Land cover change in low-warming scenarios may enhance the climate role of secondary organic aerosols

Most socioeconomic pathways compatible with the aims of the Paris Agreement include large changes to land use and land cover. The associated vegetation changes can interact with the atmosphere and climate through numerous mechanisms. One of these is emissions of biogenic volatile organic compounds (BVOCs), which may lead to the formation of secondary organic aerosols (SOAs) and atmospheric chemistry changes. Here, we use a modeling framework to explore potential future global and regional changes in SOA and tropospheric ozone following idealized, large-scale vegetation perturbations, and their resulting radiative forcing (RF). Guided by projections in low-warming scenarios, we modify crop and forest cover, separately, and in concurrence with changes in anthropogenic emissions and CO2 level. We estimate that increasing global forest cover by 30% gives a 37% higher global SOA burden, with a resulting forcing of −0.13 W m−2. The effect on tropospheric ozone is relatively small. Large SOA burden changes of up to 48% are simulated for South America and Sub-Saharan Africa. Conversely, increasing crop cover at the expense of tropical forest, yields similar changes but of opposite sign. The magnitude of these changes is strongly affected by the concurrent evolution of anthropogenic emissions. Our land cover perturbations are representative of energy crop expansion and afforestation, two key mitigation measures in 1.5 °C compatible scenarios. Our results hence indicate that depending on the role of these two in the underlying mitigation strategies, scenarios with similar long-term global temperature levels could lead to opposite effects on SOA. Combined with the complexity of factors that control SOA, this highlights the importance of including BVOC effects in further studies and assessments of climate and air quality mitigation involving the land surface.


Introduction
Many mitigation pathways compatible with the Paris Agreement ambition to limit global warming to well below 2 • C above pre-industrial levels rely heavily on negative emission technologies, such as reforestation, afforestation and large-scale deployment of bioenergy with carbon capture and storage (Harper et al 2018, Rogelj 2018, Roe et al 2019. This in turn will entail significant changes to vegetation and other features of the terrestrial ecosystem (Popp et al 2017, Doelman et al 2018, that may feed back on the atmosphere and climate through a range of other physical, chemical and biological processes. Land cover change (LCC) can therefore have substantial impacts on the atmospheric energy budget and the hydrological cycle, on local to global scales (e.g. Mahmood et al 2014), sometimes with strong regional heterogeneity. For instance, loss of boreal forest cover has a cooling effect due to enhanced albedo, while tropical deforestation reduces evapotranspiration and increases sensible heat fluxes, resulting in surface warming (Bala et al 2007, Alkama andCescatti 2016). Studies suggest that historical LCC has induced a net cooling impact through biogeophysical processes, which, although smaller than the impact of land-use CO 2 emissions over the same period on a global scale, has substantial influence on regional climate features, including climate extremes (Pongratz et al 2010, Ghimire et al 2014, Lejeune et al 2017, Hirsch et al 2018.
In this study we focus on a possibly highly important, but often less well explored, mechanism of land-climate interactions: the emission of chemically reactive biogenic volatile organic compounds (BVOCs) and subsequent impacts on the atmospheric composition of aerosols and trace gases. Once in the atmosphere, BVOCs, of which isoprene is the dominating species, may affect oxidation chemistry, and hence tropospheric concentrations of ozone, a potent greenhouse gas and harmful air pollutant. The BVOC oxidation products also react to form secondary organic aerosol (SOA) (Claeys et al 2004, Kanakidou et al 2005, Carslaw et al 2010. SOA plays a marked role in the climate system through its ability to scatter and absorb solar radiation and influence the concentration of cloud condensation nuclei (CCN) through new particle formation or particle growth. Earlier studies have suggested that organic aerosol (OA), including SOA, are comparable in abundance to sulfate aerosol in the Northern Hemisphere and that SOA may be responsible for more than half the continental CCN (Jimenez et al 2009, D'Andrea et al 2013. However, despite significant recent advances, the SOA budget, and its contribution to the spread in indirect aerosol forcing estimates, remain uncertain (Carslaw et al 2013, Rap et al 2013, Shrivastava et al 2017, Zhu et al 2019, Sporre et al 2020. The processes that control the terrestrial source of biogenic SOA are highly sensitive to climate, land cover and other environmental factors. For instance, rising surface temperatures drive increases in BVOC emissions and SOA loads, leading to negative radiative forcing (RF) and a cooling effect on the climate. This feedback mechanism has been explored through both modeling and observational studies (Paasonen et al 2013, Scott et al 2018a, Thornhill et al 2021. BVOC emission rates are also sensitive to the background CO 2 