2.1 Global climate models

Two fully coupled ocean and atmospheric models, which are identical for the ocean and the tropospheric atmosphere component, are used in this study. The horizontal resolution of the atmosphere component for both models is set at $\mathrm{0.9}{}^{\circ }\times \mathrm{1.25}{}^{\circ }$.

CESM1-CAM5 (the Community Earth System Model version 1 with Community Atmosphere Model version 5, shortened to CESM hereafter) is a global climate model that consists of coupled atmosphere, land, land ice, ocean, sea ice, and diagnostic biogeochemistry module component models, plus one coupler component. A new moist turbulence scheme explicitly simulates stratus–radiation–turbulence interactions, making it possible to simulate full aerosol indirect effects within stratus. A new shallow convection scheme uses a realistic plume dilution equation and closure that accurately simulate the spatial distribution of shallow convective activity. Computation of an updraft vertical velocity now allows for aerosol–cumulus interactions. The revised cloud macrophysics scheme provides a more transparent treatment of cloud processes and imposes full consistency between cloud fraction and cloud condensation. Stratiform microphysical processes are represented by a prognostic, two-moment formulation for cloud droplet and cloud ice, as well as liquid mass and number concentrations. The radiation scheme has been updated to the Rapid Radiative Transfer Method for GCMs (RRTMG) and employs an efficient and accurate correlated-k method for calculating radiative fluxes and heating rates. The three-mode modal aerosol scheme (MAM3) has been implemented and provides internally mixed representations of number concentrations and mass for Aitken, accumulation, and coarse aerosol modes.

Sharing most of the features described above, CESM1-WACCM (the Community Earth System Model version 1 with the Whole Atmosphere Community Climate Model, shortened to WACCM hereinafter) is a chemistry–climate model of the Earth's atmosphere from the surface to the lower thermosphere. A finite-volume dynamical core (Lin, 2004) is used for the tracer advection. The vertical resolution ranges from 1 to 2 km over the lower stratosphere and the tropopause. The land model CLM4.5 has fixed distributions of vegetation including interactive carbon and nitrogen cycles. CLM4.5 allows leaf area indexes to respond to climate changes. The Ball–Berry stomatal conductance model was utilized. The stomatal conductance is directly influenced by the relative humidity at the leaf surface, the CO₂ concentration at the leaf surface, and the soil water.

The major difference between WACCM and CESM is the stratospheric atmosphere component. WACCM has a high-top configuration that extends to about 140 km ($\mathrm{6}\times {\mathrm{10}}^{-\mathrm{6}}$ hPa) and has a comprehensive representation of stratospheric chemical and dynamic processes (Mills et al., 2017), both crucial to simulate the radiative impact of stratospheric aerosols (via volcanic or artificial injection). The sulfur injection large-ensemble simulation is only available in the WACCM version. CESM has a typical model top at around 40 km, and the CO₂ mitigation large-ensemble simulation is only available in this model version.

Note that the model structural differences here can affect simulated climate response. The climate sensitivity is 4.1 K under a doubling of CO₂ for CESM and 5.2 K for WACCM (similar to a new version recently released; Gettelman et al., 2019). Both are within the CMIP5 range and are more in line with the higher CMIP6 range. Additionally, the land components are also different between these two versions due to incremental model development. CESM used an earlier version of CLM4 and WACCM used an updated version of CLM4.5. According to Cheng et al. (2019), relative to the earlier version CLM4, CLM4.5 includes modifications of the hydraulic properties of frozen soils and thus some seasonal changes in Arctic temperature. In the following analysis, we made attempts to address the caveat of model differences by (a) applying corresponding bias corrections to future climate response simulated by these two models separately (Sect. 2.5) and (b) normalizing the response with respect to the temperature response (Sect. 5).

2.2 Numerical experiments

Two pairs of model simulations were conducted and published in the last few years, both featuring a large ensemble to enhance the robustness of examined climate responses, especially at a regional level. The WACCM set (http://www.cesm.ucar.edu/projects/community-projects/GLENS/, last access: 27 July 2020) is aimed at stabilizing the climate at its 2020 level with sulfur injection. The CESM set (low warming large ensemble; http://www.cesm.ucar.edu/experiments/1.5-2.0-targets.html, last access: 27 July 2020) is aimed at climate stabilization at 1.5 and 2 ^∘C warming levels, with aggressive emission cuts and rapid growth of negative emission technology, an extreme mitigation scenario. Land cover and land use in these two pairs of simulations are consistently the same as in RCP8.5.

