To what extent can cirrus cloud seeding counteract global warming?

We use the ECHAM6 GCM (Stevens et al 2013) coupled with the HAM2 aerosol module (Zhang et al 2012, Neubauer et al 2014). The model has a resolution of 1.875° × 1.875° with 31 vertical layers extending to 30 km altitude. ECHAM6-HAM uses a two moment cloud scheme (Lohmann et al 2007) with a cirrus microphysical scheme, which allows competition between homogeneous and heterogeneous nucleation and the deposition of water vapor on pre-existing ice crystals (Kärcher et al 2006, Kuebbeler et al 2014, Gasparini and Lohmann 2016). Convection is parameterized by the mass-flux scheme of Tiedtke (1989) with modifications for deep convection from Nordeng (1994). The model has been previously evaluated with satellite observations (Gasparini et al 2018) and was used in several process-based studies that focused on cirrus clouds and their responses to various forcings (Kuebbeler et al 2012, 2014, Gasparini and Lohmann 2016, Gasparini et al 2017). ECHAM6-HAM is run in the mixed layer ocean setup, which explicitly simulates the interactions between the atmosphere and the surface layer of the ocean and sea ice, but neglects possible responses of deep ocean currents.

We also use the NCAR Community Earth System Model (CESM) version 1.2.2, which couples separate model components for the atmosphere, ocean, land, and sea ice (Hurrell et al 2013). We use the atmospheric component Community Atmosphere Model (CAM) version 5.3, run at a horizontal resolution of 1.9° latitude by 2.5° longitude with 30 vertical levels. The standard configuration for CAM uses the Zhang-McFarlane deep convection scheme (Zhang and McFarlane 1995), with the dilute plume closure assumption by Neale et al (2008). The shallow convection parameterization follows Park and Bretherton (2009). The stratiform cloud scheme is handled by two separate components: a macrophysics scheme for grid-scale condensation and cloud fraction calculations (Park et al 2014) and a microphysics parameterization for sub-grid scale cloud processes (Morrison and Gettelman 2008). However, for ice nucleation in cirrus clouds, the parameterization scheme by Barahona and Nenes (2009) was used. The default ice cloud macrophysics scheme in CAM 5.3 is the modified Slingo (1987) scheme as in Gettelman et al (2008). The aerosol size distribution is described by a 3-mode scheme described in Liu et al (2012). The atmosphere is coupled to a Slab Ocean Model (SOM) to include the thermodynamic effects of the ocean mixed-layer. Its spatially varying depth is based on observations of the annual-mean mixed-layer depth (Kiehl et al 2006). The SOM treats the ocean as motionless and perfectly mixed throughout its depth.

We perform three simulations with each model (table 1): a reference simulation with present day CO₂ concentrations (REF), a simulation with 1.5 × present day CO₂ concentrations (1.5CO₂), and a cirrus geoengineering simulations for the 1.5 × CO₂ climate. The simulations are run for 80 years (ECHAM) or 100 years (CESM), where we always consider only the last 60 years of data with a monthly averaged output frequency. The CESM simulation is longer due to a longer equilibration time compared to ECHAM. The seeding strategy in ECHAM follows the results of Gasparini et al (2017): all cirrus clouds (clouds at temperatures colder than −35 °C) are seeded with a concentration of 1 INP l⁻¹ using 50 μm large seeding nuclei only during night. Such setup does not only decrease the amount of seeded material needed but was also shown to increase the cooling efficacy and decrease the convective precipitation responses (Gasparini et al 2017).

The CESM seeding strategy assumes a globally uniform seeding of 18 INP l⁻¹ using 10 μm large seeding nuclei as in the HOMHET_50% scenario in Storelvmo and Herger (2014). CESM1 (unlike CESM2) does not consider pre-existing ice crystals, primarily because their existence has not been documented. The differences in simulation of cirrus clouds and their responses to seeding aerosol lead to a different choice of seeding strategies between the two models. A CESM-like seeding strategy would in ECHAM lead to overseeding and a warming of climate, while seeding with only 1 INP l⁻¹ in CESM would lead only to a minimal climatic cooling effect. We note that the total mass of delivered particles is due to the cubic dependence of mass on seeding aerosol radius about 3.5 times larger in the ECHAM seeding scenario compared to the CESM one. We stress that the main purpose of this study is not to understand the subtle differences in microphysical modeling setups and underlying microphysical responses but to rather evaluate the climatic responses of the modeled maximal cirrus seeding effects in both models.

