Uncertainty in the response of transpiration to CO₂ and implications for climate change

The model used in the following analysis is version 2.9 of the UVic ESCM, a climate model of intermediate complexity, with a horizontal resolution of 3.6° longitude × 1.8° latitude. It includes schemes for ocean physics based on the modular ocean model version 2 [16], ocean biogeochemistry [14], and a two-dimensional atmospheric energy moisture balance model including a thermodynamic sea ice model [17, 18]. The terrestrial component consists of simplified versions of the Meteorological Office surface exchange scheme (MOSES) and the top-down representation of interactive foliage and flora including dynamics (TRIFFID) vegetation model [19, 20]. The land surface scheme calculates surface albedo, runoff and evapotranspiration, which is a function of canopy resistance and based on the Penman–Monteith equation [21]. The vegetation scheme calculates the state of the terrestrial biosphere in terms of soil carbon, and the structure and coverage of bare soil or five plant functional types [20, 22]. Changes in vegetation biomass and distribution are driven by net carbon fluxes, which are derived for each vegetation type using the coupled photosynthesis stomatal conductance model [24].

In the UVic ESCM the leaf conductance of H₂O (g_w) is directly proportional to the leaf conductance of CO₂ (g_c) via their molecular diffusivities:

$$\begin{eqnarray}&&{g}_{w}=\beta \ast {g}_{c}\end{eqnarray} \tag{ 1 }$$

β accounts for different molecular diffusivities of water vapour and CO₂. As in numerous other models [23], g_c is directly proportional to the amount of carbon uptake by photosynthesis (P) and the gradient between the internal (C_I) and ambient (C_A) CO₂ concentration

$$\begin{eqnarray}&&{g}_{c}=(P*\alpha )/({C}_{{\rm{A}}}-{C}_{{\rm{I}}}).\end{eqnarray} \tag{ 2 }$$

α is a unit conversion factor.

Since transpiration is an important parameter connecting the energy balance, water mass balance and the carbon cycle, the objective of the applied scaling was to investigate the relevance of changes in the CO₂-sensitivity of transpiration to future climate projections.

The scaling could have been applied to both, the stomatal conductance of carbon and water vapour, which would have kept the molecular diffusivities consistent. In models, the effect of CO₂ on photosynthesis and transpiration are highly parameterized and not well constrained by observations. The UVic ESCM underestimates the fertilization effect on photosynthesis over the historical period [25] and the transpiration response to CO₂ appears to be larger compared to other models. Therefore, we decided to scale only the strength of g_w relative to its preindustrial value, undoing the direct proportionality of CO₂ and H₂O stomatal conductance. Since we do not change the effect of CO₂ on photosynthesis we keep the model's photosynthesis response closer to observations. Simultaneously, we manipulate the CO₂-sensitivity of transpiration, in order to cover the wide range simulated by other models. While modifying only the transpiration response to CO₂ may seem to make the model inconsistent in terms of stomatal conductance, it is a simple and clean way to separate the two different effects and allows us to manipulate plant WUE.

In order to vary the sensitivity of transpiration to increasing atmospheric CO₂ concentrations, a scaling is applied to the ambient CO₂ concentration, C_A, used in calculating the stomatal conductance of water vapour. ${C}_{{\rm{A}},\mathrm{scaled}}$ can be scaled up or down, relative to preindustrial concentrations of 280 ppm (${\mathrm{CO}}_{2,\mathrm{preind}}$), using a sensitivity factor, f_sens:

$$\begin{eqnarray}&&{C}_{{\rm{A}},\mathrm{scaled}}={\mathrm{CO}}_{2,\mathrm{preind}}+({C}_{{\rm{A}}}-{\mathrm{CO}}_{2,\mathrm{preind}})\ast {f}_{\mathrm{sens}}.\end{eqnarray} \tag{ 3 }$$

This scaling is only applied to ambient CO₂ used in calculating the leaf's conductivity to water vapour. The other equations used in the model, including those for photosynthesis continue to use the unaltered ambient CO₂ concentration for all calculations.