With the two sets of paired simulations, the effect of the two geoengineering approaches can be assessed by contrasting the run with geoengineering applied with the baseline potential warming worlds, which we would refer to as the climate benefits. Of course, for specific regions and certain climate variables, it may turn out that the climate benefits are negligible or even negative; i.e., geoengineering makes climate change under RCP8.5 even worse.

Stratospheric sulfur injection

The simulations were conducted with WACCM: one with baseline emissions (RCP8.5) and the other with sulfur injection applied at 2020 and afterward (hereafter referred to as sulfur injection). The amount and location of sulfur injection were strategically selected to stabilize the temperature at the present-day level and minimize undesired perturbation to the latitudinal temperature gradient (Tilmes et al., 2018). The simulation was repeated multiple times (see Table 1 for details) to form a larger ensemble (Geoengineering Large ENSemble, GLENS; Tilmes et al., 2018), which enhances the statistical significance of the assessed climate impact, especially at a regional level.

Table 1The model experiments.

Download Print Version | Download XLSX

Carbon capture

Similarly, two simulations were conducted with CESM: one with baseline emissions (RCP8.5) and the other with CO₂ removal and emission cuts applied to start in 2015–2020 (hereafter referred to simply as carbon capture). This scenario is constructed to be a much more aggressive decarbonization pathway than RCP2.6 (see details in Sanderson et al., 2017). Net-zero carbon emission is thus reached at 2050, beyond which the net emission turns negative, which helps the planet to stabilize at the 1.5 ^∘C warming level, just slightly higher than the present-day (2010–2019) temperature level. We show the mass amount of CO₂ reduction (combining emission cuts and various forms of net emission technology) and the associated negative radiative forcing in Fig. 4 to illustrate the idealized trajectories and to compare with well-studied RCP2.6. However, the technical feasibility and required socioeconomic shift to facilitate the scale-up of capacities such as direct air capture and clean energy transition (Hanna et al., 2020) are beyond the discussion in this paper.

2.3 Hydroclimate variables examined

In this study, we focus on climate quantities over land due to their high relevance for agriculture, ecosystems, and the carbon cycle, including the following.

• a.

  T: surface air temperature at the reference height of 2 m above the ground.

• b.

  P: total precipitation, which includes rainfall and snowfall without differentiating large-scale or convective components as formulated in the model.

• c.

  PET: potential evapotranspiration, which represents the atmospheric demand for water. PET is calculated using model output, following the physically based Penman–Monteith equation (Shuttleworth, 1993), which includes key parameters of surface air temperature, surface-available energy, surface wind speed, and relative humidity.

• d.

  Aridity index (P ∕ PET): calculated as the simple ratio of annual P and annual PET (Mortimore et al., 2009; Hulme, 1996; Middleton and Thomas, 1992). P ∕ PET is highly correlated with vegetation types (Feng and Fu, 2013; Huang et al., 2017). Using the P ∕ PET classification by the United Nations Convention to Combat Desertification (United Nations Convention to Combat Desertification, 1994), drylands (characterized by P ∕ PET <0.65) are further classified into various types, including subhumid (0.5<P ∕ PET <0.65), semiarid (0.2<P ∕ PET <0.5), arid (0.05<P ∕ PET <0.2), and hyperarid (P ∕ PET <0.05).

• e.

  Other variables to depict aridity and corroborate the aridity index.

To further test the utility of P ∕ PET as a metric to depict aridity conditions, we show the similar temporal evolution of P ∕ PET with soil moisture and that between PET and actual ET (another metric commonly used) over major land regions (60^∘ S to 60^∘ N) in Fig. 1. Spatial patterns also show similar agreement between P ∕ PET and soil moisture as well as P-ET (another metric commonly used) for both present-day and future change (Fig. 2), again supporting the appropriateness of using P ∕ PET as an aridity metric.

Figure 1(a) PET (mm yr⁻¹) and ET (mm yr⁻¹) as simulated by WACCM over global major land regions (60^∘ S to 60^∘ N) only because P ∕ PET is not a good indicator of drought conditions over snow-covered cold regions. (b) P ∕ PET (unitless) and soil water in the top 10 cm layer (mm). The solid lines are for the left y axis and the dashed lines are for the right y axis. Red lines are under RCP8.5 and blue lines are under sulfur injection. Note that no bias correction or smoothing was performed for this figure. CESM results are not shown because (a) there is no reduction of soil moisture in the RCP8.5 warming, and (b) there is no separation of soil moisture in the long-term trends between the RCP8.5 and mitigated warming simulations, presumably due to model deficiency outside the Americas because within the Americas (Fig. 2), the CESM results are largely consistent with WACCM.

