Aerosol distribution

It is important to assess whether single-hemisphere SAI scenarios are feasible. The injection of aerosol into the stratosphere following a volcanic eruption can lead to radically different spatial and temporal distributions depending upon the altitude and latitude of the injection30, the representation of the quasi-biennial oscillation (QBO31) and the local meteorological conditions that prevail at the time of the eruption32. We demonstrate the sensitivity to altitude and latitude by performing volcanic eruption simulations using an atmosphere only version of the HadGEM2-CCS model32 with the model top at ~84 km. The same amount of SO 2 was emitted at various different altitudes and latitudes in the NH (the results from simulations emitting into the SH reveal a strong similarity and are not shown here). It is immediately evident from Fig. 1 that only injection strategies where SO 2 is injected into high altitudes (23–28 km) at equatorial latitudes (Equator and 15°N) lead to an aerosol distribution that is approximately hemispherically symmetric. Emissions at high altitudes (23–28 km) northward of 15°N lead to aerosol distributions that are significantly larger in the NH than the SH. Emissions at intermediate altitudes of 16–23 km altitude result in aerosol predominantly in the NH. Emissions at low altitude (11–15 km) are mostly below the tropopause leading to a much more limited lifetime of the resultant aerosol. Given current technical challenges of any deliberate SAI scheme33, it is feasible that a sub-optimal SAI strategy (ie, at mid-latitudes and/or at lower altitudes) might conceivably be pursued.

Fig. 1 Investigating the sensitivity of aerosol dispersion to the altitude and latitude of a volcanic eruption. Five-year evolution of the anomaly in sulphate (SO 4 ) aerosol optical depth (AOD) for injections of sulphur dioxide into the Northern Hemisphere. Latitudes progress from 60oN in the left column (a, f, k) to the Equator in the right column (e, j, o) and injection altitudes from 23 to 28 km in the top row (a–e) to 11–15 km in the bottom row (k–o). Simulations are with a ‘high-top’ version of the HadGEM2 model with stratospheric layers up to 80 km using the CLASSIC aerosol scheme32 Full size image

We now consider the simulations in this study performed with the low-top version of the HadGEM2 model. As in other studies11,31,34,35,36, we compensate for the lack of adequately resolved QBO owing to the limited height of the top of the model by injecting over a wide range of latitudes rather than injecting at a single point. Figure 2 shows the annual-mean sulphate (SO 4 ) aerosol optical depth (AOD) anomalies in the G4, G4NH and G4SH simulations averaged over 2020–2070. It is clear that the SO 4 aerosol is primarily confined to the hemisphere(s) of injection in all of the SAI scenarios.

Fig. 2 Aerosol optical depth anomalies in the solar geoengineering simulations. Sulphate (SO 4 ) 550 nm aerosol optical depth (AOD) anomaly 2020–2070 for a, global solar geoengineering (G4); b, northern hemisphere solar geoengineering (G4NH); and c, southern hemisphere solar geoengineering (G4SH) relative to RCP4.5 Full size image

Climate changes

Regional and global SAI applications would do much to ameliorate changes in near-surface air temperature and sea-ice evident in the RCP4.5 scenario, with the principal counteractive effect occurring in the hemisphere(s) of injection (Fig. 3). However, the impacts in the un-geoengineered hemisphere would also be significant owing to atmospheric and oceanic inter-hemispheric energy transport, for instance, in these simulations an NH cooling of 0.7 K is observed in G4SH relative to RCP4.5 (2020–2070), compared to 1 K in G4 and 1.1 K in G4NH (Fig. 3b). This result corroborates previous research suggesting that the impacts of SG would not be entirely confined to the perturbed region4,8,10,11,12.

Fig. 3 Twenty-first century temperature and sea-ice changes. a Global-mean near-surface air temperature (NSAT) anomaly relative to a 240-year pre-industrial control simulation for RCP4.5, global solar geoengineering (G4), northern hemisphere (NH) solar geoengineering (G4NH),and southern hemisphere (SH) solar geoengineering G4SH; b, c, NH NSAT anomaly and total sea-ice extent (106 km2); d, e, SH NSAT anomaly and total sea-ice extent. Sea-ice extents are smoothed by a 10-year simple moving average. Vertical dotted lines at years 2020 and 2070 indicate the start and cessation of solar geoengineering, respectively Full size image

