Tropical Volcanic Eruptions and Low Frequency Indo-Pacific Variability Drive Extreme Indian Ocean Dipole Events
Abstract
Volcanic eruptions can have significant climate impacts and serve as useful natural experiments for better understanding the effects of abrupt, externally forced climate change. Here, we investigate the Indian Ocean Dipole's (IOD) response to the largest tropical volcanic eruptions of the last millennium. Post-eruption composites show a strong negative IOD developing in the eruption year, and a positive IOD the following year. The IOD and El Niño-Southern Oscillation (ENSO) show a long-term damped oscillatory response that can take up to 8 years to return to pre-eruptive baselines. Moreover, the Interdecadal Pacific Oscillation (IPO) phase at the time of eruption controls the IOD response to intense eruptions, with negative (positive) IPO phasing favoring more negative (positive) IOD values via modulation of the background state of the eastern Indian Ocean thermocline depth. These results have important implications for climate risk in low-likelihood, high-impact scenarios, particularly in vulnerable communities unprepared for IOD and ENSO extremes.
Key Points
-
Both Indian Ocean Dipole (IOD) and El Niño-Southern Oscillation (ENSO) show oscillatory responses for up to 8 years following strong tropical volcanic eruptions
-
These eruptions induce a negative IOD followed by a positive IOD, and this response scales with eruption intensity
-
Initial Interdecadal Pacific Oscillation phasing at time of eruption influences the post-eruption response via low-frequency thermocline depth modulation
Plain Language Summary
Volcanic eruptions were previously found to affect the El Niño-Southern Oscillation in the Pacific, but little is known about their effect on the Indian Ocean. In this study, we use simulations of the strongest tropical eruptions of the last millennium to better understand the effects of volcanic activity on the Indian Ocean. We find a consistent response of sea surface temperatures and surface winds across the basin between different simulated eruptions. This response scales with eruption intensity and is long-lived, lasting for multiple years. We also find that sea surface temperatures in the tropical Pacific at the time of eruption influence the strength of the initial tropical Indian Ocean response. Understanding these connections will allow for improved climate predictability and preparedness following large tropical eruptions.
1 Introduction
In the preindustrial era of the last millennium, most variability in planetary radiative forcing can be attributed to volcanism (Otto-Bliesner et al., 2016). The largest and most explosive eruptions are capable of injecting aerosols tens of kilometers into the stratosphere with significant effects on global oceanic and atmospheric climate (Dogar & Sato, 2019; Dogar et al., 2017; Robock, 2000). These punctuated events offer an opportunity to study the Earth system's response to abrupt, external forcing and allow a fairer risk assessment of stratospheric geoengineering projects designed to mitigate anthropogenic warming (Plazzotta et al., 2018; Robock et al., 2009).
While some aspects of the post-eruption climate response are well studied, such as characteristic multi-year reduction in global mean temperatures due to the radiative effects of aerosol forcing (Fujiwara et al., 2020; Robock, 2000), others such as their impacts on internal climate variability are still a subject of debate. The El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD) are modes of interannual climate variability with far-reaching hydroclimate impacts on the global tropics (Dogar et al., 2019; Lyon, 2004; Saji & Yamagata, 2003). Most previous work on the volcanic forcing of tropical climate variability has focused on ENSO. Observed and simulated sea surface temperatures (SST) show a consistent El Niño response following large tropical and Northern Hemisphere eruptions (Dogar et al., 2023; Emile-Geay et al., 2008; Khodri et al., 2017; Pausata et al., 2015; Singh et al., 2020; Stevenson et al., 2016).
In agreement with observations and models, paleoclimate archives tend to support an ENSO response to tropical eruptions with a few exceptions. A recent compilation of ENSO reconstructions found that 12 out of 17 display an El Niño-like warming after large tropical eruptions (McGregor et al., 2020). ENSO reconstructions tend to show El Niño conditions in boreal winter of the first year after large tropical eruptions followed by La Niña conditions (Adams et al., 2003; Li et al., 2013). In contrast, coral δ18O records show little to no relationship between volcanism and ENSO (Dee et al., 2020; Zhu et al., 2022). This discrepancy might be due to issues surrounding temporal discontinuity in coral paleorecords and age uncertainties; the former precludes analysis of particular eruptions and the latter muddles signals derived from timeseries composite methods such as superposed epoch analysis (SEA) (Zhu et al., 2022).
