Full effects of land use change in the representative concentration pathways

Future land use change (LUC) is an important component of the IPCC representative concentration pathways (RCPs), but in these scenarios’ radiative forcing targets the climate impact of LUC only includes greenhouse gases. However, climate effects due to physical changes of the land surface can be as large. Here we show the critical importance of including non-carbon impacts of LUC when considering the RCPs. Using an ensemble of climate model simulations with and without LUC, we show that the net climate effect is very different from the carbon-only effect. Despite opposite signs of LUC, all the RCPs assessed here have a small net warming from LUC because of varying biogeophysical effects, and in RCP4.5 the warming is outside of the expected variability. The afforestation in RCP4.5 decreases surface albedo, making the net global temperature anomaly over land around five times larger than RCPs 2.6 and 8.5, for around twice the amount of LUC. Consequent changes to circulation in RCP4.5 in turn reduce Arctic sea ice cover. The small net positive temperature effect from LUC could make RCP4.5’s universal carbon tax, which incentivizes retaining and growing forest, counter productive with respect to climate. However, there are spatial differences in the balance of impacts, and potential climate gains would need to be assessed against other environmental aims.


Introduction
The IPCC (International Panel on Climate Change) fifth assessment report assessed 21st century projections from a new range of socio-economic scenarios for climate modelling -the representative concentration pathways (RCPs). They span a range from low to high radiative forcing, corresponding to aggressive mitigation policies or business as usual (Van Vuuren et al 2011). They also include different representations of future land use change (LUC) (Hurtt et al 2011). RCP4.5 has a decrease in agricultural land (crop and pasture) by 2100, whereas RCP2.6 and RCP8.5 both have increases (see figure 1(a) and supplementary material figure 1, available at stacks.iop.org/ERL/9/114014/mmedia). This results in a reduction of forest in RCP2.6 and RCP8.5, mainly in the tropics, and an increase in forest in RCP4.5, predominantly in the mid latitudes (Hurtt et al 2011). The RCP scenarios include land carbon emissions and uptake (biogeochemical effects) in their radiative forcing targets. However, they do not account for non-greenhouse gas changes to climate, via the surface energy balance (biogeophysical Environmental Research Letters Environ. Res. Lett. 9 (2014) 114014 (7pp) doi: 10.1088/1748-9326/9/11/114014 Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. effects, e.g. albedo, evapotranspiration etc). Earth system models (ESMs), however, include both. Therefore ESMs can be used to understand the full climate effects of LUC in the RCPs, rather than only the effects from biogeochemical changes (greenhouse gases).
Biogeophysical and biogeochemical effects can have opposite signals (Pongratz et al 2010, Matthews et al 2004 and it is spatially variable which is most important (Bonan 2008). Biogeochemical effects are often stronger in the tropics because of high biomass content of tropical forests. Biogeophysical effects tend to be stronger in the temperate and high latitudes because of differences in albedo between surfaces with and without snow (Bonan 2008). Therefore the location of the LUC affects the sensitivity of the RCPs to their respective LUC. To assess the balance of biogeophysical and biogeochemical forcings from LUC in the RCPs, we use ensembles in HadGEM2-ES (Collins et al 2011) for three of the RCP simulations (Jones et al 2011) with land cover fixed in its 2005 state. This gives 'NoLUC' simulations, which can be compared to the standard 'RCP' simulations that give the biogeophysical effects of LUC in the temperature anomaly. The biogeochemical effect is inferred from the model's land carbon anomaly. The net effect of LUC is inferred from the biogeophysical and biogeochemical effects combined.

Methods
We use the Met Office Hadley Centre's coupled earth system model, HadGEM2-ES (Collins et al 2011, Martin et al 2011 which includes the MOSES2 land-surface scheme (Essery et al 2001); the TRIFFID dynamic global-vegetation model in dynamic mode (Cox 2001); the four-pool RothC soil carbon model (Jones et al 2005); the HadGEM1 atmospheric model (Martin et al 2006); interactive ocean biogeochemistry; terrestrial biogeochemistry; dust and interactive atmospheric chemistry; and aerosols. The atmosphere component contains 38 1.875°× 1.25°levels and interacts with water, energy and carbon within the land surface scheme and the dynamic Five plant functional types (broadleaf tree, needleleaf tree, C 3 and C 4 grasses and shrubs) are simulated in the model, representing broad category generalizations. In particular, the grass plant functional types also encompass crops. The agricultural fraction in HadGEM2-ES is imposed as an area where broadleaf and needleleaf trees and shrubs cannot be grown. Therefore the agricultural fraction can only be bare soil, or C 3 and C 4 grasses. Increases in agricultural fraction within a grid box are preferentially expanded into grasses, only converting trees to agricultural fraction when other plant function types are not available.