Download

2.4 Bias correction of the raw model results

Climate model output cannot be taken at its face value, especially for future projection. The analysis, especially regarding the future projection, must consider the potential model bias in simulating the past and the present as well as the uncertainty inherent to climate models that are so far poorly constrained. Considering these, we corrected the model-simulated T, P, and PET based on the observational dataset, following the approaches in Dai (2011) and Feng and Fu (2013). The NOAA Climate Prediction Center (CPC) data (Fan and van den Dool, 2008; Chen et al., 2002) is used as the benchmarking temperature and precipitation over land. The Global Land Data Assimilation System (GLDAS) (Rodell et al., 2004) provides the surface sensible heat flux and latent sensible heat flux, specific humidity, and wind speed to derive the observation-based PET as in Feng and Fu (2013). The climate variables over the ocean are not corrected and thus not analyzed in this study.

The bias correction based on the observational dataset is only directly applied to CESM_Historical (with 30 runs from 1975 to 2005). The bias correction will force the model output to agree with the observed 1975–2005 climatology (and remove the bias of the seasonal cycle if any) but not necessarily to agree with the observation every year. In a nutshell, the temperature is adjusted using the addition or subtraction of identified biases on a monthly basis. For P or PET, the variables are adjusted using multiplication or division of identified biases on a monthly basis. A multiplication or division is used instead of addition or subtraction to avoid negative numbers for P or PET. Note that the difference between WACCM_RCP8.5 and WACCM_SulfurInjection will yield the “impact” of sulfate injection. The impact will change compared to raw data without bias correction because, for precipitation and PET, the bias correction is done via multiplication or division and not simple addition or subtraction.

The same bias correction is applied to future simulations of CESM (CESM_RCP8.5 and CESM_CarbonCapture in 2006 and afterward), which branched from the CESM_Historical in 2005. For WACCM_RCP8.5 simulations, we indirectly adjust them to agree with the corrected CESM_RCP8.5 in 2010–2019 (the “present day”). We could not adjust WACCM directly based on observations because the WACCM_RCP8.5 simulation only starts in 2010 and overlaps the observation record (ending in 2019) for less than 10 years. Similarly, WACCM_SulfurInjection was adjusted to match WACCM_RCP8.5 in 2020 (when it branches from WACCM_RCP8.5).

3.1 Temperature

In Fig. 3, we show the long-term changes simulated by all four model experiments (two baseline worlds and two mitigated climates via different geoengineering schemes). We focus on the following three periods: the present day (2010–2019), mid-century (2046–2055), and the end of the century (2086–2095). Without the bias correction, the models do not agree on T and P for the recent period of 2005–2020 (Fig. 3a–b; see also the regional map in Fig. 2, first row). Therefore, it is essential to apply the bias correction to model output before carrying out a meaningful comparison of future changes. For example, comparing Fig. 3b and d, WACCM appears to have a 25 % positive bias for precipitation over major land, and CESM has a smaller positive bias.

Figure 2In the two left columns are WACCM-simulated P ∕ PET (unitless), soil water (mm), and P-ET (mm d⁻¹) in the present day and at the end-of-century change under RCP8.5. The two right columns are the same as the left but for CESM. Note that all data here in Fig. 2 are without bias correction as in Fig. 1. The focused periods are the present day (2010–2019) and the end of the century (2086–2095). The areas with stippling in the panels labeled “End of century – present day” indicate where the differences are significant at a 95 % confidence level following a Student's T test. Similar results are presented in Fig. 5 but after bias correction and in relative terms (%).

Figure 3The annual average of major land temperature (T) (^∘C) (a, b without bias correction and c, d with bias correction) and major land precipitation (P) (mm yr⁻¹) (a, b without bias correction and c, d with bias correction). Panels (e, f) show the major land PET (mm yr⁻¹) and major land P ∕ PET (unitless) (after bias correction). Shading represents the standard deviation of ensemble members. We have applied decadal smoothing to all the time series, but note that WACCM_RCP8.5 has three ensemble members during 2030–2099 and thus has large fluctuation due to natural variation in contrast with 10 or more ensemble members in other cases.

