Trade-offs of Solar Geoengineering and Mitigation under Climate Targets

So far scientific analyses have mainly focused on the pros and cons of solar geoengineering or solar radiation management (SRM) as a climate policy option in mere isolation. Here we put SRM into the context of mitigation by a strictly temperature-target based approach. As a main innovation, we present a scheme by which the applicability regime of temperature targets is extended from mitigation-only to SRM-mitigation analyses. Hereby we explicitly account for a risk-risk comparison of SRM and global warming, while minimizing economic costs for complying with the 2°C temperature target. 5 To do so, we suggest precipitation guardrails that are compatible with the 2°C target. Our analysis shows that the value system enshrined in the 2°C target would be almost prohibitive for SRM, while still about half to nearly two-third of mitigation costs could be saved, depending on the choice of extra room for precipitation. In addition, assuming a climate sensitivity of 3°C or more, in case of a delayed enough policy, a modest admixture of SRM to the policy portfolio might provide debatable tradeoffs compared to a mitigation-only future. In addition, in our analysis for climate sensitivities higher than 4°C, SRM will be an 10 unavoidable policy tool to comply with the temperature targets.


Introduction
Since Paul Crutzen has highlighted solar radiation management (SRM) as a potential climate policy option in addition to adaptation and mitigation (Crutzen (2006)), there is and increasing research on this technique as a measure to counteract anthropogenically caused global warming (Barrett et al. (2014); Bellamy et al. (2013); Goes et al. (2011); Irvine et al. (2012); global warming as the dominant effect. The earlier studies presented trade-off results for stylized impact assumptions within the standard economic paradigm of cost benefit analysis. This is as much as possible in-line with standard economic axioms.
But, at the same time, some studies suggest that it is challenging to directly recommend climate policy through only cost benefit analysis, due to the presence of deep uncertainty about impact functions (Ekholm (2018); Kolstad et al. (2014);Kunreuther et al. (2014)). They would suggest a target-based approach, known as Cost Effectiveness Analysis (CEA), as long as no better 10 data is available (Kunreuther et al. (2014); Neubersch et al. (2014)).
Although Arino et al. (2016), Ekholm and Korhonen (2016), and Emmerling and Tavoni (2018) evaluated SRM together with mitigation applying CEA, for a true risk-risk consideration, one needs to define an explicit guardrail over side-effects of SRM.
To the best of our knowledge here for the first time we introduce and apply a concept for an integrated analysis of SRM 15 and mitigation in-line with the '2°C temperature target'. 1 The 2°C target is the cornerstone of the Paris agreement (UNFCCC, 2015[UNFCCC (2015: Adoption of the Paris Agreement. FCCC/CP/2015/L.9/Rev.1.]). The 2°C target encapsulates society's informal risk evaluation of deeply uncertain global warming impacts (Neubersch et al. (2014); Schellnhuber (2010)). Driven by the expectation that in fact costs of transforming the energy system are much more robustly to project than the aggregate impacts of global warming (Stern (2007)), a plethora of economic mitigation analyses has derived cost-minimal energy scenarios which 20 are in compliance with this target (Edenhofer et al. (2014)).
However, when SRM comes into play, global mean temperature is no longer a good proxy for regional climate impacts because SRM imprints patterns of regional precipitation and temperature change that would differ from those induced by greenhouse gas forcing (Kravitz et al. (2013); Oschlies et al. (2017)). Accordingly, and as a key innovation of this article, we suggest extending the regime of applicability of the 2°C target from mitigation only to joint SRM-mitigation portfolios when 25 temperature and precipitation are simultaneously considered. For our joint SRM-mitigation analysis we utilize the integrated energy-economy-climate model MIND (Edenhofer et al. (2005)) which provides one of the simplest possible options to distinguish the renewable sector from the fossil-fuel sector under induced technological change. We then further extend the model to include a spatially-explicit resolution in terms of 'Giorgi regions' (Giorgi and Bi (2005)) and run specific policy scenarios showing the trade-offs between mitigation and SRM. We also highlight the most important factors that derive our results. 30 The rest of the paper is organized as follows. Section 2 details on the innovated guardrails, data, and the numerical model employed. Section 3 presents the results. In Sect. 4, some sensitivity analyses are presented. And, Sect. 5 concludes the paper.
