Assessing the implications of human land-use change for the transient climate response to cumulative carbon emissions

Recent research has shown evidence of a linear climate response to cumulative CO2 emissions, which implies that the source, timing, and amount of emissions does not significantly influence the climate response per unit emission. Furthermore, these analyses have generally assumed that the climate response to land-use CO2 emissions is equivalent to that of fossil fuels under the assumption that, once in the atmosphere, the radiative forcing induced by CO2 is not sensitive to the emissions source. However, land-cover change also affects surface albedo and the strength of terrestrial carbon sinks, both of which have an additional climate effect. In this study, we use a coupled climate-carbon cycle model to assess the climate response to historical and future cumulative land-use CO2 emissions, in order to compare it to the response to fossil fuel CO2. We find that when we isolate the CO2-induced (biogeochemical) temperature changes associated with land-use change, then the climate response to cumulative land-use emissions is equivalent to that of fossil fuel CO2. We show further that the globally-averaged albedo-induced biophysical cooling from land-use change is non-negligible and may be of comparable magnitude to the biogeochemical warming, with the result that the net climate response to land-use change is substantially different from a linear response to cumulative emissions. However, our new simulations suggest that the biophysical cooling from land-use change follows its own independent (negative) linear response to cumulative net land-use CO2 emissions, which may provide a useful scaling factor for certain applications when evaluating the full transient climate response to emissions.


Introduction
The transient climate response to cumulative carbon emissions (or TCRE), defines the global temperature response to cumulative emissions of CO 2 , Gillett et al 2013 and has been identified as a useful and practical tool for evaluating CO 2 -induced climate changes (Collins et al 2013). While Earth system models may have large uncertainties when representing specific physical processes that have an impact on temperature, most of these models consistently demonstrate a linear temperature response to cumulative emissions. Sensitivity simulations further suggest that a constant TCRE value (constant linear slope) can reasonably represent the climate effect of cumulative emissions over a wide range of emission scenarios for a single model (e.g., Leduc et al 2015).
Furthermore, the TCRE remains relatively constant as models evolve over time . While the TCRE linearity is robust, most climate models also show that the temperature sensitivity to cumulative emissions decreases under very high emissions scenarios , due in part to the radiative band saturation for certain wavelengths for very high atmospheric CO 2 concentrations. Nevertheless, due to its simplicity and relative independence to the emissions scenario, the TCRE relationship has become a measure for defining policy initiatives for CO 2 mitigation , Raupach et al 2011, Matthews et al 2012.
An important component of total anthropogenic CO 2 emissions are emissions from land use change (LUC), which represent a substantial fraction of the total historical emissions. Land use emissions have the complicating effect of changes in the albedo of land surface, as well as other effects of changing vegetation types, which can have additional impacts on the climate response to land-use change (Pongratz et al 2010). Most research on TCRE to this point assesses climate response with combined fossil fuel and land use emissions, without addressing the spatial interaction of the biosphere to albedo-induced temperature changes from LUC. However, many studies have shown that the biophysical effects of land-use change are an important contributor to historical climate changes (Brovkin et al 2006, Eby et al 2013 and are also relevant to capturing the climate response to future changes in land-use (Davies-Barnard et al 2014).
It is therefore important to carefully assess the climate response to cumulative LUC CO 2 emissions within the context of land surface changes. The objective of this study is to evaluate the influence of spatially-explicit LUC on the TCRE, including both the historical period and future RCP scenarios. We use a suite of simulations of the University of Victoria Earth System Climate Model (UVic ESCM) to isolate both the biogeochemical (the CO 2 radiative warming from land use CO 2 emissions) and biophysical (land-surface changes from agriculture and pasture affecting surface albedo, roughness, and energy balance) components of temperature change, in order to evaluate the full range of impacts of LUC on the transient temperature evolution.

