Figure 2 shows simulated changes (perturbed Hunga simulations minus control simulations) in total column ozone simulated by the three chemistry-climate models (GEOSCCM, MIROC-CHASER and WACCM-MAM) in the different free-running and nudged experiments. In the free-running simulations (row 1; panels a–c), which inject both SO₂ and H₂O and include Hunga impacts on all the different chemical, radiative and dynamical processes, all three models simulate total column ozone decreases in the SH mid-latitudes and springtime high latitudes following the eruption, although with large differences in the magnitudes and duration of the effects. GEOSCCM shows the largest Antarctic ozone reductions (of up to ∼ 50 DU in late 2023), with statistically significant total column ozone reductions in the SH mid- and high latitudes persisting until 2026. MIROC-CHASER shows the smallest response, with ozone losses of a few DU in the SH mid-latitudes from late 2022 into early 2024, and statistically not significant ozone losses in the Antarctic in late 2022 and 2023. While WACCM-MAM also shows ozone losses of a few DU in the SH mid-latitudes from mid-2022 into early 2024, it simulates considerably larger ozone reductions in the Antarctic (of up to ∼ 15 and ∼ 20 DU in the zonal mean in late 2022 and 2023, respectively).

Row 2 in Fig. 2 shows the corresponding responses to SO2+H₂O injections in the nudged experiments; by design, these include only direct chemical impacts of the eruption on ozone and not any impacts driven by changes in temperatures or circulation (because the meteorology is identical for the Hunga and control simulations, and so any forced response in it cancels out). Comparison of the free-running and nudged responses (i.e., row 1 vs row 2 in Fig. 2) shows that a significant part of the total response (Fig. 2, row 1) is evident in the nudged simulations. Taking the Antarctic response in late 2023 as an example, GEOSCCM shows ozone loss of up to ∼ 50 DU in the free-running experiment and ∼ 25 DU in the nudged one, while for WACCM-MAM these ozone losses are ∼ 20 and ∼ 10 DU, respectively. This indicates that the Hunga impact on the SH extra-tropical ozone, at least as simulated by these three models, is driven by a combination of both chemical and dynamical/temperature changes, the contributions of each roughly comparable in importance. Furthermore, isolating the contributions of SO₂-only and H₂O-only injections in the nudged simulations (rows 3 and 4) shows that the total column response due to chemistry is driven primarily by the aerosol forcing, with the H₂O forcing in isolation inducing only small total column ozone losses (up to ∼ 5 DU), although the two forcings are not strictly additive (as will be discussed below).

As most of the ozone depletion from Hunga is found in the SH extra-tropics, Fig. 3 shows the vertical distribution of the 90–30° S mean ozone anomalies. The free-running simulations (row 1) show a distinct pattern of ozone anomalies that is consistent across all three models, with ozone decreasing in the upper stratosphere, increasing in the middle stratosphere (near ∼ 30–10 hPa), and decreasing again in the lower stratosphere. Comparison with the corresponding nudged response (row 2) shows that, as noted earlier, a significant portion of the total response (row 1) is found in the nudged simulations, indicating that Hunga impacts on both chemical and dynamical/radiative processes are important in determining the overall ozone distribution following the eruption.

Notably, the lower stratospheric ozone reduction extends into the troposphere and affects ozone all the way to the surface, with ozone losses of a few percent in the SH extra-tropical free troposphere (500 hPa, Fig. 4, row 1) simulated by models until at least 2024. Stratospheric ozone tracer (O3S, e.g. Banerjee et al., 2016) is a proxy for the stratospheric origin of ozone in the troposphere; in the stratosphere O3S equals ozone concentrations, while in the troposphere it has no chemical production source (but it still undergoes chemical loss). Comparison of the tropospheric ozone response with the corresponding O3S response (row 2 in Fig. 4; note that the O3S tracer was not included in GEOSCCM) shows good qualitative and quantitative agreement, indicating the predominantly stratospheric origin of the negative tropospheric ozone anomalies via weakening of stratosphere-to-troposphere ozone transport. In the nudged simulations (Fig. S3), the corresponding tropospheric O3S changes are much smaller than in the free-running simulations, suggesting the reduced transport has the dominant role in driving this response (see also Sect. 3.3), as opposed to just reduced lower stratospheric ozone levels. While O3S tracer was not included in the GEOSCCM simulations, that model shows the largest tropospheric ozone losses (Fig. 4a) among all three models; these decreases are in agreement with the largest concurrent lower stratospheric ozone reductions (Fig. 3a), thus also pointing towards stratospheric origin of the tropospheric ozone anomalies.

In the following sections, we discuss the different chemical and dynamical drivers of the simulated Hunga ozone responses.