The model setup is as for the HadGEM2-ES CMIP5 simulations (Jones et al 2011) and the HadGEM2-ES setup for the LUCID simulations (Brovkin et al 2013). The simulations are concentration driven, allowing simulations with and without LUCs to be run. This also allows decoupling of the biogeochemistry from the biogeophysics, so both can be considered individually. There are four initial condition ensemble members for each of the three RCPs considered here (RCP2.6, RCP4.5 and RCP8.5), for a combination of the standard RCP (with LUC) or NoLUC (with no LUC). I.e. 3 RCPs × 2 land use scenarios × 4 ensemble members = 24 simulations in total. The ensemble members are initialized from four historical simulations that ran from 1850-2005 (see Jones et al 2011). Each simulation is run for 95 years, from 2005 to 2100. Therefore the ensemble has an 'RCP' simulation (with LUC) and a 'NoLUC' simulation without LUC.
The total agricultural fraction (cropland and pasture) remains constant (to the 2005 value) in the NoLUC simulations. For the standard RCP, the LUC is as determined by the Integrated Assessment Model scenario for that RCP. For all the NoLUC simulations, all non land-use forcings (greenhouse gas concentrations and other aerosol forcings, etc) are prescribed as for the equivalent RCP (Meinshausen et al 2011). Therefore the NoLUC simulations do not have any greenhouse gas feedbacks from LUC included (i.e. they only include the biogeophysical impacts of the LUC).
From the anomaly between the standard RCP (with LUC) and the NoLUC simulations, two LUC effects on the radiative forcing can be seen: biogeophyiscal and biogeochemical. The biogeochemical forcing from LUC arises from the additional CO 2 put into the atmosphere by the LUC, and the climate model's sensitivity to this additional radiative forcing. The additional carbon is diagnosed from the HadGEM2-ES simulated land carbon stores (soil and vegetation) and this is converted to a resulting additional global climate change using the Transient Response to Cumulative Emissions (TRCE) (Gillett et al 2013), which is an approach that has demonstrated proportionality between carbon emissions and temperature rise. This approach has been extensively used in the IPCC fifth assessment report.
The biogeophysical forcing from LUC arises from physical changes to the land surface from LUC. From our model simulations we diagnose the result of this forcing from the anomaly of the climate mean (2070-2100) of the standard RCP ensemble (i.e. including LUC) and the equivalent NoLUC simulation. The net climate forcing of LUC is not simulated explicitly, but produced from combining the global biogeochemical and biogeophysical values, with the assumption that they combine linearly. The anomalies used here are the climate mean (2070-2100) for RCP minus NoLUC. Therefore the anomalies here give the effect (biogeophysical, biogeochemical, or net) of the LUC.

Results
The mean annual temperature of the RCP4.5 ensemble (with LUC) has a small difference of 0.19 K (land only) compared to the NoLUC ensemble, that emerges around 2070 (see figure 2(a)). In contrast, the two scenarios of deforestation (RCP2.6 and RCP8.5) have ensemble mean annual global temperatures that are almost identical to their equivalent NoLUC ensembles. The largest physiological difference between the plant functional types in the model is the difference between grasses and trees, and thus the forest fraction changes gives a guide as to what scale of change should be expected. Given that the absolute change in forest fraction is twice the size in RCP4.5 than in RCP8.5 or RCP2.6 (+2.4% and −1.2% of total land area respectively) it might be supposed that the climatic changes would also be twice the size. The change in carbon conforms to this assumption, with around twice as much change in temperature from the biogeochemical changes in RCP4.5 than in RCP2.6 or RCP8.5 (see figure 2(b)). However, the biogeophysical temperature change in RCP4.5 is 9.5 times and 4.7 times the response of the RCP2.6 and RCP8.5. Hence we are seeing around 2.5 times the size of response that would be expected if the response were simply proportional to the change in forest fraction (see figure 2(b)) and supplementary material table 1).
