Variations of biochemical and photosynthetic responses to eCO2
The direct CO2 fertilization effect occurs at leaf level and is
determined by kinetic sensitivity of Rubisco enzymes to internal leaf
CO2 concentration. In fact, the normalized short-term sensitivity
of leaf-level photosynthesis to CO2 is mainly regulated by Ci
and slightly influenced by leaf temperature, regardless of light, nutrient
availability, and species characteristics (Luo et al., 1996; Luo and Mooney,
1996). In our study, the modeled Ci/Ca ratio is approximately
constant with eCO2 for a specific PFT, and varies little
within and across PFTs in all simulations. This is in line with FACE
experimental results which show almost constant Ci/Ca values
for different PFTs under CO2 fertilization (Drake et al., 1997;
Long et al., 2004). Γ∗ varies little for different species
and only depends on leaf temperature (Luo and Mooney, 1996). Sensitivity
analysis in a previous study has shown that a ±5∘ of leaf
temperature changes caused approximately ±7 ppm changes in Γ∗, leading to variation of 0.12 to leaf-level β (Luo and
Mooney, 1996). The overall variation of leaf-level β caused by
variation in leaf temperature is still quite small compared with that of
βGPP. Therefore, biochemical and leaf-level β values
vary little within and among PFTs in this study. Our results also illustrate
that nutrient effects do not significantly change Ci and Γ∗, leading to similar biochemical and leaf-level β values
in all simulations, which is in accordance with Luo et al. (1996).
To identify the source of uncertainty of β in CMIP5 models, Hajima et
al. (2014) decomposed β into several carbon-cycle components. They
used GPP divided by LAI (GPP / LAI) as a proxy to represent leaf-level
photosynthesis for CMIP5 models, since there are no leaf-level process
outputs of these models. They found the sensitivities of GPP / LAI to
eCO2 diverged a lot among models. One possible issue of
this calculation is that it ignores different canopy structures used by each
CMIP5 model, such as big-leaf, two-leaf or multiple-layer. Our results just
show that the sensitivities of GPP / LAI are different from our
mechanistic calculation of leaf-level β for different PFTs in a
two-leaf model. β values estimated from GPP / LAI formulation are
greatly underestimated for woody trees and slightly overestimated for
C3 grass and tundra, but best match for shrub if compared with our
calculation (Fig. S8). Therefore diagnostics such as Ci and Γ∗ for leaf-level β are more desirable for woody trees.
Another advantage of our calculation of leaf-level β is that the
reason for the divergence of leaf-level β across PFTs can be traced
back to the difference from Ci and leaf temperature as shown in Fig. 2.
Variations of β at canopy and ecosystem levels
The two-leaf scaling scheme in CABLE is widely employed by many land-surface
models, such as Community Land Model version 4.5 (CLM4.5, Oleson et al.,
2013) and the Joint UK Land Environment Simulator version 4.5 (JULES4.5, Best
et al., 2011; Clark et al., 2011; Harper et al., 2016). We found the
responses of ecosystem carbon cycle to eCO2 diverge
primarily because the responses of LAI diverge within and among PFTs in all
simulations. Besides, GPP of shaded leaves responds to
eCO2 more strongly than GPP of sunlit leaves for all
C3 PFTs. This is because the portion of shaded leaves increases
exponentially with increasing LAI (Fig. S9), leading to a rapid change in
shaded leaf GPP, while for sunlit leaves, GPP shows a saturating response
because of the decreasing portion of sunlit leaves with increasing LAI (Dai
et al., 2004). Our results also indicate that saturation of GPP is not only
regulated by the leaf-level photosynthetic response, but also by the response
of the LAI-dependent scaling factor to eCO2. For shaded
leaves, the sensitivity of the LAI-dependent scaling factor contributes more
to the magnitude and trend of βGPPsha than
that of photosynthesis rate. The evidence all suggests LAI is a key process
in modeling the response of ecosystem carbon cycle to climate change.
