inferred from previous studies 555 (Franks and Beerling, 2009a, b) and the ratio of stomatal conductance between modern angiosperms and gymnosperms from in-situ experiments (Lin et al., 2015). Our results are also consistent with recent studies on coordination theory (Maire et al., 2012;Stocker et al., 2020).
Our study also suggests that proto-angiosperm vegetation with low leaf hydraulic and photosynthetic capacities was adapted to high pCO 2 , where the combination of both physiological constraints nonetheless allowed high productivity and WUE.

560
Conversely, it would not have been able to exist under low pCO 2 , where we simulated a collapse of GPP under such physiological parameter configuration. Modeling the full coupling between GPP and stomatal conductance allows to understand why increasing both structural conductance and maximum photosynthetic capacity, even at an expense of a possible increasing water loss, was a selective advantage with decreasing pCO 2 and is the likely explanation for observed increasing structural conductance of angiosperms since the Cretaceous, consistent with previous studies (Franks and Beerling, 2009a, b;Boyce and 565 Zwieniecki, 2012; Brodribb and Feild, 2010). From a low stomatal conductance (low vein density and V cmax ) similar to that of gymnosperms under high pCO 2 , angiosperms evolve towards a high stomatal conductance (high vein density and V cmax ) to counteract the effect of the pCO 2 decrease on carbon assimilation.
While this study provides clues on how to account for angiosperm evolutionary traits in paleoclimate simulations, further work is needed to assess the potential climate effects of the Cretaceous angiosperm leaf evolution, especially on the hydrological 570 cycle and the energy balance at the land surface. Furthermore, allowing dynamic vegetation would be an important future refinement of this research to model feedbacks between vegetation and climate. Also, replacing PFT morpho-physiological traits by species-specific traits (Kattge et al., 2020) as it has already been done for the aDGVM2 model (Scheiter et al., 2013), the LPJmL-FIT model (Sakschewski et al., 2015) or the JEDI model (Pavlick et al., 2013), allows plant communities to be assembled based on how plants with different trait combinations perform under a given set of environmental conditions. This 575 way, changes in functional traits would be more directly taken into account. With such a dynamical vegetation model, we may question the veracity of the reciprocity with the human-induced increase in pCO 2 . Building leaves with high vein density and V cmax induces an investment in nutrients and energy (Chapin et al., 2011;Fiorin et al., 2016;Beerling and Franks, 2010) that could be an extra cost for plants without benefit under a higher pCO 2 . That would suggest that plants should evolve back to a reduced stomatal conductance and V cmax . However, the pace of the current increase in pCO 2 is dramatically higher than the 580 million-year time scale of Cretaceous changes, and likely incompatible with a genetic plant adaptation. Whether extant plants will be able to adjust their physiological and morphological traits or not to the human perturbation is a challenging question for future studies.
Code availability. The code is available as part of the IPSL-CM5A2 earth system model, that has been made available by Sepulchre et al. (2020). The code can be retrieved through svn, with the following command lines: 585 svn co http://forge.ipsl.jussieu.fr/igcmg/svn/modipsl/branches/publications/IPSLCM5A2.1_11192019/ See also: https://gmd.copernicus.org/articles/13/3011/2020/#section11 The ORCHIDEE code, that has been modified for this study, is available as well through svn: svn co https://forge.ipsl.jussieu.fr/orchidee/browser/branches/publications/ORCHIDEE_IPSLCM5A2.1.r5307. The login/password combination requested at first use to download the ORCHIDEE component is anonymous/anonymous. Fraedrich, K., Kleidon, A., and Lunkeit, F.: A green planet versus a desert world: Estimating the effect of vegetation extremes on the