Download

Comparing the bias-corrected model output (Fig. 3c and d), because of the larger climate sensitivity, WACCM warms up faster than CESM, reaching a 6 ^∘C warming over land under RCP8.5 just within this century (T for 2086–2095 – T for 2010–2019), which is ∼7 ^∘C relative to the preindustrial, thus posing an existential threat to mankind if it becomes a reality (Xu and Ramanathan, 2017). CESM, with a climate sensitivity of 4 ^∘C, simulates a 4.7 ^∘C warming over the major land areas during the 21st century. The two types of mitigation efforts, by design, would lead to a similar amount of temperature stabilization to a level close to the present day. The sulfur injection simulation here leads to a cooling of 6 ^∘C towards the end of the century compared with the baseline warming. This larger cooling is designed to largely balance the projected warming by introducing a large amount of sulfur gas, some from locations off the tropics (Tilmes et al., 2017; Kravitz et al., 2013).

The mid-century warming over global major land projected by the two models amounts to 2.0 to 2.5 ^∘C compared to the present day (Table 2). Sulfur injection can completely offset the warming by 2.4 ^∘C, while carbon capture can only mask about 70 % (1.3 ^∘C to 2.0 ^∘C). Even though the carbon capture scheme is introduced slightly earlier in 2015 to 2020, the longer lifetime of CO₂ makes the impact of perturbation on the carbon cycle and atmospheric concentration slower to emerge (Fig. 4e vs. f) compared to sulfur injection. This additional inertia (or lag) in CO₂-emission-based mitigation is well-noted when the CO₂ emission cut was previously contrasted with the mitigation of short-lived climate pollutants (e.g., Hu et al., 2013). Towards the end of the century, the lagging effects in the carbon capture case become negligible, and both schemes can almost completely offset the projected warming in the baseline by 4.0 and 5.7 ^∘C out of the 4.5 and 5.8 ^∘C baseline warming, respectively (Table 2, column 1, right).

Table 2The changes in key hydroclimate variables (bias-corrected based on observation) over major land (60^∘ S to 60^∘ N). All numbers are derived from the time series in Fig. 3. All changes are statistically significant at the 95 % level. The definition of major land is the land regions over 60^∘ S to 60^∘ N, thus excluding cold regions where seasonal snow cover or a permanent ice sheet surface makes P ∕ PET less useful as a predictor for aridity and vegetation types. The focused periods are the present day (2010–2019), mid-century (2046–2055), and the end of the century (2086–2095). The values in parentheses are percentages relative to present-day levels. The uncertainty estimate based on the large-ensemble spread (1 standard deviation) is after ±.

Download Print Version | Download XLSX

Figure 4The changes in T, P, PET, and P ∕ PET due to two types of geoengineering schemes. The changes are shown in relative terms (except for T in degrees Celsius), which are computed from the absolute values in Fig. 3. In panel (d), the linear fit for the blue and green lines (separately for 2005–2050 and 2050–2095) is to highlight the near-term benefit of sulfur injection. (e) The mass of carbon cuts (Gt CO₂ yr⁻¹) and sulfur injection (Tg SO₂ yr⁻¹). The dashed green line is the RCP2.6 carbon cut for comparison; (f) the radiative forcing (W m⁻²) due to carbon capture and sulfate injection. The mass of carbon cuts and corresponding radiative forcing are based on Fig. 1 of Sanderson et al. (2017). The mass of sulfur injection comes from Fig. 2 of Tilmes et al. (2018). The sulfate radiative forcing is calculated based on the equations in Metzner et al. (2014).

Download

3.2 Precipitation

Similar lagging effects are also found for global land precipitation. Table 2 shows that the mid-century reduction due to carbon capture is 31 % of the projected increase (23.0 mm yr⁻¹), and sulfur injection can reduce the precipitation increase by as much as 75 %. Again, towards the end of the century, the difference between the two cases, in terms of potential mitigation benefits, becomes smaller (62 % reduction by carbon capture and 98 % by sulfur injection).

The close to 100 % reduction of precipitation increase due to sulfur injection is worth commenting on. Because of the lack of direct radiative heating in the troposphere because CO₂ or black carbon aerosols, sulfate aerosols will have a larger precipitation effect (per degree of global mean temperature change) compared to CO₂ (Lin et al., 2016, 2018). Another alternative perspective to explain the overdrying (or the slowdown of the hydrological cycle) due to geoengineering is the increase in atmospheric stability and suppressed evapotranspiration, especially in the tropics, which tends to be overcooled in previous sulfur injection model experiments. As a result, when the amount of sulfur injection is carefully adjusted to balance the CO₂ warming, total rainfall will usually be excessively suppressed (so-called overdrying) (e.g., Bala et al., 2008; Jones et al., 2013; Crook et al., 2015).

