Implementation of gs,n
Diurnal time series of canopy conductance (a, c) and transpiration
(b, d) for Ethiopia over 5 days in mid-January (a–b) and mid-July (c–d).
The control simulation (solid black line) had lower conductance and
transpiration than the Δgo simulation (dotted red line) and the
Δgmin simulation (dashed blue line). The Δgnight
simulation (dot-dashed teal line) had higher nighttime conductance and
transpiration than the control simulation, but similar daytime conductance
and transpiration, allowing for daytime conductance to fall below the
nighttime threshold. The Δgo simulation added the observed
gs,n values to the conductance calculation at every time, day or
night, which is not theoretically aligned with the function of including
observed gs,n. As a result, the Δgo simulation was
eliminated from further analyses. Note that all minimum thresholds (go,
gnight, and gmin) were adjusted using a soil moisture scalar.
Incorporating observed minimum constraints on gs in all modified
simulations increased gs and transpiration compared to the control
simulation, illustrated in Fig. 1 for a highly impacted semi-arid location
in Ethiopia (see Fig. S1 for other regions). The large variability in the
observational dataset causes substantial uncertainty in the simulations,
masking the differences among parameterizations and highlighting the impact
of gs,n on transpiration (Fig. S2). The sensitivity of gs and
transpiration to the altered go parameter in the Δgo
simulation is large
(Barnard
and Bauerle, 2013; Bowden and Bauerle, 2008). Since the higher go is
added to gs in the BWB calculation at every model time step (see Eq. 1),
altering go increases transpiration throughout the entire diel
cycle, and produces changes in the daytime evaporative flux that are not
supported by observations of gs,n. We consider that uniformly adjusting
the go parameter does not represent the correct implementation of
observed gs,n values.
Simulated average transpiration (a), runoff (d), and soil moisture
(g) for a control simulation, and percent change from control in
transpiration (b–c), runoff (e–f), and soil moisture (h–i) after including a
nighttime threshold (Δgnight; b, e, h) or a minimum gs
threshold (Δgmin; c, f, i) based on observational data. Note that
both nighttime and minimum thresholds were adjusted based on a soil moisture
scalar.
If go cannot be equated to plant minimum gs in the BWB paradigm,
this raises the possibility of whether go has a theoretical
interpretation beyond an empirical fitting parameter. It is possible that
go is equivalent to cuticular conductance (gcut), or conductance
that is not regulated by the stomatal guard cells (Caird et
al., 2007), occurring during the day and night.
Niyogi and Raman (1997) describe
go as cuticular conductance, though there is no record of go being
tested or described as gcut previously. Studies that have quantified
gcut found that gcut was a low proportion, < 10 %, of
total gs and less than measured gs,n
(Caird et al., 2007;
Zeppel et al., 2014). The values of go used in current implementations
of the Ball–Berry model for C3 plants (10 mmol m-2 s -1) fall
within the range of measured gcut values (4 to 20 mmol m-2 s-1; Caird et al., 2007). Assuming go does have
a theoretical function of representing gcut, rather than gs,n,
incorporating an observed threshold of minimum gs is necessary. Whether
go functions theoretically as gcut in the BWB model needs further
evaluation, as adjusting simulated go has large impacts on canopy
conductance and transpiration (Fig. 1;
Barnard
and Bauerle, 2013). Regardless, observed gs,n is larger than modeled
go and functions differently, and therefore should be considered
independently in model parameterizations.
The Δgmin and Δgnight simulations represent the
intended change in minimum gs with greater fidelity, by limiting the
minimum value without increasing gs at every model time step.
Interestingly, in restricting only nighttime conductance, the Δgnight simulation allows daytime gs to decrease below the
nighttime threshold during the dry season in semi-arid ecosystems (Fig. 1a).
This occurs when An nears zero in shade or low humidity, causing
gs to fall to the default (lower) go. In contrast, the Δgmin simulation restricts minimum gs at all times, and therefore
daytime values are never less than the water-adjusted gs,n. This
increases canopy-averaged daytime gs, and hence transpiration, compared
to the unmodified simulation whenever daytime gs values fall below the
minimum threshold (Fig. 1a, c).
The data in Table S1 are a compilation of all available published gs,n
data to date, and report gs,n values for 204 distinct plants. Of
these, only four plants exhibit higher gs,n than daytime gs, and
two of those are Crassulacean acid metabolism (CAM) plants, which by
definition open their stomata at night to gain carbon dioxide and close
their stomata during the day, and were not used in our parameterization.