A possible mechanism driving post-eruption El Niños is an enhanced land-sea temperature gradient between tropical Africa and the Indian Ocean which spurs equatorial westerly wind anomalies across the Indo-Pacific (Khodri et al., 2017). However, recent modeling work found land-sea gradients play a smaller role in the longer-term ENSO response than the moderating influence of the equatorial eastern Pacific upwelling region on SST anomalies (Predybaylo et al., 2020). This effect, often described as an ocean-dynamical thermostat, is equally capable of spurring westerly anomalies and giving rise to El Niño conditions. However, simulated ENSO responses to eruptions are sensitive to many factors such as eruption magnitude, eruption seasonality, and tropical Pacific initial conditions (Predybaylo et al., 2020; Stevenson et al., 2017).
The modulation of internal variability by volcanic (external) forcing has been far less studied in the Indian Ocean region. The few studies investigating this region's simulated response to large tropical eruptions find an initial negative IOD (nIOD) followed by positive IOD (pIOD) conditions (Blake et al., 2018; Izumo et al., 2018; Maher et al., 2015; Singh et al., 2020). Here, we expand upon these findings using the Community Earth System Model Last Millennium Ensemble (CESM-LME). We focus on understanding IOD responses to a range of eruption amplitudes, inter-basin connectivity with ENSO through the longer-term forced response, and the role of the Interdecadal Pacific Oscillation (IPO) in preconditioning post-eruption IOD responses.
2 Data and Methods
CESM-LME consists of transient simulations run over 850–2005 CE with all-forcing as well as single-forcing scenarios (Otto-Bliesner et al., 2016). The 13 all-forcing runs (ALL) incorporate volcanic, solar, orbital, land use, and greenhouse gas external forcings. Of the single-forcing scenarios, the 5 volcanic-only runs (VOLC) were used in this study for comparison with ALL. Anomalies are reported relative to a non-volcanic climatology generated by compositing all years when global stratospheric injection = 0 in each scenario. Analyses largely focus on September-October-November (SON) means, when the IOD is known to peak (Saji et al., 1999). Time post-eruption is represented as the following: SON(0) indicates the boreal fall in the year of an eruption, SON(1) indicates the boreal fall in the year following an eruption, etc. The IOD Dipole Mode Index (DMI), Niño3.4 index, and Interdecadal Pacific Oscillation (IPO) tripole index were calculated. Discussion of model validation against observational data as well as details regarding the calculation of climate indices can be found in Supporting Information S1.
A reconstruction of stratospheric sulfate injection which forced the CESM-LME (Gao et al., 2008) was used to identify tropical eruptions of the last millennium. Some of the largest eruptions were used in this analysis, namely Samalas (1258), Kuwae (1452), Tambora (1815), Huaynaputina (1600), an unnamed eruption (1284), and Pinatubo (1991) (Table S1 in Supporting Information S1). For those with unknown seasonality, which includes nearly all the above, events are simulated to erupt in April. For more details on tropical eruption selection see Supporting Information S1.
3 Results
3.1 The Initial IOD Response to Volcanic Eruptions
A comparison between the ice core-derived Gao et al. (2008) volcanic reconstruction and CESM-LME all-forcing and volcanic single-forcing DMI shows that large eruptions often coincide with some of the most extreme IOD events of the last millennium (Figure 1). Similarly large swings are observed in the Niño3.4 index (Figure S11 in Supporting Information S1). This relationship is clearly seen in the Samalas eruption of 1258 as well as the Kuwae eruption in 1452. Perhaps unsurprisingly, VOLC DMI has higher variance than ALL DMI because VOLC has fewer ensemble members, with potentially greater noise than ALL due to internal variability across the small ensemble size.
The Indo-Pacific's response to strong tropical eruptions. (a): CESM-LME volcanic single-forcing (VOLC) ensemble mean of DMI during SON. (b): CESM-LME all-forcing (ALL) ensemble mean of DMI during SON. (c): Ice core-derived global stratospheric sulfate injection reconstruction (Gao et al., 2008). Light gray bars highlight the eruptions in this study. (d): SST and surface wind (1000 mbar) post-eruption composite anomaly for SON(0). Anomalies are relative to the SON non-volcanic climatology (i.e., when global stratospheric sulfate injection = 0). Red boxes indicate the western and eastern DMI regions, and the Niño3.4 region. Stippling indicates SST anomalies insignificant at the 95% confidence level using single sample t-tests. Only surface wind anomalies significant at the 95% confidence level are shown.