In the upper stratosphere, the timescales of chemical processes are generally much faster than dynamical timescales, and so ozone concentrations are by and large in photochemical equilibrium. Accordingly, Fig. 3 shows that in the upper stratosphere the ozone response simulated in the nudged SO2+H₂O experiment (Fig. 3, row 2) accounts for most of the full ozone response (Fig. 3, row 1), indicating that the Hunga impact in this region is largely chemically driven. In particular, the enhanced water vapor (Figs. 1 and S2) accelerates HO_x-catalyzed ozone depletion (see also Santee et al., 2023; Wilmouth et al., 2023; Randel et al., 2024). The H₂O forcing from the eruption is the dominant driver of the ozone response in this region (Fig. 3, bottom row), with SO₂ injection having no discernable impact at these altitudes (Fig. 3, row 3). Consistent with this picture, the evolution of global-mean upper stratospheric ozone anomalies in these three models tracks closely the evolution of water vapor, with the models projecting that the Hunga water will continue to exert a small impact on upper stratospheric ozone until at least 2028 (Fig. 5; see also Zhuo et al., 2025).

In the middle and lower stratosphere, while dynamical processes become more important, the chemical impacts of the eruption are still significant. The presence of aerosols decreases active nitrogen levels (NO_x, as illustrated by changes in NO₂ – its dominant component – in Figs. 6a–c and S5) and slows down the rate of NO_x-catalyzed chemical ozone loss due to the enhancement of the heterogeneous N₂O₅ hydrolysis on their surfaces. This agrees with the MLS observational results reported by Santee et al. (2023), and previous model results (Wilmouth et al., 2023; Zhang et al., 2024a). The reduction in stratospheric NO_x (Fig. S5) and the resulting increase in middle stratospheric ozone (Fig. 3) is stronger in the nudged simulations with SO₂-only injection than in those that include simultaneous injection of both SO₂ and H₂O (Figs. 3 and S5, row 3 vs row 2). This is partly related to the fact that in the absence of H₂O injection the anomalous SAD extends to higher altitudes (Fig. S1, row 3 vs row 2) as aerosols are smaller and less prone to sedimenting out (Quaglia et al., 2026). This larger ozone increase in the middle stratosphere under SO₂-only injection offsets the ozone decrease below, leading to less negative and more positive total column ozone responses in the SH mid- and high latitudes as seen in Fig. 2 (row 3 vs row 2).

Another set of heterogeneous reactions on aerosol surfaces involves halogens (chlorine and bromine). This effect leads to a significant enhancement of active chlorine and bromine species (ClO_x = Cl + ClO +2⋅ Cl₂O2+ HOCl + OClO +2⋅ Cl₂; and BrO_x = Br + BrO + BrCl + HOBr) in the lower stratosphere (Figs. 6d–i, S6 and S7), resulting in enhanced halogen-catalyzed ozone loss under Hunga forcing. Significant perturbations in chlorine species were also observed in the MLS record (Santee et al., 2023; Wilmouth et al., 2023) and previous model studies (Zhang et al., 2024b). Notably, the simulations show substantial enhancements of BrO_x in the lower stratosphere (up to a few ppt, i.e. ∼ 30 %–80 % of the background values, see Fig. S4). Unlike the ClO_x enhancement, which is largest in absolute terms inside the polar vortex during austral winter and spring, substantial BrO_x enhancement occurs throughout the year, thus likely constituting an important chemical contributor to the SH mid-latitude ozone loss (although the ClO_x enhancement outside the polar vortex, whilst smaller in absolute terms, still constitutes a relatively large percentage increase, see Fig. S4, and is thus also important). In all three models most of the anomalous halogen repartitioning in the lower stratosphere (Figs. S6 and S7) and the resulting lower stratospheric ozone reduction (Fig. 3) is driven by the aerosol forcing (row 3 in Figs. S6, S7 and 3), with the H₂O forcing in isolation having a relatively small chemical impact (row 4 in Figs. S6, S7 and 3) based on the nudged simulations.

Aside from direct chemically driven impacts (which are isolated by the nudged Hunga minus control simulations), the free-running responses also include contributions from the Hunga-induced changes in atmospheric temperatures and transport. Changes in atmospheric temperatures themselves are driven both directly by changes in radiatively active constituents – mainly the anomalous water vapor and aerosols, but also the associated changes in stratospheric ozone itself – and indirectly via close two-way coupling with the associated dynamical changes (via the thermal wind balance, changes in meridional circulation and adiabatic heating/cooling). Both direct and indirect changes include the forced response to the eruption and contribution from interannual variability.

To varying degrees all three models show indications of SH extra-tropical cooling from the H₂O radiative forcing (Fig. 7a–c), first in the middle stratosphere and then reaching to slightly higher altitudes following the evolution of the anomalous water vapor (Fig. 1d–f). In the lower stratosphere, there is also a weak warming from aerosol absorption and water vapor long-wave heating during the first half of 2022, but later this is offset by the decrease in ozone itself acting to cool the lower stratosphere due to the reduction of its short-wave absorption. Hence, no lower stratospheric warming can be seen after about mid-2022.

