Climate and CO 2 modulate the C 3 /C 4 balance and δ 13 C signal in simulated vegetation

Abstract. Climate and atmospheric CO2 effects on the balance between C3 and C4 plants have received conflicting interpretations based on the analysis of carbon isotopic fractionation (δ13C) in sediments. But, climate and CO2 effects on the C3/C4 balance and δ13C signal are rarely addressed together. Here, we use a process-based model (BIOME4) to disentangle these effects. We simulated the vegetation response to climate and CO2 atmospheric concentration (pCO2) in two sites in which vegetation changed oppositely, with respect to C3 and C4 plants abundance, during the Last Glacial Maximum to Holocene transition. The C3/C4 balance and δ13C signal were primarily sensitive to temperature and CO2 atmospheric partial pressure. The simulated variations were in agreement with patterns observed in palaeorecords. Water limitation favoured C4 plants in case of large negative deviation in rainfall. Although a global parameter, pCO2 affected the δ13C signal differently from one site to the other because of its effects on the C3/C4 balance and on carbon isotopic fractionation in C3 and C4 plants. Simulated Plant functional types (PFT) also differed in their composition and response from one site to the other. The C3/C4 balance involved different competing C3 and C4 PFT, and not homogeneous C3 and C4 poles as often assumed. Process-based vegetation modelling emphasizes the need to account for multiple factors when a palaeo-δ13C signal is used to reconstruct the C3/C4 balance.


Introduction
Studies of past vegetation combined with climate reconstruction help to identify important factors controlling vegetation dynamics. Due to their photosynthetic pathway, C 3 and C 4 plants show different carbon isotopic fractionation (δ 13 C), around −26 and −13 respectively (Deines, 1980). The variability of δ 13 C in sediment records can thus inform on the relative abundance of C 3 and C 4 plants (Street-Perrott et al., 1997;Aucour et al., 1999;Wang et al., 2008). Based on this proxy, contrasted trends have been evidenced during the transition from the Last Glacial Maximum (LGM) to the Holocene. Interpretation of these trends led to conflicting conclusions regarding the relative influence of atmospheric CO 2 and climate on the distribution of C 3 and C 4 plants (Street-Perrott et al., 1997;Huang et al., 2001). In this paper, we use a process-based vegetation model to disentangle the effects of climate and CO 2 on the C 3 /C 4 balance and δ 13 C signal.
In intertropical African highlands, sediment sequences in peatbogs revealed decreasing δ 13 C, from high values during the LGM (> − 18 ) to low values during the late Holocene (−28.5<δ 13 C<−19.5 ; Aucour and Hillaire-Marcel, 1994;Aucour et al., 1999), indicating a decreasing contribution of C 4 plants to vegetation production after the LGM. Consistently, pollen records highlighted a shift in vegetation at the end of the LGM, from grasslands comprising C 4 species to montane forests with a majority of C 3 trees . In contrast, in continental central China, records revealed lower values of δ 13 C during the LGM than during the late Holocene (Gu et al., 2003;Wang et al., 2008). This trend was interpreted as a relative increase in the abundance of C 4 species between these two periods, also consistent with pollen records.
Published by Copernicus Publications on behalf of the European Geosciences Union.

