Insights into contrasting ENSO influence on SST variations off Australia’s southeast and west coasts

Fundamentally a coupled ocean-atmosphere phenomenon established in the tropical Pacific Ocean, El Niño–Southern Oscillation (ENSO) has teleconnections well beyond the Tropics and is the dominant global mode of interannual climate variability¹. ENSO plays an important role in influencing sea surface temperature (SST) variations in many regions around the world^2,3,4. Ocean temperature anomalies associated with ENSO, including extremes such as marine heatwaves (MHWs), have played a critical role in altering marine species, ecosystem structures and their functioning^5,6. Understanding ENSO modulations of SST variations is important for regional marine ecosystem management strategies and planning^6,7, with potential benefits from well-developed ENSO prediction systems^{8,9,10,11,12,13,14,15}.

While ENSO impacts are felt globally, specific ENSO contributions and associated teleconnection mechanisms are still not well understood in many ocean regions^3,4. Within the Earth system, internal climate variability arises from internal processes in different climate components (e.g., atmosphere, ocean, land, cryosphere), as well as their coupled interactions. In some ocean regions outside of the tropical Pacific, SST variations are simultaneously modulated by ENSO and other sources of internal climate variability which can mask ENSO contributions. Further, ENSO may modulate regional SST variations via more than one teleconnection pathway. Through some pathways, ENSO may even modulate regional SST variations indirectly by ENSO firstly impacting other climate modes (e.g., the Indian Ocean Dipole (IOD)^16,17,18 and the Southern Annular Mode (SAM)^19,20,21,22), and then together these modulating the regional SST variations. For these ocean regions, it can be challenging to disentangle the role of ENSO teleconnections from each other and from other non-ENSO internal climate variability with high confidence by using empirical methods (e.g., composite and regression analyses), especially when dealing with relatively short observational records (e.g., <50 years)^23,24.

To explore and identify ENSO modulations of regional SST variations, we set up a novel modelling approach in which tropical Pacific pacemaker ensemble simulations (e.g., refs. ^24,25) and ocean general circulation model (OGCM) perturbation experiments (e.g., ref. ²⁶) are combined. This modelling framework investigates the role of ENSO on regional SST variations, by taking account of: (1) the contributions from various ENSO forcing mechanisms and pathways that are related to different teleconnections (e.g., ENSO-driven wind stress forcing remotely from the tropical Pacific via oceanic teleconnections); and (2) the superposition of non-ENSO climate variability on ENSO forcings. For a tropical Pacific pacemaker ensemble, with the tropical Pacific SST field nudged to the reconstructed or observational product, its ensemble mean provides a dynamically based estimate of global SST and atmospheric responses to ENSO (e.g., refs. ^24,25,27; see details in “Methods”), We refer to these ensemble-mean SST and atmospheric responses to ENSO as ENSO-driven SST variations and atmospheric forcings respectively. They reflect the accumulations of ENSO impacts through all different teleconnection pathways, including those indirect pathways associated with other climate modes. The spread of ensemble members from the ensemble mean represents, and is referred to as, responses to non-ENSO internal climate variability (i.e., non-ENSO forcings due to other sources of internal climate variability). In our modelling approach, we further combine pacemaker ensemble simulations with OGCM perturbation experiments. We separate the derived ENSO-driven atmospheric forcings into several regions and undertake a suite of perturbation experiments forced by different ENSO-driven atmospheric forcings (e.g., wind stress and heat flux) in different regions. This enables us to efficiently examine the individual contributions of these forcings (and ENSO teleconnection mechanisms relevant to them) to SST anomalies in the regions of interest. Additionally, OGCM experiments forced by atmospheric forcings from individual ensemble members are undertaken as well to compare the SST response to both ENSO and non-ENSO forcings, identifying the superposition effects of non-ENSO climate variability on regional SST responses to ENSO.

In the present study, as an example case study, we applied this modelling approach to the oceans off Australia’s west and southeast coasts, identifying and comparing the contributions of physical mechanisms that explain ENSO modulations of regional SST variations. Notably, to reduce complexity, in the study, we focused on forced deterministic ENSO and non-ENSO climate variabilities which are generally driven by large-scale ocean and/or atmospheric forcings. The effects from chaotic intrinsic variability associated with mesoscale activity was not discussed and is considered beyond the scope of the study. The Massachusetts Institute of Technology general circulation model (MITgcm)²⁸, and the Community Earth System Model1 (CESM1) Pacific Pacemaker Ensemble Experiment (hereafter CESM Pacific pacemaker experiment)²⁴ were adopted for the approach (see “Methods”).