1 Based on the previous version of this article (Stankoweit et al. (2015)), Roshan et al. (2019) applied a Cost Risk Analysis and evaluated the optimal SRM in conjunction with mitigation, considering regional disparities in the precipitation risks.

Precipitation guardrails
We define a corridor for admissible (Bruckner et al. (2008)) regional precipitation values, diagnosed from a global mean temperature change of at maximum 2°C and at minimum 0°C in the absence of SRM. In the following we focus on a subset of regional climate guardrails in terms of terrestrial precipitation changes that have been highlighted as a key drawback of SRM 5 (Shepherd (2009);Bala et al. (2008); Robock et al. (2008)). Thereby, we add a necessary condition, while keeping the 2°C target in order to reflect those impacts which are not yet formulated in an equivalently explicit manner. Here we ask: 'How much regional precipitation change, as an example of a climatic change other than temperature, would someone, who has already accepted up to 2°C of global warming, accept?' If we were able to confine regional climate change to the intervals of climate variables that would be spanned by ramping the global mean temperature anomaly (as against its pre-industrial value) 10 up from zero to 2°C, we could augment the 2°C target by this exact set of intervals as the more fundamental target. Note that here we suggest that the intervals would be generated in the absence of SRM because the 2°C target has emerged from a line of argument excluding SRM (Schellnhuber (2010)). This region-based, hence more fundamental, target would then be valid also for portfolios of SRM and mitigation. Analogous to the original global target, this target also allows for bypassing the criticized monetarization of climate impacts on which a cost benefit analysis is based. 15 Figure 1 shows our suggested guardrails for two hypothetical regions r 1 and r 2 . r 1 is characterized by a positive CO 2 (Greenhouse-gas-driven) scaling coefficient (C(r 1 , co 2 ) > 0) which denotes a positive change in precipitation (P ) when the global mean temperature (T ) rises. r 2 is, however, characterized by a negative CO 2 scaling coefficient (C(r 2 , co 2 ) < 0). The green bands in panel a) and panel b) define the regular admissible area for precipitation change which is compatible with the 2°C target. As noted earlier, SRM imprints patterns of regional precipitation and temperature change that would differ from 20 those induced by greenhouse gas forcing (Kravitz et al. (2013)). Therefore, the regular admissible area would totally prohibit SRM use in the regions whose SRM scaling coefficients have the same sign as their CO 2 scaling coefficients. Therefore, an extra area of admissibility is required for any SRM use. For the extra admissible area, we pragmatically suggest adding a fraction of regional standard deviation, derived from inter-annual variability, on both ends. These areas are demonstrated in blue.
In this paper, we consider 5% and 10% of inter-annual variability. In Sect. 4.1, we conduct a sensitivity analysis by changing 25 the extra room added to the upper and lower bounds of the guardrails.
2.2 Regional scaling coefficients and natural variability For the regional resolution, we decide on a spatial resolution in terms of 'Giorgi regions' (Giorgi and Bi (2005)) (see Table 1 and Fig. 2) which brings about a markedly different image from a global average in the sign and magnitude of effects in the climatic variables under scrutiny, and at the same time, it avoids a larger number of simultaneous regional targets that might be perceived as too restrictive. 2 However, we stress that the choice of the resolution is ultimately a normative decision to be taken by society.
