*3.3. Sensitivity Experiments for Climate and Vegetation*

Increased CO2 affects LAI and GPP through climate change, through both the greenhouse effect and CO2 fertilization effect, but the CMIP5 ensemble for the RCP 8.5 scenario cannot separate the importance of the two processes [19]. To isolate their varying importance, we generate an eight-model ensemble (highlighted in Table 1), comparing results from the sensitivity experiments.

At first, we summarized the climate responses to increasing CO2 concentration in these sensitivity experiments. The esmFdbk1 experiment shows that radiative effects clearly dominate projected warming, the signature feature of climate change (compare black and red lines in Figures 7a–c and 8a,d,g). Radiative effects also drive other key aspects of climate change, such as Arctic amplification and land–sea warming contrast (Figure 8a), equatorial and high-latitude increases in precipitation (Figure 8b), and high-latitude dimming from increases in fall/winter cloud cover (Figure 8c) [34]. Meanwhile, the esmFixClim experiment revealed that, except for increases in radiation for the temperature-limited region in JJA (Figure 7g), the vegetation physiological effect has little effect on region-averaged climate (e.g., near-zero changes in Figure 7a–f). Spatially, the physiological effects

on climate are more consistent than the radiative effects, and include a slight drying (Figure 8e) and brightening because of reduced cloud cover (Figure 8f) over most land surfaces.

**Figure 7.** Additive and counteracting impacts of CO2 on climate. Simulated changes in climate and vegetation from the CMIP5 eight-model ensemble in response to increasing CO2 from 280 ppm by 1%/year for 140 years: temperature, **a**–**c**; precipitation, **d**–**f**; and radiation, **g**–**i**. The changes are the mean of the last ten years minus the mean of the first ten years (see Figures 8 and 10 for a spatial representation of the changes). In order to assess the contribution of radiative and vegetation physiological effects on climate and vegetation, three experiments were carried out: (1) CO2 has a radiative forcing on climate but no direct effect on vegetation; (2) CO2 has a vegetation physiological impact, primarily on internal CO2 concentration and stomatal conductance, but does not directly alter radiative forcing; and (3) CO2 has a combined effect on both radiative forcing and physiological impacts. Shading is used to highlight changes in the limiting factor for each of the three regions (e.g., blue shading highlights temperature changes in the temperature-limited region). Circles indicate that the trend in the 140-year ensemble mean is significant at the 95% level from a Mann–Kendall trend test.

**Figure 8.** Spatial distribution of the additive and counteracting impacts of CO2 on climate limiting factors. Temperature, (**a**,**d**,**g**); precipitation, **b**,**e**,**h**; and solar radiation, (**c**,**f**,**i**). In order to assess the contribution of radiative and vegetation physiological effects on climate and vegetation, we used an eight-model [highlighted in Table 1] ensemble to compare three CMIP5 experiments, each of which was run for 140 years and experiences a constant CO2 at pre-industrial levels and/or CO2 increasing by 1%/year to 4xCO2: (1) In the radiative experiment (**a**–**c**), CO2 increases for the atmosphere but stays constant for vegetation and the carbon cycle and hence the direct effects of CO2 on plants are suppressed; (2) in the vegetation physiology experiment (**d**–**f**), CO2 increases by 1%/year to 4xCO2 for vegetation and the carbon cycle—thereby reducing stomatal conductance and providing CO2 fertilization—but stays constant at 280 ppm for the atmosphere and thus the radiative effect is suppressed; (3) in the combined experiment (**g**–**i**), CO2 concentration increases for the full Earth system. The changes are the mean of the last ten years minus the mean of the first ten years. Stippling indicates that the trend in the 140-year ensemble mean is significant at the 95% level from a Mann–Kendall trend test.