Simulated tropical cyclones

In the historical period, the model skilfully captures observed TC frequency trends as inferred from TRACK (r = 0.71 with HURDAT) including the decline in activity through 1960–1980 and the increase in activity since 1980 (Fig. 4a). The annual-mean TC frequency in HIST is 10.4 (90% confidence intervals (CIs): 5, 16) TCs per year. In RCP4.5, TC frequency decreases between 2020 and 2070 (−0.3 TCs per decade) with an annual-mean frequency of 9.7 (90% CI: 4, 15) TCs per year, while in G4, TC frequency increases slightly relative to HIST (annual-mean = 11.2 (90% CI: 5, 18) TCs per year). The TC frequency changes between 2020 and 2070, and HIST in the RCP4.5 and G4 scenarios are not statistically significant at the 5% level (Supplementary Note 3). G4SH exhibits a marked increase in TC frequency relative to HIST (annual-mean TCs per year = 14.3 (90% CI: 8, 20)), while G4NH conversely exhibits a pronounced reduction (annual-mean TCs per year = 7.6 (90% CI: 2, 13)). The G4NH and G4SH results are consistent with observed TC activity changes following volcanic aerosol enhancements confined to a single hemisphere16,17. TC frequency swiftly rebounds to concurrent RCP4.5 levels following the cessation of SAI in G4, G4NH and G4SH in year 2070 (Fig. 4b), which confirms that the SG termination effect37 extends to North Atlantic TC activity. The TC frequency changes between 2020 and 2070, and HIST in the G4NH and G4SH scenarios are statistically significant at the 5% level (Supplementary Note 3).

Fig. 4 Observed and simulated Tropical cyclone frequency. a Historical tropical cyclone (TC) frequencies, smoothed by a 10-year simple moving average, for ERA-I28, the ensemble mean of the HadGEM2-ES HIST simulations and HURDAT2 observations29. b The same as a but for the RCP4.5 and SAI simulations. The box and whisker plots (right) show the 10, 25, 50, 75 and 90% quantiles of the HIST (‘H’, 1950–2000), RCP4.5 (‘R’, 2020–2070) and SAI (2020–2070) raw annual TC frequency. G4 refers to a global SAI scenario, G4NH refers to a northern hemisphere SAI scenario, and G4SH refers to a southern hemisphere SAI scenario. Vertical dotted lines at years 2020 and 2070 (b) indicate the start and cessation of solar geoengineering, respectively Full size image

Tropical cyclone proxies

The progression of AEWs to TCs is contingent on the ambient meteorological conditions, which may act to induce or dissipate the storm. For instance, enhanced wind shear over the MDR counteracts cyclogenesis20, whereas a warm ocean surface provides the storm vortex with energy21. Historical trends in MDR wind shear, precipitation and relative SST closely correlate with North Atlantic TC activity (Fig. 1 in ref. 13) and these indices offer an alternative tool to counting vortices for predicting future TC trends. Figure 5 shows various North Atlantic TC indices as extracted from the HadGEM2-ES simulations. It is clear that active (1950–1965, 1995–2014) and inactive (1965–1995) TC periods in the HIST simulation (Fig. 4a) were commensurate with active and inactive periods in the indices (Fig. 5). The same correlations between indices and TC frequency persist in the RCP4.5 and SAI simulations, with G4SH and G4NH exhibiting continuously positive and negative indices, respectively (Fig. 5). This suggests that meteorological conditions presently conducive to cyclogenesis remain conducive in these scenarios. Figure 6 shows maps of precipitation, wind shear, and relative SST anomalies in the G4NH and G4SH scenarios. In G4NH, aerosol-induced cooling of the North Atlantic sea surface (> 30°N) results in a southward shift and strengthening of the African easterly jet (AEJ), enhanced wind shear in the MDR, and anomalous descent and precipitation reduction over the MDR (Fig. 6a–c)13. Conversely, preferential cooling of the South Atlantic in G4SH enhances ascent and precipitation in the MDR and shifts the AEJ north, reducing wind shear over the MDR and producing favourable conditions for cyclogenesis (Fig. 6d–f).

Fig. 5 Modelled tropical cyclone-related climate indices. a June–November (JJASON) precipitation anomaly (relative to 1950–2000) averaged over the hurricane main development region (MDR) (5°–20°N, 15°–85°W). b The same as a but for inverse vertical zonal-wind shear. c The same as a but for relative SST. G4 refers to a global SAI scenario, G4NH refers to a northern hemisphere SAI scenario, and G4SH refers to a southern hemisphere SAI scenario. Vertical lines at years 2020 and 2070 indicate the start and cessation of solar geoengineering, respectively Full size image

Fig. 6 Climate anomalies in the hemispheric solar geoengineering simulations. a Northern hemisphere only solar geoengineering (G4NH) 2020–2070 JJASON precipitation anomaly relative to 1950–2000. b The same as a but for vertical zonal-wind shear. c The same as a but for relative SST. d–f The same as a–c but for southern hemisphere only solar geoengineering (G4SH). Stippled regions on the maps show where differences are outside the 90% variability of a 240-year pre-industrial control ensemble mean. The hurricane main development region (MDR, (5°–20°N, 15°–85°W)) is marked by black rectangles Full size image

