Figure 2 illustrates the composite temperature anomaly for each classification and season in the CESM model. In both the ASYMMNH and ASYMMSH cases, the hemisphere that is subjected to the strongest forcing is preferentially cooled. In the ASYMMNH results, the cooling peaks over the Eurasian and North American continents. As expected, there tends to be a much larger response over land, as well as evidence of NH winter warming in the mid-to-high latitudes, a phenomenon previously highlighted in the literature and often associated with increased (decreased) pole-to-equator stratospheric (mid-tropospheric) temperature gradients (Fig. S1 in the Supplement) and a positive mode of the Arctic–North Atlantic Oscillation (Robock and Mao, 1992, 1995; Stenchikov et al., 2002; Shindell et al., 2004; Ortega et al., 2015). This effect is weak in the ASYMMNH composite, likely because the maximal radiative forcing is located in the NH, offsetting any dynamical response, but is present in the SYMM and ASYMMSH composites in both models (see Fig.  S2 for the GISS-E2 composite).

In the SH, cooling is muted by larger heat capacity associated with a smaller land fraction, with weak responses over the Southern Ocean while still exhibiting statistically significant cooling in South America, South Africa, and Australia in all cases. In fact, the cooling in the ASYMMSH composites is largely confined to the tropics, in contrast to the polar amplified pattern that is common to most climate change experiments. The cooling in all categories is communicated vertically (Fig. S1 in the Supplement) and across the free tropical troposphere, suggesting AET toward the forced hemisphere (Sect. 3.4) for asymmetric forcing.

The cooling in the GISS-E2 model (Fig. S2 in the Supplement) displays a very similar spatial structure to CESM in all categories but with much greater amplitude due to the larger forcing. We note that the composite-mean forcing is similar between the four asymmetric panels but larger in the symmetric cases. In Fig. 3, we show the hemispheric and global average temperature response for both models after normalizing each event by a common global aerosol mass excursion, thereby accounting for differences in the average forcing among the different eruptions. This is done to highlight spread associated with internal variability and model differences, and it assumes that the response pattern scales linearly with global forcing, which is unlikely to be true across all events and for the two models. Nonetheless, the gross features of the hemispheric contrast and reduction in global-mean temperature are shared between both models.

The CESM precipitation response is shown in Fig. 4 (Fig. S3 in the Supplement for GISS-E2). For both the ASYMMNH and ASYMMSH cases, the ITCZ shows a robust displacement away from the forced hemisphere. The precipitation reduction in the SYMM composites is much less zonally coherent, instead featuring tropical-mean reductions in precipitation and a slight increase toward the subtropics (see also Iles et al., 2013; Iles and Hegerl, 2014). Despite global cooling and reduced global evaporation (not shown), the ITCZ shift in ASYMMNH and ASYMMSH tends to result in precipitation increases in the hemisphere that is least forced (Fig. 5), since the hemispheric-mean precipitation signal is largely influenced by the ITCZ migration itself.

The ensemble spread in precipitation for a selected eruption (1762 CE, NDJFM) is shown in Fig. S4 in the Supplement, corresponding to the Icelandic Laki aerosol loading (a large ASYMMNH event). We note that the Laki eruption in Iceland actually occurred in 1783 CE but is earlier in our composite due to an alignment error in the first version of the G08 dataset. Results are shown for the 1763 CE boreal winter only (the full composite also includes 1762, see Table 1; Fig. S4 also reports the winter 1763 Niño 3.4 anomaly in surface temperature for each ensemble member, and therefore we restrict the anomalous precipitation field to the same season). The ITCZ shift away from the NH is fairly robust across the ensemble members, particularly in the Atlantic basin, although internal variability still leads to large differences in the spatial pattern of precipitation, notably in the central and eastern Pacific.

The monthly time evolution of the composite temperature and precipitation responses for the ASYMMNH and ASYMMSH cases can be viewed in an animation (Figs. S9–S12 in the Supplement). The global and hemispheric difference in aerosol loadings is also shown for each timestep (at monthly resolution) in the animations. When averaged over the individual eruptions within each classification, the global aerosol mass loading remains elevated above 8 Tg for nearly 2 years, coincident with the peak temperature and precipitation response that begins to dampen out gradually and relaxes back to pre-eruption noise levels after ∼ 4–5 years. The seasonal migration of anomalous precipitation in the ITCZ domain occurs in nearly the same way as the meridional movement of climatological rainfall, highlighting important connections between the timing of the eruption relative to the seasonal cycle of rainfall at a given location.