For the scaling coefficients, we diagnose annual mean regional precipitation changes from linear pattern scaling (Ricke et al. (2010)) which are driven as a linear superposition (Ban-Weiss and Caldeira (2010)) of greenhouse-gas-induced and SRM-induced changes in global mean temperature. We use the outputs of nine atmosphere-ocean general circulation models  Table 1. Figure A1 in the Appendix also shows the variations of scaling coefficients for each region from nine AOGCMs. Obviously, for some regions, the scaling coefficients may switch the sign if a specific AOGCM is considered. Table 1 also shows the ratio R r = c SRM /c CO2 which is used as an indication for co-effects of SRM and CO 2 10 that link temperature effects to precipitation effects. All regions are characterized with SRM and CO 2 coefficients that increase or decrease precipitation in opposite directions, and hence R r is negative in all regions. In regions where 0 > R r > −1, SRM under-compensates CO 2 -induced precipitation changes. However, in regions where −1 > R r , SRM over-compensates CO 2induced precipitation changes.
An important note here is that, from the average scaling coefficients and their sample standard deviation, one can read 15 that with a normal distribution of scaling coefficients, there are chances that the signs of scaling coefficients in some regions switch. Therefore, it is also likely that for some regions R r becomes positive which means SRM and CO 2 both either increase or decrease precipitation. In Sect. 4.2, we conduct a Monte Carlo analysis by randomly choosing the scaling coefficients from the above distributions.
We determine the standard deviation of natural variability in precipitation the following way. We use three data sets of annual 20 precipitation, aggregated to Giorgi regions resolution, based on GPCC_WATCH (1901-2001 (Weedon et al. (2010)), PGFV2   (Sheffield et al. (2006)), and GPCC_WFDEI (1979 (Weedon et al. (2014)). Then for any region r and data set s we determine the precipitation means µ r,s . To also obtain the standard deviation of natural variability as distinct from global warming, for any r, s we subtract a parabolic fit of the time evolution from the time series. From the thereby detrended data we determine the inter-annual precipitation time variances (σ 2 r,s ). For each region r we average means and variances across 25 all data sets s to obtain σ 2 r and µ r . Finally the standard deviation of natural variability in percent is obtained by 100 (σ r /µ r ). The last column in Table 1 expresses the derived regional natural variability.

Model
For our joint SRM-mitigation analysis we utilize the integrated energy-economy-climate model MIND (Edenhofer et al. (2005)) which provides one of the simplest possible options to distinguish a renewable from a fossil sector and to include  (Stern (2007)) and turned out to deliver centered results in comparison to other models of that category. Compared to more advanced models that would distinguish an electricity, a household, and a transport sector, it tends to underestimate mitigation costs by a factor of two (see, e.g., Edenhofer et al. (2014)). However, it serves as one of the simplest possible models to project mitigation costs in a realistic manner and to a great deal can serve as a pedagogic model to mimic the most important economic-climatic aspects that are under investigation. We further extend the model to include a spatially-explicit resolution in 5 terms of 'Giorgi regions' (Giorgi and Bi (2005)). In addition, we upgrade the model to include SRM as a control. We assume a reduction of the solar constant as a good approximation (Kalidindi et al. (2015)) of sulfur aerosol injection that is currently discussed as the most feasible SRM scheme. As cost of SRM we took the joint upper end of the costs reported in The Royal Society's Report on Geoengineering the Climate (Shepherd (2009) and Klepper and Rickels (2011)): 0.02% gross world product as of 2010 per W/m 2 . This is at least an order of magnitude smaller number than the cost of mitigation (Edenhofer et al. 10 (2014)).