Both the west and southeast coasts of Australia have been strongly impacted by substantive MHWs in recent years^29,30,31,32, with the extreme 2011 MHW off Western Australia occurring during the large 2010/11 La Niña event^29,33 and the unprecedented 8-month-duration 2015/16 Tasman Sea MHW coinciding with the extreme 2015/16 El Niño—with the latter most likely influenced by climate change³¹. Given the global importance of ENSO, and its particular significance in the Indo-Pacific region, there is potential benefit in understanding how ENSO dynamics and ENSO teleconnections modulate and/or precondition SSTs around Australia, including informing SST predictability^11,14,15,34. The role of ENSO on SSTs is found to be strong off Australia’s west coast (eastern Indian Ocean). In this region, the variation of southward advection of warm water by the poleward-flowing Leeuwin Current³⁵ (Fig. 1a) is strongly modulated by oceanic coastal waves with forcing originating from equatorial Pacific wind stress anomalies related to ENSO^36,37,38,39. Further, via atmospheric teleconnections, ENSO induces changes in meridional wind stress and air-sea heat fluxes locally off the coast, contributing to the Leeuwin Current southward transport and regional SST variations^29,40,41,42. In contrast, SST variations off Australia’s southeast coast (southwest Tasman Sea) are found to be weakly, but nevertheless significantly, related to ENSO^{43,44,45,46,47,48}. Several mechanisms have been proposed to explain this, from atmospheric teleconnections causing local wind and heat flux changes (e.g., ref. ⁴⁹) to oceanic teleconnections leading to East Australian Current Extension and Zeehan Current variations (Fig. 1a; e.g., ref. ⁴⁸). However, the role of ENSO and the specific teleconnection pathways in the Tasman Sea are less understood than those off Australia’s west coast. While there has been no comprehensive comparative study of how ENSO modulates SSTs off Australia’s west and southeast coasts, our modelling approach is designed to find physical explanations behind the relative importance of ENSO in these regions, including the ENSO teleconnection pathways.

Monthly SSH anomaly standard deviation (shading areas) and long-term mean SSH (contour) from a MITgcm Hindcast experiment and b ECCO v4r4 product during the 1994–2016 period. Schematic of the ocean circulation around the west and southeast coasts of Australia is also shown in (a). The red boxes in (b) highlight the Western Australia box (WAbox) region off Australia’s west coast and the Tasman Sea box (TSbox) region off Australia’s southeast coast in the study. Land and non-active ocean grids for the MITgcm regional model are shaded in white in (a) and (b). For comparison, time series of SSH anomalies averaged in c WAbox and d TSbox are derived from the MITgcm Hindcast experiment (black), ECCO v4r4 (red) and AVISO (blue). “Corr” indicate correlation coefficients between MITgcm Hindcast experiment with ECCOv4r4 or AVISO. All correlation coefficients in the subplots are statistically significantly at the 95% confidence level. LC Leeuwin Current, ZC Zeehan Current, TO Tasman Outflow (undercurrent), EAC East Australian Current, EAC Ex East Australian Current Extension, TF Tasman Front.

Full size image

In this study, with the combined use of global ENSO-driven and non-ENSO atmospheric forcings derived from the CESM Pacific pacemaker experiment and MITgcm perturbation experiments, we provide a comprehensive analysis of the contributions of different forcings in different regions to SST anomalies off Australia’s west and southeast coasts. We identified the small net ENSO effect on SST variations off Australia’s southeast coast and provided information on the underpinning mechanisms. We further demonstrate the significant role of non-ENSO climate variability in weakening this identified small ENSO effect. We conclude by explaining why SSTs along the southeast coast of Australia are modulated by ENSO much less strongly than along Australia’s west coast. This successful application to Australian coasts also offers the potential for this modelling approach to be applied to other regions around the world.