432
O. Flores et al.: Simulated C 3 /C 4 balance and δ 13 C signal The shift from C 4 to C 3 -dominated vegetation in intertropical Africa was primarily interpreted as triggered by the increase in atmospheric CO 2 concentration (p CO 2 ) after the LGM (Street-Perrott et al., 1997). Simulations also showed that a change in p CO 2 alone could drive such large vegetation changes (Cole and Monger, 1994;Jolly and Haxeltine, 1997). However, climatic conditions also changed drastically after the LGM, but with different regional patterns. In South American lakes, sequences revealed opposed signals in δ 13 C (Huang et al., 2001), from which the authors concluded that local climate controlled the abundance of C 4 plants, since all sites shared the same p CO 2 . In fact, opposite signals in the C 3 /C 4 balance and δ 13 C can also occur under similar changes in mean climatic conditions. In intertropical Africa, cooler and dryer climatic conditions occurred during the LGM compared to late Holocene (Bonnefille et al., 1990;Bonnefille and Chalié, 2000). Similarly, in central China, vegetation faced cooler conditions (Zhang et al., 2003) and decreased monsoon intensity during the LGM (Liu et al., 2005).
It is thus likely that both climate and p CO 2 interact to control the C 3 /C 4 balance and δ 13 C signal, in similar or opposite ways. Increased aridity can mimic effects of low p CO 2 for instance: stomatal closure due to water stress and/or low p CO 2 induce low concentration of CO 2 in photosynthetic tissues, which favours photorespiration (the fixation of O 2 instead of CO 2 by the RuBisCo) and can alter photosynthesis. Because they decouple CO 2 fixation from light reactions, C 4 plants are less sensible to photorespiration than C 3 plants (Sage, 2004). They can thus tolerate low stomatal conductance during warm hours as well as under low p CO 2 . Another factor, the vegetation composition in plant functional types (PFT), is also likely to influence the C 3 /C 4 balance. The proportion of C 4 production represents the outcome of the competition between C 3 and C 4 plants; but whereas C 4 plants are mostly grasses and sedges, C 3 plants can be grasses, shrubs or trees with highly variable productivity.
In studies of palaeovegetation, the C 3 /C 4 balance is generally estimated using a linear relationship between the vegetation δ 13 C and the proportion of C 4 production (Gu et al., 2003;Wang et al., 2008), hereafter r C 4 : δ 13 C=r C 4 ×δ 13 C C 4 +(1−r C 4 )×δ 13 C C 3 where δ 13 C C 4 , and δ 13 C C 3 are often set to constant average values of δ 13 C in C 3 and C 4 plants. However, climate and p CO 2 can have direct effects on r C 4 as discussed above, as well as on δ 13 C C 3 and δ 13 C C 4 (Lloyd and Farquhar, 1994;Wang et al., 2008). Reconstructing the C 3 /C 4 balance and interpreting the δ 13 C signal require to take account of these effects. But decoupling climate and p CO 2 effects is difficult in empirical studies. Process-based models of vegetation dynamics can then help address this issue because they explicitly model physiological processes as well the vegetation composition in plant functional types.
In this study, we use a process-based equilibrium model (BIOME4, Kaplan et al., 2002) to simulate the effects of climate and p CO 2 on the C 3 /C 4 balance and δ 13 C signal in the vegetation. The C 3 /C 4 balance is characterized by the proportion of estimated net primary production (NPP) due to C 4 PFT (r C 4 ). δ 13 C is simulated as the NPP-weighted mean of δ 13 C of C 3 and C 4 PFT. The responses of these two variables to deviation in climatic conditions and to p CO 2 are investigated in two sites with contrasted vegetation history: one in the intertropical African highlands in Burundi, and one on the central Loess plateau in China. We specifically address the following questions: (I) is climate or p CO 2 the main forcing on the C 3 /C 4 balance and δ 13 C signal? (II) which of climate or p CO 2 had a prevailing effect on vegetation change during the transition from the LGM?

Site selection
We focus on two sites differing in climatic conditions, current and past, and vegetation history in addition of being located in areas where reconstructions of climate and vegetation have been conducted. The two sites selected for comparison are Kuruyange in intertropical Africa (Burundi), and Lingtaï on the Chinese loess plateau.
Kuruyange lies in the interlacustrine highlands of Burundi and in the current altitudinal range of montane forests. These forests are present as fragments of broad-leaved forest inbetween cultivated and grazed land . In Kuruyange, as in intertropical Africa overall, the vegetation of the LGM was essentially composed of cold grasses and scrub (Hamilton, 1982;Bonnefille and Riollet, 1988;Vincens, 1991;Taylor, 1992;Jolly and Haxeltine, 1997). After ca. 10-11 kyr BP, this vegetation was progressively replaced by tropical montane forests (Coetzee, 1967;Hamilton, 1972;Taylor, 1990Taylor, , 1993Bonnefille et al., 1991Bonnefille et al., , 1995Vincens, 1991;Jolly et al., 1994). Pollen records from the area show that Poaceae and Cyperaceae, including C 4 species, were abundant during the LGM. After 10 kyr BP, grasses tend to disappear from records, whereas C 3 trees become more abundant.
Lingtaï lies on the Loess plateau in central China. The vegetation in the area is a mixed forest of temperate coniferous and broad-leaved trees together with numerous grass species. Studies of δ 13 C in sediments showed that the vegetation at the LGM which was mostly steppe and desert vegetation, without C 4 plants (Yu et al., 2000). During the Holocene, C 4 plants became more abundant (Gu et al., 2003;Wang et al., 2008

