Potential volcanic impacts on future climate variability

Volcanism has been a major driver of past climate variability1 and will continue to affect future climate alongside human influences9. Explosive volcanic eruptions warm the stratosphere10, cool the troposphere11, cause changes in the hydrological cycle12, 13, and trigger modifications of atmospheric circulation that give rise to large regional climate responses14. The instrumental period covering the past 150 years has been relatively volcanically quiescent, and it is therefore tempting to ascribe potential volcanism a minor role in future climate impact and risk assessments. In a millennial perspective, however, there have been periods with considerably stronger volcanic activity5 (Supplementary Fig. 1). Clustered occurrence of strong tropical eruptions has contributed to sustained cold periods such as the Little Ice Age15, where the longer-term climate impacts are mediated through ocean heat content anomalies16 and ocean circulation changes17, 18, 19 that also affect global and regional sea level20 and sea-ice conditions15, 17.

Because volcanic eruptions are unpredictable events, they have generally been excluded from twenty-first century climate projection protocols. Most recent projections either specify future volcanic forcing as zero or a constant background value4, whereas considerations of more realistic volcanic effects have been limited to idealized eruption scenarios, repeating recent volcanic activity in near-future simulations21, 22. Herein we explore whether a more complete representation of volcanic forcing uncertainty that considers a range of volcanic forcing possibilities will have an impact on important aspects of probabilistic twenty-first century projections increasingly being used for adaptation planning purposes. The risk from not realistically accounting for volcanic forcing effects is that critical possible future outcomes are being discounted and maladaptation ensues.

The possibility of utilizing stochastic volcanic forcing in projections has been recognized in previous studies23 and underscored in the latest assessment report of the Intergovernmental Panel on Climate Change2. Increasing computational power, facilitating large ensemble simulations24, together with improved reconstructions of past volcanic activity5, 25 that allow for a better statistical characterization9, 23, 26, make it timely to revisit the question of volcanic effects on twenty-first century climate projections. We start by deriving plausible future volcanic forcings (Fig. 1) by sampling from reconstructed volcanic activity of the past 2,500 years5 (Supplementary Fig. 1). We next perform three twenty-first century simulation ensembles with the Norwegian Earth System Model (NorESM)6, that use the same mid-range anthropogenic forcing scenario RCP4.5 (ref. 7) but differ in their volcanic forcing: a 60-member ensemble using plausible stochastic volcanic forcing (VOLC); a 60-member reference ensemble using zero volcanic forcing (NO-VOLC); and a 20-member ensemble using 1850–2000 averaged volcanic forcing27 (VOLC-CONST). NO-VOLC and VOLC-CONST are the two approaches that were adopted across the group of models contributing twenty-first century projections to the Coupled Model Intercomparison Project phase 5 (CMIP5) (ref. 3). Hence we consider both as useful counterfactual cases here to aid reader interpretation of possible limitations in existing twenty-first century projection runs. Specifically, we assessed the volcanic influence on the climate variability and means of future projections by comparing our three ensembles for several societally relevant diagnostics.

Figure 1: Historical and plausible future volcanic forcing.

a, Stratospheric volcanic aerosol loadings in the model’s historical and twenty-first century CMIP5 simulations6, 27 (grey) and synthetic forcing realizations with statistically low (5%, green), mid (50%, cyan) and extreme (>98%, black) loadings when compared to the historical reconstructions over the past 2,500years5. The data are monthly, filtered with a 12-month running mean. b,c, Stratospheric aerosol loading time series (b) and their century means for all simulation members (c). Members are ranked according to their time-mean loadings. Colour marked realizations correspond to the three realizations displayed in a.

We start by examining the impact of future volcanic activity on Global-Mean Surface Air Temperature (GMST)—an integrated climate-change indicator of particular relevance to mitigation decision making. Figure 2a shows annual-mean GMST changes over the course of the twenty-first century as simulated in the three ensembles. The effect of volcanic forcing on the ensemble mean temperature (thick solid lines) is modest, amounting to a 5% reduction of the centennial GMST change projected under RCP4.5, with VOLC and VOLC-CONST being slightly cooler than NO-VOLC throughout the post-2005 period, as expected from the first-order response to volcanic forcing. Near-term GMST projections for the 2016–2035 period (Fig. 2b) exhibit only a small (0.05K) reduction in mean response in VOLC and VOLC-CONST, with an increased skew in VOLC leading to a 0.1K shift in the lower distribution tail. As a result, the 1.5°C warming target of the Paris agreement COP-21 (ref. 24) is exceeded on average two years later in VOLC and VOLC-CONST (Supplementary Table 1), with the upper distribution tail of VOLC being shifted by twice that amount (Fig. 2c). Models which did not include constant background forcing in their standard twenty-first century simulations prepared for CMIP5 are thus slightly overly pessimistic as to the likely time until different warming thresholds are reached.

Figure 2: Annual-mean GMST.

