Journal of Geophysical Research: Atmospheres

Can the stomatal response to higher atmospheric carbon dioxide explain the unusual temperatures during the 2002 Murray-Darling Basin drought?

Authors


Abstract

[1] In 2002, the Murray-Darling Basin (MDB) in Australia experienced a severe drought, characterized by unusually high maximum surface air temperatures. Increased atmospheric carbon dioxide (CO2) affects radiative forcing, but it also directly affects vegetation by increasing its water use efficiency (WUE). The potential contribution made by the physiological response of vegetation to the anomalously warm conditions over the basin is assessed using a coupled atmosphere-land surface model. CO2 concentrations near the leaf surface are increased from preindustrial to present day and near-future levels, without changes to the radiative forcing. Results showed that the increased WUE due to higher CO2 levels led to changes in the maximum surface air temperature and latent heat flux over the basin. The mean increase in maximum surface temperature over the whole MDB in May–October 2002 was not statistically significant, but the warming and the decreases in the latent heat flux were statistically significant at a 95% confidence level over the forested regions. The impact of elevated CO2 on the changes in maximum surface temperature may be incorrectly estimated if the physiological response of vegetation is not taken into account in future climate projections for the MDB.

1. Introduction

[2] The Murray-Darling Basin (MDB) is a Global Energy and Water Experiment (GEWEX) Regional Hydroclimate Project. It is the key agricultural region in Australia, providing about 40% of the nation's agricultural product [CSIRO, 2008]. Situated in southeast Australia, the basin also supports a diverse ecosystem. The climate over this region is highly variable, where it is subtropical in the northeast, cool and humid in the eastern uplands, temperate over the southeast, and hot, dry semiarid and arid in the far west [CSIRO, 2008]. There is a strong northwest to southeast temperature gradient and a strong east-west rainfall gradient. The northern area of the basin is dominated by summer rainfall, whereas more rainfall occurs in the south in winter [CSIRO, 2008]. Significant efforts have been undertaken to understand the climate and the observed changes in the climate of this region [Murphy and Timbal, 2007; Timbal and Murphy, 2007; Maxino et al., 2008]. This focus arises partly from the significant economic cost to the nation of changes over the MDB [Adams et al., 2002], as well as the likely vulnerability of the MDB to future climate change [Pitman and Perkins, 2008].

[3] In 2002, a severe drought occurred in southeast Australia with a significant impact over the MDB, associated with an El Niño event [Jones, 2002] and the absence of a negative Indian Ocean Dipole event [Ummenhofer et al., 2009]. This drought, defined as the difference between precipitation and evaporation, was more severe than the previous drought years, such as 1982 and 1994, which had comparably low rainfall [Nicholls, 2004]. The enhanced drought severity was attributed to higher maximum temperatures which have increased evaporation rates and soil moisture loss over the MDB [Karoly et al., 2003; Nicholls, 2004]. The significantly high temperatures in 2002 may not have been solely due to nonanthropogenic causes because of other factors which have contributed to this event, such as the recent overall warming trend in Australia because of the enhanced greenhouse effect [Karoly et al., 2003; Nicholls, 2004] and land use change [Narisma and Pitman, 2003; McAlpine et al., 2007; Deo et al., 2009].

[4] Another potential contributor to the high surface temperatures is the stomatal response of vegetation to the increased levels of atmospheric CO2. Observational studies have shown a direct impact of CO2 on plant physiology [e.g., Field et al., 1995; Ainsworth and Long, 2005], which is also influenced by factors such as the plant's photosynthetic pathway, environmental stresses and resource availability, e.g., light, water, and nutrients. In a higher CO2 environment, the general short-term response of stomates, which regulate the fluxes of water vapor and CO2 at the leaf surface, is to reduce their opening since they are able to take up CO2 more efficiently. By transpiring less, plants increase their water use efficiency (WUE). Changes in the surface energy and water balances result from the reduced transpiration, which lowers evaporative cooling and increases air temperature. Rainfall can also be affected by the decrease in water vapor flux from the surface, although the scale of these changes depends in part on the coupling strength between the land and atmosphere [Koster et al., 2004]. On the other hand, reduced transpiration can also increase soil moisture and affect the evapotranspiration and temperatures in subsequent periods. Increased temperatures can enhance water vapor demand that may induce slightly higher evapotranspiration rates and consequently lower surface temperature and reduce soil moisture. Knowing which of these responses will be dominant is important. It is most likely that the responses will vary according to season, the nature of the regional climate, vegetation type and soil type.