In both CESM and GISS-E2, the ITCZ shift is approximately scalable to eruption size. For both models, we define a precipitation asymmetry index, PAi (Hwang and Frierson, 2013), in each season as the area-weighted NH tropical precipitation minus SH tropical precipitation (extending to 20∘ latitude) normalized by the model tropical-mean precipitation, i.e., PAi=PEQ-20∘N-P20∘S-EQP20∘S-20∘N. Supplement Fig. S5 illustrates the relationship between PAi and the AOD gradient between hemispheres (AOD is inferred for the CESM model by dividing the aerosol loading by 75 Tg in each hemisphere, an approximate conversion factor to compare the results with GISS-E2). The mis-scaling in GISS-E2 results in a wider range of AOD gradients than occurs in CESM. Both models feature more tropical precipitation in the NH (SH) during boreal summer (winter) in their climatology, with more asymmetry in CESM during boreal summer. Interestingly, the most asymmetric events in GISS-E2 (those that result in equatorward precipitation movements) can be sufficient to produce more precipitation in the tropical winter hemisphere, thus competing with the seasonal insolation cycle in determining the seasonal precipitation distribution.

The meridional ITCZ shift leads to a number of important tropical climate responses. For example, an intriguing feature of the temperature pattern in Fig. 2 is the El Niño response that is unique to the ASYMMNH composites. This is unlikely to be a residual feature of unforced variability, since there are 288 events in the ASYMMNH composites (16 eruptions in Table 1, multiplied by 18 ensemble members), significantly more than in the other categories. The GISS-E2 temperature composite (Fig. S2 in the Supplement) also features a relatively weak cooling for ASYMMNH, despite the very large radiative forcing. The relationship between ENSO and volcanic eruptions has, historically, been quite complicated due to the problem of separating natural variability from the forced response and due to a limited sample of historical eruptions where ENSO events were already underway prior to the eruption. Older studies have suggested that El Niño events may be more likely 1 to 2 years following a large eruption (e.g., Adams et al., 2003; Mann et al., 2005; Emile-Geay et al., 2008). Our findings are also consistent with recent results (Pausata et al., 2015) that found an El Niño tendency to arise from a Laki-like forcing (in that study, a sequence of aerosol pulses in the high latitudes that was confined to the NH extratropics), and the El Niño response in CESM to different expressions of volcanic forcing was recently explored in Stevensonn et al. (2016). Pausata et al. (2015) attributed the El Niño development directly to a southward ITCZ displacement. Since low-level converging winds are weak in the vicinity of the ITCZ, a southward ITCZ displacement leads to weaker easterly winds (a westerly anomaly) across the central equatorial Pacific. This was shown for a different model (NorESM1-M) and experimental setup but also emerges in the ASYMMNH composite results for CESM. Indeed, a composite anomaly of ∼ 0.5 ∘C emerges over the Niño 3.4 domain, lasting up to 2 years (Fig. S6 in the Supplement) with peak anomalies in the first two boreal winters after an eruption. Consistent with the sea surface temperature (SST) anomalies, a relaxation of the zonal winds and redistribution of water mass across the Pacific Ocean can be observed in the ASYMMNH composite response (Fig. S7 in the Supplement).

Since the ITCZ shift is a consequence of differential aerosol loading, we argue that the El Niño tendency in CESM is a forced response in ASYMMNH but otherwise depends on the state of internal variability concurrent with a given eruption, as no such ENSO response is associated with the composite SYMM or ASYMMSH, although we note that El Niño does tend to develop in response to the Samalas eruption that was removed from our composite and would strongly influence the interpretation of the SYMM results due to the few events sampled (not shown, though see Stevenson et al., 2016). However, we also caution that this version of CESM exhibits ENSO amplitudes much larger than observations and also features strong El Niño events with amplitudes that are ∼ 2 times larger than strong La Niña events even in non-eruption years. Therefore, we choose not to further explore the dependence of our results on ENSO phasing.