SST responses to ENSO off Australia’s west and southeast coasts

SST responses to ENSO within a regional ___domain off Australian’s west coast (hereafter, the WAbox; bounded by (110-116°E, 22-32°S); Fig. 1b) and a separate regional ___domain off Australia’s southeast coast (hereafter, the TSbox (bounded by (147-156°E, 37-46°S); Fig. 1b) were analysed as box-average SST anomalies from the ensemble mean of the CESM Pacific pacemaker experiments (Fig. 2, black lines). Similar regions have been used in previous studies to investigate the significant SST changes, including MHWs, off Australia’s west and southeast coasts^{29,31,40,50,51}. Here, the WAbox and TSbox ensemble-mean box-averaged SST anomalies were respectively compared against, and show different relationships with, those of the individual ensemble members from the pacemaker ensemble—which can be interpreted as the combined responses to both ENSO and non-ENSO climate variability (Fig. 2, black and grey lines).

Time series of monthly SST anomalies averaged in the a WAbox and b TSbox regions. The black and red lines indicate time series of ENSO-driven SST anomalies (i.e., SST responses to ENSO) derived based on the ensemble mean of the CESM Pacific pacemaker experiment, and MITgcm PACE-EM-Full experiment respectively. The grey lines correspond to the time series of SST anomalies derived from the individual ensemble members of the CESM Pacific pacemaker experiment. The yellow lines correspond to the time series of SST anomalies derived from the ECCO v4r4 dataset.

Full size image

Despite considerable diversity in the time series, the WAbox time series of individual ensemble-member SST anomalies follow and share many features with the ensemble-mean SST anomalies, implying an overall deterministic ENSO modulation of the SST field in the WAbox (Fig. 2a). In contrast, a large spread of SST anomalies is evident across the TSbox ensemble members, with significant deviations from the ensemble mean (Fig. 2b), suggesting a weaker and considerably more uncertain ENSO modulation of the SST in the TSbox.

To further examine and compare the ENSO modulation mechanisms of SST variations in the WAbox and TSbox, a suite of MITgcm perturbation experiments was designed and performed (see details in “Methods” and Supplementary Table 1). The PACE-EM-Full (PACEmaker-EnsembleMean-Full) experiment was performed to provide model estimates of the SST response to ENSO across the entire model ___domain. We found that the WAbox and TSbox SST anomalies from the PACE-EM-Full experiment compared well with those from the CESM Pacific pacemaker experiment ensemble mean (Fig. 2, red and black lines). This suggests that the important ENSO-driven SST variations (i.e., SST responses to ENSO) for both regions are effectively captured by the model. We next undertook other MITgcm perturbation experiments to explore the ENSO teleconnection mechanisms influencing the SST variations, as well as the superposition effects of non-ENSO climate variability on SST responses to ENSO in the two regions respectively.

ENSO modulation mechanisms of SST off the west coast of Australia

The PACE-EM-Wind and PACE-EM-Buoy experiments were used to respectively identify the contributions of global wind stress and buoyancy (i.e., heat and freshwater fluxes) forcings to the total ENSO-driven SST variations in the selected regions. For the WAbox, the summed SST anomalies of the two experiments were found to be nearly equal to the SST anomalies from the PACE-EM-Full experiment (Fig. 3a, dashed and black lines), indicating a very good closure. Comparing the WAbox SST anomalies from the PACE-EM-Full, PACE-EM-Wind and PACE-EM-Buoy experiments over the 1994–2013 period, we found that the SST response to ENSO in the region was largely a consequence of ENSO-driven wind stress forcing, but moderately damped by ENSO-driven buoyancy forcing (Fig. 3a). Specifically, the WAbox SST anomalies from the PACE-EM-Wind experiment are highly correlated with those from the PACE-EM-Full experiment (correlation coefficient \(r=0.83\), \(p < 0.05\)), together with a larger standard deviation of 0.69°C (Fig. 3a; Table 1). The WAbox SST anomalies from the PACE-EM-Buoy experiment are significantly anticorrelated with those from the PACE-EM-Wind experiments (\(r=-0.74\), \(p < 0.05\)), and have a smaller standard deviation of 0.40 °C (Fig. 3a; Table 1).

a shows the SST anomalies derived from the PACE-EM-Full (black line), PACE-EM-Wind (blue line) and PACE-EM-Buoy (red line) experiments. b shows the SST anomalies derived from the PACE-EM-Wind (black line), PACE-EM-TPWind (blue line) and PACE-EM-WAWind (red line) experiments. c shows the SST anomalies derived from the PACE-EM-Buoy (black line) and PACE-EM-WABuoy (red line) experiments. All monthly SST anomaly time series are calculated by averaging with the WAbox region.