Current and past climate
Climatic data were extracted from 10×10 climatic grids (New et al., 2002) considering the nearest point from sites location. Monthly values of rainfall in mm/month, mean temperature in • C, and mean sunshine in % of day length were taken at (−3.58 • , 29.75 • , 1850 m a.s.l.) for Kuruyange, and (35.03 • , 107.58 • , 1180 m a.s.l.) for Lingtaï. Annual rainfall and mean annual temperature were respectively 1370 mm, 18.0 • C in Kuruyange, and 600 mm, 10.2 • C in Lingtaï. Kuruyange has a typical equatorial climate with consistent monthly temperatures throughout the year, high rainfall from September to May followed by three warm and dry months during summer (Fig. 1a). Lingtaï has a continental climate with cold and dry winters and low rainfall concentrated in summer (Fig. 1b). The climate on the Loess plateau is largely controlled by the East Asian Monsoon (EAM) system (Balsam et al., 2004). Deviation of climatic parameters were estimated at about −4±2 • C (Bonnefille et al., 1990) and −450±250 mm (Bonnefille and Chalié, 2000) during the LGM in Burundi. In China, the mean annual temperature deviation ranges between −7 and −10 • C compared to current conditions in northern and central areas, and −4 and −6 • C in southern areas (Zheng et al., 1998). Rainfall was between 400 and 600 mm lower in regions under EAM influence and 200 to 300 mm lower in northern areas (Zheng et al., 1998).

Simulated climate and p CO 2
We used BIOME4 to analyse the C 3 /C 4 balance response to various conditions of p CO 2 and climate. Climatic conditions were derived from current distributions of climatic parameters in order to deal with realistic scenarii. Soil parameters were obtained from the FAO database and kept constant in all simulations. Cloudiness (New et al., 2002) was kept constant as well. We conducted two types of simulations with varying climatic parameters and constant p CO 2 or varying p CO 2 and constant temperature and rainfall distributions. First, current distributions of temperature and rainfall were modified to adjust the mean annual temperature T and annual rainfall (P a ) to chosen values, while keeping the overall shape of the distributions. Monthly means were adjusted as T i =T i +δT , where T i is the current mean temperature of month i, and δT is the simulated deviation in mean annual temperature. Likewise, for a given deviation in rainfall (δP a ), monthly rainfall were adjusted as P i =(1+ δP a P a )P i , where P i is the rainfall of month i and P a is the annual rainfall. p CO 2 was kept constant using two different values: the LGM value (180 ppmv, Monnin et al., 2001) and the current value (360 ppmv).