Model description
The simulations in this paper were performed using theUVic ESCM v. 2.9 (Weaver et al 2001. The UVic model is classified as an Earth system model of intermediate complexity (Claussen et al 2002), providing a simplified mathematical representation of the climate system that is useful for studying paleoclimates and assessing potential future climate changes, such as under 21st-century representative concentration pathway (RCP) scenarios for the latest Intergovernmental Panel on Climate Change (IPCC) report (Eby et al 2013). The UVic ESCM operates all model components on a resolution of 1.8°l atitude by 3.6°longitude. These include a one-layer (2D) energy-moisture balance model (Weaver et al 2001) to calculate atmospheric surface air temperature and humidity, with prescribed winds controlling atmospheric advection and wind stress on the ocean surface. The model ocean is a 3D general circulation model with 19 vertical layers, incorporating inorganic carbon (Ewen et al 2004), organic carbon and nutrient cycling (Schmittner et al 2008) and ocean sediments , coupled to a dynamic-thermodynamic sea ice module (Weaver et al 2001).
The terrestrial biosphere component of the UVic ESCM is dynamically represented by the top-down representation of interactive foliage and flora including dynamics (TRIFFID) model, coupled to the onelayer MOSES soil model where vegetation litter is stored and respired (Meissner et al 2003). TRIFFID represents five different plant functional types (PFTs): broadleaf trees, needleleaf trees, C3 grasses, C4 grasses, and shrubs. Each PFT uniquely affects the roughness and albedo of the land surface as well as carbon uptake from the atmosphere and transfer to the soil layer. Terrestrial sequestration in the UVic model (Matthews 2007, Pinsonneault et al 2011 is enhanced by higher atmospheric CO 2 concentrations. However, the coupling of the nitrogen and carbon cycles is not represented. Agricultural LUC is implemented in the model (as discussed in section 2.2 below) by removing tree and shrub PFTs in the land use fraction of each grid cell and allowing C3 and C4 grasses to grow in their place. In regions designated as pasture areas, 50% of the existing tree or shrubs cover is removed for replacement by grasses. In grid cells where C3 and C4 grasses already predominate, LUC leads to no change in PFT biomass or fractional coverage. In the version of the model used in this paper, 30% of the removed vegetation carbon due to LUC is transferred directly to the soil carbon pool in the form of litter-fall. The remaining 70% is then distributed between direct emissions to the atmosphere (30%), and intermediate storage in two additional wood carbon pools with varying residence times: a fast-respiring wood pool (20% of harvested carbon; 2 year residence time), and a slow-respiring wool pool (50% of harvested carbon; 20 year residence time), based on the book-keeping approach described in (Strassmann et al 2008).

Experimental setup
The model experiments in this paper explore the respective contributions of spatially-explicit LUC and fossil fuel emissions to the total TCRE. The LUC data for the period from 1750 to 2005 used here is based on the HYDE 3.1 data set (Klein Goldewijk et al 2011) interpolated to the UVic model resolution, with any resulting land use fractions over the model's ocean grid cells removed rather than redistributed to nearby land grid cells. For the period from 2005 to 2100, LUC data for the RCP scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5) are taken from pastoral and agricultural land use fractions used in the EMIC model intercomparison report for the IPCC (Eby et al 2013). Annual fossil fuel emission data from the historical period  were obtained from the Carbon Dioxide Information Analysis Centre (Boden et al 2013), and harmonized fossil fuel CO 2 emissions for different RCP scenarios for the 2005-2100 period were prescribed from (Meinshausen et al 2011).
The model was spun up to equilibrium conditions with year 1750 CO 2 concentration (279 ppmv) and spatial land cover for the year 1750 based on spatial land use data from HYDE 3.1, which we then used as the initial condition for time-evolving simulations from 1750 to 2100. These transient experiments, summarized in table 1, use either prescribed fossil fuel emissions and/or spatial LUC data as forcing mechanisms. In addition to a control simulation (CTRL) with no external forcing (no fossil fuel emissions and no change in the land use data from the 1750 distribution), transient simulations were run with only LUC forcing (LUC, or LUC-only), with only fossil fuel forcing (FF), and with the LUC and FF forcing applied together (LUC+FF).
