The biogeophysical effects of idealized land cover and land management changes in Earth system models

. Land cover and land management change (LCLMC) has been highlighted for its critical role in mitigation scenarios, both in terms of global mitigation and local adaptation. Yet, the climate effect of individual LCLMC options, their dependence on the background climate and the local vs. non-local responses are still poorly understood across different Earth system models (ESMs). Here we simulate the climatic effects of LCLMC using three state-of-the-art ESMs, including the Community Earth System Model (CESM), the Max Planck Institute for Meteorology Earth System Model (MPI-ESM) and the European Consortium Earth System Model (EC-EARTH). We assess the LCLMC effects using the following four idealized experiments: (i) a fully afforested world, (ii) a world fully covered by cropland, (ii) a fully afforested world with extensive wood harvesting and (iv) a full-cropland world with extensive irrigation. In these idealized sensitivity experiments, performed under present-day climate conditions, the effects of the different LCLMC strategies represent an upper bound for the potential of global mitigation and local adaptation. To disentangle the local and non-local effects from the LCLMC, a checkerboard-like LCLMC perturbation, i.e. alternating grid boxes with and without LCLMC, is applied. The local effects of deforestation on surface temperature are largely consistent across the ESMs and the observations, with a cooling in boreal latitudes and a warming in the tropics. However, the energy balance components driving the change in surface temperature show less consistency across the ESMs and the observations. Additionally, some biases exist in speciﬁc ESMs, such as a strong albedo response in CESM mid-latitudes and a soil-thawing-driven warming in boreal latitudes in EC-EARTH. The non-local effects on surface temperature are broadly consistent across ESMs for afforestation, though larger model uncertainty exists for cropland expansion. Irrigation clearly induces a cooling effect; however, the ESMs disagree whether these are mainly local or non-local effects. Wood harvesting is found to have no discernible biogeophysical effects on climate. Overall, our results underline the potential of ensemble simulations to inform decision making regarding future climate consequences of land-based mitigation and adaptation strategies.

the average over the entire simulation period. This is needed to ensure that all ESMs have the same amount of solar energy en- EC-EARTH (f). Both land cover changes are shown as an area fraction of the land cover in that grid cell. The amount of wood harvest applied in the HARV simulation as compared to the FRST simulation is shown for CESM (g) and MPI-ESM (h) in terms of intensity of harvesting (gC m -2 s -1 ). Finally the amount of irrigation is shown as expressed in a discharge (mm year -1 ) for CESM (i), MPI-ESM (j) and EC-EARTH (k). Do note that the color bar is exponential for land management change (g-k) while it is linear for land cover change (a-f).

180
For the IRR simulation, we apply the same land cover maps as in the CROP simulation, but here, the native parameterisation of each model is activated and applied at the global scale (Figure 1i-k). Although the individual implementations of the irrigation parameterisation differ, all models follow a similar logic. Once a crop suffers a certain amount of water stress (defined differently in the models, see Appendix A), this amount is replenished by applying an irrigation flux until the water stress is relieved. In CESM and EC-EARTH, no limit is imposed on water available for irrigation. In MPI-ESM however, water The amount of wood harvesting is typically a prescribed value in ESMs, often expressed as an amount of biomass carbon extracted from the PFTs. In the HARV simulation, we force the models to use the wood harvest rates specified in the CMIP6 SSP5-8.5 scenario by the end of the century (Figure 1g-h). We let the forest grow as in the FRST simulation without harvesting 190 for the first 40 years to build up biomass before prescribing the intensive wood harvest rates. For the remaining 120 years of the simulation, the harvest rates are kept constant. It should be noted that EC-EARTH did not provide this simulation. In MPI-ESM, there is no feedback implemented of this management practice to any atmospheric processes. Therefore, only CESM can be used to investigate the biogeophysical effects due to wood harvesting. The comparison of land management between CESM and MPI-ESM shows strong differences, despite using a qualitatively 205 consistent implementation across both ESMs. For wood harvesting, the spatial pattern and intensity differ notably. In CESM the wood harvesting is generally more intense locally and less homogeneous across space than in MPI-ESM (Figure 1g-h).
For irrigation the spatial extent also differs strongly between the models. Most notably, due to the simple irrigation scheme implemented in MPI-ESM (see appendix A), this model shows high irrigation amounts in the boreal latitudes while there is no irrigation occurring in CESM and EC-EARTH at these latitudes (Figure 1i-k).