Climate sensitivity is a crucial uncertain parameter and some studies considered a log-normal probability density distribution  (2001)). In this study, the model MIND is used in its deterministic setting with a climate sensitivity of 3°C when the time scale of the climate module has a strict relationship with the climate sensitivity suggested by Lorenz et al. (2012). MIND employs the simplest climate module (one-box 15 climate model) (Petschel-Held et al. (1999)). Nonetheless, Khabbazan and Held (2019) tested the validity of a one-box climate model as an emulator for fourteen AOGCMs and showed that it is a good emulator of these AOGCMs (accurate to within 0.1°C for Representative Concentration Pathways, RCPs), provided the one-box climate model being trained according to the AOGCM's equilibrium and trasient climate sensitivity and a certain time horizon (on the order of the time to peak radiative forcing) is not exceeded (see Khabbazan and Held (2019) for a detailed discussion). Therefore, according to Khabbazan and 20 Held (2019), the results in this article can be interpreted as being influenced by a slightly larger climate response to forcing than intended. Hence, for the sake of completeness, in Sect. 4.3 we conduct a sensitivity analyses by altering the climate sensitivity. 4 c): precipitation changes when all regional constraints are binding and the extra admissible area is 5% of the standard deviation ('G0 5%'); d) similar to c) but with 10% of standard deviation ('G0 10%'). Note that there are no extra admissible area for BAU and REF. We normalize precipitation such that '1' is the constraint which corresponds to the precipitation levels of a temperature anomaly of 2°C and '0' equals the other constraint, which is determined by the preindustrial precipitation levels (see Sect. 2.1 for more details on the guardrails). Note that in the calculation of the normalized precipitation for G0 5% and G0 30 10% the extra admissible areas are taken into account. We indicate this corridor as grey band in Fig. 3. Figure  c) all Giorgi regions' precipitation guardrails being activated when the extra admissible area is 5% of the standard deviation ('G0 5%'); d) similar to c) but with 10% of the standard deviation ('G0 10%'). In BAU, for most regions, regional precipitation would surpass this corridor, indicating the need for an active policy if a 2°C framing is of interest (see Fig. 3 (a)).

Results
As a first policy scenario we simply add SRM to the option folder without activating the regional precipitation guardrails 5 while keeping the global 2°C target (REF scenario). As expected, SRM almost completely crowds out mitigation. As shown in Fig. 4 (b), with unrestricted SRM usage, the CO 2 -contribution would mimic BAU, and SRM would totally compensate for an overshoot of 2°C. However, for about half of the regions, precipitation transgresses the 2°C-compatible precipitation corridor (see Fig. 3 (b)). Now we activate the precipitation corridor for any of the regions (G0 scenarios). By construction, for any of the regions the 10 precipitation trajectories stay confined to the grey band (see Fig. 3 (c) and (d)). Comparing G0 5% with G0 10%, one notices that the upper and lower bounds in G0 5% is touched 15 years earlier, which is due to the slimmer admissible area in G0 5% as against G0 10%. Compared to REF, with regional precipitation constraints being activated in G0 5% and G0 10%, SRM usage is restricted to about 1/6 and 1/3 respectively in the G0 5% and G0 10% scenarios. Hereby temperature anomaly peaks at 2°C and then declines, for example to 1.5°C in G0 10%, which means SRM partly overcompensates the CO 2 -effects (see Fig. 4 (c) 15 and (d)).
This overcompensation can be explained from the panels c) and d) in Fig. 3 in which the following four phases along the time axes can be identified. (i) No guardrail is active, the paths mimic a BAU development (compare to panel a)). (ii) The 2°C guardrail is active and SRM is utilized (as in the case depicted in Fig. 4 (b)), and for most regions, normalized precipitation declines. (iii) For one region, either the upper ('1') or lower ('0') normalized precipitation guardrail is touched 20 and activated. In the course of time, for several regions normalized precipitation evolves towards the guardrails (either '0' or '1') and simultaneously global mean temperature decreases. (iv) For a different region, the normalized precipitation guardrail is touched and the system becomes quasi-stationary.