[5] Model simulations have demonstrated that transpiration is suppressed in response to a doubling of the stomatal resistance, resulting in a global mean increase in temperature with impacts on rainfall and soil moisture [Pollard and Thompson, 1995; Henderson-Sellers et al., 1995; Martin et al., 1999]. The physiological forcing can also offset the enhanced evapotranspiration that would have resulted if only the radiative effect of CO2 was considered [Bounoua et al., 1999; Levis et al., 2000]. Earlier work by Idso and Brazel [1984] and Aston [1984] noted that the stomatal response to the elevated CO2 can increase streamflow. There is an ongoing discussion on the impact on the terrestrial water balance, particularly on runoff [Martin et al., 1999; Gedney et al., 2006; Piao et al., 2007; Betts et al., 2007; Huntington, 2008]. Biospheric feedbacks can also affect the impact of land cover change on temperature and the latent heat flux (LHF) [Narisma et al., 2003; Narisma and Pitman, 2004].

[6] Given the recent literature that shows that the physiological response of vegetation can affect the surface climate [Cruz et al., 2008; Boucher et al., 2009; Cao et al., 2009], the question arises whether this response can explain part of the anomalously high observed maximum temperatures in the MDB in 2002. We focus on the impact of changes in the CO2 only at the leaf surface, without the additional contribution from the changes in radiative forcing. Our study is therefore idealized to determine whether the impact of increasing CO2 on stomates could partly explain the anomalously hot temperatures over the MDB in 2002. We explore the impact of increases in CO2 relative to the preindustrial level both at present day and in the near future. We do not include structural feedbacks (e.g., changes in roughness, leaf area index (LAI), root depth, etc.) as we are exploring the impact of a single year and because there is no conclusive evidence of a large-scale structural response of the vegetation to elevated CO2 over the MDB. The numerical experiments conducted using a coupled atmosphere-land surface model are discussed in section 2. Our results are shown in section 3. The discussion is presented in section 4, followed by conclusions in section 5.

2. Methodology

2.1. Global Climate Model and Land Surface Model

[7] The Conformal-Cubic Atmosphere Model (CCAM) is a variable resolution global climate model developed at CSIRO [McGregor and Dix, 2001, 2008]. This model uses a conformal cubic grid derived from projecting the panels of a cube onto the surface of the Earth. It has a stretched grid configuration facilitating regional climate modeling [McGregor et al., 2002; Nunez and McGregor, 2007; Lal et al., 2008].

[8] CCAM is a two time-level semi-implicit hydrostatic model. Its model dynamics include a semi-Lagrangian horizontal advection with bicubic horizontal interpolation, total variation diminishing (TVD) vertical advection, an unstaggered grid and reversibly staggering scheme for winds [McGregor, 1993, 1996; McGregor and Dix, 2001; McGregor et al., 2002]. Physical parameterization schemes used in the model include Geophysical Fluid Dynamics Laboratory parameterization for shortwave and longwave radiation [Schwarzkopf and Fels, 1991], a stability-dependent planetary boundary layer scheme with nonlocal vertical mixing [Holtslag and Boville, 1993], gravity wave drag and McGregor [2003] cumulus convection.

[9] The terrestrial surface in CCAM is represented using the Community Atmosphere Biosphere Land Exchange (CABLE) land surface scheme [Kowalczyk et al., 2006]. CABLE includes a representation of the canopy, soil and snow, carbon pool dynamics, and soil respiration [Kowalczyk et al., 2006]. This land surface model has been extensively tested using traditional and innovative measures of model performance [Abramowitz, 2005; Wang et al., 2007; Abramowitz et al., 2008].