Extraction of local and non-local signals
To disentangle the local and non-local effects due to LCLMC, the checkerboard approach of Winckler et al. (2017) is applied, which is described here briefly (see Winckler et al. (2017) for details). The checkerboard approach alternates LCLMC grid cells with grid cells which remain unaltered. This allows for a clean separation of local and non-local effects as the latter 215 only occur over unaltered grid cells while the grid cells where LCLMC did occur represent a combination of both local and non-local effects. In our simulations, 1 out of 2 grid cells are affected by the LCLMC and these cells are spread out in a regular checkerboard pattern. The checkerboard like LCLMC alternation is applied to all simulations except the CTL simulation. This means that for each simulation, only half of the grid cells undergo LCLMC. The remaining unchanged grid cells show the exact same land cover as the CTL simulation. The 150 year-simulation is split into 5 slices of 30 years each. To account for natural variability, we treat each slice as a member of a perturbed initial condition ensemble. A multi-year monthly mean is computed over each of these ensemble members. To extract the local and non-local signals, we subtract a land cover change member (CROP, FRST) from its corresponding CTL member. The resulting signals for grid cells where no land cover change occurred cannot be ascribed to any direct (i.e. local) land cover change effect and can therefore be ascribed entirely to non-local effects caused by LCLMC in other grid cells. We then spatially interpolate (using linear interpolation) these values to get a global 225 map of non-local effects. The differences between both ensemble members for grid cells where land cover change did occur are caused by both local and non-local effects (local effects stem from the land cover change within the grid cell, while nonlocal effects are caused by land cover change in other grid cells). Hence, these non-local effects are subtracted from the total combined effect to get a local signal. As this local signal can only be calculated over the grid cells where land cover change occurred, we again spatially interpolate this pattern to get a full global map. Finally, the local and non-local signals are summed 230 up to derive the total signal, which corresponds to the signal from an idealised global experiment without the checkerboard-like LCLMC pattern applied. The checkerboard approach is implemented to each model grid at its native resolution. Hence, grid cell sizes vary across the different ESMs. As we have five ensemble members of 30 years for each simulation, we can extract local and non-local signals for each ensemble member, which are then used as a measure of uncertainty coming from natural variability.

235
The procedure described above can be extended to land management change by using one of the land cover change simulations as a reference simulation instead of the CTL simulation. To extract the signal from irrigation expansion, the IRR simulation is compared against the CROP simulation. In case of wood harvesting, the HARV simulation is compared to the FRST simulation.

Evaluation of local signal to deforestation
The modeled responses induced by deforestation are evaluated against products from observational studies. Several studies provide global estimates of the effect of deforestation with remote sensing products (Li et al., 2015; Alkama and Cescatti, Table 2. Overview of observational products available for the different variables considered in the evaluation.(*) This data was first published in 2018 but later extended to cover a larger area in 2020, as the extended dataset is used in this study,we will refer to this dataset as DV20 from hereon. (**) Note that the sensible heat flux was obtained by the closure of the energy balance. dataset (see Table 2). The spatial extent of the observational studies varies strongly, therefore the evaluation will be performed along latitudinal bands following Meier et al. (2018) to focus on the global patterns. A description of the different observational 255 datasets used and their spatial maps are provided in appendix C.