The first binding region is 'AMZ' (Amazonia, see Fig. 2). Here it is characterized by a negative c CO2 and positive c SRM with comparatively small modulus. Although the CO 2 -and the SRM-effects on precipitation levels in 'AMZ' work in opposite 25 directions, the regional SRM-effect cannot significantly compensate for the CO 2 -effect that in phase 2 activates to compensate for any overshooting of 2°C. Therefore 'AMZ' appears a likely candidate to touch the upper bound and start phase 3. However, once precipitation in 'AMZ' reaches the upper boundary of regional normalized precipitation levels (phase 3), any CO 2 -induced change in AMZ's precipitation needs to be compensated for by an SRM-induced contribution. As AMZ's scaling coefficient has a small modulus, more SRM forcing needs to be applied per unit of CO 2 -induced radiative forcing than in phase 2, to 30 stop AMZ's normalized precipitation trajectory from growing any further. Therefore, in phase 3 SRM overcompensates CO 2 in terms of their effects on global mean temperature. Finally, one of the regions with larger, yet negative scaling coefficient ratios (in this case 'SQF' (South Equatorial Africa)) would hit the lower bound, triggering phase 4. Therefore, the 'G0' scenario is characterized by the interplay of scaling coefficients and precipitation standard deviations of two regions.
What are the economic effects of allowing for G0 (i.e. restricted SRM) instead of mitigation only? Figure 5 displays mitigation costs for a step-wise omission of regional precipitation guardrails with Balanced Growth Equivalent (BGE) values attributed to a region are the loss in comparison to the BAU scenario, if the analysis is henceforth not constrained by the precipitation guardrails of the respective region. The acronyms of regions are defined in Fig. 6. The economic losses (disregarding impacts) induced by a 2°C policy without SRM usage is displayed by the leftmost bar (hereafter we call it 'TradCEA'), then 5 for SRM included when all regional precipitation guardrails are active (G0), and continued with scenarios when the guardrails of binding regions are disregarded one by one. Hereby 'BGE' losses (y-axis of Fig. 5) can be approximately interpreted as consumption losses, or in other words the loss in Global World Production (GWP) (for details see Anthoff and Tol (2009) and Lorenz et al. (2012)). From the graph we read that 2/5 and 3/5 of mitigation costs could be saved respectively in G0 5% and G0 10%. For someone who interprets mitigation costs as large, this could be an argument for employing SRM. For someone 10 who perceives the scale of 1% GWP loss as small, the cost of mitigation would not provide a reason to become interested in SRM.
If saving 2/5 to 3/5 of mitigation costs were of interest, however, instead of utilizing SRM, one could also imagine relaxing the 2°C target instead. What would be the equivalent additional global warming? From Fig. 4 we read: about 0.1°C and 0.3°C respectively in G0 5% and G0 10%. If one interprets the 2°C target as an (academically informed) political target that does not 15 represent a singular global threshold (Neubersch et al. (2014)), society might want to discuss whether a modest transgression of the target to extents of, for examples, 0.1°C and 0.3°C, could be seen as acceptable in view of the risk of potential further side-effects of SRM such as stratospheric ozone depletion (Tilmes et al. (2008)). In that sense the extrapolation of SRM impacts to regional precipitation corridors has proven almost prohibitive for SRM.
However, if the society is willing to successively disregard the step-wise (economically) most binding corridor boundary, 20 and hence to 'sacrifice' the region that would induce the largest economic welfare gain, the picture can gradually change.
Progressively, more economic gain could be harvested (see Fig. 5, further right bars). From one bar to the next, the guardrail of that very region is omitted that would deliver the largest economic welfare gain. Figure 6 indicates the economic gain per region when precipitation is allowed to leave the respective corridor. Similar for both G0 5% and G0 10%, the order of the first four regions whose guardrail should be omitted to gain the most are AMZ, SQF, CAM, and WAF. However, while for G0 10% 25 these four regions are followed by CNA, MED, and CAS, in G0 5% CAS, CNA, and SAH follow. Note that the precipitation change in other regions are already within the guardrails in the REF scenario, and hence, there is no need for omission of their guardrail to gain more welfare (see Fig. 3). Notably, even not so significant for the analysis regarding the welfare gains, the order of regions depends on the decision about the extra room for guardrails, as shown in panel a) and b). The reason behind this observation is that with a tighter guardrail, while SRM is used less and at the same time mitigation must be employed 30 more, the interplay of different scaling coefficients would likely cause different ordering. In addition, if such extra room is tighter, then the economic gain from its omission is higher. The same rule applies to the G0 scenarios, too. For example, the BGE loss in G0 5% is almost 60% higher than the BGE loss in G0 10%, and the BGE loss when the extra room is 5% of natural variability and AMZ is omitted is nearly double the cost for the same scenario when the extra room is 10% of natural variability.