Model description
BIOME4 (Kaplan et al., 2002) is a process-based equilibrium model for terrestrial vegetation of the BIOME family (Prentice et al., 1992;. The inputs fed in the model are monthly temperatures, rainfalls and cloudiness, absolute minimal temperatures, soil texture, latitude, atmospheric pressure through altitude and p CO 2 . Incoming solar radiation is calculated using current orbital parameters and average albedo. The model uses a two-layer description of soil, with different texture and depth. Run-off www.clim-past.net/5/431/2009/ Clim. Past, 5, 431-440, 2009 is evaluated, but lateral fluxes are not redistributed. Routines are included to approximate additional risks due to canopy fire and snow. The models simulates 13 plant functional types (PFT) primarily constrained by absolute bioclimatic tolerance limits, such as the minimal supported number of growing days over 5 • C (Table 1). These constraints determine the set of NPP that are potentially present given climatic parameters. A growth procedure then optimizes the NPP and the optimal leaf area index of each PFT in this set. Growth is estimated using a photosynthesis model coupling water and carbon fluxes in the plant . Water fluxes in the model depend on soil water availability and evapotranspiration of soil and plants. Evapo-transpiration at the plant level is controlled by canopy conductance which is integrated from stomatal conductance. Stomatal conductance controls the ratio of intercellular to atmospheric CO 2 concentration ( c i c a ) and thus the availability of CO 2 for photosynthesis. Ten PFT out of 13 are exclusively C 3 , one exclusively C 4 , and two may use either pathway (temperate grass and woody desert type, Table 1).
We used a revised version of BIOME4 with enhanced carbon isotopic discrimination compared to previous versions (Hatté and Guiot, 2005). The discrimination model originally followed Lloyd and Farquhar (1994) for C 3 and C 4 plants: it estimates A , the total discrimination against 13 CO 2 during carbon assimilation from the atmosphere to photosynthetates. The C 3 and the C 4 photosynthetic pathways are considered separately (Lloyd and Farquhar, 1994;Kaplan et al., 2002). Revisions by Hatté and Guiot (2005) included a lower limit on c i c a in C 3 PFT to avoid situations of stomatal closure. The revised version also takes into account all potentially present PFT, not only the dominant as previously, and weight discrimination by the estimated NPP which accounts for the net amount of stored carbon. Finally, discrimination during photosynthesis was refined by taking temperature influence into account (see Hatté and Guiot, 2005, for details).
We studied two response variables: (I) the fraction of NPP produced by C 4 PFT: r C 4 = C 4 NPP NPP where the sum is on the three obligate or facultative C 4 PFT in numerator and on all PFT in denominator (Table 1), and (II) the carbon isotopic fractionation in vegetation (δ 13 C). δ 13 C was obtained from A as δ 13 C= δ atm − A 1+ A where δ atm is the fixed atmospheric value (−8 ). In BIOME4, A is the NPP-weighted discrimination by all PFT.

Carbon isotopic fractionation in C 3 and C 4 PFT
Carbon isotopic fractionation varied over separated ranges, between −32.9 and −24.5 in C 3 PFT, and between −18.9 and −8.9 in C 4 PFT, when all simulations were considered. These values are consistent with reported values in the literature for these two poles (O'Leary, 1988).

Variation in r C 4 and δ 13 C from LGM to current conditions
The simulated proportion of C 4 NPP (r C 4 ) was 6% at Kuruyange and 0% at Lingtaï in average LGM conditions (Table 2). From LGM to current conditions, r C 4 increased at the two sites (Kuruyange: +14%, Lingtaï: +23%, (Table 2), although r C 4 decreased with increasing p CO 2 under constant climate ( Fig. 2a and 2b). Negative variation in r C 4 were obtained at Kuruyange at the upper bound of climatic parameters (−22%, Table 2) whereas the variation was positive in Lingtaï over the whole range of parameters.
Regarding δ 13 C, a variation of −3.2 was simulated at Kuruyange and +2.6 at Lingtaï (Table 2). In both sites, the sign of variation was consistent over the range of climatic parameters for LGM (Table 2). At Kuruyange, δ 13 C responded to p CO 2 similarly in LGM and current climatic conditions (Fig. 2a). In contrast, the response at Lingtaï differed with respect to climatic conditions: δ 13 C increased from LGM to current climate (Fig. 2b).
To quantify the relative effects of climate and p CO 2 , we decompose the variation of δ 13 C in one signal due to deviating climatic parameters only, and one signal due to increasing p CO 2 under constant climate. At Kuruyange, the climate signal was +1.4 [+2.8,−1.9] with p CO 2 kept at 180 ppmv, and +0.3 [+1.6,−2.6] at 360 ppmv. The p CO 2 signal was −3.5 [−3.4, −4.0] in LGM conditions, and −4.6 in current climatic conditions. In Lingtaï, climate and p CO 2 had more contrasted effects (Fig. 2b)