These data suggest that, as expected, gs,n is typically less than
daytime gs. Most data presented in Table S1 are average values under
non-drought stressed conditions, and are likely only reported for leaves in
sunlit canopy layers. Thus, these data do not elucidate whether, at any
given time, daytime values might drop below the nighttime threshold, but
only suggest that, on average, they do not.
Average gross primary productivity (GPP) for a control simulation
(a), and percent change from control (b–c) after including a nighttime
threshold (Δgnight; b) or a minimum gs threshold (Δgmin; c) based on observational data. Note that both nighttime and
minimum thresholds were adjusted based on a soil moisture scalar.
In the context of the model simulations, low daytime gs occurs any time
that Ahr/C is low. These are conditions which are poorly illuminated
(in shade or at dawn/dusk and night), or when humidity is low. The CLM4.5SP
contains a representation of the shaded canopy, which has lower gs and
often reaches the minimum daytime threshold (go in the unmodified,
Δgo, and Δgnight simulations, and gs,n in the
Δgmin simulation). The central issue in determining whether the
Δgmin or Δgnight simulation is a better
representation of minimum gs is whether, under the same conditions in
the real world, daytime gs might be lower than gs,n. For example,
if observational data support that daytime gs is less than gs,n in
shaded canopy layers given the same water availability, then the Δgnight simulation is a better parameterization. However, if
observational data suggest that daytime gs is consistently higher than
gs,n, then the Δgmin simulation is a better
parameterization. While observational data are not available to specifically
answer this question, the available data presented in Table S1 and data from
Dawson et al. (2007), which suggest that gs,n is a fraction of daytime
gs, imply that daytime gs is on average higher than gs,n,
providing partial support for the Δgmin implementation. A
different implementation of gs,n might calculate gs,n as a
proportion of daytime gs, based on Dawson et al. (2007), who find that
gs,n is a proportion of daytime gs that changes based on days
since last rainfall. We do not test this potential method here, but
acknowledge it as a viable alternative to be considered.
The possible existence of a higher gs,n compared to daytime gs
raises an interesting question about the potential selective advantage for
leaves with a high gs,n. It is hypothesized that high gs,n may
provide a beneficial function to the plant, such as embolism repair or
phloem transport (e.g., Dawson et al., 2007). Additionally, gs,n may
contribute to xylem refilling, potentially improving carbon gain by making
water available when light conditions allow for photosynthesis (Dawson et
al. 2007). Critically, it is not clear whether these potential functions are
only relevant at night (and daytime gs can be lower than gs,n), or
whether high gs,n is representative of a general strategy of higher
overall minimum gs. We are not aware of data that exist to support
either possibility, and advocate for observations that will help determine
the functional significance of gs,n.
Terrestrial coupling for June–July–August for a control simulation
(a), and the difference from control (b–c) after including a nighttime
threshold (Δgnight; b) or a minimum gs threshold value
(Δgmin; c) based on observational data. Note that both
nighttime and minimum thresholds were adjusted based on a soil moisture
scalar.
Global values from CLM simulations and
observations*.
Simulation
gs,n
GPP
ET
Runoff
data used
(Pg C yr-1)
(103 km3 yr-1)
(103 km3 yr-1)
Control
n/a
157.83
65.6148
30.462
go
Mean
152.56
72.6555
24.2141
gnight
Mean
156.068
66.0926
30.0724
gmin
Mean
151.252
68.6843
27.8161
go
Median
153.641
71.5441
25.1739
gnight
Median
156.346
66.031
30.119
gmin
Median
152.385
67.8881
28.51
Observation
119–175
65.13
37.7521
* Global gross primary productivity (GPP), evapotranspiration (ET), and runoff values. Observed values presented in Bonan et al. (2011),
Welp et al. (2011), and Lawrence et al.
(2011). n/a – not applicable
From a model or theoretical perspective, it is important to note that the
reason that simulated gs values are reduced to as low as 10 mmol m-2 s-1 (or lower, if down-regulated for water stress) is a
function of the universal parameterization of all C3 plants with that
value of go. Given that it is unlikely that this value is universal for
all plants, we consider that the large difference between the Δgmin or Δgnight simulations is an artifact of the
poorly constrained parameterization of the daytime BWB model.