Sensitivity analysis
The results in the previous section were derived based on some specific assumptions. Here we pick up some of the most important assumptions and investigate the likely alternative scenarios. respectively. Expectedly, with the higher extra rooms, the BGE loss decreases such that when extra room is 50% of the natural variability (the rightmost bar), the BGE loss is negligible (about -0.01% of BAU). However, such a decrease in BGE loss is not linear with respect to the increase in the extra room, but it is convex. That is, for example, while the decrease in BGE loss is about 0.25% when the extra room changes from 0.05 to 0.1, the decrease in the BGE loss will amount only to nearly 0.15% 10 when the extra room changes from 0.1 to 0.15. Note that, according to the argument presented in Sect. 3, there is a chance that the order or regions change.

Scaling coefficients
As discussed earlier, the SRM and CO 2 scaling coefficients determine the order of regions as well as when to hit the guardrails in the G0 scenario. However, the scaling coefficient may vary based on which AOGCM data they are derived from. The results 15 in Sect. 3 were derived from the average value of scaling coefficients from nine AOGCM. These data can be also used to derive specific standard deviations for each scaling coefficient. Figure 8 shows the box plots for a Monte Carlo study on 1000 random, simultaneous variations in scaling coefficients and measures the BGE loss in the G0 scenario. In addition, the extra room in guardrails can increase in each scenario from 10% of natural variability to 100% of natural variability. In each G0 scenario, the sign of some scaling coefficients as well as the binding regions in the four phases can be different. 5

20
According to the Monte Carlo study, it is more likely that the random variation of scaling coefficient results in a higher BGE loss in G0 scenarios. For example, while in the deterministic results for an extra room for guardrail equal to 10% of natural variability the BGE loss is about 0.45%, the median of BGE loss in the Monte Carlo study can reach to about 0.85%.
In addition, with the higher extra rooms, the median of the BGE loss decrease. However, similar to its deterministic case, a decrease in median of the BGE loss is convex. Therefore, although for an extra room for guardrail equal to 50% of natural 25 variability, the BGE loss is negligible in the deterministic case, in the Monte Carlo study the BGE loss is about 0.4%. In other words, if the precipitation guardrails are considered, the likelihood that SRM being used may still be low if no region's guardrail is about to be omitted. Note that the decision about SRM being used may involve many more factors than just an economic study. However, our results at least call for attempts to better estimate the sensitivities of regional precipitation changes to SRM and global tmeperature increase.  Figure 9 shows the BGE loss in Traditional CEA (TradCEA, where the temperature target is active without SRM use) and the G0 5% and G0 10% scenarios with the climate sensitivity varying between 1.5°C to 5°C. As expected, with a higher climate sensitivity, BGE loss for the G0 5%, G0 10%, and TradCEA scenarios increase rapidly. The 2°C target is not attainable without the use of SRM when the climate sensitivity is equal and higher than 3.75°C. Yet, by using SRM, the 2°C target is reachable 5 with higher climate sensitivities. Nonetheless, the feasibility space depends on the extra room in the guardrails. If the extra room is 10% of the natural variability, the 2°C target is still reachable when the climate sensitivity is below 5°C (not reachable at 5°C). However, if the extra room is 5% of the natural variability, the 2°C target is only reachable when the climate sensitivity is below 4.25°C (not reachable at 4.25°C).