[10] CABLE uses a one-layer, two-big-leaf canopy model to solve the photosynthesis, stomatal conductance, leaf temperature, energy, and CO2 fluxes for sunlit and shaded leaves [Wang and Leuning, 1998]. A coupled model of the photosynthesis and stomatal conductance is used, where stomates respond to changes in the atmospheric and soil environments [Leuning, 1995; Wang and Leuning, 1998]. A radiation scheme is used to determine the amount of photosynthetically active radiation, near-infrared and thermal radiation absorbed by the sunlit and shaded leaves [Wang and Leuning, 1998; Kowalczyk et al., 2006]. Vegetation is represented above ground which allows full radiative and aerodynamic interaction among atmosphere, vegetation, and ground [Kowalczyk et al., 2006]. Air temperature and humidity within the canopy are determined using a plant turbulence model developed by Raupach et al. [1997]. The soil moisture and temperature for six layers are solved using Richard's equation and the heat conduction equation in the soil module [Kowalczyk et al., 2006].

2.2. Design of the Numerical Experiment

[11] The stretched conformal cubic grid of CCAM has been configured to provide a fine resolution of ∼50 km over southeast Australia where the MDB is situated, with coarse resolution elsewhere. Land surface data, such as vegetation type, LAI, soil albedo, and soil texture, are obtained from the International Satellite Land-Surface Climatology Project, Initiative II (ISLSCP II) data archive [Hall et al., 2005] and are reclassified then interpolated to the model grid. The climate model is forced with observed midmonthly mean sea surface temperature and sea-ice concentration, obtained from the second phase of the Atmospheric Model Intercomparison Project (available at http://www-pcmdi.llnl.gov/projects/amip/AMIP2EXPDSN/BCS/amipobs_dwnld.php) and processed following Taylor et al. [2000].

[12] The objective of this study is to determine the contribution of the physiological forcing of CO2 to the high maximum temperatures over the MDB in southeastern Australia in 2002. The coupled CCAM-CABLE model was run for 14 months (January 2002 to February 2003), with the last 12 months used for analysis. The simulation of the annual cycle allows an examination of the seasonal variation of the changes in the surface climate that result from the stomatal response to the increasing CO2. However, the longer simulation period and the limitations in computational resources constrain the number of realizations performed and CO2 levels examined.

[13] Parallel simulations have been conducted where only the CO2 concentration at the leaf level is changed. In this paper, we define the leaf level CO2 as the atmospheric concentration near the leaf surface used to calculate gradients within the leaf. Three concentrations of leaf level CO2 are considered: 280 (preindustrial), 375 (present day), and 500 ppmv (near future). A level of 500 ppmv is reached by about 2040–2060 under a range of emission scenarios (from B1 to A1F1) used by the Intergovernmental Panel on Climate Change [Meehl et al., 2007]. The atmospheric CO2 concentration is fixed at 375 ppmv, such that the changes in leaf level CO2 do not directly affect the atmosphere or net radiation at the surface.

[14] A total of 22 realizations have been conducted for each of the three CO2 levels: 280, 375, and 500 ppmv, where perturbations to the initial deep soil temperature and initial soil moisture are done. Eleven realizations are performed where the initial global soil temperature within the deepest soil layer (4.6 m below the surface) was perturbed from −0.025°C to 0.025°C in increments of 0.005°C. The perturbation to the deep soil temperature is mirrored about zero such that it does not bias the resulting changes in the surface climate. At each model time step (i.e., 20 min), this perturbation within the deep soil layer is felt by the upper layers to some degree because of the communication among layers as the soil model solves the heat conduction equation. Thus, these small perturbations affect the surface temperature enough to generate an independent realization. Over southeastern Australia, the initial soil moisture was less than the wilting point in some areas during the austral summer of 2001. Therefore, a second set of experiments was performed where in addition to the perturbation to the deep soil temperature, the initial soil moisture in all soil layers was increased by half the difference between the current soil moisture and the field capacity over southeastern Australia only. In our analyses, the soil moisture experiments are combined since simulations with low soil moisture responded similarly to those with slightly wetter soils to the increase in leaf level CO2.

[15] The averages of the monthly mean daily maximum temperature, LHF, and rainfall over all realizations are derived for each season. The difference in the ensemble monthly means from the 280 ppmv simulation (control run) is obtained from the simulations for the higher CO2 levels. A two-sample F test and t test had been conducted to determine the statistical significance of the mean differences. The F test determines if the two independent samples come from normal distributions with equal or unequal variance and determines the type of standard t test used. Both tests are two tailed using 90% and 95% significance levels.