Energy balance decomposition for changes in surface temperature
An energy balance decomposition approach is used to decompose the change in surface temperature to its driving surface processes. Here, we use this approach to understand the processes underlying the modelled effects of LCLMC. We use the 260 approach developed by Juang et al. (2007) and modified by Luyssaert et al. (2014) which has often been used in LCLM studies, notably with CLM (Akkermans et al., 2014;Hirsch et al., 2018;Thiery et al., 2017;Hauser et al., 2019;Vanderkelen et al., 2021). The energy balance equation is shown below.
Where is the surface emissivity, σ is the Stefan-Boltzmann constant (5.67 × 10 −8 W m −2 K −4 ), T s is the radiative surface 265 temperature as it is directly calculated from surface upwelling longwave radiation, α is the surface albedo, and SW in and LW in are the incoming shortwave and incoming longwave solar radiation, respectively. LHF and SHF are the latent and sensible heat flux, respectively. All fluxes are expressed in W m -2 . We take the total derivative to obtain the change in surface temperature, whereby can be assumed to be equal to 1 for the application of this equation (Juang et al., 2007;Luyssaert et al., 2014).
Here, we apply the energy balance decomposition only to the local effects derived from the LCLMC signals as these are directly linked to changes in surface properties (Winckler et al., 2017). While applying this approach, a modest imbalance of less than 0.1 W m -2 is found over all land grid cells for all different cases, indicating the general applicability of the method. This could partially be explained by opposite signs in the biases of both turbulent heat fluxes, which cancel each other out, as is likely the case over boreal latitudes for MPI-ESM and in the tropics for EC-EARTH.

310
The deforestation-induced albedo change is especially important at boreal latitudes where it dominates the overall surface temperature response (Davin and de Noblet-Ducoudre, 2010). CESM captures the observed albedo response well, except north of 40°N where it overestimates the albedo change and south of 30°S where it underestimates the albedo change ( Figure 4c).
MPI-ESM shows a similar bias in the SH. It also overestimates the brightening in the tropics and boreal latitudes following deforestation and underestimates the brightening over most mid-latitudes.

315
The bias in albedo response north of 40°N could be caused by a strong snow masking response in both ESMs, as a snow covered forest is darker than a snow covered cropland. This would also explain the strong cooling in boreal spring and summer seasons in CESM (Figure 3b,c) and the bias in annual surface temperature over the mid-latitudes ( Figure 2). In EC-EARTH the local albedo change is zero (Figure 4c), however there is a stronger non-local albedo change despite this being almost absent in other ESMs ( Figure C2). The non-local albedo change is near-zero except over boreal latitudes, where it agrees in sign with observations but strongly underestimates the magnitude. The results for CESM are in contrast to Meier et al. (2018) who showed that the previous version of CLM (CLM4.5) could reproduce the observed albedo relatively well. However the differences between our results might be due to differences in model setup as CLM was evaluated in offline mode in Meier et al. (2018) in contrast to the coupled simulations performed here.