On the other hand, the summary of the vegetation response to increasing CO2 concentration shows the opposite of the climate response in the sensitivity experiment. In spite of inducing climate changes to temperature and precipitation, the CO2 fertilization effect alone can account for much, and in some cases, almost all, of the simulated changes in GPP and LAI in the combined experiment (compare blue and red lines in Figure 9d–i). Increases in LAI and GPP in the FixClim1 experiment appear to be driven by more radiation (because of reduced cloud cover) and reduced transpiration (therefore increased soil water), both of which are consistent with stomatal down-regulation following CO2 increases [35]. As shown in the esmFdbk1 experiment, in contrast, the climate change induced by radiative forcing has near-zero effects on simulated GPP for all three regions (black line in Figure 9d–f), except for a small positive effect on GPP in the boreal spring of the temperature-limited region. Spatially, the ordinal impact on GPP and LAI often reverses between the esmFdbk1 and esmFixClim1 experiments, with much of the Southern Hemisphere switching from a reduction (Figure 10b,c) to an enhancement (Figure 10e,f).

**Figure 9.** Additive and counteracting impacts of CO2 on vegetation. Same as Figure 7 except for transpiration, **a**–**c**; GPP, **d**–**f**; and LAI, **g**–**i**.

For seven out of nine comparisons, the esmFdbk1 and esmFixClim1 experiments produce offsetting impacts on LAI and GPP (Figure 9). The largest differences are in the radiation-limited region, where the esmFdbk1 experiment slightly reduces GPP and the esmFixClim1 experiment increases GPP by as much as 0.15 kg C/m2/month in JJA (compare blue and red lines in Figure 9f); LAI changes switch from about −0.5 to 1.0 (compare blue and red lines in Figure 9i). The temperature-limited region is an exception, where, for both LAI and GPP, the esmFdbk1 and esmFixClim1 experiments are additive (Figure 9m,p). In particular, high northern latitudes are the one clear location where both the esmFdbk1 and esmFixClim1 experiments increase LAI (Figure 10c,f) and GPP (Figure 10b,e), supporting recent conclusions of a strong climate imprint on the broad region of high-latitude during the observational era [36].

The critical role of vegetation physiology is clear in the precipitation-limited region, which experiences the least easing and an almost equal area with precipitation reductions (Figure 1d). But the region also sees a statistically significant increase in simulated equivalent precipitation water in 64% of its area (equivalent to up to 50 mm/month of water in some areas, Figure 1e) and increases in LAI and GPP (Figure 9e,h), in spite of near-zero changes in simulated annual (Figure 4b) and seasonal (Figure 6b) transpiration. While these patterns depict a more efficient use of available water resources and a progressive greening, the physiological response of plants to higher CO2 in semi-arid regions appears to depend on local variations in simulated precipitation, which remain highly uncertain in CMIP5 simulations [37].

**Figure 10.** Spatial distribution of the additive and counteracting impacts of CO2 on vegetation. Same as Figure 8 except for transpiration, **a**, **d**, **g**; GPP, **b**, **e**, **h**; and LAI, **c**, **f**, **i**.

#### *3.4. Decomposing Vegetation Growth Into Three Factors*

The results indicate that the climate feedback substantially contributed to the growth of vegetation by relaxing climate constraints. Although it is well known that the climate feedback can positively or negatively influence the growth of vegetation, it has not been quantitatively assessed at the global scale. The sensitivity experiments allow us to quantitatively evaluate the climate feedback to vegetation (*f* Δ*Climf eedback* in Equation (1)). We assumed that the ratio of GPP change to change in climate variables is constant (i.e.,b=f= *const.*). At first, we calculated Δ*GPP* / Δ*Clim* (i.e., b) in the climate experiment using the linear regression of annual GPP on annual mean climate variables. Then, we calculated *a* Δ*CO*<sup>2</sup> by subtracting *f* Δ*Climf eedback* from GPP increase in Equation (2). Finally, *b* Δ*Clim* was derived from Equation (3) by subtracting *a* Δ*CO*<sup>2</sup> and *f* Δ*Climf eedback* by assuming the additive relationship of the fertilization effect and the climate change effect.