We also performed additional simulations with prescribed atmospheric carbon content (PC) to calculate the climate response to the biogeochemical-only and biophysical-only effects of terrestrial emissions. The first of these (PC(LUC) in table 1) is a simulation in which atmospheric CO 2 concentrations simulated by the LUC experiment were prescribed to a second simulation, but this time holding land-use distributions constant at the year 1750 configuration. The second (PC(LUC+FF) in table 1) is a comparable simulation but with prescribed CO 2 concentrations taken from the LUC+FF simulation. Finally we prescribed the net CO 2 emissions from LUC+FF diagnosed to a final simulation (EMIT in table 1) to represent the effect of externally-prescribed land-use CO 2 emissions without the corresponding changing spatial land-cover change. Each of these simulations was performed for the historical period, as well as for each future RCP emission scenario. Table 2 summarized the method we used to diagnose the net terrestrial CO 2 emissions from LUC, as well as the individual (bio)geochemical and biophysical effects of LUC. In order to capture both the direct LUC CO 2 emissions and their influence on global vegetation cover, we calculated as the difference in total terrestrial carbon between the PC simulations and the respective simulation with spatially-varying LUC. The results from the PC(LUC) (PC(LUC+FF)) simulation diagnoses the global potential natural vegetation distribution (which, in turn, influences planetary albedo, surface temperature, vegetation coverage, and the strength of the terrestrial carbon sink) in response to the atmosphere CO 2 concentrations obtained in the LUC (LUC+FF) simulation, respectively. As PC(LUC) and LUC (or PC(LUC+FF) and LUC+FF) have the same atmospheric CO 2 concentrations, the difference in the terrestrial carbon storage between these pairs of simulations captures the net effect of spatially-explicit land use on terrestrial carbon storage. These 'net terrestrial emissions' therefore include both the direct CO 2 emissions to the atmosphere from land-use change (agriculture and pasture), the atmospheric CO 2 sequestered by the natural potential vegetation (in the PC simulations) in LUC-affected regions, as well as the change in terrestrial carbon storage elsewhere due to terrestrial sequestration and expansions/shifts in vegetation cover attributable to temperature differences between LUC and PC(LUC) (or LUC+FF and PC (LUC+FF)). In general, afforestation reduces the net terrestrial (LUC) emissions reported here, whereas Table 1. A summary of the transient (time-evolving) simulations presented or used for calculations in this study. 'Spatial land use' refers to the application of spatial HYDE 3.1 data in the model, whereas 'prescribed emissions' denotes CO 2 emissions released directly to the atmosphere from an emissions data file (in which yearly prescribed emissions are released directly to the model atmosphere). Prescribed CO 2 refers to simulations with a prescribed, time-evolving atmospheric carbon content, in which atmospheric CO 2 is forced to follow a specified CO 2 concentration trend (to which the model's other carbon reservoirs respond). Cumulative emissions and net temperature response (ΔT) to emissions deforestation (for crops, pasture, and natural die-off) increases these emission totals for a given simulation year. These net emissions resulting (directly and indirectly) from the inclusion of spatial, time-evolving LUC in the model (as calculated in table 2) are described as 'net terrestrial emissions' in the results section and 'LUC emissions' in figure captions of this paper for the sake of simplicity. The biogeochemical (BGC) effect (i.e., the CO 2 radiative warming) of these net terrestrial emissions can be calculated as the simulated temperature difference between the two PC simulations (PC(LUC) and PC(LUC+FF)) and the corresponding simulation without any land-use change (CTRL and FF respectively). The total geochemical (GC) effect presented in the results includes the biogeochemical effect (i.e., the CO 2 radiative warming due to net terrestrial emissions alone) in addition to to the radiative warming from fossil fuels. Finally the biophysical (BP) effect of landuse changes can be calculated as the difference between the temperature change simulated by the LUC and LUC+FF simulations and the respective PC simulations.