The spatial response to the strongest tropical eruptions manifests as an intensification of the climatological SST gradient across the tropical Indian Ocean with enhanced warming near the eastern Indian Ocean upwelling region and cooling off East Africa accompanied by surface westerly wind anomalies across the basin (Figure 1d). This response is nearly identical between the VOLC and ALL scenarios, with minor differences likely due to the effects of other external forcings (Figure S4 in Supporting Information S1). Previous work found these post-eruption surface westerly winds driven by enhanced land-sea temperature gradients due to increased cooling over tropical continental Africa relative to the Indian Ocean (Izumo et al., 2018; Khodri et al., 2017). This response is at a maximum in SON when the IOD is known to peak (Saji et al., 1999) but retains strong negative SST anomalies in the western Indian Ocean and enhanced westerlies through DJF (Figure S5 in Supporting Information S1). The SST and surface wind response is considerably weaker in the Pacific relative to the Indian Ocean in this first year. The equatorial Pacific response is insignificant against the background cooling generated by eruptions. The use of relative SST anomalies (i.e., SSTs relative to the global tropical climatology) was found to bring out a stronger ENSO signal following simulated eruptions (Khodri et al., 2017). Unlike ENSO, the IOD is notable in that its initial post-eruption signal is often large enough to be apparent beyond the global tropical cooling induced by eruptions. The only area with significant warm SST anomalies in the global tropics in these post-eruption composites is the eastern Indian Ocean upwelling region, making the Indian Ocean response exceptional in this regard (Figure 1).
Strong tropical eruptions most often result in negative DMI values in the ALL and VOLC scenarios in SON(0), the peak IOD season of the first year (Table S1 in Supporting Information S1, Figure 2a). These responses mark a shift away from the range of unforced IOD variability (“No volcanism”). The initial nIOD response to intense volcanism in CESM-LME is in good agreement with a previous study which found an nIOD response to Pinatubo-style eruptions in IPSL-CM5B-LR simulations (Izumo et al., 2018). The response in SON(1) of the year following an eruption is the exact opposite of the response in SON(0), showing strongly positive DMI values (Figure 2c). This agrees with previous work which finds a pIOD response to strong tropical volcanic eruptions in the following year (Blake et al., 2018).
Initial and subsequent year's IOD response to tropical volcanism in SON. (a): Boxplots showing the initial DMI response in SON(0) to the largest tropical eruptions of the last millennium. These responses are compared to the DMI's range of non-volcanically induced variability (“No volcanism”) when global stratospheric sulfate injection = 0. Horizontal black lines in the boxplots indicate the third quartile, median, and first quartile. Whiskers extend from the boxes by 1.5X the interquartile range. (b): Scatterplot of DMI response in SON(0) versus global stratospheric sulfate injection for tropical eruptions of various sizes in the ALL and VOLC ensemble means. (c) and (d): The same as (a) and (b) but for SON(1) in the year following an eruption. Dark orange and light orange plot elements represent responses from volcanic single-forcing (VOLC) and all-forcing (ALL), respectively.
The higher the eruption intensity, the more negative the nIOD response (Figure 2b). This is observed in the all-forcing as well as volcanic single forcing scenarios, suggesting a relationship between the initial nIOD response and eruption intensity. The opposite is true in the year following an eruption, with DMI values increasing with greater eruption intensity in both scenarios (Figure 2d). Regardless of the sign, the IOD response plateaus with increasing stratospheric aerosol loading which suggests a nonlinear relationship. This might indicate a limit on zonal SST gradient extremes in response to the most intense eruptions. Though this is the first time a relationship between the IOD and tropical eruption intensity has been described, simulations show a stronger ENSO response with increasing eruption intensity (Emile-Geay et al., 2008). Moreover, the more tropical an eruption, the greater the IOD response (Figure S6 in Supporting Information S1). Contrary to tropical eruptions with relatively symmetric hemispheric injection, eruptions dominated by asymmetric Northern Hemisphere or Southern Hemisphere injection do not show as much of a relationship between eruption intensity and IOD response.