Annual-mean GMST.

a, Ensemble mean (solid) of VOLC (blue), VOLC-CONST (magenta) and NO-VOLC (red/orange) with 5–95% range (shading) and ensemble minima/maxima (dots) for VOLC and NO-VOLC; evolution of the most extreme member (black). b, Probability density function (PDF) of the 2016–2035 mean relative to pre-industrial (PI, see Methods), with 5–95% bootstrap confidence bounds. c, PDF of the time when SAT change relative to PI (20-year running average) exceeds 1.5K. d, PDF of annual anomalies with anthropogenic trend removed. The spread of VOLC-CONST is linearly shifted relative to NO-VOLC, and therefore not shown in ac.

Over the course of the simulation period, the ensemble mean difference grows, eventually saturating just below 0.1K around 2040 (Fig. 2a), after which the means are well separated. The delay highlights the role of slow-response components, particularly the ocean16, 17, in aggregating the global response to episodic volcanic forcing. The general correspondence of the VOLC and VOLC-CONST ensemble shows that the application of a time-invariant background forcing adequately accounts for long-term aspects of volcanic impacts in the ensemble mean projections28. One could use plausible low and high background values to further account for projection uncertainty stemming from uncertainty in the centennial-mean volcanic forcing (Fig. 1c and Supplementary Fig. 1). However, this would fail to capture the response to episodic volcanic forcing and attendant impacts on annual-to-decadal variability and extremes.

The interannual uncertainty range (5–95% ensemble spread) in annual-mean GMST is inflated by more than 50% (from 0.3 to 0.5K) in VOLC relative to NO-VOLC (Fig. 2a—red versus blue shading; Supplementary Fig. 9 shows GMST from individual ensemble members). Consistent with a tropospheric cooling response, the change in ensemble spread in VOLC relative to NO-VOLC is skewed towards lower GMST, leaving the higher bound largely unaltered (Fig. 2d). Reductions in frequency of extremely warm years are generally small, whereas increases in frequency of extremely cold years—relative to the moving average or ‘present-day’ climate at any point—are much more substantial. In contrast, the application of a constant background forcing merely shifts the distribution of VOLC-CONST relative to NO-VOLC, overestimating the reduction of warm years and underestimating the increase of cold years.

Decadal-scale GMST series are even more affected by future volcanic forcing uncertainty than annual temperatures (Fig. 3a). The distribution of the decadal means—with the global warming trend removed prior to the analysis—is considerably wider for VOLC than for NO-VOLC, with roughly a doubling in standard deviation (Fig. 3b and Supplementary Table 1). Anomalously cold decades become more frequent at the expense of ‘normal’ and, to a lesser degree, anomalously warm decades. As for decadal means, the spread in decadal trends is significantly wider for VOLC than for NO-VOLC (Fig. 3c). Occurrences of decades with negative GMST trend become more frequent if accounting for volcanic forcing, with the probability increasing from 10% in NO-VOLC to more than 16% in VOLC (Fig. 3d). Conversely, the widening of the upper tail of the decadal trend distribution (Fig. 3c) indicates enhanced probability of decadal-scale warming surges, due to the rebound of GMST after volcanic-induced cooling has reached its maximum (see Supplementary Fig. 4). The probability of decades with negative GMST trend more than doubles from 4% to 10% (Fig. 3e) if the analysis is limited to the first half of the century—before the stabilization period of RCP4.5—suggesting that the relative impact is sensitive to other forcings and depends on the anthropogenic scenario. Volcanic-induced cooling becomes increasingly important in facilitating neutral or negative temperature trends on longer timescales on which natural internal variability effects such as El Niño are no longer sufficient to offset anthropogenic forcings (Fig. 3f, g).

Figure 3: Decadal temperature means and trends.

Decadal temperature means and trends.

a, Decadal means of GMST relative to pre-industrial. Ensemble mean (solid) with 5–95% range (shading) of VOLC (blue) and NO-VOLC (red). b, PDF with 5–95% bootstrap confidence bounds of decadal anomalies (without overlap) relative to NO-VOLC ensemble mean. c, As b, but for decadal trends. d, Cumulative probability distribution with 5–95% confidence bounds for decadal trends (with overlap), using a 10-year window that is moved over 2006–2099. f, Probability for obtaining negative trends as function of length (solid) with 5–95% bootstrap confidence bounds (shading). e,g, As d,f, but for the shorter period 2006–2050.