[16] Probability density functions (pdfs) have also been constructed for the maximum temperature and LHF, corresponding to each level of CO2. Each pdf consists of daily values within the specified time period (i.e., May to October), averaged over the 22 realizations from all grid points which have the same vegetation type within the vicinity of the MDB, defined at 24°S–40°S and 138°E–153°E (Figure 1). A two-sample Kolmogorov-Smirnov test has also been conducted on the populations used in constructing the pdfs to determine whether or not the values come from the same continuous distribution. The pdfs enable us to examine the impact of the increasing CO2 on these variables on both the means and extremes.

Figure 1.

Map of vegetation types found over the Murray-Darling Basin in southeast Australia.

3. Results

[17] On the basis of the parameterization in CABLE, and off-line analyses [Cruz et al., 2008] changes in the surface climate can result from the stomatal response to the elevated leaf level CO2. The impact of increasing the leaf level CO2 from 280 to 375 ppmv on the ensemble seasonal mean difference maps of daily maximum surface air temperature, LHF, and rainfall over the MDB are shown in Figure 2. Differences which are statistically significant at a confidence level of 90% and 95% are noted in Figures 24.

Figure 2.

Ensemble mean seasonal differences in maximum surface air temperature (°C), latent heat flux (W m−2), and rainfall rate (mm d−1) between leaf level CO2 of 375 and 280 ppmv for (a–c) March to May, (d–f) June to August, (g–i) September to November, and (j–l) December to February. Averages are taken over 22 realizations for each season. Changes that are statistically significant at a 95% and 90% confidence levels are marked with “1” and “2”, respectively. Negative values are outlined by dashed lines.

[18] A warming of 0.1°C–0.5°C occurs over much of the forests along the east coast and a region in the southwest of the domain in autumn (March to May (MAM)), which is only occasionally statistically significant at the 90% confidence level (Figure 2a). There is a cooling of 0.1°C–0.5°C over the northern part of the domain, but this is not statistically significant. Over the forests, the warming is consistent with the changes in the LHF (Figure 2b) where there is a widespread reduction (∼5 W m−2) caused by the decreased stomatal conductance in a higher CO2 environment. There is a consistency between the changes in the LHF and temperature over most of the domain, where the changes in the LHF along the east coast and southwest are statistically significant at the 90% and 95% levels. The CO2 physiological forcing may also locally affect rainfall, partly because of the changes in the LHF. The simulated changes in rainfall are significant at the 90% level (Figure 2c), but there is little spatial coherence between the change in the LHF and rainfall. The low levels of statistical significance also do not provide convincing evidence of an impact on rainfall.

[19] In winter (June to August (JJA)), the warming is more widespread but not significant (Figure 2d). Again, there are large areas of reduced LHF over the forests (Figure 2e) but with little coherence with reduced rainfall (Figure 2f). The reduction in the LHF is statistically significant over most of the forests along the east coast at the 95% level (Figure 2e). The reduced LHF and locally increased rainfall in MAM (Figures 2b and 2c) act to increase the available moisture at the center of the domain leading to a higher LHF in JJA, but this is rarely significant at the 90% level (Figure 2e).

[20] Higher maximum temperatures are simulated in spring (September to November (SON)), where the strongest warming (0.2°C–0.5°C) occurs over the forests along the east coast, commonly significant at the 90% level, and over areas of crops in the south of the domain, locally statistically significant at the 95% level (Figure 2g). Large reductions in the LHF (1–5 W m−2) occur over coincident areas, which are statistically significant at the 95% level (Figure 2h). The changes in rainfall (Figure 2i) are locally significant at the 90% confidence level.

[21] A clear contrast in the change in the maximum temperature between the forested area and nonforested area is simulated in summer (December to February (DJF)). A large region of cooling (up to 1°C) is simulated at the center of the domain, and a warming of a similar magnitude is simulated in the north (Figure 2j); but both are not statistically significant. There is also a warming in the south, which is significant at the 90% level and locally at the 95% level. The lower maximum temperatures in the center of the domain are influenced by both the statistically significant (90% confidence level) increase in LHF (Figure 2k) and rainfall (Figure 2l) in the central regions. The increased LHF at the center of the domain is due to the slightly enhanced rainfall from the previous season, as well as changes in LHF simulated in JJA and SON, which increases the available moisture in summer. The forested areas experience large (5–10 W m−2) reductions in the LHF (Figure 2k) which is statistically significant (95% level) and associated with statistically significant warming in areas of the far south of the domain (Figure 2j). In these regions the change in rainfall (Figure 2l) is too small to offset the impact of higher leaf level CO2 on LHF.