As the atmospheric models are coupled to a shallow mixed-layer ocean, it takes approximately 20 (ECHAM) to 40 (CESM) years to come into climatic equilibrium. Thus we use only data from the last 60 simulated years. The CO₂ concentrations for the reference (present day) conditions are taken as 353.9 ppm (1990 concentrations, Taylor et al 2012), while the CO₂ concentrations in the 1.5 × CO₂ simulations are 530.9 ppm, roughly equivalent to the concentrations in the last decades of the 21st century from the RCP4.5 scenario (van Vuuren et al 2011). Therefore, we can compare our increased emission simulation result to CMIP5 model output for years 2081–2100 (Collins et al 2013). We use the double sided Student's t-test at the 95% significance level to test the robustness of our results.

We define a damage function with a quadratic dependence on mean temperature and precipitation anomalies normalised by their respective natural variability (one standard deviation range of the present day climate simulation) as

$$\begin{eqnarray}&&{\rm{Damage}}={\left(\displaystyle \frac{{\rm{\Delta }}T}{T\_{stdev}}\right)}^{2}+{\left(\displaystyle \frac{{\rm{\Delta }}p}{p\_{stdev}}\right)}^{2}.\end{eqnarray} \tag{ 1 }$$

Studies assessing climate change impacts frequently use quadratic damage functions (Keller et al 2004, Nordhaus 2008, Weitzman 2010, Nordhaus and Sztorc 2013, Kravitz et al 2014) or some higher order functional form (Goes et al 2011). However, the precise functional shape does not considerably affect the outcomes of our moderate climate change scenario, when temperatures do not exceed 2 °C of warming with respect to the reference simulation (Weitzman 2010, Kopp et al 2012).

The function is nondimensional, and defined as strictly positive (or equal to zero), where a higher value means a larger damage with respect to the present day climate. We use an arbitrary scale with no upper limit. The calculated damage serves thus only as a tool to compare the 1.5CO₂ simulation with the geoengineered simulations.

To make our damage function more relevant for the society, we consider only land gridboxes, which are divided into 21 larger geographical units (Giorgi and Francisco 2000), covering all continents except Antarctica (table S2). The damage function input values are area weighted means of temperature and precipitation, separately shown in figures S4 and S5 are available at stacks.iop.org/ERL/15/054002/mmedia.

We define cirrus clouds as all clouds at temperatures colder than −35 °C and compare the cloud radiative effect (CRE) of all clouds formed at such conditions (figures 1(a) and (b)). In both models the cirrus CRE peaks in the tropics, particularly in the convectively active Warm Pool region, and in the storm track regions. CESM simulates a higher ice water content at cirrus levels (figure S1), resulting in a more positive net CRE (6.8 W m⁻²) compared with ECHAM (4.8 W m⁻²). Nevertheless, the radiative effects of both models lie within the range of the available satellite observations of cirrus CRE (Hong et al 2016, Matus and L'Ecuyer 2017).

