Vegetation Carbon.

To facilitate comparison across simulations using all GCMs and RCPs, we express global vegetation change with respect to change in global mean land surface temperature (ΔMLT). It is important to recognize that ΔMLT is a proxy for changing magnitudes of temperature, precipitation, humidity, and CO₂, and that both climate and ecosystem inertia also play roles in the relationships between climate forcing and vegetation responses.

Baseline (i.e., mean 1971–1999) global varies between 461 Pg C and 998 Pg C, and increases with ΔMLT for all vegetation models under all 110 climate and CO₂ increase scenarios (Fig. 1) (see Materials and Methods and SI Text for details of simulations). Global increases more-or-less linearly in all models up to about +4 °C, but with different slopes. At higher temperatures four of the models saturate, whereas the remaining three continue increasing. Intermodel mean stops increasing at +4 °C, where the full range of the increase is 52–477 Pg C, depending on the GVM×GCM×RCP combination. The mean increase of 224 Pg C over all models is equivalant to of recent global fossil fuel CO₂ emissions. Within-GVM variation is due to the different GCM climates and CO₂ mixing ratios.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Future global vegetation carbon change calculated by seven global vegetation models using climate outputs and associated increasing CO₂ from five GCMs run with four RCPs, expressed as the change from the 1971–1999 mean relative to change in global mean land temperature. The annual values for each model are shown for all simulations binned into 0.2-°C–wide bins (short, horizontal stripes; n for each bin varies from 6 to 857). The means for each model (thick colored lines) and the multimodel means and SDs (black bars and boxes) are also shown. Average CO₂ in the bins increases from 370 ppm at ΔMLT = 0 °C to 911 ppm at ΔMLT = 7.5 °C.

The spatial distribution of the mean response across all 110 realizations that fall into the 3.5 °C–4.5 °C change bin is shown in Fig. 2. Despite the large spread between simulations, consistent spatial patterns are evident. Most land supports increased vegetation carbon, with simulations agreeing on this increase in many locations. Increases are particularly high across much of northern North America, northwestern and southeastern South America, the colder regions of western Europe, most of northern Eurasia, southeastern Asia, and tropical Africa.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Mean change in vegetation carbon at +4 °C global land warming from a 1971–1999 baseline. Values are changes averaged across all simulations and decades (i.e., GVM×GCM×RCP×decade, n = 110), binned into the mean-decadal 3.5 °C–4.5 °C change bin (CO₂ = 510–758 ppm; see Table S2 for details). Stippling shows where at least 90% of all realizations agree on the sign of change.

There are also regions where vegetation carbon declines, including parts of southern North America, much of central South America, the southern Mediterranean region, southwestern Africa, and southwestern Australia. Declines range to −59% of current . The realizations are less in agreement concerning these declines than they are for regions of increase, except in the northern Maghreb. The spatial extent of the areas experiencing decreased vegetation carbon increases monotonically with warming above +3 °C, as does the intermodel agreement on these reductions. At ΔMLT = +7 °C, very large areas of South America, the Mediterranean region, and Australia experience mean decreases in relative to current values. Moreover, model agreement at the 90% level extends to southwestern Africa and Australia at these higher temperature increases. In contrast, model agreement on increases at +7 °C becomes less widespread than at lower warming, with the Tibetan Plateau, the Ethiopian Highlands, northeastern Siberia, and southwestern Canada still consistently experiencing higher . In addition, we find that an extensive region of the southern Sahara/northern Sahel experiences very large relative increases in vegetation carbon, although there is less agreement on this between simulations, primarily due to variation in climate predictions.

Variation between model simulations is greatest for change between current and +4 °C-binned simulations across the central United States, northeastern South America, southern Africa, the Near East, and much of Australia. These tend to be regions with increasing future moisture stress, which varies considerably between GCMs.

Model Differences.

Focusing on the results from one of the GCM×RCP scenarios allows vegetation model behavior to be analyzed in more detail. The mean global land surface temperature in the Met Office Hadley Centre Global Environment Model version 2 - Earth System (HadGEM2-ES) RCP 8.5 simulation rises to above current values, covering almost the entire spread of all GCM×RCP forcings. Furthermore, the global responses under this forcing (Fig. 3) are close to the mean responses of each vegetation model across all forcing scenarios. In other words, this GCM simulates average relationships between global mean warming and impact-relevant climate variables such as precipitation. Global increases remarkably linearly throughout the century in five vegetation models (i.e., ORCHIDEE, JeDi, JULES, SDGVM, and VISIT). In contrast, in LPJmL and HYBRID4 global saturates by about 2060, followed by a decline.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

(A–C) Change in annual global mean vegetation carbon (A), NPP (B), and residence time of carbon in vegetation (C) under the HadGEM2-ES RCP 8.5 climate and CO₂ scenario for seven global vegetation models. Symbols are GVM outputs and lines are fitted responses, using a simple model fitted to the global NPP responses to CO₂ and temperature and residence time responses to temperature for each model as described in the main text.

