The sensitivity of the ENSO to volcanic aerosol spatial distribution in the MPI large ensemble

Using the Max Planck Institute Grand Ensemble (MPI-GE) with 200 members for the historical simulation (18502005), we investigate the impact of the spatial distribution of volcanic aerosols on the ENSO response. In particular, we select 3 eruptions (El Chichón, Agung and Pinatubo) in which the aerosol is respectively confined to the Northern Hemisphere, the Southern Hemisphere or equally distributed across the equator. Our results show that the ENSO anomalies start at the end of 10 the year of the eruption and peak the following one. Especially, we found that when the aerosol is located in the Northern Hemisphere or is symmetrically distributed, El Niño-like anomalies develop while aerosol distribution confined to the Southern Hemisphere leads to a La Niña-like anomaly. Our results strongly point to thevolcanically induced displacement of the ITCZ as the main mechanism that drives the ENSO response, while suggesting that the other mechanisms (the ocean dynamical thermostat, the cooling of tropical northern Africa or of the Maritime continent) commonly invoked to explain the post-eruption 15 ENSO response appear not to be at play in our model.


Introduction
Aerosol particles from volcanic eruptions are the largest non-anthropogenic radiative forcing that have influenced the climate system in the past centuries (Robock, 2000). Oxidised, sulfur gases (mainly in form of 2 ) injected into the stratosphere by large Plinian eruptions form sulfate aerosols ( 2 4 ) (Pinto et al. , 1989;Pollack et al., 1976) that have a time residence of 1-20 3 years (Barnes & Hofmann, 1997;Robock & Yuhe Liu, 1994). These particles both scatter and absorb incoming solar radiation as well as part of the outgoing longwave radiation (Stenchikov et al., 1998;Timmreck, 2012). For intense and sulfurrich volcanic eruptions, the net effect is a general cooling of the surface and a warming in the stratosphere where the aerosols tend to reside longer (Harshvardhan, 1979;Rampino & Self, 1984). The maximum global cooling is in general reached within 6-8 months following the eruption peak in optical depth before returning to normal values after about 3 to 4 years (Thompson 25 et al., 2009). These rapid modifications in temperature can induce dynamic changes in the atmosphere and in the ocean including a strengthening of the polar vortex (e.g., Christiansen, 2008;Driscoll et al. , 2012;Kodera, 1994;Stenchikov et al., 2006), a weakening in the African and Indian Monsoon (e.g., Iles et al., 2013;Man et al., 2014 ;Trenberth & Dai, 2007;https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License. Zambri & Robock, 2016) as well as forced changes on the El Niño-Southern Oscillation (ENSO) (e.g., Emile-Geay et al., 2008;McGregor & Timmermann, 2011;Pausata et al., 2015;Wang et al., 2018). 30 Paleoclimate archives and observations from the past centuries suggested that large tropical eruptions are usually followed by a warm sea-surface temperature (SST) anomaly in the Pacific (El Niño events ; e.g., Adams et al. , 2003;D'Arrigo et al. , 2005;Li et al., 2013;S. McGregor et al. , 2010;Wilson et al., 2010). In particular, El Niño events followed in the first or the second winter after the five largest eruptions of the last 150 years (Krakatoa in August 1883, Santa Maria in October 1902 Agung in March 1963, El Chichón in April 1982and Pinatubo in June 1991. However, the Santa Maria, El Chichón and Pinatubo eruptions occurred after an El Niño event was already developing making it difficult to determine a causal link between ENSO and these eruptions (e.g., Self et al., 1997;Nicholls, 1988).
Moreover, modelling studies initially found divergent responses for the ENSO changes after large tropical eruptions (Ding et al., 2014;McGregor & Timmermann, 2011;Stenchikov et al., 2006;Zanchettin et al., 2012). However, the most recent studies 40 have pointed to an El Niño-like response following volcanic eruption (McGregor et al, in press). In particular, the use of relative sea surface temperature (RSST) or sea surface height (SSH) instead of SST have helped to disentangle the ENSO response from volcanically induced cooling in the Pacific and to highlight the dynamical ENSO response (Khodri et al., 2017;Maher et al., 2015).