The historical period: 1750-2000
The results for the 1750-2000 period are presented separately from the RCP scenarios, as this 'historical' period represents a shift in the dominant emissions source from LUC to fossil fuels (whereas the RCP emissions are largely dominated by fossil fuels). Figure 1(a) shows the evolution of atmospheric CO 2 during this period, and figure 1(b) provides the cumulative emissions (net terrestrial emissions+fossil fuel emissions) for each simulation. Land use alone (LUC) generated cumulative CO 2 emissions of 110 Pg C, which resulted in an 18 ppm increase in atmospheric CO 2 by the year 2000. Cumulative fossil fuel emissions were 280 Pg C, which by themselves produced a 58 ppm increase, and both land use and fossil fuels together (LUC+FF) produced emissions of 410 Pg C, which led to an 84 ppm increase. An interesting feature of these results is that the sum of emissions from the LUC and FF simulations was smaller than the net emissions in LUC+FF, which results from slightly higher (20 Pg C) cumulative landuse emissions in the LUC+FF simulation as compared to LUC alone (compare the yellow and green lines in figure 1(b)). This reflects an enhancement of simulated LUC emissions associated with higher CO 2 in the LUC+FF simulation; the stronger CO 2 fertilization of the terrestrial biosphere in LUC+FF (as compared to LUC) resulted in the presence of more overall vegetation carbon in the model, which then led to higher net terrestrial emissions when this carbon was removed due to changing land-use distributions.
The LUC+FF simulation reproduces the observed atmospheric CO 2 trend after 1950 reasonably well, though, as with many other models, noticeably underestimates the observed CO 2 increase (by 5-10 ppm) during the late 19th and early 20th centuries. This may be associated with the result that annual net terrestrial emissions values (not shown) obtained in the UVic ESCM from the spatially-explicit LUC are significantly less than annual harmonized LUC emissions estimates (e.g., Meinshausen et al 2011) for this period. Low early LUC emissions may be related to the equal treatment of croplands and natural grasses in the model-agriculture and pasture fractions in areas already dominated by bare ground or grasses lead to no change in PFT distribution or biomass in the model (such as for the North American prairies). In addition, the lower-than-observed atmospheric CO 2 values may also be related to a variety of interrelated natural and anthropogenic factors, including (for example) the strength of the model's ocean carbon sink, the simulated pre-industrial vegetation distribution, an underestimation of per capita land-use for the earlier period in the HYDE 3.1. dataset (Kaplan et al 2011), and the omission of a faster soil carbon cycling to the atmosphere in agricultural regions due to tillage (e.g., Stocker et al 2011).
The temperature response to the modeled CO 2 changes and human land use are plotted in figure 1(c), which includes the simulations with prescribed atmospheric CO 2 that are used to isolate the biochemical and biophysical temperature changes shown in figure 1(d). The common assumption that CO 2 (once in the atmosphere) leads to a certain radiative warming regardless of the emissions source is represented by the temperature trend from PC(LUC+FF). However, the biophysical effects of land-use change are evident in the divergence in the temperature response between PC(LUC+FF) and LUC+FF, beginning relatively early in the simulation period (∼1850). The divergence stops increasing after 1960, when the effect of fossil fuel emissions begins to dominate the overall temperature change in the simulation (as shown by the point at which cumulative FF emissions overtake those from LUC: figure 1(b), yellow line versus purple line).
The importance of albedo changes in determining the temperature response to emissions is further illustrated in the LUC case, where, despite an 18 ppm atmospheric CO 2 increase ( figure 1(a), green line), global temperatures decreased slightly over the course of the simulation ( figure 1(c), green line). Similarly, in spite of the significantly greater emissions ( figure 1(b), red line) and higher atmospheric CO 2 ( figure 1(a), red line) in the LUC+FF scenario as compared to the FF case, the actual temperature increase produced in the LUC+FF simulation is slightly less than in the FF scenario ( figure 1(c), red line versus purple line). This means that in these simulations the biophysical effect of land-use changes over the historical period are slightly larger than the biogeochemical effect of land-use CO 2 emissions. This can be seen in figure 1(d), which shows biophysical LUC temperature change of −0.24°C between 1750 and 2000 for both the LUC and LUC+FF scenarios. By comparison, the biogeochemical effect of net terrestrial emissions was a warming of 0.2°C in the LUC simulation, and 0.22°C in LUC+FF, reflecting the slightly higher land-use CO 2 emissions in LUC+FF compared to LUC. This estimate of the biophysical effect of land-use change is comparable to the value produced by a recent intercomparison of 12 EMICs (Eby et al 2013), though is slightly larger than that simulated by an earlier version of the UVic model using a different land-use database (Matthews et al 2004).
