The role of CO2 and dynamic vegetation on the impact of temperate land use change in the HadCM3 coupled climate model

: Human-induced land-use change (LUC) alters the biogeophysical characteristics of the land surface inﬂuencing the surface energy balance. The level of atmospheric CO 2 is expected to increase in the coming century and beyond, modifying temperature and precipitation patterns and altering the distribution and physiology of natural vegetation. It is important to constrain how CO 2 -induced climate and vegetation change may inﬂuence the regional extent to which LUC alters climate. This sensitivity study uses the HadCM3 coupled climate model under a range of equilibrium forcings to show that the impact of LUC declines under


Introduction
Human-induced land-use change (LUC), such as the conversion of natural land cover to agriculture, transforms the land surface, altering its structure and influencing biogeophysical processes such as albedo, leaf area index (LAI), seasonality, surface roughness, and moisture fluxes.This has implications for the surface energy balance, altering shortwave radiation (SW) and the partitioning of latent and sensible heat (e.g., Brovkin et al. 2009;Bala et al. 2007;Boisier et al. 2012;Pielke et al. 2002).Understanding the climatic impacts of LUC is crucial for improving climate prediction and potential mitigation strategies.
Observed and modeled data show that LUC influences regional and global climate including temperature, precipitation, and humidity (e.g., Bonan 1997;Bonan 2001;Gameda et al. 2007;Ge 2010;Fall et al. 2010;Beltrán-Przekurat et al. 2012;Zhao et al. 2001;Claussen et al. 2001;Bathiany et al. 2010;Chase et al. 2000;Kalnay and Cai 2003;Nun ˜ez et al. 2008;Ezber et al. 2007;Zhao and Jackson 2014;Kumar et al. 2013;Pielke et al. 2007).Studies have shown that the regional response depends on the latitude at which it occurs, which may either enhance or attenuate CO 2 -induced warming.Tropical LUC in the form of deforestation acts to warm the climate by suppressing evapotranspiration (Claussen et al. 2001;Snyder et al. 2004;Bathiany et al. 2010), whereas temperate and boreal deforestation is dominated by an increase in albedo and snow persisting longer into the spring, which act to cool the climate (Claussen et al. 2001;Snyder et al. 2004;Bathiany et al. 2010;Davin et al. 2007;Zhao and Jackson 2014;Lee et al. 2011).The opposing temperate and tropical impact of LUC may drive a small global effect; however, its regional-scale impact is likely to be significantly larger (Lawrence and Chase 2010;Brovkin et al. 2013;de Noblet-Ducoudre et al. 2012).
Despite the large number of studies, there remain disparities in modeling results as to the magnitude and sometimes direction of regional and global effects (Pitman et al. 2009;de Noblet-Ducoudre et al. 2012;Brovkin et al. 2013).These inconsistencies have been attributed to differences in how models parameterize albedo, how LUC is implemented, and how crop phenology and evapotranspiration is represented (Pitman et al. 2009).
With levels of atmospheric CO 2 expected to increase in the coming century and beyond, it is important to constrain how the biogeophysical impact of LUC may change under higher CO 2 forcing.A previous study by Pitman et al. (2011) showed that the biogeophysical impact of LUC depended on the background state of the climate.They attributed a reduction in the winter impact of temperate LUC at higher concentrations of CO 2 to a reduced snow albedo effect.On the contrary, summertime impacts are shown to increase due to CO 2 -induced increase in precipitation and latent cooling.
Higher CO 2 concentrations are expected to not only alter the background climate, including temperature and precipitation patterns, but also the distribution and physiology (i.e., stomatal conductance and LAI) of vegetation.We refer to this as land-cover change (LCC), which differs from LUC in that it is a shift from one natural vegetation type to another (e.g., needleleaf to broadleaf vegetation).Increasing CO 2 is expected to drive a northward migration of the tree line and a shift from one dominant vegetation type to another (e.g., O'ishi et al. 2009;Scholze et al. 2006).As with LUC, LCC has implications for the initial biogeophysical characteristics of the land surface including the albedo parameters, roughness length, canopy height, LAI, and rooting depth.This in turn has implications for evapotranspiration and soil moisture availability that can influence climate (e.g., Davies-Barnard et al. 2015;Niyogi and Xue 2006).Stomatal (or canopy) conductance has been shown in laboratory (Field et al. 1995;Brodribb et al. 2009), field (Hungate et al. 2002;Ainsworth and Rogers 2007), and modeling studies (Medlyn et al. 2001;Gopalakrishnan et al. 2011;Collatz et al. 1991;Boucher et al. 2009) to decrease with higher CO 2 concentrations.This effect has been labeled ''CO 2 physiological forcing'' (Betts et al. 2007a) and acts to decrease evapotranspiration, atmospheric water vapor, and latent cooling and increase surface temperatures (Boucher et al. 2009;Niyogi et al. 2002;Cao et al. 2010).
Many of the current generation of coupled climate models include a land surface component or dynamic global vegetation model (DGVM) that dynamically simulates LCC, including vegetation composition and physiology depending on climate (e.g., Arora 2002).In multimodel studies assessing the biogeophysical impact of LUC, such as Land-Use and Climate, Identification of Robust Impacts (LUCID), only three of the seven modeling groups used dynamic vegetation, and they all used a different initial land-cover distribution (Pitman et al. 2009;Brovkin et al. 2013).This was due to the difficulty in integration and calibration of a common land-cover map.Different vegetation distribution and physiology would be expected to influence the initial conditions of the land surface; however, their relative role in amplifying or attenuating the biogeophysical impact of LUC under higher CO 2 concentrations has not been investigated.
This study will investigate the biogeophysical impact of LUC in the Hadley Centre Coupled Model, version 3 (HadCM3), a coupled climate model with dynamic vegetation [Top-down Representation of Interactive Foliage and Flora Including Dynamics (TRIFFID)] and a land surface scheme [Met Office Surface Exchange Scheme (MOSES); see section 2].There are two main aims: 1) to examine the impact of LUC under increasing CO 2 concentrations and 2) to assess the importance of dynamic vegetation and LCC simulated by the model and how this influences the extent to which LUC impacts climate.Section 2 gives a description of the HadCM3 climate model and the simulations used in this study.The results are outlined in section 3 followed by an energy balance analysis in section 4. A discussion and summary is presented in section 5.