concentration, with most studies pointing to an isoprene inhibition effect with increasing CO 2 (Arneth et al 2007, Pacifico et al 2009, Tai et al 2013. Rap et al (2018) used a modeling approach to show that there is a strong positive ecosystem feedback between global BVOC emissions and plant productivity via diffuse radiation fertilization. Studies have also demonstrated significant variability in isoprene emissions due to environmental change over historical periods (Unger 2013, Chen et al 2018, and that changes in the abundance of SOA and ozone can make notable contributions to the net climate impact of historical LCC (Unger 2014, Scott et al 2017. Future changes in BVOC emissions and SOA are expected in response to warming and increased conversion of land, although the magnitude of the response varies significantly depending on scenario and methodological framework ( Many of the recent emission and socioeconomic pathways, the shared socioeconomic pathways (SSPs) (O'Neill et al 2014), project large land cover/land-use conversion, particularly following extensive use of land-based climate mitigation measures. However, the characteristics of the projected LCC can differ substantially even in scenarios with the same global temperature level, depending on the chosen mitigation strategy and the balance between afforestation/reforestation and expansion of bioenergy crops (Bertram et al 2018, Huppman et al 2018. Concurrently, global emissions of anthropogenic aerosols are projected to decline strongly, while background CO 2 concentrations will continue to increase towards the mid-21st century even in 1.5 • C compatible scenarios. The magnitude and sign of the consequent change in BVOC emissions (affecting SOA and ozone concentrations) in these scenarios are poorly quantified, but potentially of high relevance for global and regional climate.
In the following, we explore effects on global and regional atmospheric composition and energy balance resulting from contrasting, large-scale vegetation perturbations guided by total LCC in recent lowwarming scenarios. We perform a series of model experiments with a state-of-the-art global chemicaltransport model (CTM). Alternative land cover maps are constructed, keeping the land cover perturbations idealized with the aim of spanning a broad range of possible evolution across different regions. The resulting effect on BVOC emissions, SOA and ozone is simulated, delineating the impact of LLC from the influence of changing anthropogenic emissions, and the associated impact on the energy balance quantified.

Methods
We construct three main perturbations to large-scale vegetation, as well as a range of additional sensitivity experiments. The changes in SOA and tropospheric ozone caused by these perturbations are explored using the OsloCTM3 (Søvde et al 2012), which is driven by meteorological data from the European Center for Medium Range Weather Forecast (ECMWF) open integrated forecasting system (OpenIFS) and uses the 2.25 • × 2.25 • horizontal resolution with 60 vertical layers. The OsloCTM3 includes comprehensive treatment of tropospheric and stratospheric chemistry Isaksen 1997, Søvde et al 2012), as well as modules for the main aerosol species (Lund et al 2018). The SOA parameterization, based on Chung and Seinfeld (2002), is documented by Hoyle et al (2007). A brief overview is given in the SI (available online at stacks.iop.org/ ERL/16/104031/mmedia). While the standard model configuration relies on offline data sets for BVOC emissions, we have here implemented the parameterization from the Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1) (Guenther et al 2012) in the OsloCTM3. MEGAN is driven by input of weather and land cover data and uses mechanistic algorithms to represent the major known processes that control biogenic emissions. We use ECMWF OpenIFS meteorological data, leaf area index derived from global inventory modeling and mapping studies normalized difference vegetation index (Zhu et al 2013), figure S1, and the plant function type (PFT) distribution from the Community Land Model version 4 (Lawrence et al 2011) combined with the PFT specific emission factors from MEGAN2.1. Anthropogenic and biomass burning emissions are taken from the Community Emission Data System (van Marle et al 2017, Hoesly et al 2018. The RF of the changes in SOA and tropospheric ozone is calculated offline using two different radiative transfer models; SOCRATES (Edwards and Slingo 1996, Rap et al 2013) and Oslo-RTM (Myhre et al 2017). In SOCRATES, the forcing due to aerosol-radiation interactions (RFari) is estimated using the CLASSIC aerosol scheme (Bellouin et al 2011), while stratospherically-adjusted ozone RF is calculated using the fixed dynamic heating approximation (Rap et al 2015). RFari is also calculated with the Oslo-RTM using a multi-stream model with the discrete ordinate method (Stamnes et al 1988). Biogenic SOA is assumed to be purely scattering. Additionally, the Oslo-RTM is used to derive an estimate of the RF of aerosol-cloud interactions (RFaci), using the method by Quaas et al (2006) to account for the aerosol-induced change in cloud droplet concentration, which alter the cloud effective radius and thus the optical properties of the clouds. This method has been applied in earlier studies (e.g. Myhre et al 2017).