It should be noted that all the minimum thresholds implemented in our
simulations (Δgo, Δgnight, and Δgmin)
are adjusted by a soil water scalar (βsoil). Therefore, the
nighttime (Δgnight) and the minimum (Δgmin)
thresholds are altered according to the degree of soil moisture stress. When
the daytime gs value is lower than the gnight threshold in the
Δgnight simulation (Fig. 1c), the gnight threshold is
already down-regulated for water stress. In this scenario, the daytime
minimum gs is less than the nighttime gs when water stress is
equivalent.
Responses to dry soil conditions are mediated both through the minimum
gs values, and through the impact of soil moisture on photosynthetic
capacity and leaf maintenance respiration, which are also multiplied by
βsoil. Many of the impacts of our simulations result from
feedbacks between higher transpiration rates resulting in faster depletion
of soil moisture store, and therefore greater constraint on photosynthesis.
These results are all emergent features of the model and should not be
interpreted as direct results of the altered parameterization.
Global water and carbon
Average diel canopy transpiration for the months of May, June, and
July in Castlereagh, Australia (observation, dotted black line), estimated
from sap flux measurements of Red Gum and Iron Bark, the dominant tree
species in the canopy. Average simulated canopy transpiration for the grid
cell corresponding to Castlereagh, Australia, for the control (unmodified;
solid black line), Δgo (Ball–Berry go parameter adjusted;
red line), Δgnight (minimum nighttime threshold added; teal
line), and Δgmin (minimum conductance threshold added; blue
line) simulations. Error bars corresponding to the observations (dashed) and
each simulation (solid) are colored accordingly, and are calculated as ± one standard deviation from the mean. Note that the simulations use
meteorological forcings from an atmospheric dataset (see Methods), not the
local meteorology from when the measurements were collected (some
meteorological data were collected at the site, but not all variables
required by the model). The simulated grid cell covers a much larger area
than the observational data collection site.
When averaged over 25 years, incorporating observed rates of gs,n in
the Δgmin simulation increased transpiration losses by up to
30 % in the Amazon, and > 30 % in some arid regions, in part
due to the small absolute magnitude of available soil water (Fig. 2a–c).
Semi-arid regions are primarily broad-leaf shrub and C3 grass PFTs that
have particularly high values (130 and 156 mmol m-2 s-1
respectively) of observed gs,n (Table 1), and have high nighttime vapor
pressure deficits that interact with higher minimum gs values, causing
large nighttime transpiration rates. Using median rather than mean values
caused only small (< 1.5 %) differences in global transpiration
(Figs. S3, S4). Though the magnitude of response is different depending
on parameterization used, the increases in transpiration imply that current
model estimates of plant water loss are underestimated in many regions.
Simulated higher transpiration resulting from higher minimum gs also
has ecosystem-scale ramifications for hydrology
(McLaughlin et al., 2007). For example, the
increased transpiration resulted in drier soils compared to the control
simulation (Fig. 2g–i), with Δgmin causing > 40 %
soil moisture decreases in semi-arid ecosystems like the southwestern United
States and much of Australia (> 10 % in Δgnight).
Additionally, the Δgmin-estimated changes to surface runoff are
large in some regions, such as the 10–25 % decreases in the tropics
(5–10 % in Δgnight; Fig. 2d–f), suggesting that current
runoff estimates may be too large. It should be noted that the difference
between the Δgmin and Δgnight simulations is
largely due to changes in minimum gs that affect daytime gs (see
Sect. 3.1). Hydrologic changes in soil moisture and runoff in response to
increased gs have previously been documented in catchments in
the southeastern United States (McLaughlin et al.,
2007), and our results suggest that changes to stomatal conductance have
similar consequences in CLM4.5SP simulations. Additionally, increasing
minimum gs caused gross primary productivity (GPP) to decrease
(Fig. 3) by 10 to > 25 % in many semi-arid regions. These are regions
where water availability already restricts GPP, and the decreases in soil
moisture caused by higher transpiration likely impart even more
drought-induced stomatal closure.
To more directly evaluate the potential influence of minimum gs on the
climate system, we calculate the change in terrestrial coupling to the
atmosphere. The terrestrial coupling index
(Dirmeyer, 2011) estimates the degree to which
changes in soil moisture control surface energy fluxes to the atmosphere.