Table 3Dryland area (millions of square kilometers) for semiarid and arid types. North America is 15–60^∘ N and 230–300^∘ E. South America is 60^∘ S–10^∘ N and 280–330^∘ E.

Download Print Version | Download XLSX

However, in this WACCM setup, the overdrying consequence, especially over the tropics, is dampened. This is likely due to the following reason. In the current experiment design, sulfur injection is more heavily introduced to the extratropical region and thus induces a much-reduced tropical overcooling compared to previous stratospheric geoengineering experiments. Indeed, the global mean temperature, as well as the north–south gradient, will be well-preserved while offsetting the baseline CO₂ warming. This relatively weaker tropical cooling (again, because of the larger forcing imposed off the tropics) leads to a smaller overdrying in terms of precipitation, despite a persistent suppression of large convective precipitation (Simpson et al. 2019). From Fig. 3d, sulfur injection will lead to a negligible decrease in precipitation by 4–6 mm yr⁻¹ (WACCM_SulfurInjection: 2086–2095 vs. 2020–2029) over the majority of the land. The overdrying is stronger (12–15 mm yr⁻¹) if the model output is not bias-adjusted (Fig. 3b). This overdrying, despite being small in this model, is considered to be a side effect of sulfur injection and is often used as an argument against its deployment. In contrast, carbon capture falls short in offsetting the full magnitude of projected precipitation increases (green line in Fig. 3b, d) and thus will not run into the problem of overdrying.

The applicability of this precipitation-centered perspective in assessing terrestrial aridity will be put to the test in the next subsection.

3.3 PET and P ∕ PET

It is increasingly well-recognized that precipitation alone does not reflect the full effects of the hydrological cycle on the terrestrial ecosystem. A full suite of aridity indices has been developed and is continuously being improved to provide a better depiction of aridity conditions (Mishra and Singh, 2010). It is beyond the scope of this paper, which emphasizes the contrast of two climate engineering schemes, to discuss the advantages and limitations of those approaches. However, we included more than one indicator from the model output to examine the robustness of our results largely based on P ∕ PET. We showed that ET and soil moisture are closely correlated with PET and P ∕ PET, respectively, over the global scale (Fig. 1). Moreover, Fig. 2 shows that there is broad agreement between P ∕ PET, soil moisture, and P-ET not only for the present-day climatology but also for the future increase under the warming scenarios.

The major difference between the two geoengineering approaches, when viewed on a global scale, is due to PET. Despite an increase in P, the projected PET growth that approximately scaled with the T increase will exacerbate future land aridity (Fu and Feng, 2014; Scheff and Frierson, 2014). Indeed, the projected P ∕ PET under RCP8.5 (Fig. 3f) decreases over major land in both sets of model simulations, with a larger magnitude in CESM than WACCM (also seen in the last two columns of Table 2). The two geoengineering schemes can both limit PET growth (Fig. 3e) to reduce the tendency of worsening aridity but with considerably large differences in the magnitude and timing, as detailed below.

In terms of magnitude, carbon capture can lower the projected increase in PET by 60 % to 83 % at mid-century and the end of the century, respectively, which leads to an almost complete offset of P ∕ PET decline (see green line in Fig. 3f). Therefore, the mid-century and end-of-century P ∕ PET values are mostly the same as the present-day value, except for the near-term (prior to 2040) drop in P ∕ PET due to the lagging effects of CO₂ mitigation, which will be elaborated later (Fig. 4). On the other hand, sulfur injection leads to a major decrease in PET that flips the sign of the projected increase due to GHG warming (Fig. 3e). For example, the mid-century projected PET increase is 102.2 mm yr⁻¹, but sulfur injection can reduce that by 127.8 mm yr⁻¹, which actually leads to a drop of the absolute value of PET by 25.6 mm yr⁻¹. The larger reduction in PET due to sulfur injection also is reflected in the larger increase in P ∕ PET by 0.03 to 0.05, despite a smaller projected P ∕ PET decline in the baseline WACCM warming.

The large response of P ∕ PET due to sulfur injection to more than offset the baseline changes can be clearly seen in Fig. 3f (blue line). Different from all other curves, sulfur-injection-mitigated P ∕ PET features an increase in absolute value, making the geoengineered climate over land overall “wetter” than the present day, despite a reduction in precipitation. Although both types of geoengineering can stabilize temperature at a level close to the present day, sulfur injection appears to have the additional benefit of flipping the sign of the drying trend as projected in the baseline scenarios.