We perform a baseline, present-day (year 2014) run and a set of sensitivity experiments, modifying the vegetation cover but with fixed 2014 meteorological data (table 1). Each simulation is 1 year, with 6 months spin-up. Three experiments are performed where the total global forest or crop cover is increased by 30% compared to the baseline (with corresponding decrease in other PFTs). This number is based on end-of-century LCCs from presentday in two scenarios from the IAMC 1.5 • C Scenario Explorer hosted by IIASA (Huppman et al 2018), the 1p5C_Sust and 1p5C_early (figure S2), which have the same level of global warming but different underlying policies and assumptions (Bertram et al 2018). We assume instantaneous changes, i.e. we do not treat the dynamical transition to a new land cover distribution, but explore the role of biogenic SOA and ozone in this new state. Additionally, we perform experiments with a 15% increase in global cover forest and crop, more representative of mid-century changes, and an extreme case with a 50% increase.
Our updated vegetation maps are produced as follows: (a) in each grid cell where forest currently exist, the percentage cover is increased by the respective amount (15%, 30% or 50%) (up to a maximum of 100%); (b) the amount of crop and grass/shrub is then reduced by a corresponding amount. In a similar way, for an increase in crop cover, we reduce the forest and grass/shrub cover). In reality, not all grassland or forested areas are suited for agriculture, nor can forests be planted everywhere. However, here we aim to keep the perturbations idealized and hence primarily retain the present-day geographical land cover distribution. To test the effect of this constraint, we also perform an experiment where the 30% increase in crop cover only takes place at the expense of tropical forest (i.e. equivalent to a tropical deforestation case). Figure 1 shows the baseline distribution of forests, crop, and shrub/grass and the pattern of vegetation change in the 30% forest increase, 30% crop increase and deforestation experiments, while table S1 gives total cover for each of the three land cover types in each experiment. For two of the vegetation perturbations, 30% forest and crop increase, we perform additional simulations where anthropogenic emissions of aerosols and precursor gases and background CO 2 concentrations are set to levels corresponding to a low warming scenario. CO 2 is not treated explicitly by the OsloCTM3 but is included as a parameter in the parameterization of isoprene emissions MEGAN2.1. Here we select the SSP1-1.9 ( van Vuuren et al 2017, Gidden et al 2019 and employ the year 2050 emissions and global CO 2 concentrations from this scenario (see figure S3 for selected species).

Results
In the following, we describe the global and regional effects of the large-scale vegetation perturbations on simulated isoprene, ozone and biogenic SOA, and the resulting RF.
The OsloCTM3 baseline BVOC emissions and SOA and ozone distributions (figure S4) are broadly consistent with model estimates in the literature (see section SI1 for details). Increasing the global forest cover by 15%, 30% or 50% from the baseline (i.e. experiments Forest15, Forest30 or Forest50) increases the global total isoprene emission by 14%-51% (table 2). Conversely, increasing crop cover at the expense of forest, decreases isoprene emissions, as crops have much lower emission factors. We note that for the same percentage global increase in cover, the absolute changes in total forest and crop area will be very different, due to their different baseline extents (table S1). In particular, a much smaller reduction in forest area is needed to match the respective percentage crop increases compared to the changes in area needed in the corresponding forest increase experiments. The result is changes in total isoprene emissions that are smaller in magnitude in the crop experiments than in the forest experiments.