Zoom In Zoom Out Reset image size

The larger, more positive cirrus CRE alone, implies a larger radiative impact of seeding in CESM than in ECHAM, assuming the same effectiveness of seeding strategies, which can be seen in figures 1(c), (d). However, as shown by panels e and f, the models do not only differ in the simulated unperturbed cirrus and their CRE, but also in the seeding effectiveness, defined as $-100* \tfrac{{\rm{cirrusCREanomaly}}}{{\rm{cirrusCRE}}}$. A higher effectiveness points out at a better cancellation of the (positive) cirrus CRE by seeding. The seeding effectiveness is larger in most regions in CESM compared to ECHAM, with globally averaged values of 27% and 18%, respectively. In CESM a large fraction of extratropical cirrus are dominated by homogeneous freezing and are therefore sensitive to the introduction of seeding aerosol. Seeding aerosol decrease cirrus cloud frequency and optical properties, leading to a globally averaged net CRE anomaly of −1.8 W m⁻² (figure 1(d)). In contrast, only a small fraction of cirrus simulated by the ECHAM model forms by homogeneous freezing (Gasparini and Lohmann 2016), limiting the overall seeding effectiveness and the net seeding effect to −0.8 W m⁻² in the global average (figure 1(c)). Interestingly, the seeding in ECHAM is most effective over mountains and in parts of the tropics. The seeding effectiveness in ECHAM is slightly larger in the NH, while CESM has a 7% larger effectiveness in the SH. The CESM model simulates a smaller dust burden in the SH and consequently a higher proportion of homogeneously formed cirrus clouds (Storelvmo and Herger 2014). On the other hand, negative effectiveness implies a more positive cirrus CRE due to cirrus seeding, known as 'overseeding' (Storelvmo and Herger 2014). Overseeding occurs particularly in areas dominated by heterogeneous cirrus cloud formation mechanisms, such as the Middle East and Northern Africa in CESM (figure 1(f)), or Australia in ECHAM (figure 1(e)). In summary, a larger reference (unperturbed) cirrus CRE and a higher seeding effectiveness due to more homogeneously formed cirrus clouds lead to a more than two times larger radiative response to seeding in CESM compared to ECHAM (−1.8 versus −0.8 W m⁻²).

3.2.1. Mean temperature responses

The 1.5 × CO₂ concentrations in ECHAM cause a global average warming of 1.8 °C (figure 2(a)), which falls in the middle of the likely range of the end-of-the-century warming by the IPCC models that follow the RCP4.5 emission scenario. CESM on the other hand has a higher climate sensitivity (Tan et al 2016) which results in a 2.0 °C global warming (figure 2(d)). The mean temperature responses to a combination of increased CO₂ concentration and seeding of cirrus clouds differ substantially between the two models (figures 2(b) and (e)). In ECHAM seeding counteracts about 0.6 °C of the global mean temperature increase due to the CO₂, resulting in a net global temperature increase of 1.3 °C with respect to present day conditions. Seeding is more effective in CESM, counteracting about 1.4 °C of the global meant temperature increase due to CO₂, leading to a warming of only 0.6 °C.

Zoom In Zoom Out Reset image size

Temperature responses to seeding are tightly connected to CRE anomalies resulting from seeding (figure 1) described in the previous subsection. The temperature response is always stronger in the winter hemisphere, where the scattering of SW radiation by cirrus is minimal due to low insolation. Interestingly, the effect of seeding in CESM is larger in the Southern Hemisphere (SH) than in the Northern Hemisphere (NH, figure 2(f)), while ECHAM shows the opposite response (figure 2(c)). SH high latitude cirrus form almost exclusively by homogeneous ice nucleation in CESM (Storelvmo and Herger 2014), while ECHAM preferentially forms homogeneous cirrus over mountains, which are more frequent in the NH compared with the SH (Gasparini and Lohmann 2016).

3.2.2. Mean precipitation responses

The globally averaged precipitation increase of 3.5% (ECHAM) and 3.8% (CESM) in the 1.5CO₂ climate is mainly driven by the slow, surface temperature dependent response to the CO₂ concentration increase. The rise in surface temperature increases the amount of water vapor in the atmosphere, enhances its radiative cooling and increases precipitation (Bala et al 2010, Bony et al 2013, Pendergrass and Hartmann 2014). Precipitation in both models increases mainly in the tropics and high latitudes, while many subtropical regions experience a drying (figures 3(a), (d) and 4(a), (d)), consistent with studies of precipitation responses to CO₂ forcing (Chou and Neelin 2004, Held and Soden 2006).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