For consistency, in the calculations for g_w, a scaled version of the internal CO₂ concentration (${C}_{{\rm{I}},\mathrm{scaled}}$) is calculated by replacing the C_A with ${C}_{{\rm{A}},\mathrm{scaled}}$. Note that the calculation itself was not altered.

$$\begin{eqnarray}{C}_{{\rm{I}},\mathrm{scaled}} & = & ({C}_{{\rm{A}},\mathrm{scaled}}-{C}_{*})\ast \\ & & \times {F}_{0}(m)\ast (1-Q/{Q}_{\mathrm{crit}}(m))+{C}_{*}.\end{eqnarray} \tag{ 4 }$$

Here C_* is the canopy-level photorespiration compensatory point and F₀ is the ratio of the internal to the ambient CO₂ concentration for plants that are not water stressed. Plants are not water stressed if the canopy-level specific humidity deficit, Q, equals zero. The critical humidity deficit Q_crit and F₀ are constant values depending on the plant functional type, m, their values are given in table 1. Both scaled CO₂ concentrations, ${C}_{{\rm{A}},\mathrm{scaled}}$ and ${C}_{{\rm{I}},\mathrm{scaled}}$, are used to calculate the leaf's conductivity with respect to water vapour:

$$\begin{eqnarray}&&{g}_{w}=\beta \ast (P\ast \alpha )/({C}_{{\rm{A}},\mathrm{scaled}}-{C}_{{\rm{I}},\mathrm{scaled}}).\end{eqnarray} \tag{ 5 }$$

In case C_A is scaled down, plants will only feel a reduced increase in ambient CO₂ and the difference between C_A and C_I would be reduced, causing the leaf conductivity for water vapour, and consequently transpiration, to increase. Note, that the scaling only affects the leaf conductance of H₂O and not the leaf conductance of CO₂ (figure 1). Consequently, the scaling alters the amount of transpirational water lost by the plant per unit of carbon uptake, so effectively the WUE.

Zoom In Zoom Out Reset image size

Table 1.  List of parameters used in the calculations for the internal CO₂ concentration, the leaf conductance of CO₂ and water vapour. Given are descriptions or values of the parameter and their units.

Parameter	Description/Value	Units
CA	Ambient canopy CO2 pressure	(Pa)
CI	Leafs internal CO2 pressure	(Pa)
fsens	{0.0; 0.2; 0.4; 0.6; 0.8; 1.0}	(1)
C*	Canopy-level photo-respiratory compensatory point	(mol m−3)
m	Plant functional types: broad leaf tree (BT); needle leaf tree (NT); C3 grass (C3); C4 grass (C4); shrub (S)
${F}_{0}(\mathrm{BT},\mathrm{NT},{\rm{C}}3,{\rm{C}}4,{\rm{S}})$	(0.875, 0.875, 0.900, 0.800, 0.900)	(1)
Q	Canopy level specific humidity deficit	(kg H2O/kg air)
${Q}_{\mathrm{crit}}(\mathrm{BT},\mathrm{NT},{\rm{C}}3,{\rm{C}}4,{\rm{S}})$	(0.090, 0.060, 0.100, 0.075, 0.100)	(kg H2O/kg air)
gc	Leaf conductance for CO2	(m s−1)
gw	Leaf conductance for H2O	(m s−1)
P	Net leaf photosynthesis	(mol CO2 m−2 s−1)
α	Factor for converting mol m−3 into Pa	(J mol−1)
β	1.6	(1)

The UVic ESCM was spun up with seasonal, year 1800 forcing for over 10 000 years. Since the scaling is only applied to CO₂ concentrations deviating from preindustrial, all sensitivity simulations started from the same initial preindustrial spin-up. All simulations were integrated for 500 years until 2300, using historical emissions followed by RCP8.5 and the extended concentration pathway 8.5 emissions scenario until 2250 [26]. Thereafter the atmospheric CO₂ concentrations were held constant until 2300. For the following analyses yearly output was used. Continental ice sheets, volcanic forcing and astronomical boundary conditions were held constant to facilitate the experimental setting and analyses [14]. There was no land use forcing or burning applied, in order to investigate the systems' sensitivity in an unperturbed state.