We now assess the climate response to cumulative emissions in these simulations to determine to what extent the TCRE is a useful metric of the climate response to land-use CO 2 emissions. The overall temperature response to cumulative fossil fuel and land-use change (including both the biophysical and biogeochemical land-use effects) for the UVic ESCM 2.9 is shown in figure 2 However, the net temperature response to spatially-explicit LUC is notably different. As expected from the temperature changes in figure 1, the net response to cumulative land-use emissions (green and yellow lines in figure 2(a)) follows a slight negative slope, reflecting the higher biophysical compared to biogeochemical effects in these simulations shown in figure 1(d). Consequently, the combined response to both fossil fuel and land-use emissions (LUC+FF; red line) shows a nonlinear relationship between the net temperature change and cumulative emissions, with an initially lower rate of warming, followed by an increased temperature response to increasing cumulative emissions. The lower initial temperature response reflects the earlier historical period when LUC-sources of CO 2 (and associated biophysical cooling)  predominate, whereas the larger temperature response emerges with increasing fossil fuel contribution to the total emissions. Thus, the red line (LUC+FF) tracks closer to the green and yellow lines for low emissions values and more closely parallels the purple and brown lines for higher emissions where fossil fuel CO 2 emissions predominate, leading to poorer overall linearity (table 3). By the year 2000, while the slopes of the red (FF+LUC) and brown (EMIT) temperature curves are similar, the overall warming in FF+LUC remains about 30% smaller than when LUC emissions were prescribed externally. However, despite the very different overall temperature response to land-use CO 2 and fossil fuel CO 2 emissions, figures 2(b) and (c). show that both the isolated biophysical and biogeochemical effects of landuse change can be linearly related to cumulative landuse CO 2 emissions. The net biogeochemical effect of land-use CO 2 emissions (figure 2(c)) follows a positive linear response to cumulative emissions, which is also nearly equivalent to the geochemical warming from fossil fuel CO 2 (table 3). This suggests that the TCRE for fossil fuel emissions can be readily applied to the biogeochemical effect of cumulative land-use emissions.
With respect to the biophysical temperature change ( figure 2(b)), despite some variations, land use emissions follow similar negative slopes (table 3) in LUC and LUC+FF for the historical period (i.e., increasing emissions are associated with larger biophysical temperature decreases). While biophysical temperature changes are influenced by roughness length, evapotranspiration, and other modifications of the Earth's surface due to ecosystem conversion, figure 3(a) suggests that the linear relationship between biophysical temperature and land use emissions is dominated by surface albedo changes in both the LUC and LUC+FF simulations. Though total albedo change in both simulations is increasingly influenced by the effects of geochemical warming (which decreases surface albedo due to reduced snow and ice cover) for greater LUC emissions, the biophysical component of the albedo change follows the linear relationship established in figure 2(b) (for the LUC+FF simulation, the albedo-emissions regression yields a slope of 2.65°C/Tt C and r 2 of 0.969, to compare to table 3). The albedo difference between the LUC and LUC+FF simulations (figure 3(a), yellow line versus green line) are related to the greater sequestered fossil fuel carbon in the terrestrial emissions of LUC+FF, which lead to reduced albedo change sensitivity to terrestrial emissions (as shown for temperature in table 3).