With higher emissions of isoprene and other BVOCs, the formation of oxidation products and, subsequently, SOA, increases. The global-mean SOA burden is 18%, 37%, and 65% higher in the Forest15, Forest30 and Forest50 experiments, respectively, than in the baseline. In Crop15, Crop30, and Crop50, we find a global-mean reduction in SOA burden of 6%, 12% and 18% following lower BVOC emissions. In Deforest, where the entire 30% increase in global crop cover takes place at the expense of tropical forests with high isoprene emissions, the SOA reduction increases to 20%. In general, the changes in isoprene emissions scale with the magnitude of global change of the respective vegetation type. Hence, we focus on the experiments with 30% increase in the following.  While the size of the forest cover change in northern hemisphere temperate and boreal regions is similar (or larger) to those in the tropics, their impact on isoprene emissions and SOA burden is smaller than in the tropics. This is as expected given the lower BVOC emission rate and stronger seasonal cycle at higher latitudes. In Forest30, the largest tropospheric ozone increase is seen between 20 • N and 50 • N, while a reduction occurs over tropical regions (see also figure  S5). The opposite pattern is found for Crop30 and Deforest. Over Africa and South America (SAM), ozone also exhibits opposite seasonal responses (not shown). The influence on ozone involves complex chemical processes, with dependence on e.g. background NO x levels and meteorological conditions. Ozone also has a longer atmospheric lifetime than SOA and atmospheric transport may also influence the simulated changes. The spatial pattern of tropospheric ozone changes therefore shows less direct relationship with the LCC than SOA. We note, however, that the tropospheric ozone changes are relatively small everywhere.
Over key regions (figure 2(d)), such as SAM, Asia (ASIA) and Sub-Saharan Africa, the estimated SOA burdens are 1.1-2.4 mg m −2 (i.e. 15%-50%) larger in the Forest30 experiment than the baseline (figure 2(e)). Absolute changes are smaller in higher latitude regions, but still constitute notable relative increases compared to the baseline (31%, 45% and 33% over North America, Europe, and Russia, respectively). In SAF and SAM, the increase in biogenic SOA translates to approximately 15% larger total aerosol load (total here being the sum of fine mode below particles, excluding dust and sea salt). Corresponding numbers for other regions, where SOA contributes less to total aerosol, are 3%-10%. In Crop30, the lower SOA burden translates to 1%-6% lower total aerosol burden, depending on region.
Hence, large scale land cover perturbations, especially in the form of increased forest cover, can have notable impacts on the overall aerosol pollution levels. For the tropospheric ozone column, the differences between the baseline and vegetation perturbation experiments are less than than 1% (figure S6). These results assume anthropogenic aerosol and precursor emissions at the present-day level.
Using year 2050 SSP1-1.9 anthropogenic emissions instead (Forest30_SSP119anthro and Crop30_SSP119anthro experiments) has a notable influence on our results. In the case of the Forest30_SSP119anthro and Forest30 experiments, the latitudinal pattern of SOA burden change remains similar in, however, the magnitude of the increase is substantially reduced ( figure S5). This is due to the projected decline in primary organic aerosol emissions (figure S3), and hence fewer particles for oxidation products to condense on. In all regions considered here, except SAM and SAF, this effect offsets the impact of the LCC alone, resulting lower SOA burdens in Forest30_SSP119anthro compared to the baseline (figure 2(e), dark green bars). In SAM and SAF, the SOA burden still increases, but less so than if forest increase is the only change taking place. Using year 2050 SSP1-1.9 CO 2 concentration further offsets the SOA burden increase (figure 2(e), light green bars), due to the isoprene inhibition effect (Arneth et al 2007). In the case of crop increase, the lower anthropogenic emission enhances the SOA burden reductions already estimated due to the LCC (figure 2(e), dark red bars). These results demonstrates the spatial heterogeneity and complexity of the factors that influence future pathways for SOA. For tropospheric ozone, the change in anthropogenic precursor emissions dominates and result in up to 25% lower concentrations compared to the baseline (figures S5 and S6).