Although a consensus is emerging, different aerosol spatial distributions may give rise to different ENSO responses. Stevenson 45 et al., (2016) investigated the impact of NH, SH and tropical volcanic eruptions using the Community Earth System Model (CESM 1.1) Last Millennium Ensemble. They concluded that while NH and tropical eruptions tend to favour El Niño-like conditions, SH eruptions enhance the probability of La Niña events within one year following the eruptions. Conversely, Liu, Li, et al., (2018) through a millennium simulation performed with CESM 1.0 and Zuo et al., (2018) using the Community Earth System Model Last Millennium Ensemble (CESM-LME) concluded that SH, NH and tropical eruptions all resulted in 50 El Niño-like conditions in the second year after the eruption.
The mechanisms that trigger a change in the ENSO state following volcanic eruptions is still debated. One of the most frequently used hypotheses is the "ocean dynamical thermostat" mechanism (ODT) (Clement et al., 1996), where a preferential cooling in the western Pacific relative to the eastern Pacific takes place due to the underlying dynamics of the Pacific Ocean.
Such a differential cooling weakens the zonal SST gradient along the equatorial Pacific which causes a relaxation of the trade 55 winds, leading to a temporary weakening of the ocean upwelling in the eastern Pacific. This process is then amplified by the Bjerknes feedback, yielding an El Niño (Bjerknes, 1969). A related mechanism for the preferential El Niño anomalies following volcanic eruptions is based on the recharge-discharge theory of ENSO, including changes I n the wind stress curl during the eruption year as one of the triggering factors (McGregor & Timmermann, 2011;Stevenson et al. 2017). However, through a set of sensitivity experiments, Pausata et al., (2020) have questioned the existence of the ODT mechanism in coupled 60 https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License. climate models following volcanic eruptions, pointing to the Intertropical convergence zone (ITCZ) displacement and extratropic-to-tropic teleconnections as key mechanisms in affecting post-eruption ENSO. The ITCZ-shift mechanism was originally proposed for the ENSO response to high-latitude eruptions (Pausata et al., 2015; and then suggested to also be at work for tropical asymmetric eruptions (Colose et al., 2016;Stevenson et al., 2016). In general, the ITCZ shift away from the hemisphere that is cooled (Kang et al., 2008;Schneider et al., 2014). Consequently, for an eruption with aerosol 65 concentrated in the NH, the ITCZ location moves equatorward, weakening the trade winds and leading to an El Niño-like anomaly via the Bjerknes feedback (Bjerknes, 1969). In contrast, the ITCZ moves northward following a larger SH cooling, strengthening the trade winds and triggering La Niña-like anomalies as seen in Colose et al. (2016) and Stevenson et al. (2016). eruption El Niño-like anomalies. A similar mechanism has been suggested based on the cooling of the Maritime continent or south-eastern Asia instead of tropical Africa (Eddebbar et al., 2019;Ohba et al., 2013;Predybaylo et al.,2017). However, there is yet no consensus as to which of these proposed mechanisms is the main driver of the ENSO response after large volcanic eruptions.