The regional distribution of albedo changes ( figure 3(b)) demonstrates that the resulting biophysical temperature change is not limited to areas experiencing land use conversion but has a global impact due to a high latitude amplification of land use-related cooling by the ice-albedo feedback, with biophysical cooling leading to more snow in mid and high latitude regions and greater Arctic sea ice extent (not shown, see albedo changes figure 3(c)). Similarly, mid-latitude and high altitude regions undergoing LUC (such as central North America and central Eurasia in figure 3(d)) tend to experience greater local albedo changes (figure 3(c)), due in part to the greater cold-season effect of snow on grassy (i.e., agricultural and pastoral) surfaces (compared to forested regions, where snow has a smaller impact on surface albedo due to taller vegetation masking the snow albedo). In addition, most other regions demonstrating biomass gains ( figure 3(d), blue-shaded grid cells) experience a decrease in albedo (figure 3(c), gray-shaded grid cells). Thus, the linearity between albedo and net terrestrial emissions is partly related to the way the net LUC emissions totals are calculated in this study as the sum of afforestation (which decreases both albedo and net terrestrial emissions) and deforestation (which increases albedo and net terrestrial emissions). For these simulations, the strongly linear response (table 3) of biophysical temperature to net terrestrial emissions for the historical period suggests that, for certain TCRE applications, an additional linear scaling factor can be used to represent the biophysical land-use effects. Table 3. The linear slopes of the curves represented in figure 2, with the linear correlation coefficient provided in parentheses. The slopes represent the line of best fit passing through the point (0, 0), i.e., zero emissions associated with no temperature change.  figure 4, with the temperature change and emissions plotted relative to the year 2000 to highlight changes during the 21st century alone. Each of the panels in figure 4 are also plotted with different temperature and emissions scales so as to be able to show the relative differences between simulations for a given RCP scenario, due to the very different overall emissions across the four RCPs. These plots show several common features as compared to the historical runs. In all cases, there is a consistent linear temperature response to externally-prescribed emissions (brown and purple lines), and in all cases the combined response to fossil fuel and land-use CO 2 produced by spatially-explicit LUCs is decreased as a result of the inclusion of biophysical land-use effects (red lines). There is also a consistent pattern of enhanced net terrestrial emissions when fossil fuel emissions are included (yellow versus green lines), which can be understood in general as the re-emission by land-use change of fossil fuel emissions that had been previously sequestered in the terrestrial biosphere due to CO 2 fertilization (calculated as the difference in net terrestrial emissions between the LUC+FF and LUC experiments). Furthermore, the relative importance of this effect increases according to the magnitude of fossil fuel emissions across the four RCP simulations; this re-emission of fossil fuel CO 2 accounts for 40% (45 Pg C) of total 21st century LUC+FF terrestrial emissions in RCP2.6, 45% (46 Pg C) in RCP4.5, 51% (69 Pg C) in RCP6.0, and 57% (82 Pg C) in RCP8.5. There are some interesting differences from the historical simulations. In the LUC+FF simulation, the net temperature response to land-use changes is now positive (table 4), reflecting an increase in the biogeochemical relative to biophysical effects of land-use change over the 21st century. Furthermore, the total temperature response in the LUC+FF simulation follows an overall more linear response to cumulative combined fossil fuel and land-use emissions, owing in part to this decreased overall prominence of the biophysical land-use effect. It can also be seen that, as the Figure 3. The spatial and temporal relationship between albedo change, biophysical temperature, and net terrestrial emissions. In (a), annual-average global albedo change (%) due to the biophysical impacts of LUC is plotted against corresponding net terrestrial emissions in the LUC and LUC+FF simulations, with total albedo change (dark lines) represented separately from the albedo change due to biophysical impacts of LUC. In (b), the spatial distribution of biophysical temperature change in the LUC+FF simulation is plotted for the 10 year average of the period 1990-2000. Correspondingly, in (c) the surface albedo change (1990s ten-year average) due to biophysical effects (calculated as in the legend for (a))is shown, and in (d) the spatial distribution for net terrestrial emissions (TC PC(LUC+FF) −TC (LUC+FF) ) to the end of the historical period (in Tg C) for the LUC+FF simulation is provided for comparison. overall fraction of total emissions that comes from fossil fuels increases from RCP2.6 (74%) to RCP8.5 (93.3%), the overall temperature response also becomes more similar to the linear temperature response to externally-prescribed emissions ( figure 4,  table 4). In addition, RCP4.5 has a less pronounced spatial extent of land-use changes compared to the other scenarios, with less land-use CO 2 emissions (103 Pg C, 11% of the total emissions) by the end of the century. For comparison, cumulative land-use emissions were 26% of the total 21st century emissions in RCP 2.6 (113 Pg C), 9.7% (135 Pg C) in RCP6.0 and 6.7% (144 Pg C) in RCP8.5.