Regarding timing, the increase in P ∕ PET due to sulfur injection is mostly achieved in the near term (before 2050) due to the quick response time. Fig. 4b–d show the mitigation effects (the difference between the paired simulations) in percentage relative to the present day, which highlights the response due to two approaches in the near term and long term. During the first 30 years of the deployment (2020–2050), sulfur injection leads to a quicker reduction of PET of 2.8 % per decade than the decrease in P at 1.2 % per decade and thus causes the P ∕ PET to increase (note the y axis for Fig. 4d is flipped to aid visual comparison). After 2050, when the precipitation decreases at a greater rate of 0.9 % per decade, the P ∕ PET increase in response to sulfur injection will slow down (with the rate dropping by more than half from 1.5 to 0.7 % per decade). This nonmonotonic behavior in sulfur injection induced a P ∕ PET response that is highly distinct from the carbon capture case in terms of timing. The latter, in contrast, always falls behind the sulfur injection changes in inducing climate responses (green vs. blue in Fig. 4a–d). The P ∕ PET enhancement due to carbon capture (compared to baseline warming, not to the present day) only starts to be significant toward the end of the century with a growth rate of 1.4 % per decade after 2050, almost 3 times larger than the decades before 2050.

3.4 Summary: faster and stronger benefits due to sulfur injection

In terms of mitigation at a global scale, we emphasize that P ∕ PET is closely tied to land aridity conditions and vegetation response (Huang et al., 2017) rather than P itself. Note that we select P ∕ PET among other drought indicators to avoid the complexity of the analysis because the main goal of our analysis is to compare highly extreme mitigation scenarios. Figure 2 reveals regional discrepancies suggesting that for a more comprehensive drought assessment, more metrics should be included in addition to P ∕ PET.

Our results here suggest that sulfur injection not only offsets the projected worsening in aridity in the 21st century but also, surprisingly, leads to recovery from the present-day condition, which is already worse than preindustrial conditions or even just decades ago. More encouraging, the recovery (“wetting” over major global land) is detectable in the near term (MacMartin et al., 2019). Carbon capture, in contrast, has a weaker (even in relative terms) and slower mitigation benefit and does not lead to the full reversal of P ∕ PET as in the response to sulfur injection. The response time of these two approaches is illustrated in Fig. 4e and f. Although the CO₂ reduction (Fig. 4e) has a quick increase after deployment (2020), radiative forcing due to the reduced CO₂ concentration (Fig. 4f) takes a slower trajectory to increase and is always falling behind the negative forcing due to sulfur injection.

The quicker and stronger response to sulfur injection suggests a benefit from deployment, providing a further incentive for adopting this scheme. However, because of the quicker response, one could counterargue that sulfur injection should be reserved as the “last resort” option only used when the CO₂ warming becomes too large towards the end of the century.

Figure 5P ∕ PET in the present day, the relative changes (%) in the future under RCP8.5, and mitigation due to carbon capture geoengineering using CESM. (a, b) The reanalysis and the bias-corrected output of the model in the present day (unitless). (c) The projected baseline changes relative to the present day (%); (d) the mitigation of the baseline changes relative to the present day. (e, f) Same as (c, d) but for the end of the century. The similar P ∕ PET changes in absolute values (unitless) can be found in the top row of Fig. 2.

Figure 6Same as Fig. 5, but for sulfur injection using WACCM. (a) P ∕ PET in the bias-corrected output of the model in the present day (unitless). (b) The projected baseline changes at mid-century (%) relative to the present day; (c) the mitigation of the baseline changes relative to the present day. Panels (d, e) are the same as (b, c) but for the end of the century. The similar P ∕ PET changes in absolute values (unitless) can be found in row 1 of Fig. 2.

4.1 North America

The main reason for focusing on North America is the arid land expansion from the American west into the Great Plains. The so-called 100th meridian, which traditionally separates the dry and mild climate, will likely shift eastward in future climate (Seager et al., 2018). This is clearly shown in the projection of both models, with a major decrease in P ∕ PET over the Great Plains regions by 10 % to 20 % (Figs. 5 and 6). The corresponding change in absolute terms of P ∕ PET is shown in Fig. 2. The slight increase (light blue) over the intermountain western US regions does not pass the significance test in being different from the baseline simulations.

For North America, it is clear that sulfur injection has a stronger effect than carbon capture, especially for the central US in the mid-century (Figs. 5 vs. 6). Towards the end of the century, sulfur injection increases P ∕ PET by 0.05 (7.2 %), while carbon capture can eliminate the projected decrease in P ∕ PET (Figs. 5 and 6). This is largely consistent with the results for the eastern US (although to a lesser extent for the latter) and the responses over global major land (Fig. 6, also illustrated in the time series in Fig. 3).