Because the ITCZ responds differently to the three eruption classifications, there are implications for best practices in assessing the skill of climate model output against proxy evidence. For example, Anchukaitis et al. (2010) noted discrepancies between well-validated tree-ring proxy reconstructions of eruption-induced drought in the Asian monsoon sector and the precipitation response following volcanic eruptions derived from the NCAR Climate System Model (CSM) 1.4 millennial simulation. However, we note that monsoonal rainfall responds differently to ASYMMNH, ASYMMSH, or SYMM events in both GISS-E2 and CESM. Figure S8 in the Supplement shows a histogram of boreal summer (MJJAS) Asian Pacific rainfall anomalies for all events in both models. ASYMMNH and SYMM eruptions generally lead to reductions in rainfall over the broad region averaged from 65 to 150∘ E and 10–40∘ N (see also the spatial patterns in Fig. 4 for CESM and Fig. S3 in the Supplement for GISS E2-R). Because of the southward ITCZ shift in ASYMMNH, the most pronounced precipitation reductions occur for events within this category. In contrast, for ASYMMSH events, the northward ITCZ shift and associated monsoon developments are such that precipitation changes are relatively muted, and often the anomalies are positive.

An ITCZ shift away from the forced hemisphere will manifest itself in several other components of the tropical hydroclimate system that are important to consider from the standpoint of both impacts as well as the development of testable predictions. One such important component of the hydrologic cycle is global streamflow, a variable that is related to excessive or deficient precipitation over a catchment. Rivers are important for ecosystem integrity, agriculture, industry, power generation, and human consumption. Streamflow anomalies associated with volcanic forcing in observations and models have previously been documented for the historical period (Trenberth and Dai, 2007; Iles and Hegerl, 2015). Here, we discuss this variable in the context of our symmetric and asymmetric composites.

The hydrology module of the land component of CESM simulates surface and subsurface fluxes of water, which serve as input into the CESM River Transport Model (RTM). The RTM was developed to route river runoff downstream to the ocean or marginal seas and enable closure of the hydrologic cycle (Oleson et al., 2010). The RTM is run on a finer grid (0.5∘ × 0.5∘) than the atmospheric component of CESM.

Figure 6 shows the river discharge anomalies in our different forcing categories. The southward ITCZ shift in ASYMMNH results in enhanced discharge in central and southern South America, especially in the southern Amazon and Paraná river networks. These territories of South America, along with southern Africa and Australia, are the primary regions where land precipitation increases in the tropics for ASYMMNH, and the river flow in these areas tends to increase. Our results are also consistent with Oman et al. (2006), who argue for a reduced Nile River level (northeastern Africa) following several large high northern latitude eruptions, including Laki and the Katmai (1912 CE) eruption. Their results were viewed through the lens of weakened African and Indian monsoons associated with reduced land–ocean temperature differences; our composite results suggest that regional precipitation reductions may also be part of a zonally coherent precipitation shift.

In ASYMMSH, the ITCZ moves northward, resulting in reduced river flux in the Amazon sector and increases (reduction) in the Niger of central and western Africa during boreal summer (boreal winter). Interestingly, the Nile flow is also reduced in this case, although to a lesser extent, despite very modest precipitation increases during MJJAS for a Southern Hemisphere biased aerosol forcing. There are also modest discharge increases in southern Asia. However, there is simply very little land in regions where northward ITCZ shifts result in enhanced precipitation, suggesting less opportunity for increases in discharge to a SH biased eruption. For the SYMM eruptions, river discharge is reduced nearly everywhere in the tropics, consistent with the precipitation reductions that occur (Fig. 3). The response is weaker or even reversed in the subtropics, such as in southern South America, where precipitation tends to increase (Iles and Hegerl, 2015).

Another important variable that integrates several aspects of the tropical climate system is the isotopic composition of precipitation. Here, we focus on the relative abundance of 1H218O versus the more abundant 1H216O, commonly expressed as δ18O, such that δ18Op≡VSMOW-1Omp18Omp16-1×1000, where Omp18 and Omp16 are the moles of oxygen isotope in a sample, in our case precipitation (denoted by the subscript mp). Delta values are given with respect to the isotopic ratio in a standard sample, the Vienna Standard Mean Ocean Water (VSMOW = 2.005 × 10-3).