This study uses root-zone soil moisture, rather than soil moisture over
spatially constant soil depth, to highlight the direct impact of vegetation
and minimum gs on surface fluxes. Here we calculate the terrestrial
coupling index during boreal summer months when warmer temperatures allow
for the highest gs rates. We find that the terrestrial coupling
strength increases when using the Δgmin implementation, but is
generally unchanged for Δgnight (Fig. 4), meaning root-zone
soil moisture exerts a greater control on surface flux variability for
Δgmin, largely due to the impact this simulation has on daytime
gs. This increased terrestrial coupling to the atmosphere largely
mirrors the reductions in GPP and soil moisture in semi-arid ecosystems, and
may reinforce climate extremes such as droughts or heat waves
(Hirschi et al., 2011;
Miralles et al., 2014).
Evaluating gs,n
Evaluating the performance of the new gs,n parameterizations is
challenging for numerous reasons. First, our model scales from leaf-level
gs and gs,n estimates to canopy transpiration. The best way of
evaluating the model is to compare simulated and observed canopy
transpiration because the model captures the average of an entire canopy,
which is comprised of multiple plant functional types, rather than
individual plant functional types. Incorporating realistic minimum gs
increases global evapotranspiration and decreases global runoff compared to
globally scaled observations, while estimates of GPP from all simulations
fall within the range of global GPP estimates from observations (Table 2;
Bonan
et al., 2011, 2012; Li et al., 2011). However, these comparisons should be
used with caution, since eddy covariance data used in estimating the GPP and
evapotranspiration observations are susceptible to errors at night
(Fisher
et al., 2007; van Gorsel et al., 2008; Kirschbaum et al., 2007; Medlyn et
al., 2005) due to a lack of sufficient canopy turbulence that precludes
detection of nighttime transpiration using this measurement methodology, and
are not useful for evaluating the changes in water fluxes tested in this
study. Other data for evaluating model responses to minimum gs on large
spatial scales are not yet available.
A comparison of simulated canopy transpiration to transpiration calculated
from sap-flux data in Australia (Fig. 5) illustrates that a minimum gs
threshold changes transpiration estimates during the early part of the
night, though simulated nighttime rates are still low compared to
observations. All model parameterizations fall within the observational
range of uncertainty, but under-predict nighttime and midday canopy
transpiration during May and June, and over-predict midday canopy
transpiration in July. The lack of fidelity between the various model
parameterizations and the observations is likely affected by the fact that
observed meteorological data were unavailable to force the model. Therefore,
key parameters driving both daytime and nighttime transpiration fluxes, such
as VPD and soil water availability, were likely different in the model
simulations compared to the actual meteorological conditions at Castlereagh,
Australia, during data collection. Additionally, because sap flow is measured at the
base of the tree, there is typically a lag between when sap flow is measured
and when the canopy transpires, and this lag is also notable in comparing
observed sap flow with simulated estimates of transpiration. Estimating
nighttime transpiration using sap-flow methodology is also convoluted with
the refilling of aboveground water stores depleted during the day, and thus
is not directly comparable to our simulations. It should also be noted that
the model does not have a semi-arid plant functional type, so semi-arid
plants are typically represented in the model as deciduous plant functional
types.
Given that our study focused only on one aspect of the gs formulation
within a land-surface model, evaluating daytime gs and other aspects of
the BWB model function (i.e., photosynthetic drivers of daytime gs,
feedbacks to water availability, etc.) are all subject to pre-existing
deficiencies in the representation of a host of other model processes. For
example, there are only two values of the g1 (slope) parameter in the
BWB model, one for C3 and one for C4 plants
(Sellers et
al., 1996), and this parameter has not been modified or comprehensively
evaluated within the context of the CLM4.5SP. Indeed, the use of the BWB
model at all is currently the subject of some debate
(Bonan
et al., 2014; De Kauwe et al., 2015), and this study additionally highlights
how the empirical nature of the BWB model leads to difficulties when
attempting to implement mechanistic processes. Further, daytime gs is
also dependent on the photosynthetic capacity, and observations of
Vcmax and Jmax
(Bonan
et al., 2011; Kattge and Knorr, 2007) indicate very wide ranges of plant
functional type variation in these properties, also limiting our confidence
that the globally averaged parameters used in the default model will lead to
accurate gs and transpiration at most locations. We choose not to focus
on these and other parameters that effect daytime gs, as it does not
directly impact the representation of gs,n, and is therefore beyond the
scope of this paper.