Realizing that the UVic ESCM's sensitivity of transpiration to ambient CO₂ is at the high end of current models [15] we decided to scale down its sensitivity. Hence, we implemented the scaling factor with values of ${f}_{\mathrm{sens}}=\{0.0;0.2;0.4;0.6;0.8;1.0\}$, allowing us to scale down the sensitivity of transpiration to increasing CO₂ concentrations. This is expected to cause a relative increase in terrestrial evapotranspiration and precipitation compared to the default setting (${f}_{\mathrm{sens}}=1.0$). In addition, for the comparison with the CMIP5 models three simulations following the CMIP5 forcing protocols were performed with values for ${f}_{\mathrm{sens}}=\{0.0;0.5;1.0\}$. In the default simulations following the CMIP5 forcing protocols, terrestrial precipitation trends between 1961–1990 and 2071 and 2100 are higher by 20 mm yr⁻¹ compared to the runs forced by CO₂ only. This increase results in an overall positive global precipitation trend, and can be explained by the reduction in vegetation cover due to the implementation of land use changes.

The applied scaling has a strong impact on the spatial patterns of simulated future precipitation changes (figure 2). The default model (${f}_{\mathrm{sens}}=1.0$) simulates an increase in precipitation at higher latitudes between 50–80 °S and 50–80 °N, of about 50–200 mm yr⁻¹ by the end of the century (figure 2(a)). At high latitudes, spatial patterns are relatively independent of the land-ocean distribution and hence zonally coherent. For the mid and low latitudes a distinction can be drawn between areas over land and ocean. Regions of strongly reduced future precipitation in the default simulation lie mainly over continental areas. The strongest simulated decrease occurs over Australia with end-of-the-century precipitation decreasing by up to 270 mm yr⁻¹. In contrast, simulated precipitation increases by approximately 220 mm yr⁻¹ over adjacent oceanic regions. Reducing the sensitivity of transpiration of plants towards higher CO₂ concentrations by applying lower scaling factors, f_sens, this pattern shifts to an increasing trend for precipitation values over tropical land areas, while oceanic and desert areas remain largely unchanged.

Zoom In Zoom Out Reset image size

At high latitudes simulated future precipitation is independent of the applied scaling factor, f_sens, indicating that these areas are less sensitive to variability in transpiration. In mid to low latitudes, the ${f}_{\mathrm{sens}}=0.0$ model simulates increased future precipitation over all vegetated land areas relative to the default simulation. The largest increase in terrestrial precipitation of up to 500 mm yr⁻¹ (increase by 60% relative to the ${f}_{\mathrm{sens}}=1.0$ simulation) is seen over the northern part of South America, Central Africa and Southeastern Asia. These areas are mainly covered by broad leaf trees in the UVic ESCM, corresponding to tropical rain forest or savannah.

When the scaled model simulations of the UVic ESCM are compared to the range of the CMIP5 simulations (figure 3(a), model details in supplementary table D), the general shape of the latitudinal changes in simulated precipitation is similar among the three sensitivity simulations in the extra-tropical regions. There is an increase of precipitation in the high latitudes and a decrease in the mid latitudes of both hemispheres. The latitude of transition between these two trends, however, depends on the applied scaling and varies between 45 °N for a scaling of ${f}_{\mathrm{sens}}=0.0$ and 60 °N for the default simulation. The tropical latitudes reveal large differences between the differently scaled model simulations. Precipitation between 10 °N and 10 °S either increases by 400 mm yr⁻¹ for the scenario, where transpiration is calculated with preindustrial CO₂ levels, or it decreases by 90 mm yr⁻¹ for the default sensitivity of transpiration to CO₂ implemented in the UVic ESCM. Similar to the UVic ESCM, the CMIP5 models show consistent terrestrial precipitation trends in the extra-tropical latitudes, with a slightly more positive trend compared to the default UVic ESCM simulation. In the tropics, however, the different CMIP5 model results show an even wider range from −290 to 450 mm yr⁻¹. This range of terrestrial tropical precipitation changes in the CMIP5 global warming simulations indicates the large uncertainty associated with this climate variable. It is remarkable that the range within the three scaled UVic ESCM model simulations covers almost the complete range of future tropical precipitation changes simulated by the CMIP5 models.