Methods
HadCM3 is a coupled Earth system model comprising a 3D dynamical atmosphere and ocean components and includes a thermodynamic/free-drift sea ice model (Gordon et al. 2000).The resolution of the atmospheric component is 3.758 3 2.58 with 19 vertical levels, and the ocean resolution is 1.258 3 1.258 with 20 vertical levels.
The model comprises an interactive dynamic vegetation model (TRIFFID; Cox 2001).TRIFFID simulates as a percentage of each grid box with five plant functional types (PFT)-C3/C4 grasses, shrubs, and broadleaf/needleleaf trees-and four nonvegetation types-bare soil, inland water, urban, and ice.Parameters such as albedo and LAI vary for the five PFTs (Table 1), which in turn influence land surface properties such as the surface energy balance.The PFTs vary according to a number of factors including atmospheric CO 2 concentration, temperature, soil carbon availability, and moisture.Soil carbon is calculated by litterfall and microbial respiration, the latter of which is controlled by temperature and soil moisture.Soil carbon and the configuration of PFTs in each grid box are updated every 10 days based on competition from other plants (i.e., trees top the hierarchy before shrubs and grasses) and the fluxes of carbon, which are calculated by the land surface exchange scheme MOSES, version 2.1 (Essery et al. 2003).The updated vegetation distribution is then reentered into MOSES to update land surface parameters such as albedo and surface roughness.This process maintains a consistent hydrological state between the vegetation and atmosphere (Cox 2001).The radiation scheme used in HadCM3 is that of Edwards and Slingo (1996).
MOSES models the physiological processes of transpiration, respiration, and photosynthesis.This links vegetation to CO 2 concentration and atmospheric conditions and impacts the partitioning of surface water.Higher concentrations of CO 2 act to generally decrease canopy conductance that in turn reduces evapotranspiration rate (Boucher et al. 2009).Conductance is also influenced by temperature, soil moisture, and humidity, with the overall effect scaled by LAI.Overall canopy conductance decreases with humidity and increases with soil moisture (Cox et al. 1999), photosynthetically active radiation (PAR) (Sellers et al. 1992), and LAI.
To simulate LUC and conversion to crop and pastureland, natural vegetation was replaced with C3 and C4 grasses (e.g., wheat/rice and sugarcane/maize, respectively) or bare soil using a mask originally derived from Betts et al. (2007b;Figure 1).The mask represents the global disturbed fraction as of the year 1990 constructed with combined pasture values from Goldewijk (2001) and crop fractions from Ramankutty and Foley (1999).The final LUC fraction for the model grid squares is the area designated as per Figure 1, minus that of the three remaining Table 1.Albedo and LAI values for PFTs simulated by TRIFFID.Albedo values are split into three parameters; sfree is the maximum value for calculating snow-free albedo, and smin and smax are the respective minimum and maximum snow albedos used to calculate the cold deep snow albedo (see Essery et al. 2001).PFT values are based on the International Geosphere-Biosphere Programme (IGBP) classes (Jones 2004).nonvegetation types simulated by TRIFFID-that is, inland water, urban, and ice.The model does not include a harvesting or irrigation scheme.
The model was spun up for 1000 years and run at equilibrium for a further 300 years at four CO 2 concentrations: 350, 700, 1050, and 1400 ppm.For each CO 2 concentration, the model was run with and without LUC to isolate its effect.The long spinup period has resulted in the atmosphere reaching an equilibrium state after approximately 1100 years as shown by stabilized global-mean surface air temperatures (SATs; not shown).The LUC mask (Figure 1) is fixed throughout the run, whereas remaining natural vegetation can change.These simulations will hereon be labeled 13, 23, 33, and 43, respectively, and those with LUC will have the suffix LU.Analysis is performed on the final 50 years of each run, and all anomalies shown are 99% significant as calculated by a Student's t test.
To understand the impact of vegetation distribution, a further set of simulations was run at 43 CO 2 but with 13 vegetation distribution (Figure 2a), this is labeled 43VEG13.The model was run for 300 years with and without the LUC mask (Figure 1): the former has the suffix LU.The simulations were initialized from the 43 and 43LU simulations; however, the background vegetation was replaced and initialized with that simulated by the 13 experiment, as shown in Figure 2a.As a result of the initialization, the atmospheric component of the model reaches a relative state of equilibrium after 150 years, as indicated by stabilized global-mean surface air temperatures (not shown).The 43VEG13 setup represents a simulation with dynamic vegetation switched off and no climatic influence on the distribution of LCC.The 43 CO 2 concentration was used with the 13 vegetation distribution as these represent the extreme end members of our experimental setup.A summary of experiments is shown in Table 2.