Full size image

The WAbox SST anomalies from the PACE-EM-TPWind and PACE-EM-WAWind experiments were next examined to isolate the remote and local contributions from ENSO-driven wind stress forcing in the tropical Pacific and in the eastern Indian Ocean off Western Australia (arising from ENSO teleconnections). It was found that the WAbox SST anomalies from both experiments were highly correlated with the regional SST response to ENSO derived from the PACE-EM-Full experiment (\(r=\mathrm{0.85,0.77}\) respectively, \(p < 0.05\); Fig. 3b; Table 1). Further, the SST anomalies due to tropical Pacific wind stresses as revealed by the PACE-EM-TPWind experiment (standard deviation of 0.44 °C) account for a larger part of the SST anomalies from the PACE-EM-Wind experiment (standard deviation of 0.69 °C), compared to SST anomalies arising from the local wind stress forcing in the PACE-EM-WAWind experiment (standard deviation of 0.24 °C; Fig. 3b, Table 1). For the effect of ENSO-driven buoyancy forcing, the WAbox SST anomalies from the PACE-EM-WABuoy and PACE-EM-Buoy experiments show high correlation coefficient and close standard deviations with each other (Fig. 3c; Table 1). Indicated by the above results, the SST response to ENSO in the WAbox off the west coast of Australia is significantly driven by tropical Pacific wind stress forcing, while the local wind stress and buoyancy forcings due to ENSO teleconnections considerably enhance and damp this effect respectively.

The significant positive contribution of tropical Pacific wind forcing shown by our results is consistent with the ENSO oceanic teleconnection mechanisms discussed in previous studies, including the 2011 MHW case studies^33,36,40. During El Niño (La Niña), equatorial wind stress anomalies generate upwelling (downwelling) signals which propagate westward as Rossby waves in the tropical Pacific. These upwelling (downwelling) signals later pass through the Indonesian seas and propagate down the west coast of Australia as coastal waves, weakening (strengthening) the southward advection of warm water through the Leeuwin Current and thus leading to negative (positive) SST anomalies off the west coast of Australia.

As for the moderate contributions of local wind stress and buoyancy forcings due to ENSO teleconnections, we further identified their underpinning mechanisms by examining their lagged regressions onto the Niño 3.4 index (Fig. 4a, b). For the local wind stress forcing, we found that with El Niño (La Niña), atmospheric anticyclonic (cyclonic) anomalies (Fig. 4a; contours) were generated in the eastern Indian Ocean with southerly (northerly) alongshore wind stress anomalies off the west coast of Australia (Fig. 4a; vectors). The induced enhanced (suppressed) southerly wind stress (Fig. 4a; shading) further leads to weaker (stronger) southward advection of warm water through the Leeuwin Current, and negative (positive) SST anomalies in the WAbox. These results were found to be consistent with discussions in previous studies^{33,36,40,41,42} and could be related to ENSO atmospheric teleconnections and the local air–sea interaction called the “coastal Bjerknes feedback”^40,41.

The lagged regression coefficients of sea level pressure (blue (red) contours for positive (negative) values), wind stress (vectors), wind speed (shading in a) and downward net surface heat flux (shading in b) anomalies onto the Niño 3.4 index in the eastern Indian Ocean. The contour interval is 0.15 hPa for sea level pressure anomalies. The negative (positive) numbers in the titles indicate the months that the Niño 3.4 index lags (leads). Stippling in (a) and (b) indicate the areas where the regression coefficients of wind speed and heat flux anomaly fields are statistically insignificant at the 95% level, while only significant sea level pressure and wind stress anomalies are shown. Niño 3.4 index is a commonly-used ENSO index and derived as the averaged SST anomalies over the region (170°-120°W, 5°S-5°N). c shows the standard deviations of downward latent and sensible heat flux, shortwave and longwave radiation components of net surface heat flux anomalies, while d shows the correlation coefficients between these components and the net surface heat flux anomalies. e shows the standard deviations of the reconstructed downward latent heat flux anomalies and their components (see Eq. (3) in “Methods”), while f shows their correlation coefficients with the downward latent heat flux anomalies. Stippling in (d) and (f) indicate the areas where the correlation coefficients are statistically insignificant at the 95% level.