3.2 The Long-Term IOD Response to Volcanic Eruptions
The decade following the most intense tropical eruptions exhibits continued IOD extremes (Figure 3a). The post-eruption timeseries composite features a damped oscillatory response with multiple consecutive years of positive and negative DMI anomalies peaking in SON of each year until the externally forced signal returns to baseline. It takes roughly 7–8 years to return to stable, non-volcanic levels of variability, suggesting a long-lived forced response of the IOD. This long-term response is nearly identical in the VOLC scenario (Figure S7 in Supporting Information S1). Comparing the forced composite with composites of unforced nIOD event evolution, nearly all IOD events following strong eruptions exceed the range of non-volcanic IOD variability with the exception of the initial nIOD. It is possible the model's bias for overly strong IOD events could account for this discrepancy (Figure S1 in Supporting Information S1).
The long-term Indo-Pacific response to strong tropical eruptions of the past millennium in ALL. (a): Monthly post-eruption timeseries composites of the DMI (black) and non-volcanic nIOD evolution (blue). Shading represents one standard error from composite means. Non-volcanic events below 0σ (light blue) and −1.5σ (dark blue) shown as dashed lines. (b): Post-eruption timeseries composite of the Niño3.4 index (black) with non-volcanic El Niño event evolution (purple). Non-volcanic events below 0σ (light purple) and −1.5σ (dark purple) shown as dashed lines. Non-volcanic events are those when global stratospheric sulfate injection = 0. Shading represents one standard error from composite means. Years are labeled at the start of a given calendar year and minor tick marks indicate individual months. April of the first year is shown by the vertical black line in year 0, marking the start of many eruptions in the composite. Note that control composites are largely insensitive to the event thresholds used (Figure S10 in Supporting Information S1). (c) and (d): Scatterplots of the SON(0) IOD response with the DJF(1) ENSO response (c), and the DJF(3) ENSO response (d). In the scatterplots, Niño3.4 values are reported for DJF, when ENSO events are known to peak (Izumo et al., 2010).
In agreement with spatial composites, ENSO does not show nearly as large of a response as the IOD in the eruption year (Figures 1d and 3b). However, a strong El Niño develops at the end of the first year after eruption which lags the pIOD response by about two months, consistent with the peak seasons of IOD and ENSO events. And nIOD events in years 3–5 are concurrent with strong La Niña conditions. These IOD and ENSO events exceed the range of internal, non-volcanic variability. The strength of the IOD response in the year of an eruption is positively (negatively) correlated with the ENSO response in DJF of year 1 (year 3), suggesting a role for the IOD as the primary driver of these ENSO extremes (Figures 3c and 3d). This is also supported by initial nIOD post-eruption responses occurring under neutral ENSO initial conditions (i.e., without ENSO as a possible driver) (Figure S9 in Supporting Information S1).
This inter-basin connectivity is consistent with observations and model experiments which show nIOD (pIOD) events facilitating the development of El Niño (La Niña) conditions 14 months after the initial IOD event (Izumo et al., 2010). The ENSO response simulated here agrees with previous studies finding a post-eruption El Niño followed by La Niña conditions (Adams et al., 2003; Li et al., 2013; Pausata et al., 2015). More specifically, the El Niño response to tropical eruptions in DJF(1) also reinforces previous work that finds little ENSO response beyond background cooling in the eruption year but a notable response in the subsequent year (Stevenson et al., 2016). And interestingly, the long-lived ENSO response closely mirrors a post-eruption timeseries composite from ENSO-sensitive tree ring records, potentially offering proxy-model validation (Li et al., 2013).
3.3 Modulation by Low-Frequency Pacific Variability
While nIOD conditions often follow strong tropical eruptions, the magnitude of the initial Indian Ocean response can vary considerably between model runs (c.f. the ALL and VOLC SON(0) DMI and their uncertainties in Table S1 in Supporting Information S1). Multidecadal Pacific variability can precondition IOD events via low-frequency changes in eastern Indian Ocean thermocline depth transmitted through the Indonesian Throughflow (Ummenhofer et al., 2017). The modulation of IOD variability by Indo-Pacific Warm Pool thermocline depth is also seen in CESM-LME simulations (Abram et al., 2020). To better understand how multidecadal Pacific variability modulates Indian Ocean post-eruption responses, composites of thermocline depth in SON(0) divided according to positive IPO (n = 47) and negative IPO (n = 31) initial conditions at the time of eruption for various ensemble members and the difference between them were calculated (Figures 4a–4c).