Impact of LUC at increasing CO 2 concentrations
Temperature and precipitation anomalies due to the biogeophysical effects of LUC for Northern Hemisphere summer [June-August (JJA)] are shown in Figure 3. Mean annual, JJA, and NH winter [December-February (DJF)] anomalies are shown for Europe, North America (NA), and globally for the different model simulations in Figure 4. Table 3 gives the global-mean annual, JJA, and DJF temperature and precipitation values and the anomaly due to LUC in brackets.LUC acts to cool global SATs for all CO 2 concentrations.The impact at 13 and 23 is comparable but declines globally and regionally for the 33 and 43 simulations.Annual global cooling is in the region of 20.318C for 13, decreasing to 20.268C for 43 (Figures 3, 4).The crop mask has a significantly greater regional and seasonal impact.Cooling is heavily focused in the Northern Hemisphere (NH) midlatitudes peaking at around 508N (Figure 5), correlating with areas of highest LUC fraction, that is, Europe, NA, and China.Cooling across this midlatitude band peaks in JJA due to higher levels of insolation that drive a greater anomaly in the net absorbed shortwave energy at the surface.European and NA JJA cooling is significantly greater than the global average, decreasing by 21.438 and 21.398C for 13 and 20.798 and 20.808C for 43, respectively.
The overall decrease in SATs due to LUC for all CO 2 concentrations is primarily a response to an increase in surface albedo and reduced SW energy at the surface.Replacement of natural vegetation with C3 and C4 gases increases the maximum canopy/snow albedo parameters in the model (Cox 2001) and reduces LAI.The net surface shortwave energy anomaly peaks in the summer months during periods of highest insolation, whereas the surface albedo anomaly peaks in the winter months due to a significant increase in snow cover.The way in which CO 2 alters the surface energy balance is discussed in section 4.  It is worth noting the contrasting effect seen in India (Figure 3) that experiences a JJA warming peaking at 2.38C.This pattern was identified in Singarayer et al. (2009) and discussed in more detail by Singarayer and Davies-Barnard (2012), who attributed this to a negative feedback on the monsoon hydrologic cycle.The LUC-driven cooling causes a reduction in convective cloud cover, reducing atmospheric albedo and increasing incoming shortwave (InSW) energy at the surface by up to 3.6 W m 22 at 13 CO 23 .This acts to counteract the initial cooling impact   of LUC and also reduces summer precipitation by an average of 20.42 mm day 21 (26%).The potential for LUC to reduce Indian monsoon precipitation has also been shown in the observational study of Niyogi et al. (2010).LUC influences the surface hydrology and the main flows and partitioning of water.Within HadCM3, precipitation is initially split into canopy interception and throughfall at the surface.Removal of forest and the consequent reduction in LAI decreases canopy interception resulting in a negative annual canopy evaporation anomaly in areas of significant LUC.The increase in throughfall drives a positive anomaly in surface runoff and soil evaporation.Soil moisture anomalies are negative in the top layers, shifting to positive with depth.This reflects the shift from the deep rooting depth associated with forest cover to shallow depths of grasses and therefore greater water extraction in the top layers.
LUC has a small impact on global precipitation, with levels decreasing by 20.0079 mm day 21 for 13 and 20.0053 mm day 21 for 43.There is a greater regional and seasonal impact, which again is reduced at higher CO 2 concentration.The European precipitation anomaly peaks in the summer (JJA) at 0.16 mm day 21 compared to 0.0087 mm day 21 change for 43.The JJA increase in precipitation is most likely a response to the albedo-driven decrease in surface temperature and thus an increase in relative humidity.We would expect that the conversion of natural vegetation to crops would decrease the evapotranspiration rate due to reduced LAI and rooting depth.However, in the HadCM3 climate model, the increase in relative humidity and consequently precipitation counteracts this and drives a net increase in evapotranspiration.There is a reduction in DJF precipitation with LUC in northern temperate and boreal latitudes that may at least in part reflect an increase in snow cover due to reduced SATs and so more water stored on the surface.
The negligible change in precipitation due to LUC at 43 CO 2 is likely in part due to the reduced cooling impact of LUC.However, the influence of CO 2 on canopy conductance is also expected to play a role.Conductance decreases by 20.0016 m s 21 globally with a greater impact over the tropics.This decrease is enhanced in the northern temperate and boreal regions during Northern Hemisphere summer.This is likely to decrease the rate of evapotranspiration and therefore counteract any temperature-humidity-driven increases in precipitation.