Modeling studies have investigated the impact of volcanic eruptions on ENSO using different approaches. Many studies have 75 used a superposed epoch (SEA) or composite analysis, in which they used a window of a few years before the eruption to create a reference to compare with the post-eruption period (e.g. Liu, Li, et al., 2018;Zuo et al., 2018). The significance of the response to volcanic eruptions is then assessed using a Monte Carlo method. However, this statistical methodology has some shortcomings as it is not able to fully remove the signal of internal variability: ENSO anomalies can still be seen in the reference period (see for example figure 4 in Liu, Li, et al., (2018)). Other studies use a limited number of ensemble members with the 80 volcano and no-volcano members starting from different initial conditions (e.g. McGregor & Timmermann, 2011;Predybaylo et al., 2017;Sun et al., 2019). However, when starting the two ensemble sets from different initial conditions, a large number of ensemble members (equivalent to at least 150 years reference period/climatology) is needed to isolate the internal variability of ENSO Wittenberg, 2009)  The anomalies calculated are the difference between the reference (3 years before the eruption, which represents a total of 600 years climatology) and periods after the eruption. A Student t-test is used to estimate the significance of the mean changes before and after the eruptions at the 95% confidence level. 105 Large tropical eruptions induce a global cooling so that El-Niño response may be partly masked, and the La-Niña response amplified (Maher et al., 2015). Furthermore, some climate models amplify the volcanically induced cooling (e.g. Anchukaitis et al., 2012;Stoffel et al., 2015). To better highlight the dynamical changes, we remove the tropical SST mean from the original SST, this is known as the relative sea surface temperature (RSST) (Vecchi & Soden, 2007). In this study, we use the RSST to isolate the intrinsic ENSO signal (Khodri et al., 2017). 110

ENSO response and its link to the ITZC-shift mechanism
The volcanic eruptions analyzed in the present study show three distinct aerosol plumes. While the aerosol distribution from the Pinatubo eruption is symmetrical around the equator, Agung and El Chichón eruptions both created to a large extent confined distribution in respectively the SH and the NH (Fig. 1). For the Pinatubo and El Chichón the aerosol optical depth 115 peaks in the winter that follows the eruptions. For the Agung this maximum is reached in the winter of the second year after the eruption. These different distributions are due to the location of the volcanoes, the season and the strength of the eruption (Stoffel et al., 2015;Toohey et al., 2011). in the winter of the year after the eruption reaching a maximum of approximately +0.3℃ for El Chichón and the Pinatubo and a minimum of -0.2 ℃ for Agung (Fig. A3).
The asymmetric forcing caused by the three eruptions induces a different cooling of the surface temperature in the two 125 hemispheres (Fig. 4). Although Pinatubo is the most intense eruption and has the largest global temperature decrease (Fig.   A2), El Chichón shows the strongest hemispherical cooling (Fig. 4). The hemispherical cooling associated to Agung (SH) is in absolute values comparable to Pinatubo even if Agung's magnitude was twice as small (Bluth et al., 1992;Self & Rampino, 2012). Furthermore, while the aerosol distribution of the Pinatubo eruption is symmetric, the cooling is not and is concentrated in the NH, which is likely due to uneven distribution of landmass between hemispheres (i.e. reduced heat capacity in the NH). 130 The maximum cooling for all three eruptions occurs at the beginning of year two and so does the temperature difference between the two hemispheres (Figs. 4 and A1).
Our results are consistent with the displacement of the Hadley cell associated with the differential cooling between the hemispheres, showing a northward shift of the ITCZ for the eruption of the Agung, and a southward shift for both El Chichón and the Pinatubo, for the three years following the eruption (Figs. 5 and 6). Additionally, the Pinatubo's rainfall anomaly is 135 the largest even though the difference in the cooling of each hemisphere is stronger for El Chichón. This could indicate that the on-going global warming could amplify the rainfall anomaly following the volcanic eruptions through modulation in the ocean stratification and near-surface winds amplifying the response as suggested in a recent study   (Fig.   A9). We find that the ITCZ displacement and associated rainfall anomalies peak the second year after the eruption, when the differential cooling of the hemisphere is larger (Figs. 4, 5 and 6). The ITCZ displacement is associated with a strengthening of 140 the trade winds for the Agung eruption and a weakening for the El Chichón and Pinatubo eruptions as expected by the direction of the ITCZ movement in each case. These wind anomalies affect then the ENSO state: a change in the strength of the trade winds along the equatorial Pacific alters the ocean upwelling in the eastern side. This leads to a change in the east-west temperature contrast across the tropical Pacific, which is amplified by the Bjerknes feedback (Bjerknes, 1969) thus altering the ENSO state. All our results (the evolution of the Nino 3.4 index, the precipitation anomalies or the temperature anomalies) 145 show an almost perfect symmetry between the tropical/NH distribution and the SH distribution (Figs. 2, 3, 4, 5, 6 and A2), which strongly suggests that the volcanically induced ITCZ displacement is the key mechanism to explain the ENSO response to the volcanic forcing in agreement with others studies (Colose et al., 2016;Pausata, et al., 2016;Stevenson et al., 2016;Pausata et al,. 2020).

