We used the Norwegian Earth System Model [NorESM1-M (27, 28)], which has a horizontal resolution of 1.9° (latitude) × 2.5° (longitude) and 26 vertical levels. NorESM1-M uses a modified version of Community Atmospheric Model version 4 [CAM4 (36)], CAM4-Oslo, to simulate the atmospheric circulation with an updated module that simulates the life cycle of aerosol particles, and primary and secondary organics. NorESM1-M includes treatment of the direct effect of aerosols and the first and second indirect effects of aerosols on warm clouds (37). The atmospheric model is coupled to the Miami Isopycnic Coordinate Ocean Model (MICOM), which has a horizontal resolution of ~1.125° along the equator and 53 vertical levels. A detailed description of the model used in this study can be found in Bentsen et al. (27) and Iversen et al. (28).
In the experiments used in the present study, a tropical eruption has been simulated in which 60 Tg of SO2 was injected mostly between ~15 and ~21 km (upper troposphere/lower stratosphere) over a period of 3 days to mimic a Tambora-like eruption. Although the injection height was likely much higher than what we used [likely more than 40 km (38)], we lowered it to overcome an increased residence time of the aerosol particles (for details, see the Supplementary Materials).
We then performed two experiments starting from a particular instant in time in the control run (1911–1964): 1 June 1923. In one experiment, a volcanic eruption was simulated in the subtropics (17°N) of the NH (TrNH), and in the other experiment, an eruption was simulated in the subtropics (17°S) of the SH (TrSH; see Fig. 1). Each experiment included an ensemble of 20 simulations, each with a small stochastic perturbation applied to the otherwise identical initial conditions. The perturbation, of the order of 10−14°C, was applied to the surface temperature. A third experiment was an ensemble of 20 simulations—also constructed starting from 1 June 1923 and with identical initial conditions as the TrNH and TrSH members—but with no volcanic eruption. We distinguished the impact of the volcano from intrinsic noise by comparing the 20-member average of each volcano experiment to the 20-member average of the control. Each ensemble member was run for 4 years, ending on 31 May 1927.
We chose 1 June 1923 as our starting date because the tropical Pacific is in an ENSO neutral state (Niño3.4 index = −0.1°C in June), but in the absence of an eruption is trending to La Niña conditions 3 months later (Niño3.4 index = −0.4°C in September; Fig. 3). To explore the sensitivity of the response to the volcanic eruption to the particular phase of ENSO, we constructed two additional experiments with identical NH and SH tropical eruption emission scenarios, but starting from a different year in the control simulation together with the reference NV experiment (1 June 1927). Unlike for 1 June 1923, the June 1927 volcanic eruption occurs when the ENSO is in a slightly warm state (Niño3.4 index = +0.4°C), and in the absence of a volcanic eruption, it would have remained warm for the next 18 months (see Fig. 2). We denote the three experiments (two volcano and one NV) that start on 1 June 1923 (1 June 1927) as the ENS01 (ENS02) set of experiments. Early summer (1 June) eruptions have been chosen since it is the ENSO developing season, and it has been shown that the associated radiative forcing is most likely to affect the ENSO response (15).
Every volcanic eruption shares an identical sulfate emission scenario, which is tailored after the Tambora eruption of 1815—the largest tropical eruption in the past 500 years (39).
This experimental design with interactive aerosols allows investigating the two-way interaction between aerosol and climate to explore the associated feedbacks in a more realistic framework compared with approaches where there is a one-way effect of aerosols on climate: While focusing on the climate response, we consider the whole process of forcing generation. This setup, thus, allows to demonstrate that volcanic forcing asymmetries around the equator affect the spatial structure of the forcing as well as the magnitude of the forcing itself, yielding a more realistic quantification of associated climatic impacts.
We have also performed a highly idealized experiment to test the ODT, in which the SO4 mixing ratio has been prescribed exclusively over the equatorial Pacific (10°S to 10°N; 100°E to 80°W). The prescribed SO4 mixing ratio for the EqPAC simulations has been created to have a peak anomaly in radiative forcing in July over the equatorial Pacific similar to the asymmetric simulations but with approximately double their strength (fig. S1). The prescribed SO4 gradient at the borders of the forcing region (equatorial Pacific) is not dissimilar from the gradients that develop in experiments where the volcanic plume is allowed to evolve [see Fig. 2, A and B, in the TrNH eruptions]. Therefore, it does not, in itself, present an additional or unique unrealistic forcing that could unduly affect the results.
To test whether the ODT mechanism is at play in response to a tropical volcanic eruption, we prescribed an aerosol forcing that is symmetric about the equator in the tropical Pacific rather than simulate a volcanic eruption at the equator. This experimental design precludes extratropical forcing and hemispheric asymmetry in forcing that would accompany any tropical eruption by way of the Brewer-Dobson circulation (26) and, thus, isolates the impact of the ODT.
The volcanically induced anomalies (Δvolc) are calculated as the difference between the climate state induced by the eruption (VENS01 or VENS02) and the unperturbed climate state (NVENS01 or NVENS02): Δvolc* = VENS0* – NVENS0*, where * stands for the ensemble identifier (1 or 2). Each mean climate state (VENS0*, NVENS0*) is defined for each given ensemble (i.e., the averages of TrNHENS01, TrNHENS02, TrSHENS01, TrSHENS02, EqPACENS01, EqPACENS02, NVENS01, or NVENS02) as the ensemble average of all its members.
We have tested the ability of NorESM in simulating the SO4 peak concentration and e-folding time against the Pinatubo eruption (June 1991), which is currently the best observed large tropical eruption (see the Supplementary Materials and fig. S10). NorESM is able to reproduce the main features of the Pinatubo eruption, which is sufficient for the scope of this study, that is, to delve into the mechanisms that trigger an ENSO response following a uniform radiative forcing.
Although each volcanic eruption shares an identical sulfate emission scenario, the amount of SO4 produced and the strength of the forcing are larger for the TrNH eruptions (Fig. 1 and fig. S1). This is because the eruption occurs during the boreal summer, when higher concentrations of OH radical are available to turn SO2 gas into sulfate aerosol, as compared with winter.
The general validity of the conclusions presented in this study must account for possible limitations of our experimental design, in particular (i) the representativeness of the chosen forcing as the 1815 Tambora eruption corresponds to a strong and rare volcanic event and (ii) the limitations of the single model used here and its skill in simulating ENSO and the mean climate state (i.e., climatological biases). The 1815 Tambora eruption is a test case to study the climatic response to volcanic forcing (35, 40, 41). While we expect the separation between different mechanisms highlighted in this study to be affected by a lower signal-to-noise ratio, they should still be applicable for more frequent but weaker eruptions, such as the 1991 Pinatubo.
NorESM is among the best coupled climate models to represent ENSO concerning the mean climate state of the tropical Pacific and the spectrum of ENSO variability (42). It is also weakly affected by the double ITCZ issue as the double ITCZ is less pronounced in NorESM than in other climate models (27, 43).