Full size image

For the local buoyancy forcing, the comparable WAbox SST anomalies from the PACE-EM-WAHeat and PACE-EM-WABuoy experiments indicate that its contribution mainly comes from heat flux forcing (see “Methods” and Supplementary Fig. 1a). By decomposing the ENSO-driven net surface heat flux anomalies \({Q}_{O}^{{\prime} }\) (see “Methods”), the latent heat flux anomalies \({Q}_{{LH}}^{{\prime} }\) were found to be the primary contributor off the west coast of Australia as they had largest standard deviations and highest correlation coefficients among all components (Fig. 4c, d). The decomposition of \({Q}_{{LH}}^{{\prime} }\) (see “Methods”) further indicates that off the west coast of Australia, \({Q}_{{LH}}^{{\prime} }\) is largely contributed by \({Q}_{{LH}-\triangle q}^{{\prime} }\) driven by air–sea humidity gradient anomalies, with a much smaller role from \({Q}_{{LH}-W}^{{\prime} }\) due to wind speed anomalies (Fig. 4e-f). The dominant role of \({Q}_{{LH}-\triangle q}^{{\prime} }\) demonstrates that the local ENSO-driven heat flux forcing is mainly generated by changes in SST anomalies and thus ultimately linked to wind stress forcing remotely from the tropical Pacific (via oceanic teleconnections) and locally off the west coast of Australia. During El Niño (La Niña), negative (positive) SST anomalies generated by wind stress forcing decrease (increase) the humidity gradient and evaporation, suppressing (enhancing) the heat loss from ocean to the atmosphere (Fig. 4b; shading) and in turn damping the negative (positive) SST anomalies. The smaller role of \({Q}_{{LH}-W}^{{\prime} }\) suggests that while increased (decreased) southerly wind speed anomalies can lead to increased (decreased) evaporation and thus negative (positive) SST anomalies off the coast during El Niño (La Niña), this effect is generally weak and/or offset by the effect of \({Q}_{{LH}-\triangle q}^{{\prime} }\). For the overall SST variations off the west coast of Australia, many previous studies discussed the positive feedback mechanisms related to heat flux anomalies, including the wind-evaporation-SST feedback (e.g., ref. ²¹) and cloud-radiation-SST feedback (e.g., ref. ⁵²). However, for ENSO-driven SST variations, we found that the damping effect of latent heat flux anomalies induced by coastal SST anomalies themselves played an important role. This also provides an interesting case of ocean-driven heat flux variability as indicated by ref. ⁵³.

The superposition of non-ENSO climate variability on ENSO modulations of SST off the west coast of Australia

Since in addition to ENSO, the WAbox SST anomalies were also influenced by non-ENSO climate variability (Fig. 2a), the PACE-M13 group and PACE-M14 group experiments (see “Methods” and Supplementary Table 1)—which were driven by individual ensemble member rather than the ensemble mean—were undertaken to investigate the relative importance of ENSO and non-ENSO climate variability to the WAbox SST variations. Interpreted as the response to ENSO plus non-ENSO climate variability, the WAbox SST anomalies from the PACE-M13-Full (PACE-M14-Full) experiment is nearly equal to the sum of the PACE-M13-TPWind (PACE-M14-TPWind), PACE-M13-WAWind (PACE-M14-WAWind) and PACE-M13-WABuoy (PACE-M14-WABuoy) experiments (Supplementary Fig. 2a, b). While the effects of non-ENSO climate variability from ensemble members #13 and #14 are expected to be different, we found that the WAbox SST anomalies from the PACE-M13-Full and PACE-M14-Full experiments were still highly correlated with each other (\(r=0.59\), \(p < 0.05\); Fig. 5a). The standard deviations of the SST anomalies from the two experiments (0.60°C and 0.46°C) are also comparable to those from the PACE-EM experiment (standard deviation of 0.47 °C; Fig. 5a; Table 1). Moderate correlations were found between SST anomalies from the PACE-M13-WAWind and PACE-M14-WAWind experiments (\(r=0.44\), \(p < 0.05\)), and SST anomalies from the PACE-M13-WABuoy and PACE-M14-WABuoy experiments (\(r=0.64\), \(p < 0.05\); Fig. 5c, d). These together demonstrate that while non-ENSO climate variability in the eastern Indian Ocean may contribute to the overall SST variability in the WAbox, this contribution is secondary and ENSO plays the dominant role in the region.

a–d show the time series of monthly SST anomalies averaged within the WAbox derived from the MITgcm PACE-M13 and PACE-M14 group experiments. e–h are the same as (a–d), but for the TSbox region. The black lines in the subplots indicate SST anomalies from the experiments forced with corresponding atmospheric fluxes of the CESM Pacific pacemaker experiment ensemble member #13, while the red lines indicate SST anomalies from the experiments forced by atmospheric fluxes from ensemble member #14. “Corr” in the subplots indicate correlation coefficients between time series represented by black and red lines, with (*) indicating those statistically not significant at 95% level.