The seasonal biogeophysical changes in temperature over land in RCP4.5 (figures 1(b) and (c)) are correlated with areas of albedo change and subsequent changes to the surface turbulent fluxes (latent and sensible heat). This is consistent with other LUC experiments that find a decrease in albedo under afforestation because the differences in the albedo of crops, shrubs and trees are large in the summer because of higher leaf areas in the growing season (Brovkin et (Bonan 2008, Claussen et al 2001, Hallgren et al 2013. However, in this model, the evapotranspirative cooling from afforestation is not sufficient to cancel out the warming effect of decreased surface albedo changes locally in the tropics. Therefore the global warming is larger than might otherwise be expected, as there is no tropical cooling from afforestation to offset the warming from afforestation in the mid to high latitudes. As well as warming over land, there are significant temperature changes over the ocean, (mainly warming) especially in the Arctic winter (December/January/February, DJF) in RCP4.5 (figure 1(c)) from sea ice changes (figure 3). Including temperature changes over the oceans, the global biogeophysical annual mean temperature response (+0.14 K) in RCP4.5 is a little lower than the land-only mean though proportionally larger than RCP2.6 and RCP8.5 (−0.01 K and −0.015 K) (see supplementary material table 1). The Arctic warming is strongly linked to changes in sea ice coverage in RCP4.5 ( figure 3(b)), and appears across all the ensemble members. The changes are driven by northward feedbacks from European and North American LUC, which have been found previously to affect the sea ice extent and temperature (Bonan et al 1992, Rogers et al 2013. The decrease in midhigh latitude land albedo in JJA results in decreases to the mean sea level pressure in that region, driving changes to circulation northward and resulting in strong warm air advection over the Arctic and resulting reductions of summer sea ice (see figure 3 and supplementary material figure 4). In DJF, the pressure anomaly reverses and the circulation changes bring warmer wetter air from the north Atlantic. This results in further sea ice loss, which further intensifies the warming, mainly through the loss of the insulating effect of sea ice in reducing heat loss from the ocean surface. This feedback between temperate forest, sea ice and albedo gives a polar amplification of the effect of the LUC, especially for RCP4.5 (see figure 3 and supplementary material figure 5).
In the Southern hemisphere there are consistent changes in sea ice extent and temperature between RCP2.6, RCP4.5 and RCP8.5, that suggest that LUC is also resulting in changes to circulation in the southern hemisphere. RCP2.6 has little LUC in the southern hemisphere, but RCP4.5 has afforestation over Uruguay and southern Brazil and RCP8.5 has increased agricultural land over Australia, southern Africa and Argentina (see figure 1(a)). Though a causal link is for the biogeophysical (as above), biogeochemical (from TRCE) and net temperature changes (biogeophysical + biogeochemical). The error bars for the biogeophysical represent the standard deviation of the whole RCP run ensemble for 2005-2100, standardized to the rolling ensemble mean. The error bars for the biogeochemical represent the whole range of resultant temperatures for TRCE values all the CMIP5 models (these are off center because HadGEM2-ES is near the top of the CMIP5 range). The net has the same error bars as the biogeophysical. The values shown here vary slightly from other quoted values due to different averaging periods. The colours used here are as used by the IPCC AR5 report, to enable a frame of reference. difficult to establish, there are some similarities to the Northern hemisphere. Circulation changes around Antarctica are seen in RCP4.5 and RCP8.5, but not RCP2.6. The spatial patterns in mean sea level pressure changes are almost exactly opposite between RCP4.5 and RCP8.5 and correlate well with changes in sea ice and 1.5 m temperature (see figures 1(b) and (c), and supplementary material figure 4). Combined with the lack of forcing and response in the southern hemisphere in RCP2.6, this suggests a possible link between the circulation and temperature changes and the southern hemisphere LUC that is different in each of the simulations and not directly proportional with the original forcing.
The carbon emissions from LUC are proportional to the changes in forest fraction in all the simulations considered here (supplementary material table 1). This is because the biogeochemical effects of LUC are not included in the NoLUC simulations, and thus the change in land carbon is primarily from the change in woody vegetation deforest/ afforested. The carbon emissions of 18 PgC and 20 PgC from LUC in RCP2.6 and RCP8.5 are equivalent to +0.04 K of warming globally ( figure 2(b)), using the HadGEM2-ES transient climate response to emissions (TRCE) of 2.1 K EgC −1 (Gillett et al 2013) (for the entire CMIP5 TRCE range, +0.01 K to +0.05 K). Cooling from the land carbon uptake of 40 PgC in RCP4.5 is about twice the size of the warming in RCP2.6 and RCP8.5, at −0.08 K, globally, (for the entire CMIP5 TRCE range, −0.03 K to −0.1 K). The change in land carbon storage between the RCP and NoLUC simulations is dependent on the total amount of change in woody plants, especially trees, rather than LUC. The implementation of LUC in the model gives a small change in plant functional type from LUC initially because the model uses up all natural grass in a grid box before converting forest to agricultural land. Since grasses and crops are physiologically identical in the model, there is no effect from LUC in a grid box until all natural grasses within a grid box have been converted to agricultural land. The implementation of LUC in ESMs is one of the biggest uncertainties in the magnitude of LUC forcing (De Noblet-Ducoudré et al 2012, Brovkin et al 2013, Pitman et al 2009, and other models which equally, or preferentially convert forest to cropland, may see higher biogeophysical and biogeochemical impacts, and may simulate the balance between them differently. It should also be noted that the land carbon changes from LUC are different between models by as much as an order of magnitude, and that HadGEM2-ES has some of the smallest land carbon changes due to LUC (Brovkin et al 2013).