Response to climate at fixed p CO 2
Changes in rainfall had weak effects on r C 4 at both sites (Fig. 3). At Kuruyange, r C 4 hardly responded to deviation in rainfall, except in case of large negative deviation (δP a <−500 mm, Fig. 3a and c). At Lingtaï, negative rainfall deviation below −300 mm favoured C 4 plants at 360 ppmv (Fig. 3b). Increasing temperature implied higher C 4 NPP in both sites. Low p CO 2 favoured C 4 plants at both sites: r C 4 reached 45% at Kuruyange at current p CO 2 (Fig. 3a), and 61% at LGM level (Fig. 3c). At Lingtaï, these figures were respectively 43% and 85% ( Fig. 3b and d). No C 4 NPP occurred below −5 • C of temperature deviation at Kuruyange (Fig. 2c), whatever the deviation in rainfall (Fig. 3a and c). At Lingtaï, the threshold on r C 4 was lower, about −7 • C (Fig. 2d).
The response of δ 13 C showed similar patterns compared to r C 4 (Fig. 4)  . Phe.: phenology; E, evergreen; S, summergreen; R, raingreen; g C : maximal value for minimum canopy conductance; E max : maximum value of supported daily transpiration rate; SM out : soil moisture below which raingreen leaves drop; SM in : soil moisture above which raingreen leaves appear; R top : fraction of roots in top soil layer, 30 cm from Jackson et al. (1996); LL: expected leaf longevity in months; GDD5: annual growing degree-days (GDD) above base temperatures of 5 and 0 • C required for full leaf out; P.P.: photosynthetic pathway; C 3 , C 4 , or both.

Label Type
Phe. g C E max SM out SM in R top LL GDD 5 GDD 0 P.P.  Solid and dotted lines correspond to current pCO 2 and rainfall, and mean LGM climatic conditions respectively.

Fig. 2.
Top: Simulated response profile of the proportion of C 4 NPP (r C 4 , black lines) and carbon isotope fractioning (δ 13 C, gray lines) to p CO 2 at Kuruyange (Burundi, a) and Lingtaï (China, b). Solid and dotted lines correspond to current climatic conditions (δT =0, δP a =0), and mean LGM climatic conditions respectively. Bottom: Simulated response profile of r C 4 (black lines) and δ 13 C(gray lines) to deviation in mean annual temperature (δT ) at Kuruyange (c) and Lingtaï (d). Solid and dotted lines correspond to current p CO 2 and rainfall, and mean LGM climatic conditions respectively.     Fig. 3. Proportion of total NPP produced by C 4 PFT (r C 4 , in %) simulated at Kuruyange (Burundi, a and c) and Lingtaï (China, b and d) as a function of deviation in mean annual temperature (δT ) and deviation in annual rainfall (δP a ) with respect to current values. Two cases of p CO 2 are presented: current level (360 ppm, a and b), and Last Glacial Maximum level (LGM, 180 ppm, c and d). Black lines indicates isolines in r C 4 value with a 5 %-step. Asterisks indicate current (δT =0, δP a =0) and LGM climatic conditions. White lines indicate climatic parameters range for the LGM, taken from the literature.

Vegetation composition in PFT
The composition of the vegetation showed important changes in response to temperature. Increased p CO 2 from LGM to current level had an overall fertilization effect, except on the tropical grass PFT (obligate C 4 ; Fig. 5). Tree PFT had the highest NPP (tet, tft, tee, cot; Fig. 5a) at Kuruyange in current conditions, followed by the "temperate grass" type (teg;  LGM climatic conditions. White lines indicate climatic parameters range for the LGM, taken from the literature. Fig. 5a) for δT < −2 • C, and the "tropical grass" (trg) above for δT > −2 • C. In LGM conditions, the "tropical grass" type (trg) was highly productive compared to tree PFT. In Lingtaï, the "temperate deciduous tree" type (ted; Fig. 5b) and the "temperate grass type" (teg) were highly productive. The "temperate deciduous tree" was the only present tree PFT when temperature deviated between −1 and +1 • C. The "boreal tree" types (bet, bdt) occurred below −1 • C, and were replaced by the "cool conifer tree" type above +1 • C.