The percentage ratios of the contribution of each term (i.e., *a* Δ*CO*2, *b*Δ*Clim*, and *f* Δ*Climf eedback*) were shown for each limited region in Figure 11. In the temperature-limited region, all the three terms substantially contributed to the increase in GPP, and the annual average of the climate feedback contribution was 17%. The snow-albedo feedback can account for the climate feedback [38,39]. The climate feedback added 37% more increase in GPP than the radiative warming effect alone. The total contribution of climate feedback and climate is 63%, which is the highest contribution among the three climate-limited regions. The contribution is much higher in winter than summer because the temperature did not limit GPP in the summer.

In the precipitation limited area, there was almost no contribution of the climate feedback effect. This result can be explained by the relatively low water-recycling ratio compared to the humid area [40]. The moisture from the other regions controls the precipitation trend in the water-limited region, so that the influence of changing water use efficiency on the region is negligible.

In the radiation-limited area, the contribution of the climate feedback is 7%, while the radiative climate change negatively affects 24% of the increase in GPP. The feedback was caused by the decreasing trend in cloud cover through change in water use efficiency [35]. The magnitude of the contribution of the climate feedback changes with model selection due to the difficulty in modeling clouds in GCMs. Thus, the feedback contribution can be underestimated, especially when a low resolution GCM cannot

represent well the increase in regional convective clouds caused by the enhanced water cycle that results from added vegetation growth.

**Figure 11.** Monthly contribution of climate feedback and radiative climate change to vegetation growth in increases in CO2 by 1%/year to 4xCO2 experiment (1pctCO2) for each climate-limited area. The monthly contribution was calculated for each climate-limited region. The green and purple bars show the contribution of climate feedback and radiative climate change, respectively. In each region, the total of the contribution (CO2 fertilization, climate feedback, and radiative climate change) was summed up to 100%.

#### **4. Discussion**

Our analysis suggests fundamental future increases in the amount of vegetation and photosynthesis (LAI and GPP in Figure 1), mainly arising from relaxing climate constraints on vegetation growth. This feedback effect can explain the discrepancy between the models and the observation in the β factor [14]. We argue that the results are consistent with three lines of observational evidence and a considerable body of paleoclimatic evidence of dramatically different vegetation composition during past high-CO2 periods [41,42].

First, if our central claim that vegetation physiological process reduces transpiration, reduces cloud cover, and increases radiation is correct, then cloud cover, particularly low-level clouds, which strongly influence the planetary shortwave radiation budget, should decrease. Recent climate modeling studies indeed simulate a decrease in low-level cloudiness due to the vegetation physiological effect [23,29,35]. Further, several modeling studies indicate that the rapid adjustments of the troposphere for the combined radiative and physiological effects of increased CO2 are associated with a decrease in low-level cloud cover over land, but increased boundary layer cloud cover over oceans [29,43,44]. We re-analyzed the NDP026 ground observation of cloud cover [26] and show that the modeled processes have indeed occurred over the period 1971 to 2005 using the Mann–Kendall trend test,. During this period, cloud cover significantly decreased by a few percent points per decade over much of the land surface, and increased over the ocean (Figure 12).

**Figure 12.** Changes in observed (NDP026) and CMIP5-simulated cloud cover are consistent with a CO2-induced down-regulation of stomatal conductance, resulting in reduced transpiration. Trends from observations are shown as decreasing (red) or increasing (blue) annual average cloud cover for 1971–2005: Red would tend to support our hypothesis of reduced transpiration and cloud cover as a result of stomatal down-regulation. If the CMIP5 trend and the observed trend have the same sign, the corresponding box is hatched. Location and shape of boxes corresponds to the coverage in the observational dataset, the 1971–2005 comparison period represent the overlap between the observational dataset and the historical CMIP5 runs. A change during the mid-1990s in cloud cover observation methodology in the US precluded their use in trend evaluation in NDP026. A chi-squared test between the two data sets rejected the null-hypothesis that they are independent (*p* = 0.05). Chi-squared tests performed on the two data sets at seasonal scales yielded the following *p*-values: December January February (0.22), March April May (0.05), June July August (0.26), and September October November (0.01).