The net result of the combined biogeophysical and biogeochemical LUC effect is warming in all the RCPs considered here ( figure 2(b)) despite differing signs of change in agricultural area ( figure 1(a)). The net LUC global impact is very small in RCP2.6 and RCP8.5 (+0.01 K and +0.003 K respectively) and inside of the variability. For these two scenarios, the LUC effect is dominated by the biogeochemical effect of the tropical carbon emissions from deforestation for agricultural land. In RCP4.5 the net global impact is about three to five times larger (+0.11 K) than would be expected if the temperature sensitivity to LUC were the same as RCP2.6 and RCP8.5, and is outside of the expected variability. However, all the changes in temperature are small compared to the overall climate change by 2100. The increased temperature sensitivity seen in RCP4.5 originates from the strong biogeophysical effects which mean that the biogeophysical warming is larger than the cooling from the carbon sequestered by the afforestation. The LUC in this ESM is actually quite conservative because of a low number of plant functional types in the model (meaning less changes than models where plant functional types are further distinguished), and a relatively slow regrowth time for trees. This is particularly true for RCP4.5, as shown by the much smaller tree cover change in the ESM simulation of trees than in the Integrated Assessment Model simulation. The Integrated Assessment Model that created the original RCP4.5 scenario for the CMIP5 simulations also calculates the change in tree cover, (not used in the ESM in these simulations) as well as the change in agricultural fraction (used by the ESM in these simulations) and assumes faster tree growth than the dynamic vegetation component of HadGEM2-ES. The tree cover change for RCP4.5 in HadGEM2-ES is less than half of the amount by the Integrated Assessment Model simulation (which has a 24% increase in tree cover) (Di Vittorio et al 2014). If the model used the land cover change directly from the integrated assessment model, this differential in LUC between the RCP scenarios could potentially be even larger. Overall, this suggests that the net effect of LUC in the RCPs is likely to be dependent on the balance of biogeophysical and biogeochemical impacts at particular latitudes, rather than LUC scenario itself.

Discussion and conclusions
The small net warming from LUC in RCP4.5 found here potentially makes it a perverse incentive with regards to climate. The afforestation in RCP4.5 is the result of a universal carbon tax, which counts land carbon emissions towards a country's carbon emission budgets and results in afforestation. In the absence of a universal carbon tax, there is considerably more deforestation, especially in the tropics, (which has its own climatic implications, see for instance, Jones et al 2012, Davies-Barnard et al 2014). The intention of this policy is to cost-effectively mitigate warming by sequestering carbon via afforestation (Strengers et al 2008) and prevent the loss of forest carbon, especially tropical forest (Thomson et al 2010(Thomson et al , 2011. Instead, the LUC in RCP4.5 may have the opposite effect, enhancing the warming because of the large biogeophysical effect from mid latitude afforestation. Hence the incentive (gaining carbon credits) for afforestation may be perverse (it has the opposite effect to that intended, because instead of mitigating climate change the afforestation contributes to global temperature increase). However, the avoided tropical deforestation in RCP4.5 would have major benefits in terms of preserving primary tropical forest and biodiversity (Gibson et al 2011, Thomson et al 2010, Powell and Lenton 2013. A spatially differentiated tax might be more appropriate, which did not incentivize mid-high latitude afforestation, thereby avoiding the strongest warming effects. The land carbon tax could also be reposed as an environmental tax, avoiding the sole focus on climate change. A small, uncertain, projected increase in temperature from afforestation should be balanced against the more definite benefits a universal carbon tax could offer by retaining and extending forests, such as maintaining ecosystem services (Chiabai et al 2011, Gilroy et al 2014. Our work highlights that LUC in the RCPs is not only variable between the different projections, but that the temperature effect from LUC also varies significantly. This has considerable implications for Integrated Assessment Models that create the RCP scenarios (Van Vuuren et al 2011). They urgently need to include some representation of the biogeophysical effects. Whereas carbon emissions are well mixed in the atmosphere, giving them a global impact, biogeophysical effects are much more locally focused. Therefore the more important effects from LUC are likely to be regional scale, especially in mid-high latitude areas of afforestation. However, as shown here, the biogeophysical effects of LUC can still affect the global climate in the RCPs. This is particularly the case for higher latitude land cover changes for which sea ice feedbacks may lead to disproportionately more temperature change per unit area of LUC. The implication of these strong biogeophysical LUC effects is that neither the total net emissions nor the scale of the LUC are good indicators of the resultant temperature change from LUC in the RCPs. The latitude of the LUC may be a better determinant of the LUC effect on temperature, with higher latitude changes resulting in larger local and global changes. Use of Integrated Assessment Models in collaboration with spatially resolved, process-based ESMs is required to more reliably assess the complex implications of future land use policies.