ENSO response and its link to other mechanisms 150
The most commonly invoked mechanism that explains the ENSO response after large tropical volcanic eruptions is the ODT (Clement et al., 1996) and the preferential cooling of the warm pool relative to the eastern equatorial Pacific, leading to an El Niño-like response (e.g. Emile-Geay et al., 2008;Mann et al., 2005). However, in our simulations even if there is a volcanic aerosol over the equatorial Pacific and a surface cooling for all the eruptions (Figs. 1, A3, A4 and A5), we see a negative phase https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.
of ENSO and an anomalous easterly wind stress developing after the Agung eruption (Figs. 2 and 3). These results thus 155 suggest that the ODT is not a dominant mechanism in our model.
Another mechanism often used to explain the post-eruption El Niño-like response is related to the cooling of the Maritime Continent first proposed by Ohba et al. (2013) and also suggested in recent studies (e.g., Eddebbar et al.,  Recently, Pausata et al., (2020) proposed an additional mechanism related to the extratropical-to-tropical teleconnections that tends to favour an El Niño-like response for both NH and SH eruptions, hence playing in synergy (NH eruptions) or against 175 (SH eruptions) the ITCZ-shift mechanism. However, our qualitative analysis of the sea level pressure (SLP) anomalies does not match the changes expected by this mechanism (cf. Fig. A10 to Fig. 4 (a-d) in Pausata et al., (2020)). In this recent study, the volcanic aerosol alters the meridional temperature gradient of the atmosphere that eventually causes a poleward shift of the Pacific jet stream and a strong cyclonic surface pressure anomaly over the midlatitude to subtropical Pacific basin in both NH and SH eruptions in the first summer following the eruptions. In our simulation, the response is opposite for the Agung 180 and El Chichón or Pinatubo eruptions, suggesting more that the simulated extratropical anomalies are induced by the ENSO changes due to the eruption rather than affecting ENSO (Fig. A10). The reason of the disagreement could lie in the fact that the El Niño/La Niña-like response following the volcanic eruptions peak in the first winter in Pausata et al., (2020) modeling study, while in our model in the second winter (Figs. 2 and 3). The extratropical-to-tropical teleconnection could make the El Niño development following NH/symmetric eruptions occur faster than in our case where such a teleconnection appears not 185 to be present. Ad hoc sensitivity experiments are necessary to rule out the above-mentioned mechanisms in our model. https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.