It has been reported that different CMIP5 models have simulated diverse LAI
during 1985–2006. And modeled LAI values
in most CMIP5 models have been overestimated according to satellite products
(Anav et al., 2013). Many global vegetation models simulated increasing LAI
trends globally in response to eCO2 during the historical
period (Zhu et al., 2016). Our modeling study also shows that LAI responds
positively to eCO2 for all C3 PFTs in all
simulations. But experimental results are not consistent. In one review paper
with 12 FACE experimental results, trees had a 21 % increase in LAI and
herbaceous C3 grasses did not show a significant change in LAI
(Ainsworth and Long, 2005). Some studies reported that LAI dynamics did not
significantly change in specific FACE experiments, such as in a closed-canopy
deciduous broadleaf forest (ORNL FACE, Norby et al., 2003) and in a mature evergreen broadleaf forest
(EucFACE, Duursma et al., 2016). The
negligible change in LAI at the EucFACE probably leads to an insignificant
response of productivity at this site, even though leaf photosynthesis rate
significantly increases under eCO2 (Ellsworth et al.,
2017). Besides the impact of LAI on the global carbon cycle, the increasing
trend of LAI exerts profound biophysical impacts on climate by altering the
energy and water cycles on the Earth's surface (Forzieri et al., 2017; Zeng
et al., 2017). But there is a great uncertainty in the relationships between
LAI and biophysical processes among land-surface models (Forzieri et al.,
2018).
In this study, modeled nutrient-unlimited βGPP and
βNPP values are higher than leaf photosynthetic responses
for all C3 PFTs in C-only simulation (Fig. 3a). Nutrient-limited
βNPP are still higher than photosynthetic responses for
many PFTs in C–N and C–N–P simulations (Fig. 3b, c). However, it is
generally observed in experiments that the leaf-level response is
consistently larger than the whole plant response (Long et al., 2006;
Leuzinger et al., 2011). One possible reason is that models overestimate the
response of LAI to eCO2, as this study has shown that LAI
is an important factor in driving ecosystem response to CO2
fertilization. And it is also likely the overestimation of the response of
LAI to eCO2 is responsible for the overestimation of
CO2 fertilization in ESMs reported by previous studies (Smith et
al., 2015; Mystakidis et al., 2017).
Temporal trends of βGPPsun
(sensitivity of sunlit leaf GPP; red squares), βGPPsha (sensitivity of shaded leaf GPP; green
squares), βSsun (sensitivity of scaling fatcor for sunlit
leaf; pink triangles), βSsha (sensitivity of scaling fatcor
for shaded leaf; dark blue triangles), βpsun (photosynthetic
response for sunlit leaf; purple diamonds) and βpsha
(photosynthetic response for shaded leaf; sky blue diamonds) for C3
PFTs from CABLE-C-only simulation.
The sensitivities of GPPsun and
GPPsha tend to approach zero through time because the
decomposing factors βpsun, βpsha,
βSsun and βSsha all decline with time.
βSsha determines the magnitudes and trends of βGPPsha for almost all PFTs. Abbreviations are the
same as Fig. 1.
The overall response of LAI to eCO2 depends on several
processes in this study: (1) NPP increase, (2) change in allocation of NPP to
leaf, (3) change in specific leaf area (SLA) in response to
eCO2, and (4) PFT-specific minimum and maximum LAI values
prescribed in the model. First, the low responses of LAI to
eCO2 for deciduous needleleaf forest and tundra can be
attributed to smaller NPP enhancements in cold areas. The large divergence of
the response of LAI within PFTs is mainly due to the large range of NPP
increment across different geographical locations. The reduced magnitude of
βLAI under nutrient limitations is the direct outcome of
reduced βNPP. Accurate estimate of response of GPP and NPP
is therefore fundamental to realistic LAI modeling. Second, diverse
allocation schemes influence the responses of LAI for different PFTs. And,
results from two FACE (Duke Forest and Oak Ridge) experiments indicate that
the carbon allocated to leaves is decreased and more carbon is allocated to
woods or roots at higher CO2 concentration (De Kauwe et al., 2014).