[22] Figure 3 shows the changes in the surface climate when the leaf level CO2 is raised from 280 to 500 ppmv. Comparison with Figure 2 shows that the patterns of temperature and LHF change are similar, but the magnitude and extent have increased. While there are differences in the rainfall response, there are also intriguing similarities which will be discussed in section 4.

Figure 3.

As in Figure 2, but the difference is between 500 and 280 ppmv.

[23] Figure 3a shows warmer temperatures (0.1°C–0.5°C) during MAM along the east coast, coincident with the forests and statistically significant at the 95% level. The warming is associated with similarly significant decreases in the LHF (Figure 3b). Over some central and northern regions, the maximum temperatures are lower but not significant. There is a statistically significant increase in coastal rainfall in the northern half and a decrease in the southern half of the domain (Figure 3c). In JJA, the warming is more widespread where it is not only statistically significant over the forests (commonly at a 95% level) but also over the northern part of the domain (mainly at the 90% levels). There is also a pattern of reduced rainfall along the east coast (Figure 3f) which is locally statistically significant and coincident with the significantly reduced LHF (Figure 3e).

[24] The spring (SON) impact of increased leaf level CO2 is 0.5°C–1°C over the east, south and southeastern regions, mainly at a 95% significance level (Figure 3g). Statistically significant reductions in the LHF are widespread over the forests and crops, reaching 20 W m−2 in some regions (Figure 3h). This is associated with small declines in rainfall over the southwest of the domain, significant at the 95% level (Figure 3i). Small significant increases are also found elsewhere along the east coast (Figure 3i).

[25] Figure 3j shows that the southern region and the east coast are still statistically significantly warm in summer (DJF) but a larger region of cooling occurs in the northern and central areas that is only locally significant at the 90% level. The warm areas occur in coincident areas where significant reductions of 10–20 W m−2 in the LHF are widespread (Figure 3k). There is also a small area of reduced rainfall over the southern region that is statistically significant at the 90% level (Figure 3l). The region of cooling, which locally exceeds 1°C (Figure 3j), is associated with increases in the LHF (Figure 3k) and is likely linked to increased rainfall (Figure 3l) in the center of the domain. It may also be partly related to the moisture saved due to the increased WUE of plants over preceding seasons.

[26] The ensemble seasonal mean difference maps have shown contrasting responses over different sections of the domain, which vary in each season. However, there appears to be a consistent warming along the eastern and southeastern coasts, coincident with reduced LHF (Figures 2 and 3) and coincident with the distribution of forests. The seasonally invariant response over this area shows the strong physiological response to the elevated CO2 over forests, compared to grass and shrubs which cover most of the region (Figure 1). Changes in the local circulation are minimal (not shown) which suggest these do not explain the behavior over forests. It is more difficult to identify the impact of the stomatal response over the short, shallow-rooted vegetation such as grass and shrubs, where the energy and water exchanges at the surface are more influenced by the seasonal variability.

[27] Nicholls [2004] considered the impact of the drought on the regional climate of the MDB during May to October (MJJASO), which is the growing season for many crops and pasture over the basin. Figure 4 shows the impact of the increase in the leaf level CO2 over this period. In Figure 4a, the increase in the leaf level CO2 to 375 ppmv apparently warms the regions of forests and regions of crops and shrubs by 0.1°C–0.2°C, which is commonly significant at the 90% level and locally at 95% level. Similarly, the increase in leaf level CO2 reduces the LHF significantly (95% level) over the forest along the east coast (Figure 4b). The impact on rainfall is negligible and not statistically significant. At 500 ppmv the statistically significant warming becomes substantially more widespread, particularly over the areas of forest and crop (Figure 4d) and is coherent with the widespread reduction in the LHF (Figure 4e). Some areas indicate rainfall changes significant at the 90% level, and there are hints of small reductions in rainfall along the east coast (Figure 4f).

Figure 4.