Discussion and conclusions
Our study used the largest ensemble simulation (200 ensembles) currently available of the historical period performed with the MPI-ESM model to better understand the impact of the volcanic eruptions on ENSO. Our results strongly point to the volcanically induced ITCZ displacement as the primary driver of the ENSO response following volcanic eruptions. In our 190 simulations, the ENSO response after the eruptions critically depends on the distribution of the aerosol plume. When the volcanic aerosol distribution is confined to the NH or its distribution is symmetrical across the hemispheres the ENSO state tends towards a positive phase (El-Niño like conditions ; Fig. 2 (d-i)), while when the aerosols are confined to the SH the ENSO state is pushed towards a negative phase (La Niña-like conditions ; Figs. 2 (a-c)). The displacement of the ITCZ following the eruptions, caused by the asymmetric cooling of the hemisphere that pushes the ITCZ towards the hemisphere 195 that is less cooled (Kang et al., 2008;Schneider et al., 2014). Both the eruptions with aerosol confined to the NH and symmetrically distributed across the hemispheres preferentially cool the NH, consequently shifting the ITCZ southwards, weakening the trade winds over the equatorial Pacific and triggering an El Niño like response through the Bjerknes feedback (Bjerknes, 1969). The eruption with the aerosol plume confined to the SH instead cools exclusively the SH, pushing the ITCZ northward and strengthening the trade winds, leading to a La Niña-like response. 200 The ITCZ mechanism we see at play in our model is supported by other recent studies performed with different climate models (Pausata,et al., 2015(Pausata,et al., , 2020Stevenson et al., 2016;Colose et al., 2016). Pausata et al. (2020) through a set of sensitivity experiments in which the volcanic aerosol forcing is confined to either the northern or the southern hemisphere show the key role of the ITCZ displacement in driving the ENSO response. They also highlighted the presence of another mechanism related to the extratropical-to-tropical teleconnections that no matter the type of eruption (NH or SH) tends to favour an El-Niño like 205 response. Hence, it plays in synergy (NH eruptions) or against (SH eruptions) the ITCZ-shift mechanism. However, the simulated SLP changes in the extratropics in our model seem to be in response to the volcanically induced ENSO changes rather than affecting the ENSO response (cf. Fig. A10 to Fig. 4 (a-d) in Pausata et al., (2020)).
Our work also pointed out that the ODT (Clement et al., 1996), the cooling of the Maritime Continent (Ohba et al., 2013) and of the tropical Africa (Khodri et al., 2017) mechanisms are not at play. These previous studies that have suggested a different 210 mechanism to explain the ENSO response to volcanic eruptions are based on a limited number of ensemble members (e.g. 5 members for 3 eruptions for the SH plume in Zuo et al. (2018)) or they heavily rely on statistical tools (e.g. SEA in Liu et al. (2018)). Consequently, those results may be biased by the use of a restrained number of ensemble members. Hence, our study points out the importance of a large number of ensemble members when investigating the ENSO response to volcanic eruptions. The absence of these mechanisms following volcanic eruptions is also in qualitative agreement with the modeling 215 experiments in Pausata et al. (2020).
Finally, our results are consistent with the predominance of post-eruption El Niño events (Adams et al., 2003 McGregor et al, in press) and it can provide an explanation on why the majority of both observations and reconstructions are displaying El https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.
Niño events instead of La Niña events. However, the ENSO responses discussed in this study are only tendential (El Niño-like or La Niña-like response), i.e., intrinsic variability evolving toward a La Niña at the time of the eruption would not necessarily 220 lead to a post-eruption El Niño event even for a NH or symmetrical eruption, but rather to a dampening of the ongoing La Niña. Furthermore, our model suggests the peak in ENSO anomalies to be on the second or third year after the eruption as in most modeling studies (Khodri et al., 2017;Lim et al., 2016;McGregor & Timmermann, 2011;Ohba et al., 2013;Stevenson et al., 2017), which is at odds with the reconstructions and observations that see a peak in ENSO anomalies in the winter following the eruption. The delayed ENSO response in our model simulations relative to reconstructions and observations may 225 be related to the apparent lack of extratropical-to-tropical teleconnections (Pausata et al., 2020) that could favour El Niño-like response already on the first winter following the eruption.
In conclusions, our results provide further insights into the mechanism driving the ENSO response to volcanic eruptions, highlighting in particular the role of the ITCZ shift. However, further coordinated efforts with specific sensitivity studies are necessary to delve into the other proposed mechanisms and to unravel the difference between modeling studies and 230 reconstructions with regards to the peak of the ENSO response. Given that ENSO is the major leading mode of tropical climate variability, which has worldwide impacts, these types of studies are also necessary to help improve seasonal forecasts following large volcanic eruptions.    https://doi.org/10.5194/esd-2020-63 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.