Figures 5 and 6 suggest that enhanced TC activity is related to certain climatic conditions in the MDR, in particular enhanced precipitation, attenuated vertical wind shear and a warmer sea surface (relative to the tropical mean). It is important to investigate these relationships using observations and reanalyses to ascertain their practical robustness. Figure 7 shows time series for TC frequency29, precipitation and vertical wind shear (U 850 –U 25 )38, and relative SSTs39 from reanalyses and observations. From comparing Fig. 7a with Fig. 7b, c, d, it appears that periods of enhanced TC activity in the 20th century coincided with enhanced precipitation and relative SSTs and attenuated vertical wind shear13, which substantiates our modelling results (Fig. 5). The closest relationship in terms of active and inactive periods in Fig. 7 is between TC activity and relative SST. Statistical models for count data using a Poisson distribution framework can be developed to quantify the observed relationships between TC activity and MDR meteorology (Supplementary Note 4)21. Figure 8 shows time series of TC activity from the HadGEM2-ES simulations as determined by applying the statistical relationships from the historical observations (Fig. 7) to the simulated meteorology, where the covariates are anomalies from the 1900 to 2005 mean values. The covariate trends suggest enhanced (attenuated) TC activity in the G4SH (G4NH) simulation between 2020 and 2070 relative to HIST and RCP4.5 (Fig. 8), which substantiates the results of the explicit storm tracking (Fig. 4). We find little evidence to support the hypothesis that the simulated TC frequency changes (Fig. 4) are the result of an El Nino Southern Oscilation response, which further supports the ITCZ-TC connection theory (Supplementary Note 5).

Fig. 7 Twenty-first century North Atlantic climate indices from observations and reanalyses. June–November (JJASON) observations and reanalysis data time series for: a, TC frequency29; b, hurricane main development region (MDR, (5°–20°N, 15°–85°W)) precipitation and c, wind shear38; and d, sea-surface temperature (MDR minus the Tropics)39. Diamonds indicate JJASON average quantities for each year in the MDR or North Atlantic basin, and thick black lines denote 10-year moving averages. Red (blue) regions indicate where moving averages are above (below) the 1900–2010 mean values Full size image

Fig. 8 Tropical cyclone frequency inferred from various statistical relationships with climate indices. TC frequency inferred from HadGEM2-ES meteorology using statistical relationships developed from historical observations and reanalyses. The covariates used are: a, MDR precipitation; b, MDR wind shear; c, relative sea-surface temperatures (MDR minus tropical mean); and d, MDR and tropical SSTs as separate covariates. Vertical dotted lines at years 2020 and 2070 indicate the start and cessation of solar geoengineering, respectively Full size image

Statistical-dynamical downscaling

Statistical-dynamical downscaling models are able to simulate the observed intensity distribution of North Atlantic TCs40, whereas explicitly simulated storms are not as intense as those observed (Supplementary Note 6). Therefore, we employ a downscaling model to investigate changes to the most intense storms under global warming and SAI. Forced by HadGEM2-ES meteorology, the model is clearly able to reproduce TC trends in the recent historical period (Fig. 9a), although the frequency of major hurricanes (max windspeed > 96 m s−1) is undersimulated in the 1960s compared to HURDAT observations (Fig. 9c). In contrast to the results of the explicit storms (Fig. 4), the model shows a steadily increasing trend in TC frequency in the RCP4.5 scenario over 2020–2070 (Fig. 9a), in agreement with the results of applying downscaling to the CMIP5 ensemble using the RCP8.5 scenario24. SAI generally counteracts the intensification of TC activity relative from RCP4.5, except in the interesting case of the first ~10 years in G4SH, which exhibits an increase in major hurricane, hurricane and TC activity (Fig. 9c). G4SH consistently produces the most TCs per year relative to the other SAI scenarios, with 2020–2070 mean frequencies of 12.6, 8.8 and 3.3 TCs, hurricanes and major hurricanes per year, respectively. This can be compared to 10.5, 7.3 and 2.6 for G4NH; 10.8, 7.2 and 2.5 for G4; and 15.3, 11.1 and 4.3 for RCP4.5. The TC frequencies in the SAI simulations are significantly different to RCP4.5, and the G4SH TC frequencies are significantly greater than G4 and G4NH (Supplementary Note 7). Following the cessation of SAI in 2070, TC activity rebounds to the baseline RCP4.5 activity within ~15 years (Fig. 9a), again confirming that the SG termination effect extends to North Atlantic TCs31.