δ18Op is a variable that is directly obtained from many paleoclimate proxy records. Therefore, rather than relying on a conversion of the local isotope signal to some climate variable, the explicit simulation of isotopic variability is preferred for generating potentially falsifiable predictions concerning the imprint associated with asymmetric volcanic eruptions. Indeed, δ18Op variability is the result of an interaction between multiple scales of motion in the atmosphere, the temperature of air in which the condensate was embedded, and exchange processes operating from source to sink of the parcel deposited at a site.

Water isotope tracers have been incorporated into the GISS-E2 model's atmosphere, land surface, sea ice, and ocean and are advected and tracked through every stage of the hydrologic cycle. A fractionation factor is applied at each phase change and all freshwater fluxes are tagged isotopically. Stable isotope results from the lineage of GISS-E2 models have a long history of being tested against observations and proxy records (e.g., Vuille et al., 2003; Schmidt et al., 2007; LeGrande and Schmidt, 2008, 2009).

Figure 7 shows the δ18Op response in the GISS-E2 model. Seasonal calculations are weighted by the precipitation amount for each month, although changes in the seasonality of precipitation are not important in driving our results (not shown). The literature on mechanistic explanations for isotope variability has a rich history of being described by several “effects” such as a precipitation amount effect in deep convective regions or a temperature effect at high latitudes (Dansgaard, 1964; Araguás-Araguás et al., 2000), so named as to reflect the most important climatic driver of isotopic variability at a site or climate regime. Notably, δ18Op tends to be negatively correlated with precipitation amount in the deep tropics and positively correlated with temperature at high latitudes (see, e.g., Hoffman and Heimann (1997) for a review of mechanisms). However, isotope–climate relations are generally complex. In our experiments, the δ18Op spatial pattern in the tropics (Fig. 7) exhibits a similar pattern to precipitation changes induced by the ITCZ shift (Fig. S5 in the Supplement for GISS-E2), particularly over the ocean. The meridional movement of the ITCZ leads to an isotopic signal that is more positive (enriched in heavy isotopes) in the preferentially forced hemisphere. The hemisphere toward which the ITCZ is displaced on the other hand experiences increased tropical rainfall and a relative depletion of the heavy isotope (more negative δ18Op). Thus, the paleoclimatic fingerprint of asymmetric volcanic eruptions is characterized by a tropical dipole pattern, with more positive (negative) δ18Op associated with reduced (increased) rainfall.

Over land, South America stands out as exhibiting a palette of isotopic patterns depending on forcing category and season. The South American monsoon system peaks in austral summer, and the largest precipitation reductions occur in ASYMMSH when the ITCZ moves northward. There is a dipole pattern, characterized by isotopic enrichment (depletion) in 18O in the northern (southern) tropics of South America in ASYMMNH during NDJFM, while the opposite pattern emerges in ASYMMSH, both associated with Atlantic and east Pacific ITCZ displacements. During the austral winter, climatological South American precipitation peaks in the northern part of the continent, and precipitation in this region is reduced in both the SYMM and ASYMMSH composites, leading to a large increase in δ18Op. This is consistent with recent results in Colose et al. (2016), who used the isotope-enabled GISS-E2 model to form a composite of all large (AOD > 0.1) LM tropical volcanic events based on the Crowley and Unterman (2013) dataset. The eruptions analyzed in that study were smaller in amplitude due to differences in the scaling during implementation, as well as the fact that G08 tends to have larger volcanic events in the original dataset to begin with. In regions where tropical South American precipitation does not exhibit very large changes, such as in the NDJFM SYMM composites, temperature may explain much of the isotopic response, again consistent with findings in Colose et al. (2016).