Zoom In Zoom Out Reset image size

For the global mean terrestrial precipitation changes (figure 3(b)) the UVic ESCM covers the complete range of the CMIP5 models by implementing different CO₂-sensitivities of transpiration. Simulated changes in terrestrial precipitation range from −55 mm yr⁻¹ for the default simulation to +128 mm yr⁻¹ when the plants' transpiration is insensitive to increasing CO₂ concentrations. In comparison all CMIP5 models show a positive trend in simulated terrestrial precipitation with a mean annual increase from 15 to 105 mm yr⁻¹. The UVic ESCM's terrestrial precipitation trends seem to be more consistent with the CMIP5 models for the model configuration with applied low CO₂-sensitivities of transpiration.

Latent heat flux is the flux of heat from the Earth's surface to the atmosphere, that is associated with evapotranspiration of water at the surface. An increase in transpiration, therefore would cool the land surface. Globally the land surface temperature is reduced by 0.3 K in 2100 in case of a higher transpiration in the ${f}_{\mathrm{sens}}=0.0$ simulation relative to the ${f}_{\mathrm{sens}}=1.0$ simulation. This amounts to local cooling of the soil temperature in the tropical regions of up to 1.3 K (figure B1).

The scaling of the CO₂-sensitivity of transpiration also influences the simulated vegetation and the corresponding carbon fluxes. The carbon response is driven by the photosynthetic response and the transpiration response. As plants become more water stressed, due to a higher rate of transpiration, they open their stomata less often. This affects how much carbon can be taken up. More carbon is taken up in simulations with less transpiration, because stomata can stay open longer, allowing for more photosynthesis for the same amount of water loss. This increases the plant's WUE (figure 4(a), calculation of the WUE in supplementary material A). In our analysis there is an increase of the WUE for all simulations, with the strongest increase evident for the default simulation, in which transpiration is strongly reduced and the NPP has the largest increase. However, as expected, the increase in the WUE for the ${f}_{\mathrm{sens}}=0.0$ simulation is much lower compared to the default model and is mainly driven by increased carbon uptake, rather than by reduced transpiration.

Zoom In Zoom Out Reset image size

A lower WUE, as for the ${f}_{\mathrm{sens}}=0.0$ simulation, makes it more likely for the plant to become water stressed. The reduced evapotranspiration also influences the terrestrial water availability, the residual of precipitation and evapotranspiration on land, P–E, (figure 4(b)). Under a high emission scenario, the default UVic ESCM simulates a decrease in terrestrial evapotranspiration of 8.7% ${{\rm{K}}}^{-1}$, which is higher than the rate of decrease in precipitation over land, resulting in an overall increase in the terrestrial water availability. For all scaling factors the global average terrestrial water availability still increases with time. However for the simulation, in which transpiration is insensitive to CO₂, P–E is reduced by 20% relative to the default simulation in 2100. This reduction in terrestrial water availability relative to the default simulation leads to a relatively higher plants' water stress level for the ${f}_{\mathrm{sens}}=0.0$ simulation. Correspondingly, the expected future increase in terrestrial net primary productivity, and hence total vegetation carbon, is reduced if transpiration is less sensitive to CO₂ (figure 4(c)). The total terrestrial carbon pool consists of vegetation and soil carbon. An increase in net biological primary production causes an increase in the terrestrial pool, while an increase in soil respiration causes a decrease in terrestrial carbon storage. On longer time scales, soil respiration mainly depends on the soil carbon pool, which is decreased by the scaling relative to the default simulation.

The resulting carbon fluxes between atmosphere and land show that the terrestrial system remains a sink for atmospheric carbon until 2100 for all applied scaling factors (figure 4(d)). The simulated transition from a land carbon sink to a source, happens earlier for lower CO₂ sensitivities than for higher ones. Simulated future carbon uptake by the terrestrial system varies from 1.68 to 3.44 Pg C yr⁻¹ over this century, depending on the applied CO₂-sensitivities, with a reduced uptake for lower CO₂-sensitivities of transpiration.