SEED alone cannot compensate for the CO₂ driven precipitation responses—however, the global average precipitation increase is about 40% weaker in ECHAM and 20% weaker in CESM compared to the respective 1.5CO₂ simulations (figures 3(b) and (e)). Seeding has in absolute terms a smaller impact on precipitation in ECHAM compared to CESM. Interestingly, the Intertropical Convergence Zone (ITCZ) in the CESM SEED simulation shifts northwards compared to both REF and 1.5CO₂ simulations (figures 3(e), (f) and 5(b)), leading to a drier South America and Maritime Continent, and at the same time increasing precipitation in the Sahel and Central America. The driver of the shift is the temperature imbalance between the two hemispheres (figure 2(f)), which pushes the ITCZ towards the warmer hemisphere, as shown in previous work on anthropogenic and volcanic aerosol emissions (Rotstayn and Lohmann 2002, Haywood et al 2013). The drying of the SH subtropics in CESM is centered over oceans, limiting its potential impact on agriculture and society. ECHAM, on the other hand, does not simulate significant hemispheric precipitation shifts, as its radiative and temperature responses between the two hemispheres are more balanced (figures 2(b) and (c)).

We additionally compare normalized hydrological responses of both models to a unit of warming by CO₂ (figures 4(a) and (d)) and cooling by cirrus seeding (figures 4(b) and (e)). The sign of the precipitation anomalies and their regional pattern for both models changes when comparing the warming with the cooling case. Precipitation responses to seeding in ECHAM are of similar magnitude but opposite sign compared to the CO₂ warming case. Interestingly, in CESM seeding leads to one third the change in globally averaged precipitation compared to the CO₂ warming simulation, which may be caused by either the direct microphysical enhancement of precipitation due to seeding or an increased convective activity due to seeding. The zonally averaged extratropical precipitation anomalies in CESM are, on the other hand, similar for the warming and cooling case (figure 4(f)). In the tropics, however, the signal is dominated by the northward ITCZ shift.

3.2.3. Precipitation extremes

Seeding perturbations rapidly change cirrus cloud properties, potentially leading to a direct microphysical perturbation of precipitation. Moreover, a decreased occurence of cirrus clouds leads to a decreased upper tropospheric relative humidity and temperatures. The resulting increased upper tropospheric radiative cooling implies a precipitation increase (Pendergrass and Hartmann 2014), as observed in previous cirrus seeding studies (Storelvmo and Herger 2014, Gasparini et al 2017). However, as seeding cools surface temperatur, moisture decreases, decreasing both the mean and the extreme precipitation (O'Gorman and Schneider 2009). It is therefore interesting to look at how extreme precipitation events respond to seeding in a 1.5 × CO₂ climate. We define all precipitation events with precipitation rates larger than 1 mm h⁻¹ as extreme precipitation, sampled at every model timestep. We sort the precipitation events in 29 predefined bins between 0.001 and 100 mm h⁻¹. Figure 5(a) shows relative changes in the frequency of high precipitation events over land between SEED and 1.5CO₂ simulations for CESM model (defined as $\tfrac{100\cdot ({\rm{SEED}}-1.5{{\rm{CO}}}_{2})}{1.5{{\rm{CO}}}_{2}}$), isolating the seeding signal. In the global average, the frequency of extreme precipitation decreases, particularly over the SH and areas north of 35°N, as expected due to the decreased atmospheric moisture content (figure 5(b)). The SEED simulation shares a lot of the signal with the 1.5CO₂ simulation, which leads to a shift of precipitation distribution to higher rates (figure 5(c)). Interestingly, the effect of seeding alone leads to a narrowing of the precipitation distribution, decreasing the frequency of high precipitation events compared to 1.5CO₂, and increasing the frequency of moderate precipitation events. The distributions shifts due to seeding result from the interplay between the temperature-mediated decrease in the intensity of the hydrological cycle on one hand, and an increased convective precipitation frequency due to enhanced atmospheric cooling on the other hand (figure S2).

Zoom In Zoom Out Reset image size

The precipitation extremes significantly increase over central and Northern Africa and parts of the Middle East, Central America, and northernmost South America. This increase is connected to the northward shift of the ITCZ (figure 3(f)), and the associated increases in updraft velocities and mean precipitation (figure 5(b)). Figure 5(d) shows changes in the precipitation distribution for the Sahel region (10°–20° N, 20°W–40°E), in which seeding increases precipitation rates for precipitation rates of about 1 mm h⁻¹ by up to 1%. However, at the extreme tail of the distribution, for precipitation rates beyond 2 mm h⁻¹, the frequency of precipitation is similar or even slightly higher in the 1.5CO₂ simulation. We speculate that this is a response to lower moisture availability in the colder SEED simulation. Changes in moisture content were shown to be the dominant driving mechanisms of extreme precipitation events, increasing its relative importance the higher the precipitation intensity (O'Gorman and Schneider 2009, Sugiyama et al 2010).