The overarching purpose of this work was to consider the influence of asymmetric volcanic forcing on the energetic paradigm outlined in Sect. 1. This framework of analyzing ITCZ shifts in the context of asymmetric forcing predicts a net AET anomaly toward the hemisphere that is preferentially forced by explosive volcanism, with anticorrelated dry and latent energy fluxes both contributing to drive the ITCZ away from the forced hemisphere. To examine this relationship in CESM, we first write a zonal-mean energy budget for the atmosphere (Trenberth, 1997; Donohoe and Battisti, 2013): 12πa2cos⁡ϕ∂AET∂ϕ=ASRTOA-OLRTOA+SWsfc↑-SWsfc↓+LWsfc↑-LWsfc↓+LHsfc+SHsfc+LfSn-1g∫0ps∂cpT+Lvq+k∂tdp, where ASRTOA is the absorbed solar radiation, OLRTOA is outgoing longwave radiation at the top of the atmosphere (TOA), SWsfc↑ is reflected surface shortwave radiation, SWsfc↓ is shortwave received by the surface (sfc), LWsfc↑ is longwave radiation emitted (or reflected) by the surface, LWsfc↓ is longwave radiation received by the surface, LH is the latent heat flux, SH is the sensible heat flux, Sn is snowfall rate, q is specific humidity, k is kinetic energy, ϕ is latitude, a is the radius of the Earth, T is temperature, cp is specific heat capacity, Lv and Lf are the latent heats of vaporization and fusion, p is pressure (p=ps at the surface), and g is the acceleration due to gravity. All terms are defined positive in the atmosphere, and the subscripts denote TOA or surface flux (sfc) diagnostics. Equation (3) effectively calculates MSE transport (Sect. 1) as a residual of energy fluxes in the model.

The last term (∂∂t) on the right side of Eq. (3) is the time-tendency term, representing storage of energy in the atmosphere. (Hereafter, STORL and STORD for latent and dry energy, respectively. The time derivative is calculated using finite differencing of the monthly-mean fields. The term in the parentheses is the moist enthalpy, or MSE minus geopotential energy. The kinetic energy is calculated in this study but is several orders of magnitude smaller than other terms, and hereafter is folded into the definition of STORD.) The tendency term must vanish on timescales of several years or longer but is important in our context. We explicitly write out the snowfall term since CESM (and any CMIP5 model) does not include surface energy changes associated with snowmelt over the ice-free ocean as part of the latent heat diagnostic and must be calculated to close the model energy budget.

Integrating yields an expression for the atmospheric heat transport across a latitude circle: AETϕ=2πa2∫-π2ϕRTOA+Fsfc-STORL-STORDcos⁡ϕdϕ, where we have combined the TOA terms into RTOA and the snowfall and surface diagnostics have collapsed into a single variable Fsfc. Similarly, the latent heat flux HL across a latitude circle is HLϕ=2πa2∫-π2ϕLHsfc-LvP-STORLcos⁡ϕdϕ, where P is precipitation in kg m-2 s-1. We note that transport calculations are presented for CESM and were done for only 17 ensemble members, since there are missing output files for the requisite diagnostics in one run.

Figure 8a shows the annual-mean climatological northward heat transport in CESM, as performed by the atmosphere, in addition to the dry and moisture-related components of AET. The total CESM climatological poleward transport is in good agreement with observational estimates (e.g., Trenberth and Caron, 2001; Wunsch, 2005; Fasullo and Trenberth, 2008), peaking at ∼ 5.0 PW and ∼ 5.2 PW in the SH and NH subtropics, respectively (1 PW = 1015 W). In CESM, the SH receives slightly more net TOA solar radiation than the NH (by ∼ 1.3 W m-2 in the annual-mean), and the NH loses slightly more net TOA longwave radiation to space (by ∼ 0.89 W m-2). However, the CESM annual ocean heat transport is northward across the equator (not shown), keeping the NH warmer than the SH by ∼ 0.98 ∘C. As a consequence, AET is directed southward across the equator (red line). Moisture makes it more difficult for the tropical circulation to transport energy poleward, and the transport of moisture in the low-level equatorward flow is directed northward across the equator and associated with an annual-mean ITCZ approximately co-located with the atmospheric energy flux equator (EFE), the latitude where AET vanishes. This arrangement of the tropical climate is consistent with satellite and reanalysis results for the present climate (Frierson et al., 2013; Kang et al., 2014).

In response to asymmetric volcanic forcing, anomalous AET is directed toward the preferentially forced hemisphere (Fig. 8b, c), along the imposed temperature gradient. Results are shown for the annual-mean AET anomaly in ASYMMNH and ASYMMSH for 1 year beginning with the January after each eruption, although averaging the first 2–3 years yields similar results with slightly smaller amplitudes. The equatorial AET (AETeq) anomaly averaged over all events and ensemble members for ASYMMNH (ASYMMSH) is approximately 0.08 (-0.06) PW, defined positive northward, with much larger near-compensating dry and latent components. The anomalous moisture convergence drives the ITCZ shift away from the forced hemisphere. Anomalies in AETeq when considering each unique volcanic event (after averaging over the 17 ensemble members) are strongly anticorrelated with changes in the energy flux equator (r = -0.97, not shown), the latitude where AET vanishes.