Role of LCC on the impact of LUC
The shift in LCC for 13 and 43 CO 2 is shown in Figure 2. Broadly, there is a northward migration of broadleaf forest, primarily in place of needleleaf vegetation and shrubland.The shift in the tree line is a pattern seen in numerous modeling studies (Bala et al. 2006;Lucht et al. 2006;Port et al. 2012;O'ishi et al. 2009;Willeit et al. 2014).There is complete dieback of the Amazon region, a feature apparent in past studies using HadCM3 (Betts et al. 2004;Cox et al. 2000).Other notable impacts include dieback across much of Australia and the decline in the extent of grasses with the exception of sub-Saharan Africa.
The mean temperature and precipitation values for the 43VEG13 simulation (without LUC) are shown in Table 3.Compared to the 43 simulation, the 43VEG13 run has lower mean global temperatures on the order of 20.768C annually.Precipitation levels show a small annual and JJA decline of 20.06 and 20.08 mm day 21 , respectively.The cooler global temperatures for 43Veg13 are primarily driven by higher surface albedo due to the replacement of predominantly broadleaf vegetation with shrubland and needleleaf.There is enhanced cooling impact in the northern boreal regions that may in part reflect an amplification of the positive sea ice albedo feedback.The reduced SATs show that dynamically simulated vegetation that responds to CO 2 and climate change causes a decrease in surface albedo, acting as a positive feedback on warming.
In Europe and North America, altering vegetation distribution for 43VEG13 results in a change from predominantly broadleaf vegetation to needleleaf.In Europe, this results in a cooling of 21.348C; however, the albedo parameters for these forest types are the same in HadCM3.Instead, this cooling may be driven by a reduction in LAI, which is also likely to contribute to a reduction in soil evaporation, transpiration, and consequently precipitation.This also acts to increase soil moisture in the deepest soil level (2 m), the rooting depth of trees.
The annual and seasonal anomalies for the 43VEG13 simulation are shown in Figure 4, and the zonal impact of LUC for all simulations is shown in Figure 5.The cooling impact of LUC is significantly greater for 4xVEG1x relative to 43, indicating that natural vegetation distribution simulated by the model has a big effect on the impact of LUC.When background vegetation is allowed to dynamically adjust to CO 2 concentration, the cooling impact of LUC is reduced.The global, JJA, and DJF annual cooling is 20.338, 20.328, and 20.348C, respectively.The more significant winter anomaly is likely to be driven by an increase in high northern latitude sea ice cover, which responds to the greater cooling impact of the crop mask.There is a bigger regional impact, with annual European and North American SAT anomalies of 21.008 and 20.888C, respectively, peaking at 21.198 and 21.048C in the summer months.