325
The near-surface air temperature is often a preferred metric compared to the surface temperature, as it represents the perceived temperature and is considered in most policy-relevant metrics including those used to measure global warming (Arias et al., 2021). For local near-surface air temperature change, CESM and EC-EARTH show a response of similar sign to the observations in the SH and tropics. The observations diverge north of 40°N, where the DV20 dataset confirms the cooling which is simulated by CESM and MPI-ESM. In contrast, the AL16 dataset shows no temperature change, which is also the case for  EC-EARTH (Figure 4d). The near-surface air temperature in MPI-ESM is relatively insensitive to deforestation except north of 40°N as was also shown in (Winckler et al., 2019c). However, it should be considered that near-surface air temperature is a highly contested measure as its definition tends to vary strongly across different ESMs, especially over grid cells or grid cell fractions covered with tall vegetation (Boysen et al., 2020;Winckler et al., 2019c). Therefore, in the remainder of this study, we will focus on the response of LCLMC on surface temperature, while the maps for near-surface air temperature are added in 335 appendix D for reference.
Some biases exist within the evaluation approach as the modelled surface temperature does not exactly match the radiative surface temperature measured in the observational estimates. For instance, the satellite measurements have an inherent sampling bias as they only measure during cloud free conditions. Also, the different observational estimates have different 340 and often non-overlapping spatial coverage. Nevertheless, these observational studies using a diversity of approaches show a large consistency among themselves and thus can act as a benchmark for the representation of land cover change within ESMs (Winckler et al., 2019b, a).
3.2 Local and non-local effects of LCLMC on surface temperature (ii), the MPI-ESM model shows a strong decrease in annual boreal cloud cover (see Figure C5), which is especially strong well stem from internal climate variability rather than an actual response to land management change. These results imply that the biogeophysical effects of wood harvesting, as simulated here, are too weak to have a significant imprint in global and local climate conditions at the grid scale in the represented ESMs. This does not imply that the biogeophysical effects cannot play a role locally, but simply suggests that these effects are not strong enough to be discerned at the currently used grid scale level and with the process-detail of current ESMs. An analysis comparing the simulation results at the tile level (within a grid cell) would provide an alternative approach to analyse possible local effects due to wood harvesting.  (Figure 9a and b). This is in strong contrast to EC-EARTH, where an increase in latent heat flux causes a cooling which is offset by a stronger decrease in sensible heat flux (Figure 9c). This is most likely caused by over productive cropland in the tropics in EC-EARTH as was also found for grasslands in Boysen et al. (2020). The simulated decrease in sensible heat flux in MPI-ESM reduces the heat transport away from the surface, therefore amplifying the warming, while in CESM an increase in sensible heat contributes to a cooling. In MPI-ESM the tropical warming is slightly offset by an albedo increase. In all ESMs, local 445 changes in shortwave and longwave radiation increase the warming signal, however, in EC-EARTH the contribution from enhanced incoming longwave radiation is especially strong, which could indicate that atmospheric properties such as high cloud cover or atmospheric moisture have a strong influence on surface temperature in this model. In CESM over boreal latitudes, the increase in albedo dominates the surface temperature response causing a local cooling which is partly offset by a warming induced by a decrease in sensible heat flux. In MPI-ESM, this boreal albedo effect is much weaker causing no clear local cooling.

450
In EC-EARTH, the energy balance components do not explain the simulated warming over boreal latitudes, which is most likely related to the fact that EC-EARTH uses the temperature of the first whole soil layer as surface temperature. As a consequence, other processes that are not related to the surface energy balance (e.g. permafrost thawing) also affect the surface temperature in this model. Finally, contrasting to the other models, the albedo in EC-EARTH does not influence the local 455 surface temperature changes, as there is no change in local albedo (see Figure C2).
The cooling effect of albedo due to cropland expansion has a pronounced seasonal response both in both MPI-ESM and CESM (Figure 10a and b). It is most outspoken during NH spring as a consequence of the reduced snow masking effect. In