As a 1st order estimate of the climate impact of our large-scale land cover perturbations, we calculate the RF of the changes in SOA and tropospheric ozone (figures 3 and S6). SOA is predominantly scattering and increasing concentrations thus exerts a negative  RF. We estimate a global, annual mean net (i.e. sum of RFari and RFaci) SOA RF of −0.13 W m −2 for the Forest30 experiment. For Crop30 and Deforest, where aerosol loads are lower than the baseline, we estimate positive RFs of +0.04 and +0.08 W m −2 , respectively. Of this, RFaci constitutes 23%-28%, depending on the experiment. The SOCRATES RFari estimate is 40% stronger than that from Oslo-RTM (figure S7), however, their regional patterns are very similar. The strongest regional net RF in the Forest30 experiment is found over SAF (−0.54 W m −2 ), followed by SAM (−0.32 W m −2 ) and ASIA (−0.23 W m −2 ) (figure 2, table S2). In the Deforest experiment, we estimate the strongest positive RF of 0.35 W m −2 over SAF. The radiative efficiency, i.e. RF per SOA burden change, is similar across regions, although somewhat stronger over SAM for RFaci. There are notable regional differences in both sign and magnitude of the ozone RF, in particular in Forest30, where the RF is negative over the tropics and positive over the northern hemisphere ( figure S6). Globally, these regional contributions lead to a relatively small tropospheric ozone RF, i.e. 0.001, −0.0017 and −0.0019 W m −2 in Forest30, Crop30 and Deforest, respectively.

Discussion
Both the present study and previous literature demonstrate that LCC can have important climate impacts through atmospheric chemistry, and emphasize the importance of carefully considering both the characteristics of the LCC in the underlying scenario and the anthropogenic emission trend. While a comprehensive comparison with previous studies is challenging due to differences in methodology and scenarios, the overall responses to LCC in our study are in good agreement with existing literature. Combining satellite retrievals with modeling, Chen et al (2018) estimated isoprene emission increases of 5%-10% in response to recent forest cover changes in China and India that are comparable to that in our Forest15 experiment. Using the previous generation global scenarios, the representative concentration pathways (RCPs), Ward et al (2014) estimated end-of-century changes in BVOCs attributed to LULCC activities of 6%-16% depending on scenario. Scott et al (2018b) suggested that changes in atmospheric chemistry constitute 8% of the net climate LCC impact in an extreme deforestation case. Hantson et al (2017) found globally increasing isoprene emissions under RCP4.5 where substantial reforestation efforts were projected, but a decreasing trend in RCP8.5 and RCP2.6 due to a higher degree of deforestation and bioenergy demand, respectively. SOA formation depends on the background OA abundance. Using an older scenario with increasing anthropogenic emissions, Heald et al (2008) projected an increase in global SOA, whereas the strong emission decline projected in recent scenarios give the opposite effect in our study. Biomass burning sources of OA are more uncertain and difficult to project. Additionally, background aerosol levels affect the cloud albedo effect of biogenic SOA .
Here we have focused on one specific vegetationclimate interaction process. LCC will also influence the energy balance and climate through other biogeophysical and biogeochemical effects such as albedo and evapotranspiration. For instance, looking at historical cropland expansion, Unger (2014) found the global mean RF of albedo changes to be comparable in magnitude to the net effect of LCC-induced changes in atmospheric composition. Moreover, existing literature indicate that LCC may have substantial further local to regional-scale climate implications beyond energy balance perturbations. Studies of LCC, including changes of comparable magnitude to those in our experiments, find important effects on both regional mean and diurnal temperature range, as well as temperature extremes (Alkama and Cescatti 2016, Lejeune et al 2017, Hirsch et al 2018. These responses cannot be quantified within our current framework but need to be considered for comprehensive assessments of the full consequences of land management and mitigation strategies. The main motivation behind our study is to explore the impact of large-scale and constrasting LCC and we have thus kept our perturbations idealized and spanning a wide range. To place our numbers into context, our 30% increase in global forest cover corresponds to nearly 1500 Mha (table S1), roughly 3.5 times the extent of forests reported lost through conversion since 1990 (FAO/UNEP 2020). Furthermore, global cropland area is estimated to have increased fivefold (1200 Mha) between 1700and 1990(Goldewijk 2001 and by 15% since the 1960s (Arneth et al 2019). The latter is similar to our smallest crop perturbation experiment, albeit with regional differences. There is a wide spread in projected future global LCC and temporal evolution in available scenarios. In 89 scenarios from the IAMC 1.5 • C Scenario Explorer that reach 1.5 • C, including with overshoot (Huppman et al 2018), changes in total forest (crop) cover from 2010 to 2100 range from a decrease of 240 Mha (550 Mha) to an increase of more than 1700 Mha (900 Mha). With the exception of the Forest50 experiment, individual perturbations in our study fall within these end-of-century ranges.