Surface energy balance
To understand the mechanism by which the biogeophysical impact of LUC is altered by CO 2 concentration and the background vegetation distribution, the following analysis focuses on changes in the surface energy fluxes.We concentrate specifically on Europe to provide a more focused analysis; the following results are comparable to the North American region that is not shown here.
Figure 6 shows the JJA and DJF surface energy anomalies due to LUC for the 13, 43, and 43VEG13 simulation.The fluxes of outgoing shortwave (OutSW) are increased due to LUC for all simulations due to an increase in surface albedo.LUC also drives an increase in summer latent energy flux due to increased evapotranspiration, although this effect is small at 43 CO 2 .This acts to increase the Bowen ratio (i.e., ratio of latent to sensible heat fluxes) and reduce sensible energy at the surface.The conversion of forest to grassland has been shown in studies to actually decrease latent heat flux (Pitman et al. 2009); however, the increase in precipitation due to the albedo-driven cooling and subsequent increase in relative humidity counteracts this (see section 3.1).The 13 and 43VEG13 anomalies also show a decrease in incoming SW (not shown) that indicates an increase in atmospheric albedo driven by greater cloud cover due to increased relative humidity, reducing SW reaching the surface and enhancing the cooling impact.
We can compare the JJA and DJF anomalies for the different simulations to indicate how CO 2 concentration and changing LCC influence the impact of LUC.There is a greater cooling effect due to LUC at 13 CO 2 compared to 43 CO 2 in both the summer and winter months; however, they appear to be driven by different factors.In summer, LUC significantly increases latent cooling at 1x CO 2 with only a small impact at 43 CO 2 .In the winter months, it is not a difference in latent cooling but a greater OutSW anomaly (i.e., albedo) at 13 CO 2 compared to 43 that drives enhanced cooling.
The lesser winter albedo anomaly at 43 CO 2 relative to 13 is likely driven by a weakened snow albedo feedback in a warmer world, an effect highlighted in other climate modeling studies (e.g., Colman and McAvaney 2009).Figure 7 shows the seasonal cycle for European snow cover for the simulations and the anomaly due to LUC.Mean European DJF snow water equivalent decreases from 21.33 to 2.83 kg m 22 for 13 and 43 CO 2 due to warmer temperatures.The impact of LUC is similarly reduced, with European mean DJF snow cover increasing by 1.99 kg m 22 for 13LU and 0.39 kg m 22 at 43LU.This contributes to a smaller increase in winter albedo and indicates a weakened winter cooling effect of LUC under 43 CO 2 concentrations.
In contrast to DJF, the JJA temperature discrepancy is driven not by interactions with snow albedo but with a reduction in latent cooling at higher CO 2 .This may in part be driven by CO 2 -induced reduction in JJA precipitation over Europe, which declines by 43.2% at 43 CO 2 relative to 13.This decline may in part be driven by the reduction in canopy conductance due to higher CO 2 concentrations.Consequently, LUC occurs in a more moisture-limited environment at higher CO 2 concentrations.This reduces the rate of evapotranspiration, suppresses latent heat flux, and decreases the JJA cooling impact of LUC.The reduction in precipitation with CO 2 is also expected to decrease cloud cover, increasing InSW at the surface and further counteracting cooling.These results show that CO 2 -induced changes in climate, that is, snow cover and precipitation, reduce the impact of LUC in the winter and summer months.To understand the influence of background vegetation, we can compare the 43VEG13 and 43 anomalies.The JJA cooling impact of LUC is enhanced at 43VEG13, driven primarily by an increase in latent cooling.This shows that there is a greater anomaly in latent cooling when LUC is modeled with a 13 CO 2 background vegetation distribution instead of 43, irrespective of the CO 2 concentration.The enhanced latent cooling is not due to an increase in overall precipitation as it is with 13 and 43 (there is a reduction of 4.3% for 43VEG13-43).Instead, this is likely a response to the way in which water is partitioned at the surface by the vegetation.The 13 vegetation distribution is primarily needleleaf forest that has a lower LAI than broadleaf, influencing canopy height and increasing albedo; consequently, there is a reduction in soil evaporation, transpiration, and latent cooling that produces a positive soil moisture anomaly.
These results show that the decreased JJA impact of LUC under higher CO 2 concentrations is driven not only by the CO 2 -induced reduction in overall precipitation but also by the CO 2 -induced shift in vegetation and the way in which water is partitioned at the surface.This presents a strong case for the need for models to incorporate dynamic vegetation that is sensitive to climate to accurately simulate the impact of LUC.
It is worth noting that there is a negligible difference in the DJF anomalies between 43VEG13 and 43.This is likely due to only a small increase in snow depth and enhancement of the snow albedo effect for 43VEG13 (Figure 7).This shows that in the winter months, LCC has minimal effect, and it is CO 2 -driven changes in snow cover that dominate the overall impact of LUC.