Afforestation
In the case of afforestation, all models show a reduction of the surface temperature in the SH and tropics (Figure 9d,e,f). In MPI-ESM and CESM, this is caused by the cooling effect of increasing turbulent heat fluxes, which is partly counteracted by a warming effect due to an albedo decrease. This albedo effect becomes dominant when moving northward and causes a 470 local warming in CESM starting from the mid-latitudes and in MPI-ESM starting from the boreal latitudes. In EC-EARTH, the cooling is caused by changes in latent heat flux and incoming longwave radiation, but is counteracted by a decrease in sensible heat flux. At boreal latitudes, the albedo-induced warming is partly counteracted by an increase in sensible heat flux in CESM, and by an increase in latent heat flux and a decrease in incoming shortwave radiation in MPI-ESM. The decrease in incoming shortwave radiation might be caused by an afforestation-induced local increase in cloud cover (as shown in Figure C10). This cloud cover, with CESM even showing a slight decrease in cloudiness over boreal latitudes ( Figure C10).
The albedo-induced effect of afforestation has a clear seasonal peak during NH spring for both MPI-ESM and CESM (Figure 10c and d). The turbulent heat fluxes seem to follow a similar seasonality. This indicates that extra-tropical afforestation is dominating the global picture for these models due to a strong albedo response largely counteracted by the turbulent heat fluxes. In EC-EARTH, a similar seasonal pattern is visible with larger fluxes in NH summer and smaller fluxes in NH winter as was also the case for cropland expansion. Overall, all models show limited local effects due to afforestation, being quasi 0 K 485 in CESM, -0.15 K in MPI-ESM and 0.2 K in EC-EARTH. Although we have harmonised the land cover and management representation across the different models, strong differences remain, most notably in the implementation of irrigation expansion and afforestation (Figure 1). This implies that the compar-525 ison of the different simulations across ESMs is not perfect and inconsistencies can be caused by disparity in model structure and by spatial differences and differences in extent of the applied LCLMC. As for afforestation, the differences found here were mainly caused by the technical difficulty of implementing this in the dynamic vegetation model LPJ-GUESS used in EC-EARTH. However, the differences regarding land management are a direct consequence of these implementations being fairly recent in the various ESMs. There is no consistency in the implementation approach for land managements such as 530 irrigation expansion across ESMs, as was also the case in the early land cover change inter-comparison projects (De Noblet-Ducoudré et al., 2012). Over the last decade several improvements were made regarding land cover change to make the ESMs more consistent. For example, using common datasets (Hurtt et al., 2020) and common simulation protocols like the LUMIP experiments under CMIP6 (Lawrence et al., 2016). The same issues that ESMs faced before for land cover change are now apparent for land management change as well. As more ESMs are implementing land management change (Blyth et al., 2021), 535 it is crucial that common datasets and simulation protocols are set up in order to ensure comparability across the various ESMs.

Irrigation expansion
However, despite these limitations our results show that there remain similarities in the LCLMC response in the different ESMs, most notably regarding the local effects. A consensus is emerging regarding the local effects of deforestation/afforestation with a clear cooling/warming at boreal latitudes and a warming/cooling in the tropics, as is in line with 540 observational evidence. The cooling potential of irrigation (both local and non-local) is confirmed by both MPI-ESM and CESM. However, more research is needed to understand the full implications of these biogeophysical effects. The cooling effects induced by irrigation might be offset by the increased humidity and overall induce an increase in heat stress (Mishra et al., 2020). The effects on warm and cold extremes remain to be investigated as well, but lie beyond the scope of the current study.
Our results highlight the importance of including possible local biogeophysical effects in future land-use and land management policies. The current policies underpinning large-scale climate mitigation plans such as the European Green Deal are set up to only take into account the biogeochemical effects of LCLMC strategies such as afforestation. The European Green Deal plans (European Commission, 2020) rely heavily on afforestation as a possible negative emission technology to enhance the 550 land sink by planning to plant up to 3 billion trees within the EU. However, beyond the positive consequences of afforestation on carbon storage, its biogeophysical effects should also be considered in order to plan for (or avoid) side-effects for regional temperature induced by local processes (as shown in Figure 6a,e,i). The local biogeophysical effects imply some positive sideeffects over specific regions, such as the tropics and mid-latitudes, especially during the summer season; however they could also imply some negative side-effects over the boreal latitudes and part of the mid-latitudes during the winter season. These Irrigation clearly decreases temperature in both CESM and MPI-ESM, which tends to constitute another indication that irrigation could serve as an adaptation strategy. However, it remains unclear whether the irrigation induced cooling is predominantly local (induced by turbulent heat fluxes) or non-local (induced by cloud effects), and what the combined effect is of 575 irrigation-induced changes in temperature and humidity patterns on heat stress. Nevertheless, these results help assess the future climate consequences of irrigation expansion. Irrigation has been projected to increase in the future as a means to increase agricultural productivity (van Maanen et al., 2021;Rosa et al., 2020) but it may also aggravate future water stress (Haddeland et al., 2014). It should be noted that irrigation is implemented in a highly idealised way in these simulations, with 2 out of 3 ESMs not being constrained by water limitations. These water limitations should be assessed before irrigation expansion can 580 be considered as a viable adaptation option in any region.
Overall, our results show that future land-based mitigation strategies will need to consider the non-local biogeophysical consequences of LCLMC patterns, as large scale afforestation is a key strategy in intensive land-based mitigation scenarios (Smith et al., 2015;Humpenöder et al., 2014), especially in those compatible with a 1.5 K world (Roe et al., 2019). In particular, 585 the robust non-local biogeophysical warming from global afforestation presented in this study indicates that future land-based mitigation strategies would lead to an even more extensive unintended warming than the local biogeophysical warming that has been widely reported for boreal regions and the mid-latitudes in winter. More research is needed to the bridge knowledge gaps regarding which regions would be mostly responsible for this non-local warming if afforested and what would be the magnitude of this warming in realistic afforestation scenarios.