Unfortunately, CABLE has fixed allocation coefficients and likely
overestimates LAI response, leading to overestimated responses of GPP, NPP
and total carbon storage. Third, we fixed SLA to calculate LAI in CABLE. But
a reduction in SLA is a commonly observed response in eCO2
experiments (Luo et al., 1994; Ainsworth and Long, 2005; De Kauwe et al.,
2014). Tachiiri et al. (2012) also found SLA and β values are most
effectively constrained by observed LAI to smaller values in a model.
Therefore, the fixed SLA may also lead to over-prediction of the response of
canopy cover to eCO2. Fourth, in our results, LAI values
for most C3 PFTs are below the maximum LAI limits with
eCO2 in C-only simulation. With only one exception, LAI
values of many evergreen broadleaf forest patches saturate at the prescribed
maximum value under high CO2 concentration (Fig. S1a and Table S1).
That is why the sensitivity of LAI for evergreen broadleaf forest is low and
thus leads to small relative GPP enhancements. If the preset LAI upper limits
are narrowed, β values are expected to be significantly reduced. Hence
model parameters related to LAI need to be better calibrated according to
experiments and observations in order to better represent the response of
ecosystem productivity to eCO2 (De Kauwe et al., 2014; Qu
and Zhuang, 2018).
In this study, the almost identical values and variance of βNPP to those of βGPP within and across
C3 PFTs in C-only simulation suggest carbon use efficiency (CUE)
does not change with eCO2, as autotrophic respiration is
calculated from GPP and plant carbon. In C–N and C–N–P simulations,
magnitudes of βNPP for all C3 PFTs except
evergreen broadleaf forest all decline compared with those of
βGPP, indicating CUE also declines with
eCO2 under nutrient limitations. However, FACE
experimental results indicate that CUE values under eCO2
are not changed in the N-limited Duke site (Hamilton et al., 2002;
Schäfer et al., 2003), increase in the fertile POPFACE site (Gielen et
al., 2005) or decrease in the fertile ORNL site (DeLucia et al., 2005). Thus,
representations of nutrient effects on GPP and autotrophic respiration in
land-surface models should be carefully calibrated with experimental data
(DeLucia et al., 2007). Our results also show that βNPP
values diverge more than βGPP values across different PFTs
in nutrient-coupled simulations, because the different nutrient-limiting
effects on autotrophic respiration introduce additional variation across
different PFTs. Although β values at ecosystem levels are more
variable with nutrient effects, LAI responses are still linearly correlated
well with βGPP, βNPP and
βcpool across C3 PFTs in nutrient-coupled
simulations as in C-only simulation, confirming the dominant role of LAI in
regulating carbon-cycle response under CO2 fertilization.
The reduced magnitudes of βcpool compared with those of
βGPP and βNPP in all simulations
indicate carbon turnover processes make ecosystems respond to
eCO2 less sensitively due to the slow allocation and
carbon turnover processes. A previous study using seven global vegetation
models identified carbon residence time as the dominant cause of uncertainty
in terrestrial vegetation responses to future climate and atmospheric
CO2 change (Friend et al., 2014). The response of soil carbon
storage to eCO2 also depends on soil carbon residence time
(Harrison et al., 1993). In this study and many other models, allocation
coefficients were fixed over time
(Walker et al., 2014). But allocation patterns to plant organs with different
lifespans have been reported to change in response to eCO2
in experiments, thereby altering carbon residence time in plants and soil (De
Kauwe et al., 2014). Therefore, the fixed allocation scheme we adopted in
this study might lead to some biases in simulating the response of carbon
residence time to eCO2. In our study, soil decomposition
rate is assumed not to be affected by CO2 level, as in most other
conventional soil carbon models (Friedlingstein et al., 2006; Luo et al.,
2016). However, recent synthesis of
experimental data suggested that replenishment of new carbon into soil due to
eCO2 increases turnover rate of soil carbon (Van
Groenigen et al., 2014; Van Groenigen et al., 2017). Within a certain PFT,
the variation of βcpool across different geographical
locations is usually smaller than that of βNPP, while the
greater variation of βcpool than that of
βNPP across different C3 PFTs in C-only
simulation suggests other processes such as different carbon allocation
patterns, plant carbon turnover, and the soil carbon dynamics of various PFTs
are responsible for the additional divergence. In nutrient-coupled
simulations, the variations of βcpool across different
C3 PFTs are only slightly larger than those of
βNPP, indicating that nutrients do not bring many
differential effects on carbon turnover processes for different PFTs.