The large decrease in PET in response to sulfur injection is the main reason for a slight increase in P ∕ PET compared to present-day conditions in the geoengineered cases. The P ∕ PET changes over the US are consistent with the tendency of the global major land (Fig. 6d) and also largely supported by other metrics (soil water and P-ET in Fig. 6e, f).

Another way to examine the severity of aridity evolution is by further classifying the dryland into a few subtypes. These subtypes are classified by P ∕ PET values, which are shown in the color bars in Fig. 5a and b. The global arid area (P ∕ PET ranges of 0.2 to 0.5) would increase from 15 million km² in the historical period to 16.6 million km² under warming at the end of the century. Carbon capture can reduce it to 15.1 million km², but sulfur injection can more than offset it and lower it to 13.5 million km². Over North America, the results are generally similar, with a stronger reversal over the semiarid area (P ∕ PET ranges of 0.5 to 0.65) due to sulfur injection than carbon capture.

4.2 South America

The main reason for focusing on South America is that drying over the Amazon (by more than 30 % using P ∕ PET as the metric) could lead to die-out of the rainforest, decrease the carbon sink, and cause an increase in forest fires – major positive feedback that is missing in most of the global model projections so far, including the ones presented here.

While the changes over North America are largely consistent with global land change, South America's responses are more complicated. Notably, there is a major difference between northern Brazil, where the Amazon rainforest is mostly located, and the southern part of the nation (green and purple boxes in Fig. 5a).

Northern Brazil will see a major decline of P ∕ PET of 0.2 to 0.4 (15 % to 30 % relative to the present day) in the baseline warming cases (Fig. 5), the largest among the four regions we considered in this study, transitioning the wetland into semiarid dryland. This large decline of P ∕ PET is a result of both PET increase as in other regions and also a P decrease that is unique in this tropical land, in contrast to other regions examined. If the RCP8.5 projection turns out to be true for the 21st century, a major consequence is the fate of the rainforest and its carbon uptake capacity. Given the high stakes, a relevant question is whether the two geoengineering schemes are able to reverse the drying trend over the Amazon.

The CESM results show that carbon capture can essentially overturn the baseline drying and keep the P ∕ PET at the same level as the present day (Fig. 5). This is achieved by offsetting both the precipitation decrease and, more importantly, the projected PET increase by 80 % to 90 % (Fig. 7b, c). The carbon-capture-mitigated (green in Fig. 7b) climate will lead to a small increase in precipitation (relative to the present day) in northern Brazil, associated with a much smaller regional warming (<0.5 ^∘C) than baseline (4.5 ^∘C) (Fig. 5). Note that the regional precipitation change can be hard to project at the subcontinental scale, and thus the CESM-simulated carbon capture response could be model-dependent.

Figure 7Regional changes at the end of the century relative to the present day. The western and eastern US are separated by 100^∘ W, and northern and southern Brazil are separated by 10^∘ S. See the boxes in Fig. 5a for the domains. The zero lines of the y axis indicate the present-day level.

Download

On the other hand, the WACCM results show that sulfur injection falls short of completely overturning the drying tendency over northern Brazil in the baseline warming but can mitigate the drying magnitude by 60 % by the end of the century. The reason for the insufficiency of sulfur injection is that, despite a substantial reduction in PET (Fig. 7c, from 0.8 to 0), the WACCM results continue to produce a stronger precipitation reduction (relative to the present day) in the case of sulfur injection (blue dots in Fig. 7b) but is unable to reverse the precipitation reduction in the warming baseline. This additional suppression of tropical rainfall due to sulfur injection is also previously reported in Simpson et al. (2019), who concluded that the dynamical response to stratospheric heating played an important role during JJA (but not in DJF).

Although the simultaneous reduction of PET in response to sulfur injection leads to the end result of wetting (in terms of P ∕ PET increase; brown and blue points in Fig. 7d), the magnitude (roughly a 50 % offset) is small compared to that due to carbon capture (an offset by almost 100 %; green and red dots in Fig. 7d), for which the precipitation is flipped in sign to increase slightly rather than just with the baseline decrease mitigated (green in northern Brazil; Fig. 7b). Overall, there is a weaker mitigation of the P ∕ PET drying trend in response to sulfur injection compared to carbon capture over northern Brazil. However, interestingly, the opposite is true for the sub-Amazon region, which is detailed below.