Limitations and outlook
The idealised simulations performed in this study give an overview of the potential biogeophysical effects from LCLMC. We were able to separate local and non-local effects due to the application of a checkerboard like LCLMC perturbation to our idealised land cover maps (Figure 1). The local effects are only caused by changes occurring within the grid cell. Hence, they 595 represent the most extreme possible outcome of the application of a certain LCLMC within that single grid cell, without accounting for other LCLMC around the globe. In contrast, the non-local signals are a compound response caused by the LCLMC around the globe. These represent an underestimate in magnitude of the non-local effects in a simulation of global LCLMC, as due to the checkerboard pattern, non-local effects are the consequence of LCLMC applied to only half of the grid cells around the globe. It should be noted that the application of the checkerboard approach has some methodological implications, as the 600 resulting local and non-local signals intrinsically contain an interpolation error. Although we tried to minimise this error by using a checkerboard pattern of 1 out of 2 grid cells, this error can still reach up to 0.3 K based on previous simulations with MPI-ESM (Winckler et al., 2017). Moreover, the approach has limitations due to the size of a grid cell in the different ESMs.
The land cover change needed to get a local effect as presented here remains highly unrealistic (around 100 km). As ESMs are becoming computationally more efficient and their resolution gets increased, the validity of this assumption could be tested showed that Amazonian deforestation could induce a drying of the region (Lejeune et al., 2015). The local effects diagnosed from these extreme sensitivity experiments could also be used as training data for less computationally expensive statistical 615 models to emulate biogeophysical effects arising from less extreme and more realistic LCLMC scenarios. Overall, we hope that the results of the simulations presented here can help increase the present understanding of LCLMC and build towards a framework that facilitates the inclusion of biogeophysical effects of LCLMC in future policy frameworks.

620
In this study, we showed the first results of a new slate of fully-coupled ESM simulations within a multi-model framework For afforestation, a clear tropical cooling is consistent across ESMs. The non-local effects carry more uncertainty which may be due to a wider variety of mechanisms at play and due to the strong natural variability intrinsic to atmospheric processes.
However, this would require further investigation to be confirmed. All ESMs show a strong non-local warming as a consequence of large-scale afforestation. Irrigation expansion cools the climate both through local and non-local effects, although 640 the contribution of local and non-local effects to this cooling is inconsistent across ESMs. Finally, the effect of extensive wood harvesting is shown to be too small to have a clear imprint on the grid-scale climate.
The driving processes underlying the local surface temperature effects were analysed using an energy balance decomposition technique. The local surface temperature effects of land-cover change (both cropland expansion and afforestation) are that are not related to the surface energy balance, such as permafrost thawing. Moreover, the strong influence of incoming longwave radiation indicates that atmospheric properties (such as cloud cover and moisture content) are strongly related to local surface temperature changes. Both CESM and MPI-ESM agree that the main local surface temperature response due to irrigation is driven by a strong increase in latent heat flux which is only partly counteracted by a decrease in sensible heat flux.