Discussion and summary
We have used the HadCM3 coupled climate model to show that LUC cools global climate primarily due to an increase in albedo in temperate regions.The overall global SAT anomaly at a CO 2 concentration of 350 ppm is 20.318C with a significantly enhanced regional effect in areas of high LUC fraction: Europe, North America, and China.This acts to increase regional precipitation due to an increase in relative humidity.The impact on SAT agrees in sign with a number of previous studies although varies in amplitude (e.g., Brovkin et al. 1999;Govindasamy et al. 2001;Matthews et al. 2004;Betts et al. 2007b).
A simplified flowchart outlining the key processes by which LUC cools climate in HadCM3 is shown in Figure 8a.There are two seasonally dependent positive feedback cycles that act to enhance the cooling impact of LUC.In the winter months, LUC-driven cooling further increases snow cover and surface albedo.In the summer months, cooling increases relative humidity and precipitation, which enhances latent cooling and amplifies cloud albedo.
The first part of this study suggests that the cooling impact of LUC will decrease with higher concentrations of CO 2 due to the breakdown of these positive feedback cycles.The mechanism by which this occurs is shown in Figure 8b.Higher concentrations of CO 2 and warmer conditions decrease winter snow cover (Figure 7) and reduce the degree to which it increases with LUC.The overall effect is to weaken the snow albedo feedback and reduce the cooling impact of LUC in the winter.In the summer months, an increase in CO 2 is simulated to decrease overall precipitation and consequently reduce the evapotranspiration rate.This reduces the partitioning of latent to sensible energy flux and reduces the cooling effect of LUC.The model also simulates a CO 2 -induced decrease in cloud cover, increasing incoming SW energy at the surface and further counteracting cooling.This effect may in part be a response to a reduction in simulated canopy conductance under higher CO 2 concentrations.
These results show that the biogeophysical impact of LUC is dependent on CO 2 concentration and the background climate, an idea initially put forward by Pitman et al. (2011).Their study similarly concluded that the snow albedo mechanism dominates the impact of temperate and boreal LUC in the winter, which is set to diminish with warmer conditions.However, our results presented here disagree with the Pitman et al. (2011) study as to the sign of the summer change under higher CO 2 conditions.They found an increase in summer precipitation under higher CO 2 that drove an increase in latent cooling, the opposite effect to that simulated in HadCM3.Again, this highlights the uncertainty that remains in the modeled response of precipitation to climate change, specifically on a regional scale (see Schaller et al. 2011).However, the general conclusion that the impact of summer LUC is controlled by latent and sensible heat fluxes that are set to change with increasing CO 2 still holds.
The second part of this study investigates the importance of the background vegetation state in quantifying the impact of LUC.Under higher CO 2 concentrations, natural vegetation is expected to change with a general shift of the tree line and conversion of needleleaf and shrubland to broadleaf vegetation.This has implications for the biogeophysical, biogeochemical, and physiological characteristics of the land surface.