Importantly, both the temperature and the ozone changes are strongly coupled with and indicative of the associated circulation changes. We illustrate the latter in Fig. 7d–f using changes in zonal winds averaged over 70–30° S; this averaging region was chosen to encompass both the equatorward and the central part of the SH stratospheric jet to capture changes in both its position and its strength. Similar to the temperature response, the zonal wind response reflects both the forced component due to Hunga (driven by the thermal wind response to the forced radiative anomalies or associated changes in wave-mean flow interactions) and interannual variability. GEOSCCM and WACCM-MAM, and to some extent MIROC-CHASER, simulate strengthening of the SH stratospheric zonal winds in the second part of 2022 (Fig. 7d–f). This strengthening of the stratospheric jet is associated with reduced adiabatic warming (hence relative cooling) of the high-latitude stratosphere due to weakening of the Brewer-Dobson circulation  (e.g., Wang et al., 2023), which also reduces ozone transport from the tropical mid-stratosphere, thereby amplifying the extra-tropical stratospheric cooling (Fig. 7a–c) and lower stratospheric ozone reductions (Fig. 3a–c). These circulation changes also lead to the reduction in stratosphere-to-troposphere ozone transport and tropospheric ozone in Fig. 4. The dominance of the transport effect can be inferred from the significantly smaller tropospheric ozone and O3S anomalies simulated in the nudged simulations, see Fig. S3. Similar strengthening of the SH stratospheric jet is also found in WACCM-MAM in 2023 and in GEOSCCM in 2023, 2024 and 2025 but not in MIROC-CHASER, which helps to explain the varying degrees of persistence of the lower stratospheric ozone reductions across these models seen in Fig. 3a–c.

Below we focus on the contribution of the large natural interannual variability characterizing the extra-tropical stratosphere. As shown in Bednarz et al. (2026) using the 30-member WACCM-MAM ensemble, the forced dynamical response to the Hunga eruption in both hemispheres is likely relatively weak compared to interannual variability. The role such large interannual variability plays in the diagnosed ozone response to Hunga is illustrated in Fig. 8 (again using the 30-member WACCM-MAM ensemble), exemplified by the Antarctic total column ozone response in September–October–November 2023. A total column ozone response of ∼ 20 DU was diagnosed from the model based on all 30 members (black solid line), with half of it attributed to chemical changes (black dashed line; as given by the corresponding nudged simulations) and half to dynamical changes. However, and importantly, a very large spread of potential ozone responses spanning both positive and negative values is found if only a small subset of the ensemble is used for the diagnosis (blue and red lines). This suggests that (i) large model ensembles are needed to confidently diagnose the forced ozone response to the eruption, and (ii) such forced response is unlikely to be detectable in the observational record, which essentially constitutes just one realization.

An important limitation of inferring the direct chemical response to the Hunga eruption based solely on the perturbed minus control pairs of nudged simulations, as done in Sect. 3.2, is that such an experimental setup, by design, does not include any chemical changes caused by the significant stratospheric cooling induced by the anomalous water vapor (as the same temperature and wind data are used in the perturbed and control simulations). For that reason, Fig. 9. includes ozone responses simulated by three models that performed free-running simulations with H₂O injection only (CMAM, GSFC2D, and MIROC-CHASER). As expected, all models show an H₂O-induced reduction in ozone in the upper stratosphere. However, all three models show increases in ozone in the middle stratosphere, and CMAM and GSFC2D also show substantial ozone decreases in the lower stratosphere. Notably, such middle stratospheric ozone increases and lower stratospheric decreases were largely absent in the H₂O-only simulations carried out by models in the nudged configuration (Fig. 3, row 4) which, by design, do not include H₂O-induced temperature changes. Furthermore, the nudged H₂O-only experiment in GSFC2D also shows a lack of middle stratospheric ozone increase and a reduced magnitude of the lower stratospheric ozone decrease compared to its free-running counterpart (Fig. 10b, d). These ozone responses in the free-running simulations forced by Hunga water only are thus likely largely indicative of the radiative impact of increased H₂O on concentrations of chemical species of relevance for ozone chemistry, including middle stratospheric NO₂ decrease and lower stratospheric ClO_x and BrO_x increase (Fig. S9), driven by temperature dependence of their gas phase chemistry and/or temperature dependence of the relevant heterogeneous chemistry, including uptake coefficients and the formation of binary aerosols and PSCs (e.g. Carslaw et al., 1997; Solomon et al., 2015).

Finally, while the changes in chemical tracers described above could also be caused by a dynamical response to the H₂O-induced cooling (Fig. S8), the GSFC2D model, by design, does not simulate any internal dynamical response and as such changes in circulation in this model are minimal and due only to the small direct response to the Hunga radiative forcing (Fig. S8, middle). In addition, both Wang et al. (2023) and Yook et al. (2025) showed that the use of H₂O-only forcing is insufficient to capture the Hunga dynamical response, and that simultaneous injection of both SO₂ and H₂O is needed to obtain a SH dynamical response resembling that found in reanalyses. Nonetheless, we note that dynamics do play a role in CMAM, which shows a significant strengthening of zonal winds in late 2022 (Fig. S8, left); this amplifies the stratospheric cooling and also reduces ozone transport, thus contributing to the lower stratospheric ozone decrease seen in the model in late 2022 (Fig. 9).