Full size image

The dominant role of ENSO on SST variations off the west coast of Australia can be further understood in terms of the moderate-to-high ratios of ENSO-driven variance over total variance for the wind stresses and buoyancy fluxes in the tropical Pacific and off Australia’s west coast. The ENSO-driven variance ratios quantitively describe the relative strengths of ENSO-driven forcings compared to non-ENSO forcings (Fig. 6; see “Methods”). For zonal wind stress, most ocean areas in the tropical Pacific exhibit a large ratio fraction (close to 1), consisting with the fact of the dominance of ENSO over other sources of climate variability there (Fig. 6a). The meridional wind stress and buoyancy fluxes off the west coast of Australia due to ENSO atmospheric teleconnections were also found to have a moderate ENSO-driven variance ratio (Fig. 6b-d).

Ratio maps for: a zonal wind stress, b meridional wind stress, c heat flux and d freshwater flux (see “Methods” for definition of variance ratio).

Full size image

Notably, while previous studies have identified that the IOD can modulate WAbox SST variations^54,55, we note that in our study, the IOD variability should not be solely considered as non-ENSO climate variability. Since ENSO is known to affect the IOD^17,18, the IOD variability driven by ENSO was captured as part of ENSO variability and discussed in the above section. The remaining IOD contribution was included into the contributions from non-ENSO climate variability and discussed in this section. This also applies to the TSbox in the following sections while investigating the impacts of ENSO with other climate modes (e.g., SAM) involved.

ENSO modulation mechanisms of SST off southeast Australia

Off Australia’s southeast coast, the TSbox SST anomalies from the PACE-EM-Wind experiment were found to be positively correlated with the total ENSO-driven SST anomalies (i.e., SST response to ENSO) from the PACE-EM-Full experiment (\(r=0.57\), \(p < 0.05\); Fig. 7a; Table 1). While wind stress forcing makes a positive contribution to the SST response to ENSO in the TSbox, the contribution was found to be largely offset by buoyancy forcing. This offset is evident by the high anticorrelation between SST anomalies from the PACE-EM-Wind and PACE-EM-Buoy experiments (\(r=-0.84\), \(p < 0.05\)), as well as their comparable standard deviations (0.32 °C and 0.28 °C; Fig. 7a; Table 1). The counteracting effects of wind stress and buoyancy forcings result in a weaker SST response to ENSO in the TSbox off Australia’s southeast coast (standard deviation of 0.17 °C; Fig. 7a; Table 1). Furthermore, it is notable that in both the PACE-Wind and PACE-EM-Buoy experiments, the standard deviations of the TSbox SST anomalies are smaller than those of the WAbox SST anomalies (0.69 °C and 0.40 °C; Table 1). This indicates that the individual effects of ENSO-driven wind stress and buoyancy forcings are also weaker in the TSbox than in the WAbox.

a shows the SST anomalies derived from the PACE-EM-Full (black line), PACE-EM-Wind (blue line) and PACE-EM-Buoy (red line) experiments. b shows the SST anomalies derived from the PACE-EM-Wind (black line), PACE-EM-TPWind (blue line) and PACE-EM-SPWind (red line) experiments. c shows the SST anomalies derived from the PACE-EM-Buoy (black line) and PACE-EM-SPBuoy (red line) experiments. All monthly SST anomaly time series are calculated by averaging with the TSbox region.

