The Cretaceous physiological adaptation of angiosperms to a declining pCO 2 : a trait-oriented modelling modeling approach 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 peculiar (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) speciﬁc (cid:58) environmental conditions of the Cretaceous remains challenging. Here, we address this issue by combining a paleo proxy-based model with a fully (cid:58)(cid:58)(cid:58) full atmosphere-vegetation model that couples stomatal conductance to carbon assimilation. Based on the fossil record, we build and evaluate three consistent pre-angiosperm proto-angiosperm vegetation parameteri- zations under two end-members scenarios of pCO 2 (280 ppm and 1120 ppm) for the mid-Cretaceous : a reduction of (cid:58)(cid:58)(cid:58) 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 generates generate (cid:58) a reduction of transpiration that is predominantly the result of plant productivity variations , modulated by light, water availability in the soiland (cid:58) , atmospheric evaporative demand (cid:58)(cid:58)(cid:58) and (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) pCO 2 . The high pCO 2 acts as a fertilizer on 10 plant productivity that bolsters (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) strengthens plant transpiration and water-use efﬁciency. However, we show that pre-angiosperm proto-angiosperm (cid:58) 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 modelling (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) modeling approach stresses the need to better represent paleovegetation physiological traits. It also conﬁrms the hypothesis of a likely evolution of angiosperms from a stage of low of hydraulic and photosynthetic capacities at high pCO 2 to a stage of high state of leaf hydraulic and photosynthetic capacities linked to leaves with (cid:58) more and more densely irrigated veins together with a more efﬁcient biochemistry at low pCO 2 . the stomatal conductance Our results show that (cid:58) fcpl angiosperms at pCO does not change plant photosynthesis S2c 5c S6c) but slightly decreases transpiration 7c), 8c) compared to the modern angiosperm prescription. Hence, a lower maximal stomatal conductance at high pCO 2 appears as an advantage compared to modern angiosperm angiosperms because of a better optimization of carbon uptake over water loss. At low pCO 2 , both transpiration 7d) and photosynthesis are decreased because of the positive 420 feedbacks of the LAI on the entails canopy the coupling between stomatal conductance and plant productivity leaf to the canopy scale. Furthermore, we show that decreasing 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-time 3-fold lower than the reference. We have a complex response because of the coupling between stomatal conductance and assimilation the fossil record. Combining a reduction of leaf (cid:58) 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 taken together (cid:58) in (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) combination (cid:58) demonstrate that under high pCO 2 the reduced stomatal conductance of the pre-angiosperm (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) 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, the combining decrease of (cid:58) a (cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58)(cid:58) combination

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 from 500 ppm to 2000 ppm. Values peaked during the mid-Cretaceous (Cenomanian-Turonian, ca. 95 Ma), then 80 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.
2.2 Fossil evidence of increasing angiosperm ::: leaf hydraulic and photosynthetic capacities 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) : 12760 ν 12760 V P D ::::: Where ν :::: VPD : is the leaf-to-air vapor pressure deficit (MPa), ∆Ψ leaf is the leaf water potential gradient (MPa), D v is the vein density (mm mm −2 [ ::: leaf]) and y is the distance from vein terminals to epidermis (µm). ν :::: 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 ).
A lower value for y indicates a lower stomata-to-vein distance and a higher g max anat is. The highest increase in g max anat is obtained 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 S1 :: 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   195 S1 :: 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 , 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 .
LAI changes described earlier act as a feedback :::::::: 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 330 V cmax under low pCO 2 implies a reduction of the plant capacity to assimilate carbon (Eq. (3 : 4)) and directly impacts the GPP (Fig. in S2f ::: S6f and h) 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. S2d ::: 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. S2a ::: S6a, c and g). The latter lessens the 335 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 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 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 angiosperms :::::::::: angiosperm ::: leaf : hydraulic and photosynthetic capacities systematically 350 leads to lower transpiration rates (Fig. 7).

20
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 355 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 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 360 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.