That volcanic influence is not limited to GMST projections becomes evident from assessing selected global and large-scale climate indicators that all have previously been found to be sensitive to volcanism13, 17, 18, 19, 20 (Fig. 4 and Supplementary Table 1). The radiative forcing at the top of the atmosphere is reduced by 0.05Wm−2 on average (Fig. 4a), whereas its decadal standard deviation, including the anthropogenic RCP4.5 signal, is increased by 80% in response to volcanic forcing. The distribution of decadal radiative anomalies is widened, with a skew towards lower values (Fig. 4b) and a slight occurrence of more positive extremes resulting from reduced radiative surface cooling in post-eruption years. Global sea-level rise is on average slowed by 4% (relative to RCP4.5) in VOLC compared to NO-VOLC (Fig. 4c) as a direct consequence of reduced heat uptake by the oceans. The distribution of decadal sea level anomalies is significantly widened (doubling of standard deviation after subtracting global warming trend) with the lower uncertainty tail being affected most (Fig. 4d). Contrary to GMST, the volcanic forcing is generally not strong enough to halt global steric sea-level rise by offsetting anthropogenic-driven ocean warming on decadal and longer timescales. Asian summer monsoon precipitation shows consistent, albeit small reductions (Fig. 4e), with all decades featuring lower ensemble means, and a 20% overall increase in ensemble standard deviation (Fig. 4f). The Atlantic Meridional Overturning Circulation at 26°N shows a relative strengthening of 0.2Sv in VOLC compared to NO-VOLC (Fig. 4g), with all decades exhibiting increased ensemble means, and a 20% overall increase in ensemble standard deviation (Fig. 4h). Similarly, Arctic sea-ice volume shows a 1–2% relative increase for most decades (Fig. 4i) and a 15% increase in ensemble standard deviation (Fig. 4j), with more overlap between the spread of subsequent decades in VOLC compared to NO-VOLC, indicating enhanced probability for a temporary halt in Arctic sea-ice decline.

Figure 4: Decadal means of large-scale climate indicators.

Decadal means of large-scale climate indicators.

a, Top-of-atmosphere net radiation balance. c, Global steric sea level. e, May–September precipitation, averaged over Asian continent box (60°–135°E, 5°–55°N) (see Supplementary Fig. 5). g, Atlantic Meridional Overturning Circulation (AMOC) strength at 26°N. i, Northern Hemisphere sea-ice volume. Ensemble mean relative to pre-industrial (solid) of VOLC (blue) and NO-VOLC (red) with 5–95% range (shading). b,d,f,h,j, Corresponding PDFs, normalized by maximum value and NO-VOLC ensemble mean subtracted prior to computation, with 5–95% bootstrap confidence bounds. The results of VOLC-CONST are shifted relative to NO-VOLC but otherwise similar and therefore not shown.

To address if the inclusion of volcanic forcing variability has local implications we performed a time-of-emergence (ToE) analysis8 on seasonally averaged surface air temperature (Fig. 5). The ToE is formally defined as the mean time at which the signal of global warming emerges from the noise of natural climate variability (see Methods). The simulated impact of volcanic forcing variability on the ToE changes is distinct but small. The ToE is delayed almost everywhere as a consequence of the inclusion of volcanic forcing (Fig. 5a, b). The distribution of the simulated delay is strongly skewed, with delays of up to a decade in some locations, and of three years on average (Fig. 5c, d).

Figure 5: Time of emergence of anthropogenic GMST changes.

Time of emergence of anthropogenic GMST changes.

a,b, VOLC–NO-VOLC difference in ToE relative to 1985–2005 mean for boreal winter (a) and boreal summer (b) in years. Regions with no apparent emergence before 2100 are left blank. c,d, Areal distribution of ToE differences for boreal winter (c) and boreal summer (d), with dashed red lines denoting the corresponding weighted global means. The analysis includes only regions that show emergence between 2006 and 2099 in both ensembles.

Our results highlight the importance of representing volcanic forcing uncertainty in probabilistic future climate projections, in particular for risk assessments with focus on variability and certain extremes. Counter to earlier findings of destructive interference of volcanic forcing with internal climate variability modes18, our stochastic volcanic forcing generally amplifies the annual-to-decadal scale climate variability in our model. A sharp increase in simulated decades with negative GMST trend exemplifies the effect of volcanic forcing uncertainty on projections of climate extremes with direct socio-economic consequences—such as Arctic sea-ice extent, monsoon precipitation, mid-latitude storminess and temperature. Extreme volcanic activity can potentially cause extended anomalously cold periods. This will, however, not help to mitigate long-term global warming impacts as the surface climate is likely to rebound, leaving its long-term trajectory unaltered (Fig. 2a, black curve).

This study is just a step towards incorporating current knowledge on global volcanic activity in probabilistic future climate projections in realistic and systematic ways. It serves as a proof of concept for a statistical representation of potential volcanism in twenty-first century climate projections and demonstrates the importance of such a representation for the projections of future climate variability. Our ensemble analysis, based on a single model and a single anthropogenic scenario, provides only a conditional assessment of the volcanic contribution to climate projection uncertainty29. Additional uncertainties in the volcanic forcing reconstruction, in other external forcings, and in the model warrant further experiments. Our simulations may still underestimate future volcanic impact due to a too low sensitivity to volcanic forcing and the omission of small eruptions that cannot be detected in ice cores. The relative impacts further depend on the simulated global warming and amount of unforced variability in the model (details in Supplementary Information). Since simulated regional impacts are less distinct than global impacts, and have larger model uncertainty29, quantifying volcanic impacts on regional climate projections and their socio-economic aspects should be a priority of future work. Improved characterization of past volcanic forcing, improved representation of volcanic impact in models, and coordinated multi-model efforts using the same plausible forcings are essential ingredients for advancing the utilization of volcanism information in future climate assessments. The newly established Model Intercomparison Project on climatic response to Volcanic forcing (VolMIP)30 presents the ideal platform for integrating these efforts.