Second, the same models we used for future projections under RCP 8.5 produce simulations of the historical climate and vegetation that are broadly consistent with independent observations (Figure 13). Although historical skill does not guarantee future performance, region-level simulations of climatological temperature, and precipitation are statistically indistinguishable from the Climate Research Unit (CRU) product [25] in all months (Figure 13a,b). Simulated radiation, assessed for the radiation-limited regions of the Northern hemisphere against the CRUNCEP radiation dataset, has climatological differences of up to 15 W/m2, but a similar seasonal cycle (Figure 13e). The Earth system models also capture seasonal satellite-observed variations in LAI in all three regions (Figure 13b,d,f).

**Figure 13.** Performance of Earth system models used in CMIP5 against observations (1982–2005) in each of the climate-limiting regions. CMIP5 seasonality for climate and vegetation are ensemble means of each of the parameters. Shading represents standard deviation around the ensemble mean from the CMIP5 models. Observed seasonality in climate and vegetation over the same period is calculated from Climate Research Unit (CRU) (temperature, precipitation), CRU National Centers for Environmental Prediction version (CRUNCEP) (radiation), and Global Inventory Modeling and Mapping Studies Leaf Area Index (GIMMS) LAI.

Third, the projected changes in LAI and climate are already apparent in the observational era. Satellite data show that LAI has increased from 1982 to 2005 for all three regions (Figure 14b,d,f); the CRU product shows warming in the temperature-limited region (Figure 14a), increased precipitation in the precipitation-limited region (Figure 14c), and reduced cloud cover in the radiation-limited region (Figure 14e). Consistent with these changes in climate and the biosphere, terrestrial ecosystems have been shown as net sink for carbon in recent decades [45,46]. Thus, the 21st century changes to climate and greening do not appear anomalous or implausible, when viewed in the context of recent history.

Numerous processes, including extreme climatic events [47], could reduce the projected changes in LAI and GPP. But the paleoclimate record also shows that profound changes in vegetation have occurred in the past, particularly in high latitudes, where the temperature-limited region appears to benefit the most from physical climate changes, mediated through vegetation physiological mechanisms. The early Eocene greenhouse climate, for example, supported redwoods at 78◦ N paleo-latitude under CO2 levels that are similar to the modern levels [42]. The deep-time perspective, albeit associated with different time scales and continental configurations, therefore, does not appear to rule out the sort of major changes to vegetation seen in the 21st century projections [48].

**Figure 14.** Performance of Earth system models in capturing the observed trends in climate and vegetation. Observed trends in annual mean climate and satellite-derived LAI from 1982 to 2005 are shown as average responses over the three climate-limiting regions. Ensemble mean of CMIP5 models over the same period do not capture the inter-annual variability, but appear to capture the overall trends in climate and vegetation. The first year of each series is set to zero to emphasize the magnitude of the trend and deviation between the trend lines. The *p*-values indicate the level of significance for the trends in observations only (CRU, CRUNCEP, GIMMS, and cloud cover data from NDP026). Shading around the CMIP5 ensemble mean indicates 25–75 percentile.

We focused on two elemental components of terrestrial ecosystems—the amount of leafy material and gross carbon fixation—but do not provide insights into respiratory and net carbon fluxes, carbon stocks, such as biomass and soil carbon, and vegetation dynamics. The CMIP5 models, especially low-resolution models, cannot count the extreme events, such as forest fires or hurricanes. The FACE experiments also suggest that non-climate limiting factors, such as nitrogen and phosphorous [49], might supersede climate limitations in the future (although the inclusion of a nitrogen cycle produces results that are within the uncertainties of the full ensemble; see red lines in Figures 3 and 4). The available state-of-the-art Earth system models; however, depict a late 21st century world in which vegetation physiology interacts with pervasive changes to annual and seasonal climate to create a greener land surface.