Our results show that within the HadCM3 climate model, the regional summer impact of LUC is strongly affected by LCC and the background vegetation type.Again the simplified mechanism for how this occurs is shown in Figure 8b.The shift in natural vegetation with increasing CO 2 alters albedo, LAI, and canopy conductance.These consequently influence the partitioning of water at the surface, reducing soil moisture and transpiration, decreasing latent to sensible energy flux, and counteracting the cooling impact of LUC.When simulating the impact of LUC at 43 CO 2 but with a 13 CO 2 vegetation distribution, cooling is enhanced specifically in the northern temperate regions during summer.This is a response to an increase in latent cooling due to the replacement of predominantly broadleaf vegetation with needleleaf, a reduction in LAI, a suppressed transpiration rate, and higher soil moisture content, enhancing the positive feedback cycles shown in Figure 8a.This shows that the reduced cooling impact of LUC with higher concentrations of CO 2 is driven not only by how the model simulates precipitation patterns, but also how the model dynamically simulates natural vegetation distribution and the way in which this partitions water at the surface.
There remains a high level of uncertainty associated with the impact of LUC in climate models that needs to be addressed in order to assess regional climate change (Pielke et al. 2011).We are aware that there are a number of limitations to our results, namely, that this is a single-model equilibrium experiment using a simplified representation of crops and a relatively simple DGVM.This is particularly problematic when interpreting the impact and role of precipitation, which remains highly uncertain, yet is crucial to the mechanisms presented in this paper.As a result, the extent and even the sign of changes are likely to be dependent on both the model used and the way in which climate and land-use change have been simulated.As with all modeling studies, a multimodel ensemble would be appropriate in order to validate these mechanisms with a range of climate models and DGVMs.Also, a perturbed physics ensemble would help determine which specific variable(s) (soil moisture, LAI, albedo, etc.) drive the feedback mechanisms shown in Figure 8b.Building on this, a more comprehensive analysis should also include transient experiments incorporating a range of climate scenarios, such as the representative concentration pathways (RCPs) used in the IPCC report, with different LUC projections.This would not only account for the inconsistencies in models, but also with the uncertainty associated with LUC projections such as the total area and spatial pattern of change.Future projects that may help to address such issues include the Land-Use Model Intercomparison Project (LUMIP), part of the sixth phase of the Coupled Climate Intercomparison Project (CMIP6) with results planned for 2018 to 2019 (see https://cmip.ucar.edu/lumip).This will focus on outstanding uncertainty associated with LUC and may help to elucidate the above mechanisms.Despite the uncertainty associated with specific mechanisms presented, this study again highlights the potential sensitive of LUC to CO 2 concentration and background climate.This is specifically associated with patterns of precipitation and snowfall and the way in which they may change under higher concentrations of CO 2 .As a result, in order to accurately assess the impact of LUC in future projections, modelers need to be aware that the impact of LUC may be influenced by the amount and spatial distribution of these variables at increasing CO 2 concentrations.We also show the potentially important role of background vegetation in quantifying the impact of LUC, a variable that has been largely overlooked in previous studies.As discussed, only three of the seven models used in the LUCID experiment (HadGEM2-ES, MIROC-ESM, and MPI-ESM) incorporated a dynamic vegetation scheme (Pitman et al. 2009;Brovkin et al. 2013).These three models still demonstrate a varied response to LUC due to a range of other inconsistencies, including the way in which land surface processes (i.e., albedo and evapotranspiration) are parameterized and the way LUC is implemented (see Pitman et al. 2009).As such, the direct role of including a dynamic vegetation scheme in these models is not certain.The results of our study indicate that it may potentially be important, and it would be pragmatic to investigate its impact more specifically in future LUC modeling studies.