Discussion
Vegetation modelling allowed to decouple the effects of climate and p CO 2 on the balance between C 3 and C 4 plants, and on the resulting δ 13 C signal. The model could reproduce actual trends in the abundance of C 4 plants which are usually favoured at high temperature and in condition of low moisture and atmospheric CO 2 concentration (Sage, 2004). Overall, our results evidenced a prevailing effect of climate over CO 2 . Temperature was the most influencing factor on the C 3 /C 4 balance, while rainfall influence was weaker. However, strong negative deviation in rainfall amount, of the magnitude observed during the LGM, could favour C 4 plants. These results are consistent with studies of palaeo-sequences in central China (Zhang et al., 2003;Liu et al., 2005) and in central America (Huang et al., 2001) which concluded that local climate change was the main driver of the C 3 /C 4 dynamics during the LGM-Holocene transition. The low temperature of the LGM limited C 4 production of the simulated vegetation in China accordingly to empirical evidence (Zhang et al., 2003). In Africa, less limiting conditions during the LGM allowed C 4 plants to sustain limited production. However, our results confirmed that low p CO 2 also shifted the balance in favour of C 4 plants in the two studied sites. These findings are consistent with other studies which addresses the effects of both climate and CO 2 simultaneously (Ehleringer et al., 1997;Boom et al., 2002).  LGM climatic conditions respectively. PFT are labelled according to The model reproduced the overall variation in the C 3 /C 4 balance and δ 13 C following the transition from LGM to current conditions. The simulated δ 13 C signal matched the decrease in δ 13 C observed at Kuruyange (Aucour et al., 1999) and the increase observed at Lingtaï (Gu et al., 2003) following the transition from LGM to current conditions. Local conditions could thus lead to opposed effects on the δ 13 C signal, despite similar deviations in climatic parameters from the LGM to current conditions. Regarding the C 3 /C 4 balance however, the simulated variation at Kuruyange was between +20% and −22%, showing high sensitivity to incertitude in LGM climatic parameters. These results suggest that the deviation in mean temperature during the LGM was probably closer to the upper bound (δT =−2 • C, Bonnefille et al., 1990) than to the lower, as no C 4 occurred at low temperature (δT < −5 • C) in Kuruyange.
Palaeosequences obtained in swamps in Burundi showed that grasses (Poaceae) and sedges (Cyperaceae) including C 4 species dominated pollen records in the area during the LGM (Aucour et al., 1999). But the contribution of C 4 plants during the LGM was probably higher in the swamp biomass than in the surrounding vegetation (Aucour et al., 1999). Our simulations suggest however that C 4 plants represented no more than 36% of the vegetation. In Lingtaï, the simulations are consistent with Gu et al. (2003) who concluded that no C 4 production occurred during the LGM, although uncorrected values of carbon isotopic fractionation were used in their reconstruction. Lower rainfall and p CO 2 could not lead to C 4 plants expansion in the absence of favourable temperature condition.
Climate and CO 2 both controlled the δ 13 C variation from LGM to current conditions. The transition from low p CO 2 during the LGM to higher current level implied a decrease in δ 13 C in both sites. But, this effect differed from one site to the other. These results show that, although p CO 2 is a global parameter, its influence differed between sites, because of its effects on the C 3 /C 4 balance and carbon isotopic fractionation in C 3 and C 4 plants. In Kuruyange, the climate signal was lower than the p CO 2 signal, and the two strongly interacted. In contrast, in Lingtaï, the signal was stronger with climate than with p CO 2 , which led to the overall simulated increase in δ 13 C. The two effects however interacted less than in Kuruyange. Moreover, climate and CO 2 effects on the C 3 /C 4 balance were modulated by specific bioclimatic constraints limiting the potential composition of the vegetation in PFT at each site. Depending on the range of climatic conditions considered, the competition between C 3 and C 4 plants may involves PFT with different productivities. Consequently, the nature of PFT present at the site had strong effect on the reconstructed C 3 /C 4 balance: a shift from low producing to high C 3 PFT decreases the proportion of C 4 NPP all other things being equal.

Conclusions
Process-based models allow to disentangle climate and CO 2 impact on vegetation using a mechanistic approach. In the examples studied here, climate had a prevailing effect, with temperature imposing strong constraints of the C 3 /C 4 balance, rainfall being influential in case of strong limitation. Atmospheric CO 2 acted as an overall fertilizer of photosynthesis. However, its effects on the C 3 /C 4 balance and δ 13 C signal depended on local conditions because of interaction with climatic and vegetation composition effects.