The frequency of both convective and large-scale precipitation extremes increases after an instantaneous seeding perturbation, when the surface temperatures did not yet have time to adjust to the resulting radiative imbalance (figure S2, years 1–2 of the simulation). The origin of this increase could be related both to a direct microphysical perturbation or to a rapid adjustment to seeding leading to increased atmospheric cooling. The two sources of rapid precipitation changes cannot be separated by the current set of simulations. Consistently with figure 5(c), the global land precipitation extremes in SEED show a decrease for model years 40–100 compared to REF, following a surface temperatures cooling of 1.4 °C.

3.2.4. Normalized temperature and precipitation responses

Figure 6 shows a perspective on the relative size of the annual mean precipitation and temperature anomalies for the land regions in 4 selected latitudinal bands. Temperature experiences a large positive shift of about 5–7 standard deviations with increased CO₂ concentrations. Interestingly, the normalized temperature responses to both CO₂ and seeding do not show a polar amplification pattern due to the large temperature variability in the high latitudes compared with the tropics. This is consistent with studies on the time emergence of climate signals which first detect a significant climate change signal in low latitude regions (Mahlstein et al 2011, Hawkins and Sutton 2012). The precipitation changes are less pronounced and more uneven: NH high latitudes are most sensitive to both CO₂ and seeding forcing with increases of about 1.5 standard deviations in the 1.5CO₂ simulation, compared with changes smaller than 1 standard deviation in other regions. The NH high latitudes also respond to seeding with the largest precipitation decreases of about 1 (CESM) or 0.5 (ECHAM) standard deviations. Interestingly, the seeding in CESM leads to a small increase in precipitation over tropical land, differently from responses of other regions to seeding. This is related to the northward shift of the ITCZ (figure 3(e)) and the tropical land mass distribution, where a larger fraction of tropical land lies north of the ITCZ. Those regions experience a significant wetting that is driving the increase in mean tropical precipitation. CESM SEED simulation brings both temperature and precipitation in the selected latitudinal bands close to the range of ±2 standard deviations from the present day climate, only within a small distance in temperature space from the mean climate goal (shaded area in figure 6). In ECHAM SEED simulation the temperature deviations in these land regions remain noticeably outside the mean climate goal.

Zoom In Zoom Out Reset image size

In order to better evaluate the avoided warming and precipitation increase by increased CO₂ and by cirrus geoengineering, we assess the damages with respect to the present day REF climate simulation with the help of the quadratic damage function for the 21 global land regions (Giorgi and Francisco 2000) similarly to what Kravitz et al (2014) did for the SRM case. The annual mean damages are the largest in the 1.5CO₂ simulation (figure 7). Most of the damage is related to changes in temperature: the regional precipitation anomalies are small and fall within the natural variability, while the surface temperature signal often emerges out of the natural variability range (figures S4 and S5). The damage by increased CO₂ concentrations is largest in the Sahara, East Africa, and parts of South East Asia for ECHAM. The damage is about 30% larger in the CESM model due to its stronger temperature and precipitation response to the CO₂ forcing. The most affected regions are Equatorial Africa, South and Central America.

Zoom In Zoom Out Reset image size

SEED offsets about 50% of the annual average damage compared with the 1.5CO₂ simulation in the ECHAM model. The damage pattern remains similar as in 1.5CO₂, with Africa being most affected by changes in climate. However, as the seeding is most effective in high latitudes, the regions of Greenland, Northern Asia, and Alaska are subject to considerably smaller damage compared with the 1.5CO₂ simulation. Seeding in the CESM model on the other hand decreases climate damage to small values, showing only residual temperature-related damage in several tropical regions, as well as in northern North America and Greenland.