Figure 1 .
Figure 1.Map showing the proportion of each grid square in HadCM3 that has undergone LUC according to the land-cover mask of Betts et al. (2007b).Final disturbed fraction is calculated with this mask minus the nonvegetation types simulated by TRIFFID.

Figure 2 .
Figure 2. Distribution of natural vegetation simulated by the model occupying greatest proportion of grid square at (a) 13 CO 2 and (b) 43 CO 2 .

Figure 4 .
Figure 4. Mean global, European, and NA SAT anomalies due to LUC for the different model simulations (seeTable 2 and text for details).Annual, NH, summer (JJA), and NH winter (DJF) are shown.The European region is defined as 358-608N and 52.58E-11.258W.The North American region is defined as 308-558N and 67.58-123.758W.Error bars represent the standard error.

Figure 5 .
Figure 5. Zonal temperature anomaly during NH summer due to LUC for the five simulations.The dashed black line represents the 43VEG13 anomaly.

Figure 7 .
Figure 7. Seasonal cycle of European snow depth (kg m 22 ) for the four CO 2 concentrations and the 43VEG13 simulation.Dashed lines in the upper chart are with LUC.The European region is defined as 358-608N and 52.58E-11.258W.

Earth Interactions d
Volume 20 (2016) d Paper No. 10 d Page 13

Figure 8 .
Figure 8. Simplified conceptual flowcharts outlining the key impacts of temperate LUC in HadCM3 and how these are altered by CO 2 .(a) The general processes by which LUC causes temperate cooling, and (b) how this cooling impact is reduced under higher concentrations of CO 2 and the role of changing vegetation distribution.The red and blue boxes outline processes that are more prevalent in (although not exclusively) the summer and winter, respectively.
Earth Interactions d Volume 20 (2016) d Paper No. 10 d Page 4

Table 2 .
Summary of model simulations.Each simulation was run for 300 years with and without LUC (suffix: LU).Analysis is on the final 50 years of each run; all anomalies are 99% significant as calculated by a Student's t test.Earth Interactions d Volume 20 (2016) d Paper No. 10 d Page 6

Table 3 .
Mean annual/JJA/DJF SAT and precipitation for experiments.Anomalies due to LUC are shown in brackets.Earth Interactions d Volume 20 (2016) d Paper No. 10 d Page 8