10
We have performed a CEA study where SRM and mitigation act as substitutes in idealized policy scenarios of immediate action. Without risk-risk accounting, SRM would crowd out mitigation due to comparatively low costs, thereby lowering the costs of achieving the 2°C target to a negligible value (compared to the original mitigation costs). However, if only one single regional climate variable (in our case 'precipitation') is confined to regional bounds compatible with 2°C of global warming 2/5 and 3/5 of mitigation costs could be saved respectively within an accuracy of 5% and 10% of the standard deviation of 15 natural variability. Furthermore, the additional amount of carbon dioxide that could be released to the atmosphere corresponds to only about 0.1°C to nearly 0.3°C of further global warming. Society might debate whether for that rather small amount of allowed additional carbon emissions, the risks of SRM should be taken. However, a significantly larger role for SRM would be possible in case the guardrails of a few regions were relaxed. The ordering of regions presented in this article might provide a way to support the then necessary trade-off to what extent relaxing global versus regional targets. Nonetheless, the order of 20 regions whose omission brings about the most economic welfare gain depends on the magnitude of CO 2 -and SRM-induced effects on precipitation as well as the normative decision on the extra rooms on the upper and lower bounds of the precipitation guardrails.
We also see SRM and mitigation as complements if a global climate policy were put in action only within decades. The results showed that in our model the 2°C target is not attainable without the use of SRM when the climate sensitivity is equal 25 and higher than 3.75°C. Then we would necessarily need some sort of climate engineering in order to comply with the 2°C target. If the potential for carbon dioxide removal was exhausted, some amount of SRM would become indispensable if the 2°C target still should be complied to. Nonetheless, the feasibility space depends on the extra room in the guardrails. The tighter the extra rooms on the upper and lower bounds of the precipitation guardrails, the earlier the use of SRM become useless to comply with the targets. 30 We need to point to a series of caveats of the analysis. The assumptions made in the construction of regional scaling patterns for precipitation may be oversimplifying the complex hydrological effects of greenhouse gases and SRM. Hence we need to emphasize that regional precipitation guardrails can only be interpreted as necessary, not as sufficient conditions for a decision about SRM use. Yet, by employing a Monte Carlo study, we showed that it is more likely that the random variation of scaling coefficients results in larger economic losses when all temperature and precipitation targets are active. Furthermore, annual mean precipitation is only one possible driver of regional impacts in addition to temperature or evaporation, or intra-annual changes. Finally, our model does not include carbon dioxide removal options yet. Hence above-mentioned economic gains through SRM must be interpreted as upper limits of costs savings. 5 Here we demonstrate that someone who pushes for SRM in view of the 2°C target should carefully consider the consequences this target would have when extended to SRM. Already a single climate variable (such as precipitation) when explicated regionally almost completely bans SRM from a joint SRM-mitigation portfolio.

DATA
The output of nine AOGCMs were supplied by the scientists from GCESS, Beijing Normal University (John Moore's Group). 14 https://doi.org /10.5194/esd-2020-95 Preprint.  . Mitigation costs for a step-wise omission of regional precipitation guardrails. 'TradCEA' is a 2°C policy without SRM usage; 'G0' is a 2°C policy when SRM is included and all regional precipitation guardrails are active; BGE values attributed to a region are the loss in comparison to a no policy (BAU -business as usual) scenario, if the analysis is henceforth not constrained by the precipitation guardrails of the respective region. The acronyms of regions are defined in Table 1. 19 https://doi.org/10.5194/esd-2020-95 Preprint. Discussion started: 18 December 2020 c Author(s) 2020. CC BY 4.0 License.     Figure 9. Sensitivity analysis on climate sensitivity. The grey bars show the BGE losses for the G0 10% scenarios. The blue bars show the BGE losses for the G0 5% scenarios. The red bars show the BGE losses for the TradCEA scenarios.