Implication for understanding β in other models
Although we analyzed a single land-surface model in detail, the patterns of
and mechanisms underlying the variability of β we found may be
generally applicable to other models. The basic Farquhar photosynthesis model
and two-leaf scaling scheme in the CABLE model are shared by many
land-surface models. Some models use variants of the Farquhar photosynthesis
model such as the co-limitation approach described by Collatz et al. (1991).
Inflection point from Rubisco- to RuBP-limited processes is an important
control of the absolute photosynthetic response to eCO2
(Rogers et al., 2016). However, the relative photosynthetic responses for
different ecosystems will converge to a small range because the normalized
photosynthetic response to eCO2 only depends on estimates
of intercellular CO2 concentration (Ci), Michaelis–Menten
constants (Kc, Ko) and CO2 compensation point (Γ∗), and the relative photosynthetic responses are similar for
either Rubisco- or RuBP-limited photosynthesis (Luo et al., 1996; Luo and
Mooney, 1996). Soil moisture availability is another key constraint on
photosynthetic response. Water stress on plants is generally alleviated under
eCO2 due to reduced stomatal conductance (Leuzinger and
Körner, 2007; Fatichi et al., 2016). Different models simulate diverse
levels of water stress on productivity (De Kauwe et al., 2017). Water stress
is simulated in many models to regulate stomatal conductance (Rogers et al.,
2016; Wu et al., 2018). For example, the CABLE model represents water stress
by an empirical relationship based on soil texture and limits the slope of
the coupled relationship between photosynthesis rate and stomatal conductance
as Eq. (S11). The influence of water stress is reflected by Ci.
Synthesis of many empirical study results and our results in this study all
show that the ratio of Ci to Ca is relatively constant,
probably due to homeostatic regulations through photosynthetic rate and
stomatal conductance (Pearcy and Ehleringer, 1984; Evans and Farquhar, 1991).
Wong et al. (1979) showed plant stomata could maintain a constant
Ci/Ca ratio across a wide range of environmental conditions,
including the water stress condition. Land-surface models might simulate
relatively constant Ci/Ca ratios under water stress as well
since photosynthesis and stomatal conductance are theoretically depicted
based on experimental results. Moreover, Luo and Mooney (1996) found that
changing the Ci/Ca ratio from 0.6 to 0.8 caused a variation of
less than 0.08 in the sensitivity of leaf photosynthesis to a unit of
increase in Ca. Kc and Ko are variable among species,
but only slightly affect leaf-level response (Luo and Mooney, 1996).
Different leaf temperature will exert a limited influence on the variability
of leaf-level β, as we discussed above. Therefore, leaf-level β
values for different C3 PFTs are more likely to converge in other
land-surface models.
A recent study used 16 crop models to simulate rice yield at two FACE sites
(Hasegawa et al., 2017). These models have diverse representations of primary
productivity. Their results showed that the variation of yield response
across models was not much associated with model structure or magnitude of
primary photosynthetic response to eCO2, but was
significantly related to the estimations of leaf area. This is consistent
with our conclusion and highlights the great need to improve prognostic LAI
modeling. Other land-surface modeling groups may benefit from a similar
analysis to identify major causes of variability of β across the
hierarchical levels from biochemistry to land carbon storage. Candidate
causes that can make substantial contributions to the variability include
changes in leaf area index, changes in carbon use efficiency and changes in
land carbon residence times. If modeling groups can add leaf-level
diagnostics in the next inter-model comparison project, it will greatly help
disentangle the uncertainty of concentration–carbon feedback.