IPO preconditioning of the IOD response to tropical volcanism. (a–c): SON(0) 20°C isotherm depth anomaly for eruptions with (a): positive IPO initial conditions, (b): negative IPO initial conditions, (c): The difference between the positive and negative IPO initial condition 20°C isotherm depth anomaly composites. Positive (negative) anomaly values indicate deepening (shoaling) of the 20°C isotherm. Stippling indicates thermocline depth anomalies insignificant at the 95% confidence level using single sample t-tests. (d): Post-eruption timeseries composite of ALL subset according to positive IPO and negative IPO initial phasing. Shading represents one standard error from the mean, represented by the darker lines. (e): Scatterplot of SON(0) DMI versus global stratospheric sulfate injection for each ensemble member of ALL (i.e., non ensemble-means) for every tropical eruption. Points are color-coded by initial IPO phase. Probability density functions (PDFs) are provided on the scatterplot axes and share the same color scheme as the points. The y-axis PDFs are both individually normalized to one to account for different numbers of negative IPO and positive IPO samples, and the x-axis PDFs are together normalized to one.
Eruptions in ensemble members with positive IPO (negative IPO) initial conditions have a shallower (deeper) thermocline in the Indo-Pacific Warm Pool region and a deeper (shallower) thermocline in the western Indian Ocean and East Pacific in SON(0). Under negative IPO initial conditions, a deeper thermocline in the eastern Indian Ocean and a shallower thermocline in the western Indian Ocean strengthens the climatological SST gradient via the Bjerknes feedback and preconditions the basin for stronger post-eruption nIOD events. In contrast, positive IPO initial conditions weaken the climatological SST gradient by shoaling the eastern Indian Ocean thermocline with a damping effect on the post-eruption response, giving rise to more neutral IOD conditions.
Moreover, the ALL post-eruption timeseries composites of the strongest tropical eruptions separated by initial IPO phasing show that IPO preconditioning is most important for the first IOD event after an eruption (Figure 4d). Distinct responses are also seen in the year following an eruption but to a lesser extent, with initial positive IPO leading to slightly stronger pIOD in SON(1) than in the negative IPO composite. Similar responses were found with El Niño, La Niña and neutral ENSO eruption preconditioning, suggesting that although the initial IOD response does not require ENSO, it can be modulated by ENSO phasing (Figure S9 in Supporting Information S1).
There is a statistically significant difference between the SON(0) IOD response with positive IPO and negative IPO initial conditions using a two-sample Student's t-test (p < 0.001) and a Kolmogorov-Smirnov test (p < 0.001). This difference can also be seen in the probability density functions (PDFs) shown on the y-axis of Figure 4e, with less positive SON(0) DMI under initial negative IPO conditions. Note that even though there is a slight positive IPO bias in this sampling (x-axis PDFs in Figure 4e), the stronger SON(0) IOD response under negative IPO initial conditions is still a prominent feature. This separation in post-eruption responses would likely be even greater under a less positive IPO-biased sample. In VOLC scenarios, the difference in IOD responses with positive IPO and negative IPO preconditioning is even more apparent, perhaps due to the lack of confounding forcings and/or less of a positive IPO sampling bias (Figure S8 in Supporting Information S1). Altogether, these results suggest the Pacific mean state over multidecadal timescales can explain some of the spread in IOD responses through the first few years after large tropical eruptions.