The change in cross-equatorial energy transport for the SYMM ensemble/eruption mean (not shown) does not exhibit the coherence of the asymmetric cases for either AET or the individual dry and moist components and in all cases does not emerge from background internal variability.

Quantifying the ITCZ shift is nontrivial, since the precipitation field is less sharply defined than the EFE, and climate models (including the two discussed here) exhibit a bimodal tropical precipitation distribution (often called a “double ITCZ”), often with one mode of higher amplitude in the NH (centered at 8–9∘ N in CESM). However, despite pervasive biases that still exist in the climatology of tropical precipitation in CMIP5 (e.g., Oueslati and Bellon, 2015), the anomalous precipitation response is still characterized by a well-defined ITCZ shift (or a shift in the bimodal precipitation distribution; e.g., Fig.  9 in Stevenson et al., 2016), and the gross features presented here are in agreement with theoretical considerations. In our analysis, a movement in the latitude of maximum precipitation is not found to be a persuasive indicator of our ITCZ shift. In fact, the meridional shift is better described as a movement in the center of mass of the precipitation distribution, including changes in the relative amplitude of the two modes (e.g., a heightening of the SH mode for a southward ITCZ shift). Different metrics to describe the shift in the center of mass have been presented in the literature (e.g., Frierson and Hwang, 2012; Donohoe et al., 2013; Adam et al., 2016).

Here, we first adopt the precipitation median ϕmed (e.g., Frierson and Hwang, 2012) defined as the latitude where area-weighted precipitation from 20∘ S to ϕmed equals the precipitation amount from ϕmed to 20∘ N, i.e., where the following is satisfied: ∫20∘SϕmedPcos⁡(ϕ)dϕ=∫ϕmed20∘NPcos⁡(ϕ)dϕ. When considering the spread across eruption size (regressing the different events in all three categories together after averaging over ensemble members), we find a ∼ -8.9∘ shift in ITCZ latitude per 1 PW of anomalous AETeq (Fig. 9). The sign of this relationship is a robust property of the present climate system, although it is higher than other estimates (Donohoe et al., 2013) that analyzed the ITCZ scaling with AETeq to a number of other time periods and forcing mechanisms (not volcanic), including the seasonal cycle, CO2 doubling, Last Glacial Maximum, and mid-Holocene. It was argued in Donohoe et al. (2013) that the ITCZ is “stiff” in the sense that a large AETeq is required to move the ITCZ. However, the sensitivity of this relationship may vary considerably depending on the ITCZ metric considered (Fig. 9 presents a scaling with different indices), based on the following equation (Adam et al., 2016): ϕITCZ=∫20∘S20∘Nϕ(Pcos⁡(ϕ))Ndϕ∫20∘S20∘N(Pcos(ϕ))Ndϕ. Here, N controls the weighting given to the modes in the precipitation distribution. Typically ϕITCZ moves toward the precipitation maximum as N increases, but importantly, the sensitivity of a ϕITCZ migration to a given anomaly in AETeq also changes. Figure 9 shows the regression of anomalous ϕmed and ϕITCZ(N=5) against anomalous AETeq (r = -0.94). ϕITCZ(N=3) yields a high correlation (r = -0.95) and best follows a 1 : 1 line with the EFE (Fig. 9, bottom left). The slope of the relationship between ITCZ location and AETeq may vary by a factor of 4–5 depending on the relationship used. For example, there is approximately a -11.7∘ shift in ITCZ latitude per 1 PW of anomalous AETeq using ϕITCZN=3. Thus, we interpret our results as suggesting that, energetically, it is not necessarily difficult to move the ITCZ and urge caution in characterizing past ITCZ shifts as being difficult to reconcile with paleo-forcing estimates (Donohoe et al., 2013). Indeed, as many studies have used a “precipitation centroid” or a similar variant to quantify tropical precipitation migrations, we recommend exploring the sensitivity of ITCZ shifts to different ways of characterizing the movement in precipitation mass unless the community can agree upon a well-defined N that suitably characterizes the precipitation distribution in both climate models and observations.