655
Overall, our results confirm that the biogeophysical effects of LCLMC are an important factor to consider in future land planning strategies, especially as they reveal the robust importance of non-local climate responses in the context of mitigation potential of land cover change. In the case of large-scale afforestation specifically, the non-local response could lead to globalscale unintended warming, in particular over the boreal and mid-latitude regions. separation of the 3 ESMs, the evaluation and the energy balance decomposition can be found on the github page of the hydrology department of VUB (). The simulation data used in this paper will be made available online upon publication of this paper (URL/DOI to be added). to field capacity if field capacity was not reached and if enough irrigation water is available in storage.
-irrigation water is stored each time step when the reservoir drops below 0.2 m and filled up with all available water from (surface) runoff and drainage during that time step. CESM:

670
-Irrigation is applied daily at the first timestep after 6AM only when the soil moisture over all soil layers containing roots falls below a defined target soil moisture which is defined in order to match present day irrigation. If soil moisture falls below the target soil moisture it is replenished until at the target soil moisture level.
-The water needed for applying irrigation is taken from river water storage, however when this is inadequate to meet water demand it can also be subtracted from the ocean model, therefore no real water availability limit is applied within CLM.

675
-Irrigation is only applied when the crop leaf area >0, i.e. this means that crops are only irrigated when they are in there vegetation state (during the growing season).

EC-EARTH:
-In LPJ-GUESS the amount of irrigation is the deficit a crop plant is experiencing. So if a crop needs an additional amount of 680 water, it is added to the top of the soil column.
-The water comes from nowhere (i.e. unlimited water source).
-The water flux is not communicated to IFS, i.e. irrigation does not affect the surface water fluxes within the atmosphere. The only effect is that an irrigated crop would have a higher leaf area index and cover fraction compared to a non-irrigated crop of the same type.

685
Appendix B: Surface temperature in observational datasets The comparison of the ESM data and the different observational datasets has some inconsistencies as was already described before by Winckler et al. (2019a). From Figure A1 it is apparent that the different datasets do not have the same spatial coverage. Besides this the calculation of the temperature signal differs across studies. In Alkama and Cescatti (2016) the observed 690 signal is extracted by looking at changes over time in contrast to the other studies where this was extracted by comparing after the year 2000 (hence representing present day conditions)and the total duration each estimate is based on are similar. All studies provide an estimate of the response of surface temperature to a full deforestation except Alkama and Cescatti (2016) where actual deforestation was considered and which had to be converted to a full deforestation signal by weighting with the deforestation fraction, in order to get robust results only grid cells where selected where more than 1% of actual deforestation had occurred over the analysis period considered. For Bright et al. (2017) only data was provided for conversions from specific 700 forest species, to allow for a consistent comparison to the ESMs these values had to be weighted using the weights of each forest PFT within the specific ESMs. Therefore, an estimate of the Bright et al. (2017) data was created representing the different ESMs there PFT distributions, however, these differed only slightly so an average was taken over all estimates to be compared across all ESMs.

705
For the creation of the evaluation plots, the signals from the different datasets was calculated over all grid cells where data was available as most have a sufficient amount of grid cells in each latitudinal band. Each dataset was retained at its original resolution for the calculation of the latitudinal averages in order to avoid interpolation errors. The observational data could be directly compared to the output from the CROP-FRST signal separated data as in most grid cells almost a full deforestation occurs as is shown in Figure B1. The corresponding maps showing the local, non-local and total surface temperature effects 710 are shown in Figure B2.