Unlike sulfur injection, carbon capture is consistently capable of mitigating the projected drying trend almost entirely over South America (similar to the global and eastern US cases). Despite being smaller than sulfur injection, there is sufficient capacity of carbon capture to overturn the drying trend over the sub-Amazon (from a 0.08 reduction of P ∕ PET to almost zero), again due to a large reduction in PET despite slightly weakened precipitation. This suggests again that focusing on P alone may lead to a biased interpretation of the projected and mitigated trends. Notably, sulfur injection induces a much larger increase in P ∕ PET (wetting) over the sub-Amazon compared to carbon capture, actually offsetting the projected P ∕ PET decrease by 150 %. This is due to a regional increase in precipitation, which adds to PET reduction to cause a larger increase in P ∕ PET. The mechanism is exactly the opposite of sulfur injection's smaller capacity (by only 60 %) to reverse the P ∕ PET decline over the Amazon (northern Brazil), where the precipitation is projected to further decrease in response to sulfur injection despite a strong local cooling effect.

Synthesizing the precipitation changes over the Amazon (northern Brazil) and sub-Amazon (southern Brazil), the precipitation (especially in JJA as in Fig. 5 of Cheng et al., 2019) is shifted from the deep tropics to the subtropics in response to sulfur injection, leading to an actual increase in precipitation over the sub-Amazon by 36.5 mm yr⁻¹ rather than just offsetting the 18.2 mm yr⁻¹ decrease in the baseline warming case.

A substantial reduction in PET over the sub-Amazon (consistently true for other subregions explored in Fig. 7), working together with the precipitation increase, would induce a large increase in P ∕ PET (0.2) relative to the baseline case, which is more than enough to offset the projected decrease of 0.15 and actually to lead to P ∕ PET increase by 0.05. Considering that this region in the present day is already subject to a detectable drying, such an overcompensation in P ∕ PET in response to sulfur injection should be viewed as beneficial to the regional climate and ecosystem, a point we have stressed for global mitigation results but want to echo again here for sub-Amazon regions.

4.3 Summary: unique responses over the Amazon (northern Brazil)

Overall, over three out of four regions examined here (western US, eastern US, and southern Brazil but not northern Brazil), sulfur injection has a larger capacity to not only mitigate but also flip the projected drying tendency. This is evidenced by the large response seen in the spatial maps in Fig. 6 (especially the larger response in the mid-century due to the time lag effects discussed in Sect. 3.1). A complete reversal to even an increase in P ∕ PET drying highlights the benefit of sulfur injection, at least for these three regions. On the other hand, carbon capture can almost completely offset the P ∕ PET change by 100 %, which is also the desired outcome. The mechanisms to achieve this complete offset can vary from one region to another and are sensitive to the regional precipitation projection in the baseline and mitigated scenarios (e.g., over the western US); they are thus potentially sensitive to model configurations. The responses over these three regions are found to be generally representative of the global mean (e.g., Fig. 7 major land, results with smaller error bars), which is presumably less subject to model diversity issues.

The South America regions illustrate a clear argument that a precipitation-centered viewpoint may mistakenly overestimate sulfur injection side effects in causing aridity by ignoring the accompanying suppression of PET. The contrasting results for sulfur injection to overcome the global drying tendency, when using precipitation only or using P ∕ PET, are clearly demonstrated for the Amazon region but are also similarly true for other regions such as the eastern US (to a lesser extent). For example, in Fig. 6, the precipitation change over the eastern US is slightly negative compared to the present day (overdrying) but becomes slightly positive in P ∕ PET, suggesting an actual wetting, which is an improvement from the drying conditions already detectable in the present day (2010–2019) compared to the preindustrial era.

One exception to the summary above is northern Brazil (Amazon region). Sulfur injection, but not carbon capture, can only partially offset the projected drying due to a further decrease in precipitation locally in the deep tropics (Fig. 8b), which features a rainfall shift from the deep topics to subtropics (Fig. 6). In response to sulfur injection, there is a smaller amount of PET reduction from the baseline increase (Fig. 7c) compared to the other three regions and the global mean, presumably due to extensive cloud cover over the tropical land, but it could also be due to the vegetation difference between northern Brazil and southern Brazil.

Figure 8Similar to Fig. 5c–f but showing normalized P ∕ PET change in percent per degree Celsius (% ^∘C⁻¹). Panels (a–c) show mid-century changes, and panels (d–f) represent end-of-century changes. The normalization is by scaling the global temperature change in (a, b, d, e), for example 4.2 ^∘C due to carbon capture in panel (e) and 6.0 ^∘C due to sulfur injection in panel (f).