Changing results from changes in NPP and the residence time of carbon in living vegetation,where τ is carbon residence time. This formulation is applied here both locally and globally as a difference equation over annual time steps, enabling annual residence time to be inferred from simulated NPP and . In the vegetation models used here, NPP is responsive to climate and atmospheric CO₂, both directly and through indirect effects on vegetation development. Carbon residence time depends on the turnover rates of plant parts and the mortality rates of individuals, processes modeled using baseline rates, climate sensitivities (including fire), and competitively induced mortality, and are affected indirectly through shifts in vegetation composition, although not all these processes are treated in all models (SI Text).

All models except HYBRID4 display a linear increase in global NPP with time (Fig. 3). This is due to a saturating effect of CO₂ combined with a near-linear negative impact of climate change. In contrast, global mean annual residence time declines (JeDi, VISIT, JULES, and LPJmL), increases (HYBRID4 and ORCHIDEE), or does not change (SDGVM). Residence time in LPJmL remains fairly constant until about 2050 and then declines rapidly, whereas in VISIT it stabilizes in 2050 and in HYBRID4 it then increases more rapidly. These results suggest that the primary causes of the differences in the trajectories of future global between the GVMs are the different ways in which residence time responds to climate and CO₂, at least for the non-HYBRID4 models. Analyses of differences in model behavior should therefore focus not only on the processes of carbon acquisition (i.e., photosynthesis and NPP), but also on the dynamics of vegetation carbon turnover.

Using additional simulations with each GVM in which the CO₂ experienced by the vegetation was held constant, these results were further analyzed by fitting to each GVM globally, a simple two-parameter model for the relationship between NPP and CO₂ [i.e., , where is the change in CO₂], combined with linear models for the relationships between NPP and temperature (i.e., MLT) and residence time and temperature (i.e., MLT). The fitted global model is plotted as lines in Fig. 3 and is used to ascribe sources of uncertainty to different processes. The variance in final caused by differences in fitted residence time relationships between models was found to be 30% higher than that caused by differences in the fitted NPP responses when all models were considered. Leaving out HYBRID4, which displays a different NPP response, increases the effect of residence time to 151% more than that of NPP. In other words, for the non-HYBRID4 models, differences in residence time relationships with climate between the models are responsible for more than twice the variation in modeled global change to 2100 than are differences in NPP relationships with temperature and CO₂.

The HYBRID4 vegetation model includes a nitrogen cycle, and so this might be expected to be the reason for the decline in NPP, and hence , above 4 °C of warming because of the potential for nitrogen to limit CO₂ fertilization (e.g., ref. 15). However, further analysis showed that the decline in NPP is in fact due to increased vapor pressure deficits causing stomatal closure and increased evaporative demand over temperate and tropical forests, with increasing nitrogen mineralization reducing the potential constraint from N feedbacks (Fig. 4). There is considerable variability in the ways in which stomata respond to humidity between the models.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Change in mean-decadal vegetation carbon, annual NPP, and vegetation carbon residence time simulated by seven GVMs under HadGEM2-ES RCP 8.5 forcings between A.D. 2005 and 2099. (Right) Letters indicate relevant processes for residence time and NPP behavior included in each model: D, dynamic vegetation; F, fire; N, N cycle; P, permafrost; V, vapor pressure deficit affects stomatal conductance; S, temperature affects senescence; and M, temperature affects mortality.

In contrast, the decline in after 2050 in LPJmL is driven by increases in turnover across the temperate and boreal forests (Fig. 4). Tree mortality increases as a consequence of increasing tissue mortality due to high-temperature periods and in response to water stress in these regions, with subsequent increasing transient dominance by C₃ grasses during slow regrowth of better-adapted tree types.

Declining residence times in JeDi, LPJmL, JULES, and VISIT with warming, as shown in Fig. 4, would by themselves lead to reduced , but these changes are more than compensated for by increases in NPP at the global scale (Fig. 3). Declining residence times in these models are due to temperature dependencies of turnover rates (processes included in the models and important for production and residence time differences are indicated symbolically in Fig. 4). In contrast, global mean residence time in HYBRID4 and ORCHIDEE increases. In HYBRID4 this is due to reduced tree mortality across the boreal forest as higher CO₂ concentrations and temperatures improve their carbon balance. Residence time of temperate and Amazon forest carbon in HYBRID4 is reduced with warming due to heat-induced increases in vapor pressure deficits leading to unfavorable carbon balances, which increase rates of individual mortality (Fig. 4). In ORCHIDEE residence time increases across much of the boreal forest and many tropical regions. These changes are due to a greater biomass of younger leaves, which increases with NPP and has lower intrinsic rates of turnover than those of older leaves.

In JeDi, residence time falls over much of the northern boreal and arctic region, as well as the tropics, whereas it increases across temperate latitudes (Fig. 4). In contrast, in JULES residence time decreases in midlatitude deciduous forests, but changes little elsewhere. These changes are driven by changing vegetation type mixtures, with grasses and smaller shrubs in JeDi decreasing boreal residence times, and greater proportions of trees increasing residence times in the southeastern United States and China. Large reductions in residence time in northeastern Amazonia in SDGVM, and more generally across the tropics in VISIT, occur in response to increased fire frequency.