400
NOANGIOh(280) depicts a lower WUE than 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. (3 : 4), Fig. S2b ::: 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. S2f :: S6f : and h). Once the ::: leaf : photosynthetic 405 capacity is decreased, changes in GPP are the main contributor to the changes in WUE whatever the pCO 2 level prescribed.
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 410 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 pre-angiosperm ::::::::::::::: proto-angiosperm vegetation is to account for the decrease of :::: lower : maximal anatomic stomatal conductance by a factor of 5, consistent with the fossil records, directly by applying this factor to fcpl in 415 the calculation of the leaf stomatal conductance (Eq. (1)). Our results show that decreasing : a :::: lower : fcpl of angiosperms at high pCO 2 does not change plant photosynthesis ( Fig. 5c and S2c :: 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 angiosperm :::::::::: 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 of the positive 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 pre-angiosperm ::::::::::::::: 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 425 gain against water loss (Bonan, 2015) and is made possible by the coupling between ::: leaf stomatal conductance 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 S2e : 5e :::: and :::: S6e) or even collapse at low pCO 2 ( Fig. 5f and S2f : 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 430 modern vegetation. Hence, taking into account stomatal conductance reduction only through V cmax reduction does not appear to be 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 435 improved considering also the cost related to stomatal conductance (Stocker et al., 2020). This V cmax limitation is related to the high 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 strat-440 egy between modern angiosperm trees and gymnosperm trees. They argue that modern angiosperm trees have 2 times higher stomatal 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. S2g ::: S6g), that remains enough to sustain the LAI ::::::: sufficient :: to ::::::: sustain :::: LAI ::::: values ::::: close :: to ::::: those :: of ::: the :::::: control :::::::: scenario (Fig. 5g), and a relatively high WUE (Fig. 8g) 445 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. S2h, 7h and 8h ::: S6h, ::: 7h ::: and ::: 8h) as a response to carbon assimilation drop.
4.2 Do the high ::: leaf : hydraulic and photosynthetic capacities of angiosperms provide a selective advantage compared 480 to the other plants under decreasing pCO 2 ?
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), plants had to adjust their stomatal conductance. Our study suggests that having a high stomatal conductance, which means a large vein 485 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 of basal :::: early angiosperms show that vein density was as low as the ::: 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). At that time, we confirm ::: For :::: this :::: time, ::: our :::::: results :::::: support : the hypothesis that angiosperms evolved towards leaves more and more densely irrigated together :::: with ::::::::: increasing :::: vein :::::: density ::::::::: combined with a more efficient biochemistry that allowed them to have an increasingly :::::::: increasing : stomatal conductance and photosynthetic capacity to counteract the effect of pCO 2 decrease on carbon assimilation. Among others :::: other :::::: factors, this evolution of physiological leaves ::: leaf : traits 495 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 that :::: 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 strengthen GPP , meanwhile a potential decrease of transpiration rate by the closing of the stomata :::::::: enhanced :::: GPP ::::: while :::::::::::: simultaneously :::::::: allowing : a :::::::: reduction :: of :::::::::::: transpirational ::::: water :::::: losses ::: due :: to ::::::: reduced ::::::: stomatal ::::::::::: conductance. They also show that 500 even after the evolution of angiosperm leaf morphology and 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 505 invested increasingly :::: more : energy in building more and more veins to sustain the higher stomatal conductance and then carbon assimilation, while other plants did not. The innovation of angiosperms in densely :::: dense : water transport networks could have become a necessity to support higher stomatal conductance and prevent plant desiccation (De Boer et al., 2012).
As a first step toward understanding the impact of trait evolution on the ::: leaf : hydraulic and photosynthetic capacities of an-510 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 515 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 520 (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 fully :: full : dynamical vegetation model, i.e. by allowing PFTs to spatially settle in regions where the simulated climate is the most appropriate.
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 535 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).
Finally our ::: Our : parameterization of stomatal conductance in ORCHIDEE from Yin and Struik (2009) (Eq. (1)) is semi-540 empirical. A refinement of our modelling ::::::: 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.

28
of our study is to better represent past vegetation in earth system models by implementing :::::::: emulating : "paleo-traits" in the vegetation parameterizations. Our approach involves an atmosphere-vegetation model, which couples stomatal conductance and carbon assimilation, with :::::::: motivated :: by : an ecophysiological model based on angiosperm fossil records. Here, it allows us to evaluate three different paleovegetation 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 565 trait prescribed. Only accounting for ::: leaf : hydraulic capacity reduction provides no significant change in LAI, GPP and transpiration, while slightly increasing WUE at 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 not recorded in the fossil :: in ::::::::::: contradiction :: to ::: the :::::: fossil ::::: record. Combining a reduction of ::: leaf : hydraulic capacity with that 570 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 taken together : in ::::::::::: combination : demonstrate that under high pCO 2 the reduced stomatal conductance of the pre-angiosperm :::::::::::::: 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, the combining decrease of : a ::::::::::: combination 575 :: of :::::::::::::::: lower-than-modern ::: leaf : hydraulic and photosynthetic capacities seems the most realistic physiological parameterization for pre-angiosperms ::::::::::::::: proto-angiosperms : in the specific high pCO 2 context of the Cretaceous. This is supported by evidence of coevolution inferred from previous studies (Franks and Beerling, 2009a, b) and the ratio of stomatal conductance between modern angiosperms and gymnosperms from in-situ experiments (Lin et al., 2015). This result is ::: Our :::::: results ::: are also consistent with recent studies on coordination theory (Maire et al., 2012;Stocker et al., 2020).

610
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. Competing interests. The authors declare that they have no conflict of interest.
supported by the Commisariat à l'énergie atomique et aux énergies alternatives (CEA). PS is supported by Centre national de la recherche scientifique (CNRS).