Ensemble mean seasonal differences in maximum surface air temperature (°C), latent heat flux (W m−2), and rainfall rate (mm d−1) (a–c) between leaf level CO2 of 375 and 280 ppmv and (d–f) between 500 and 280 ppmv for May to October. Averages are taken over 22 realizations for this season. Changes that are statistically significant at a 95% and 90% confidence levels are marked with “1” and “2”, respectively. Negative values are outlined by dashed lines.

[28] To examine the variation of the physiological impact with vegetation type, pdfs were constructed from the ensemble mean daily values of LHF and maximum surface air temperature corresponding to each level of CO2 in MJJASO. For all vegetation types, the values of maximum temperature at 375 ppmv and at 500 ppmv come from distributions different from the values at 280 ppmv. This difference is statistically significant at a 95% confidence level based on a two-sample Kolmogorov-Smirnov test. The same is true for LHF, except for LHF values at 375 ppmv over crop, which are not significantly different from the values at 280 ppmv. Figures 5 and 6 show the overall pdf and the upper tail of each pdf for LHF and maximum surface air temperature. Over the forested areas, there is a clear shift in the upper tail of the pdf of the LHF toward lower values as the leaf level CO2 is increased (Figures 5a and 5b). This is consistent with the shift in the pdf of temperature toward higher values (Figures 6a and 6b). The impact of the physiological response on the LHF and on maximum temperature is very similar for crops, where the shifts to lower LHF (Figures 5c and 5d) and to higher maximum temperature (Figures 6c and 6d) are clear. Over grass, the results are very much less clear in both the LHF (Figures 5e and 5f) and maximum temperature (Figures 6e and 6f).

Figure 5.

Probability density function and its upper tail of latent heat flux (W m−2) for each CO2 (280, 375, and 500 ppmv) for (a–b) forest, (c–d) crop, and (e–f) grass for May to October. Each pdf consists of ensemble mean daily values within the time period from all grid points having the same vegetation type within the region. The minimum value on the x axis of the upper tail of the pdf is determined from the 95th percentile value of latent heat flux at the 280 ppmv level.

Figure 6.

As in Figure 5 but for daily maximum surface air temperature (°C).

4. Discussion

[29] Using a coupled atmosphere-land climate model, simulations have been conducted to determine the role of the physiological forcing of CO2 in explaining the anomalously high temperatures observed during the 2002 drought over the MDB. Only the leaf level CO2 was increased (from 280 to 375 ppmv and from 280 to 500 ppmv) without changes in radiative forcing. This isolates the impact of the increasing atmospheric CO2 on just the physiological response.

[30] The simulated changes in temperature and LHF are in broad agreement with results from many earlier doubled CO2 [Sellers et al., 1996; Bounoua et al., 1999; Levis et al., 2000; Betts et al., 2007] or doubled stomatal resistance model experiments [Pollard and Thompson, 1995; Henderson-Sellers et al., 1995; Martin et al., 1999]. This was anticipated [Cruz et al., 2008]. Changes in the mean and in the upper tails of the pdfs were also identified indicating particular sensitivity at the upper tails of both maximum temperatures and LHF.

[31] Our results showed variability in the changes depending on season and vegetation type. A stronger and more significant impact of the stomatal response occurs in the austral spring to summer, when vegetation is most active, in agreement with previous studies [Henderson-Sellers et al., 1995; Bounoua et al., 1999; Martin et al., 1999]. Seasonally varying changes in LHF are simulated over most of the domain, consisting of grasses and shrubs, which may be linked with the strong influence of soil moisture variability. Since changes in the vegetation structure have not been considered, the increased WUE of plants and its indirect impact on rainfall tend to increase the soil moisture in the following season. This enhances the capacity to sustain evaporation in a subsequent season during high evaporative demand and therefore causes cooler temperatures. On the other hand, higher maximum temperatures are simulated over the forest and to a lesser degree, crops, throughout the year, resulting from the significant reductions in LHF. The strong physiological effect of CO2 over forests may be associated with the high LAI, which is among the physiological and environmental factors influencing transpiration rates. Deep roots may have also contributed to a reduction in the influence of soil moisture variability on the CO2 impact on the WUE of trees.

[32] The purpose of this paper is to determine whether the increased WUE due to increasing atmospheric CO2 can partly explain the anomalously high temperatures observed over the MDB, which was said to have enhanced the drought in 2002. We note that this physiological effect alone cannot explain the drier conditions over the basin, at least not during the same season. As discussed in section 1, the first effect of increased WUE is to reduce transpiration which lowers the latent cooling and increases temperature. This warming can consequently slightly increase evapotranspiration, but this secondary increase will always be smaller than the original decrease.