4 Discussion
These results spanning the largest tropical eruptions of the last millennium expand upon previous work which has often only focused on single eruptions. The CESM-LME all-forcing and volcanic single-forcing simulations feature robust SON(0) nIOD and SON(1) pIOD responses to strong tropical volcanic eruptions beyond the range of internal variability in agreement with previous studies (Blake et al., 2018; Izumo et al., 2018). IOD eruption responses scale with eruption intensity. Strong westerly anomalies across the tropical Indian Ocean are accompanied by a strengthening of the SST gradient in SON(0), with positive SST anomalies in the eastern Indian Ocean upwelling region exceeding the widespread volcanically-induced surface cooling in the global tropics. These patterns are consistent with mechanisms involving enhanced land-sea gradients between tropical Africa and the Indian Ocean driving surface westerly anomalies (Izumo et al., 2018; Khodri et al., 2017). Post-eruption timeseries composites show a long-term damped oscillatory response of the IOD and ENSO. Unlike the IOD, ENSO shows little response in year 0, though a robust El Niño event occurs at the end of year 1 and a La Niña at the end of year 3. Correlations between the SON(0) nIOD response and ENSO responses in subsequent years suggest the role of inter-basin interaction and the IOD as a primary driver of post-eruption Indo-Pacific variability. The initial IOD eruption response is modulated by the mean state of the Indo-Pacific. Eruptions with positive IPO initial conditions experience a relative shoaling of the eastern Indian Ocean thermocline in SON(0) relative to eruptions with negative IPO initial conditions. The result of this low-frequency thermocline modulation is a dampening of the initial negative IOD response under positive IPO conditions, and a strengthening of the initial negative IOD response under negative IPO conditions. This is an enhancement of the known effect of IPO preconditioning on IOD phasing (Ummenhofer et al., 2017).
Additional work can investigate the influence of eruption seasonality on the IOD response, an aspect that was found to influence ENSO responses in CESM simulations (Stevenson et al., 2017). Though the IOD response found in this work is robust, it is only applicable to eruptions occurring in boreal spring due to the often-unknown seasonality of eruptions. It is likely there is a minimum amount of time required between eruptions and IOD/ENSO peaks for there to be a detectable response, perhaps due to adjustments in land-sea temperature gradients which may require a certain number of months to deepen. From this study, it is clear that an April eruption provides enough time for a sufficient tropical ocean-atmosphere response in SON and DJF of the same year, though additional nuance would be beneficial. Future work could also explore the IOD-ENSO inter-basin connections following strong tropical eruptions, an aspect of the post-eruption response presented in this study but not sufficiently characterized in terms of its dynamics or causality. Targeted pacemaker experiments could shed light on the nature of forced pantropical responses. Model biases act as an additional limitation, possibly making IOD response amplitudes too large. Reducing known model biases will help increase confidence in future results. Finally, improvements in the continuity and distribution of paleorecords, particularly corals, will increase the fidelity of tropical climate mode reconstructions and allow for better comparison with model and observational data.
This work has important implications for climate risk assessments of low likelihood, high impact scenarios. These results suggest that interannual climate variability associated with ENSO and the IOD will substantially increase after large tropical eruptions, which is expected to cause whiplash in precipitation extremes across the Indo-Pacific (Cai et al., 2014; Saji & Yamagata, 2003). In light of the IPO's role in modulating eruption responses, the high decadal predictability of the tropical Indian Ocean (Guemas et al., 2013) will be useful for evaluating post-eruption climate risk. With decadal predictability improving, and a better understanding of the external forcing of internal variability, it will be possible to better prepare those most vulnerable for the impacts of large tropical eruptions.
Acknowledgments
Use of the following data sets is gratefully acknowledged: HadISST data from the Met Office Hadley Centre, ERSST V5 from the Physical Sciences Laboratory at NOAA. We acknowledge the CESM1 Last Millennium Ensemble Community Project and supercomputing resources provided by NSF/CISL/Yellowstone. This work was supported by NSF award AGS-1702691 (to BHT), NSF award AGS-2002083 (to CCU) and the James E. and Barbara V. Moltz Fellowship for Climate-Related Research at WHOI (to CCU). Helpful discussion with K. Thirumalai and constructive comments by two reviewers on earlier versions of the manuscript are gratefully acknowledged.
Open Research
Data Availability Statement
All data used in this study are publicly accessible. The ERSST, OISST and HadISST data used for model validation can be accessed at https://psl.noaa.gov/data/gridded/data.noaa.ersst.v5.html, https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.html, and https://www.metoffice.gov.uk/hadobs/hadisst/, respectively. The CESM-LME simulations are available through NCAR at https://www.earthsystemgrid.org/dataset/ucar.cgd.ccsm4.cesmLME.html.