In figure 5, we combine the historical and future simulations to show the net temperature response to cumulative CO 2 emissions in the case of externally prescribed (EMIT simulations) versus dynamicallycalculated (LUC+FF simulations) land-use CO 2 emissions ( figure 5(a)). Also plotted are the biophysical-only and biogeochemial-only temperature  Table 4. The linear slopes of cumulative emissions versus total ΔT curves with the historical period and RCP scenarios together (1750-2100). The linear correlation coefficient is provided in parentheses. As in table 3, the slopes represent the line of best fit passing through the point (0, 0), i.e., zero emissions associated with zero temperature change. response to net terrestrial emissions ( figure 5(b)). As can be seen in figure 5(a), in all cases there is a slight negative deviation from a linear temperature response to cumulative emissions with increasing total emissions, which has also been shown in previous analyses of the TCRE (Gillett et al 2013, Herington and Zickfeld 2014, Leduc et al 2015. The divergence between the EMIT and LUC+FF simulations is especially notable for lower emissions values, where LUC emissions and the associated biophysical cooling have a more prominent effect. Correspondingly, the temperature response to cumulative emissions in EMIT and LUC+FF largely parallel each other at higher cumulative emissions which are dominated by fossil fuel CO 2 . Figure 5(b) shows the individual response of biophysical and biogeochemical land-use effects to cumulative land use CO 2 emissions. In the simulations with spatially-explicity land-use changes, but no fossil fuel emissions (LUC simulation), both the biogeochemical and biophysical effects of land-use change can be well described as a linear (positive and negative respectively) function of cumulative emissions. Again, this suggests that the TCRE for externally prescribed (i.e. fossil fuel) CO 2 can be used to also estimate the biogeochemical warming associated with land-use CO 2 emissions, and that an additional linear scaling factor can be used to incorporate the corresponding biophysical effects of land-use change. The linearity appears to hold less well for the LUC+FF simulations, which again can be seen to have higher land-use CO 2 emissions associated with the re-emission of previously sequestered fossil fuel CO 2 emissions. In the case of the biogeochemical effect, however, the deviation from linearity seen in the LUC+FF simulations is actually consistent with the overall negative deviation of the temperature response relative to a constant TCRE at very high cumulative emissions (Leduc et al 2015), recalling that the land-use emissions in the LUC+FF simulations occur in conjunction with increasing fossil fuel emissions and are therefore subject to the same diminishing temperature response at high total emissions.
For the case of biophysical effects, the increase in albedo associated with land-use change does not depend on whether land use occurs in conjunction with fossil fuel emissions. Accordingly, the LUC simulations ( figure 5(b), blue-shaded lines) suggest a very linear biophysical cooling response to cumulative CO 2 emissions, as discussed for the historical period in section 3.1. However, this biophysical linearity breaks down for the LUC+FF simulations ( figure 5(b), redshaded lines). Most of difference in land use emissions between the LUC and LUC+FF simulations results Figure 5. The temperature response to emissions for the historical period and the four RCP scenarios. In (a), the annual total net temperature change is plotted against total emissions for the LUC+FF and EMIT cases. In (b), the biogeochemical and biophysical temperature changes are plotted against LUC emissions for the LUC and LUC+FF experiments. from fossil fuel carbon absorbed by the terrestrial biosphere and then re-emitted to the atmosphere during land cover change. Plotting the biophysical cooling from LUC+FF against the LUC emissions from LUC (not shown) for each of the RCP scenarios provides a linear trend, indicating that much of the biophysical nonlinearity seen in figure 5(b) for the 21st century LUC+FF simulations results from the high proportion (40%-57%) of biosphere-sequestered fossil fuel carbon in the LUC+FF net terrestrial emissions.