[33] Figure 2 highlights large-scale decreases in LHF over the forests throughout the year, and statistically significant warming in some regions in some seasons. However, Nicholls [2004] focused on May to October as illustrative of the drought. Figure 4 reproduces this period from the model. The increase in CO2 from 280 to 375 ppmv resulted in a simulated 0.08°C mean increase in maximum temperature over the basin (Figure 4a), resulting from reduced LHF (Figure 4b). However, an examination of the warming over areas with a common vegetation type indicates a higher mean warming of 0.14°C over the forest along the east coast and 0.13°C over crops in the south, which are both commonly significant at the 90% level and locally at 95% level (Figure 4a). There is also a clear potential for increased leaf level CO2 to affect this region at higher CO2 levels (Figures 4d and 4e). The warming over grass and shrubs is lower at 0.06°C and 0.05°C, respectively. Both are not statistically significant, but they lower the mean temperature over the basin due to their larger spatial coverage.

[34] There are a suite of limitations to our work. The first uncertainty is the robustness of the simulated changes as CO2 is increased from 280 ppmv. Given the perturbation is at leaf level and is not directly affecting the full atmosphere, sea surface temperatures, sea ice, etc., in ways an atmospheric perturbation would, it is reasonable to expect some similarity in the response to the perturbation from 280 to 375 ppmv compared with 280 to 500 ppmv. Comparing Figures 2 and 3 points to some similarity: it is by no means a linear response and there are clear regional differences. However, a comparison of Figure 2b to Figure 3b, or Figures 2e and 3e, or Figures 2k and 3k points to clear similarities in the patterns of change in the LHF. These extend clearly to similarities in the pattern of the temperature response. The changes in the rainfall are not as similar, but this is to be expected as this is a less deterministic field. However, over the forests there are similarities (Figure 2c compared to Figure 3c or Figure 2f compared to Figure 3f). Overall, the coherence of the results for the two CO2 levels and given the large number of realizations undertaken give confidence that the model response is consistent between CO2 levels. The simulated changes may be small (e.g., about a tenth of a degree warming) but statistically robust, which also indicates that the response is driven by the physiological response of vegetation and not a result of internal model variability.

[35] The second uncertainty relates to the land surface model and the parameterization of the response of the stomates to increased CO2. The parameterization in CABLE is based on the work of Leuning [1995] and Wang and Leuning [1998] and is well suited to Australian ecosystems. Field data were used in the development of this model and innovative evaluation [Abramowitz et al., 2008] has provided confidence in the model's performance. Additional field experiments that explore this response over Australian native ecosystems to elevated CO2 would add further confidence in the results, but we are not aware of serious limitations in CABLE on the timescales considered in this paper. For example, it is unlikely that the structural response to elevated CO2 which has an opposing effect to the physiological response [Betts et al., 1997; Levis et al., 2000; Betts et al., 2007; Calvet et al., 2008] could entirely compensate for the increased WUE of plants in 2002.

[36] A further issue is the form of the relationship used in CABLE to associate the leaf level CO2 to stomatal conductance. As CO2 increases, the stomates reduce their opening; but, is the form of this relationship right and is the amount of closure correct for a given CO2 concentration? It is possible that CABLE is overly sensitive to leaf level CO2, but it is also possible that it lacks sensitivity and that the responses shown in Figure 3 (for 500 ppmv) more accurately reflect the ecosystem response. Again, this will require the integration of recent elevated CO2 experimental data into CABLE to resolve.

[37] There are certainly other caveats to our results relating to the surface parameterization. The coupling of the soil moisture and stomatal response on seasonal timescales is not well understood in water-limited environments. We would not suggest we know the root depths, water table supply, soil characteristics, or nutrient relationships to the degree that we would defend the precise warming induced by the physiological response. We do not know whether the atmospheric model and the land model are strongly or weakly coupled [Koster et al., 2004] which can affect the impact of surface phenomena on the atmosphere in some regions [Seneviratne et al., 2006]. The version of CABLE as used in this study does not explicitly represent the phenology of crops or irrigation and this casts doubt on the reliability of the response of the models over cropped regions. CABLE is also unable to capture the response of forest canopy LAI to drought, such as the shedding of leaves in native Australian trees during drought, which suggests that the simulated warming due to the reduced LHF over the forests may be overestimated.