Furthermore, our experiments are run with present-day meteorology to disentangle the influence of LCC alone. Research has suggested a small but non-negligible warming-induced effect on biogenic emission rates (Unger 2013, Hantson et al 2017. For instance, Unger (2013) estimated a 3% increase in global isoprene emissions due to physical climate change alone from 1880 to 2000. Accounting for projected temperature increases could hence increase the BVOC change in our forest increase experiments or, conversely, reduce the decline in the crop experiments. Moreover, changes in vegetation may themselves affect local temperature and hence emissions. Other influences that can alter the vegetation distribution, and hence BVOCs, include wildfire activity, active species management and changing vegetation in response to climate change. Physical climate change can also alter the distribution of aerosols and ozone via e.g. removal rates and oxidation capacity (Heald et al 2008), as well as influence the indirect aerosol effect through changes in clouds. Finally, our modeling framework does no enable us to study feedbacks on plant productivity from changes in SOA and ozone through diffuse radiation or deposition (Rap et al 2018). Thus, further studies of LCC in lowwarming scenarios are needed to fully understand the interactions with the natural aerosol system and subsequent climate implications.
SOA formation involves complex and interdependent processes with non-linear effects, many still poorly understood (Shrivastava et al 2017). For instance, the fraction of BVOCs that is transformed to SOA is suggested to be overestimated in global models due to the coarse resolution and subsequent insufficient level of detail of land surface characteristics and oxidant concentrations. Moreover, not all known oxidation pathways or biogenic precursors are represented in current models, potentially resulting in an underestimation of SOA formation. Finally, the SOA yield has been shown to vary between high and low NO x environments (Hoyle et al 2011). NO x -dependent yields are not included in the Oslo-CTM3 but could affect the SOA response to LCC under rapidly declining anthropogenic emissions. Combined with the potential changes in SOA that may play out over the coming decades, this underlines the need for further research in this area.

Conclusions
The land surface interacts with the atmosphere and climate through a multitude of mechanisms that need to be quantified to assess implications of land-based climate mitigation. Here we have studied one such mechanism: effects of large-scale LCC in line with 1.5 • C warming scenarios on BVOC emissions and abundances of SOA and tropospheric ozone, and the associated RF.
By imposing idealized, contrasting vegetation cover perturbations in a modeling framework, we have demonstrated that LCC, e.g. from afforestation and reforestation or crop expansion, can have substantial regional and global effects on atmospheric composition and energy balance through biosphere-atmosphere interactions involving natural aerosols. We estimate that a 30% increase in global forest cover (corresponding to approximately 1500 Mha), imposed under current climate conditions, atmospheric composition and background emissions, results in 25%-48% higher regional SOA burdens, with the largest changes in SAM and sub-Saharan Africa. In our model, this corresponds to up to 15% increase in total regional biogenic plus anthropogenic aerosol levels. Furthermore, we estimate that these regional SOA changes induce regional RF ranging from −0.09 to −0.54 W m −2 . Conversely, increasing global crop cover at the expense of forests and/or grassland reduces BVOCs emissions and SOA burden. For a 30% increase of cropland in tropical regions (460 Mha), we estimate up to 30% lower regional SOA burdens.
Concurrently with LCCs, significant declines in anthropogenic emissions of aerosols and ozone precursors are projected in many low warming scenarios. We find the subsequent reduction in particles for condensation of BVOC oxidation products significantly affects SOA production and burden, in some regions offsetting the effect of increased forest cover on burden. This demonstrates the complex interplay of factors that shape future evolutions of atmospheric SOA. Nevertheless, SOA may come to play a relatively larger role for atmospheric pollution levels in many regions in the future. The impact of vegetation changes alone on tropospheric ozone is relatively small in our experiments and overwhelmed by the impacts of changing anthropogenic emissions.
Our results highlight the importance of considering the biogenic emission pathway in landclimate interaction assessments under low warming scenarios, demonstrating that different strategies for land management and land-based mitigation can have opposing effects even under scenarios with the same long-term effect on global temperature. Further coupled modeling studies, including transient and more realistic land cover evolution, are required to fully quantify and contrast the biogeochemical and biogeophysical climate impacts and feedbacks due to LCC in low warming scenarios.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: 10.6084/ m9.figshare.14561826.