Focusing more specifically on the individual cases, the LUC+FF RCP2.6 scenario ( figure 5(b), dark red line) has the smallest fraction (40%) of biospheresequestered fossil fuel carbon in the net terrestrial emisisons, and it is the closest curve to the linear LUC scenarios. The LUC+FF RCP4.5 simulation ( figure 5(b), magenta line) has 10 Pg C less terrestrial emissions than RCP2.6, but with a greater proportion (45%) of sequestered fossil fuel carbon, and it demonstrates the most biophysical nonlinearity of all the LUC+FF simulations. From LUC+FF RCP6.0 ( figure 5(b), dark pink line) to LUC+FF RCP8.5 ( figure 5(b), red line), the biophysical linearity gradually increases again despite greater proportions (51%-57%) of fossil fuel sequestered carbon in the net terrestrial emissions. This is, in part, because the biophysical temperature change (ΔT BP =T LUC+FF −T EMIT , from table 2) is dependent on temperature change in response to fossil fuel emissions. As the temperature sensitivity to CO 2 decreases for higher emissions scenarios (figure 5(a), brown/orange lines), the magnitude of ΔT BP correspondingly increases. Therefore, the reduced geochemical influence on transient temperature changes for higher emissions scenarios contributes to the gradually steeper biophysical slope from RCP4.5 to RCP8.5 in the LUC+FF simulations, in a similar manner as the deviation from linearity for the biogeochemical warming effect (figure 5(b), brown/ orange shaded lines).

Conclusions
Recent literature has highlighted the strong linearity between temperature change and cumulative carbon emissions, both in the observational record, and in a range of Earth system models , Gillett et al 2013. This relationship, as defined by the TCRE metric, represents a simple, yet powerful, framework for understanding contemporary climate change and informing climate mitigation targets (Friedlingstein et al 2014). However, these studies consider only the radiative effect of land use carbon emissions in the atmosphere, thus neglecting the additional radiative influence of anthropogenic land use change (LUC, i.e., horticulture, agriculture, and pasture) at the Earth's surface. By increasing global surface albedo and thus allowing more incoming solar radiation to be reflected back to space, human land use has both a cooling (biophysical) and a warming (biogeochemical) effect on the climate, with the former able to reduce the temperature sensitivity to total cumulative carbon emissions.
In this paper, we incorporated spatial human land use patterns (from HYDE 3.1) into the UVic ESCM (v. 2.9) in order to evaluate the combined biogeochemical and biophysical influence of LUC in the model on the climate response to cumulative carbon emissions for the period from 1750 to 2100. We found that the biophysical cooling in our simulations was of comparable magnitude to the biogeochemical warming over the historical period, with the result that the net temperature response to land-use change did not follow the linear TCRE relationship associated with externallyprescribed emissions. Prior to 1950 in particular, when land use emissions predominated over fossil fuel emissions, this led to a net warming response to all emissions that was substantially lower than what would be expected from the TCRE. As fossil fuels and industry became the dominant emissions source (largely for the higher emissions values reported for the 20 to 21st centuries), the slope of temperature-emissions relationship increased gradually and therefore more closely paralleled the TCRE for fossil fuel emissions alone. It is worth noting that we have included only CO 2 emissions from fossil fuel combustion in our study; further research analyzing the climate effect of non-CO 2 gases and aerosols emitted with industrial CO 2 emissions would also be useful, in order to diagnose their influence on the transient temperature response (Rogelj et al 2015), in a similar manner to what we have done here for the non-CO 2 effects of LUC.
Our simulations indicate that the biogeochemical effect of land-use CO 2 cumulative emissions is equivalent to the temperature response to cumulative fossil fuel emissions, with a consistent linear response well represented by a constant TCRE value. Furthermore, we also found that the biophysical effect of land-use change can also be well represented by a linear scaling factor, due to a strong linear relationship between albedo and net terrestrial emissions, though we did find evidence that this linearity became less robust when combined with high fossil fuel CO 2 emissions owing to the effect of enhanced land-use emissions associated with CO 2 fertilization of the terrestrial biosphere. Overall therefore, we conclude that the TCRE remains a useful metric to estimate the biogeochemical effect of cumulative land-use CO 2 emissions, but that the additional biophysical cooling should also be accounted for as this has a non-trivial effect on the overall temperature response to changing spatial patterns of human land-use.