[38] Despite these limitations and uncertainty some interesting results appear to be robust over a large sample of realizations. According to Nicholls [2004], the observed mean maximum temperature anomaly for May to October over the MDB was 1.98°C relative to the 1961–1990 climate. It is difficult to make a comparison of our results with this observed warming since our numerical experiments are designed to consider only a time slice representing the impact of the change in CO2 from preindustrial to near-present concentrations and to examine the climate model's sensitivity to the changes in leaf level CO2. Clearly, the anomalous warming in 2002 cannot be attributed to just the increased WUE of plants since drought-affected years in the 1990s were subject to very similar CO2 concentrations and were not anomalously warm to the same extent. We are not suggesting some threshold was crossed such that the stomatal response suddenly became important in 2002. Clearly, 2002 was anomalously hot for a series of reasons as highlighted by Nicholls [2004]. Our results, shown in Figure 4a, suggest that the increased WUE of vegetation has the capacity to contribute a few tenths of a degree of warming, particularly focused on the forests of eastern and southeastern Australia. Figure 4d shows that there is a potential under higher levels of CO2 for this vegetation response to contribute up to 1°C. This is not comparable with the impact of increasing atmospheric CO2 because it does not affect all regions significantly, it is confined to some seasons, and it is not as large an impact as the increase in atmospheric CO2. However, it does appear, over southeastern Australia, to be an additive effect and may therefore explain a portion of the anomalous warming experienced in 2002 in this region.

5. Conclusions

[39] Simulations have been conducted for investigating the role of the physiological response to increases in leaf level CO2 on the high maximum temperatures observed in 2002, which exacerbated the drought experienced over the MDB. Our results have shown that reduced LHF due to the increased WUE of plants led to higher maximum surface temperatures over the MDB. Over the forested regions, a strong response was seen, where higher maximum temperatures due to CO2-induced suppression of transpiration are simulated throughout the year.

[40] Over the basin during May to October, the impact of the physiological response of vegetation is statistically significant at the levels of CO2 experienced to date on maximum temperature and LHF, particularly over the forest. The magnitude of these changes suggests that projections of temperature over the basin may be incorrectly estimated if the physiological response is not taken into account. It is therefore important to incorporate these feedbacks, especially in future climate projections for the MDB.

[41] The caveats and uncertainties noted earlier need to be in mind when reaching a final conclusion. Does the stomatal response explain the anomalously high maximum temperatures over the MDB in 2002? We have little confidence in the response over the crops due to CABLE not including explicit phenology and irrigation. Further, the response over the grasses is likely particularly sensitive to soil-water-plant interactions, and we have relatively little confidence in this aspect of the model. However, the response over the forest is more likely robust because the deep roots minimize the sensitivity to near-surface soil moisture variability. CABLE also used forest-based observations in its development. We suspect that the overall warming of 0.08°C due to the increased WUE is an overestimate although the scale of this overestimate is hard to quantify without additional observational data of the vegetation response to elevated CO2. The dependence of the physiological response on the vegetation type may be further investigated by doing a land cover change experiment, for example, by replacing the forests along the east coast with crops. This can be explored in future work.

[42] In conclusion, we present evidence that the stomatal response to elevated CO2 can partially explain why the MDB was anomalously hot in 2002. The amount of warming is about a tenth of a degree. The impact on the LHF, via the increased WUE of plants, is an important contributor to regional scale hydrometeorology of the MDB. Given that the impact of the physiological response is additive to regional warming caused by the radiative effects of atmospheric CO2 over forests, it is likely that existing estimates from climate models of projected warming over the MDB are underestimates.

Acknowledgments

[43] This research was funded by a grant from the Australian Research Council (DP0662887) and supercomputing supplied by the Australian Partnership for Advanced Computing. The authors thank John McGregor for help with CCAM and Gab Abramowitz and Eva Kowalczyk for help with CABLE. Martin Dix and Kim Nguyen provided the input files for CCAM. The authors also thank our anonymous reviewers for their comments which improved this manuscript. F.T. Cruz is a recipient of the UNSW PhD scholarship.