Full size image

The TSbox SST anomalies from the PACE-EM-TPWind and PACE-EM-SPWind experiments were next examined to understand the contribution of ENSO-driven wind stresses from the tropical Pacific as well as contribution from the extratropical South Pacific (arising from ENSO teleconnections). High correlation was found between the TSbox SST anomalies from the PACE-EM-TPWind and PACE-EM-Wind experiments (\(r=0.93\), \(p < 0.05\); Fig. 7b; Table 1). This indicates that the effect of ENSO-driven wind stress forcing on the TSbox SST variations is significantly due to forcing from the tropical Pacific, as the WAbox SST variations. The TSbox SST anomalies from the PACE-EM-SPWind experiment were found to be less correlated with those from the PACE-EM-Wind experiment (\(r=0.33\)) and with a much smaller standard deviation (0.11 °C), suggesting a smaller effect of the extratropical South Pacific ENSO-driven wind stress forcing on regional SST variations (Fig. 7b; Table 1).

The TSbox SST anomalies from the PACE-EM-SPBuoy experiment was also examined. As evident by the high correlation and comparable standard deviations between the TSbox SST anomalies from the PACE-EM-Buoy and PACE-EM-SPBuoy experiments, the influence of the ENSO-driven heat and freshwater fluxes on regional SST variations was found to mainly result from the extratropical South Pacific (\(r=0.97\), \(p < 0.05\); standard deviations of 0.28 °C and 0.25 °C; Fig. 7c; Table 1).

The dominant positive contribution of ENSO-driven tropical Pacific wind stress forcing raises the question: What pathway connects the remote ENSO-driven wind forcing in the tropical Pacific with the TSbox SST response off Australia’s southeast coast? We investigated this question by carrying out the PACE-EM-TPWind-Sponge experiment (see “Methods” and Supplementary Table 1). In this experiment, the oceanic channel between the tropical Pacific and the Tasman Sea via the west and south coasts of Australia was mainly blocked by an artificial sponge wall (red box in Fig. 8b). This acted to negate the previously sizeable TSbox SST anomalies shown in the PACE-EM-TPWind experiment (Fig. 8a). The spatial pattern of the root-mean-square differences of the SSH anomalies between the PACE-EM-TPWind and PACE-EM-TPWind-Sponge experiments further shows that the largest SSH differences are located along the west and south coasts of Australia (Fig. 8b). This suggests the important role of coastal Kelvin waves in “delivering” the effect of the tropical Pacific ENSO-driven wind stress forcing into the southwest Tasman Sea.

a Time series of SST anomalies in the TSbox region derived from the PACE-EM-TPWind (black line) and PACE-EM-TPWind-Sponge (red line) experiments. b Root-mean-square differences of SSH anomalies between the PACE-EM-TPWind and PACE-EM-TPWind-Sponge experiments. The red box in (b) highlights the inserted sponge wall applied in the PACE-EM-TPWind-Sponge experiment.

Full size image

During El Niño (La Niña), upwelling (downwelling) signals are generated by equatorial wind stress anomalies and propagate westward as Rossby waves in the tropical Pacific. These signals later pass through the Indonesian seas and propagate along the west and south coasts of Australia as coastal waves until reaching the southeast coast of Australia. Such anticlockwise propagations of oceanic waves then lead to decreased (increased) northward advection into the southwest Tasman Sea, transporting less (more) cold waters into the TSbox. This is consistent with the fact that the equatorward-flowing Zeehan Current off southeast Tasmania is the final component of a continuous boundary current stretching back to the Leeuwin Current that is strongly influenced by ENSO^48,56, and previous empirical studies of coastal wave propagation along Australia’s coastline⁵⁷. The propagation of coastal waves may also explain why the TSbox SST anomalies are smaller than in the WAbox for the PACE-EM-Wind (and PACE-EM-TPWind) experiment. With the same wind stress forcing from the tropical Pacific, the coastal waves reach the TSbox through a longer and more complicated propagation pathway than for the WAbox. This ENSO teleconnection mechanism, together with the importance of ENSO-driven wind stress forcing in the tropical Pacific, have received little attention previously due to the large offset caused by ENSO-driven buoyancy forcing.

Compared to the tropical Pacific, ENSO-driven wind stress forcing in the extratropical South Pacific has a smaller contribution to the SST response to ENSO in the TSbox. This contribution was found to be primarily due to wind stress forcing locally in the Coral and Tasman Sea, according to the comparison of TSbox SST anomalies between the PACE-EM-SPWind and PACE-EM-CTWind experiments (see “Methods” and Supplementary Fig. 1b). Examining the lagged regressions onto the Niño 3.4 index, it was found that for El Niño (La Niña), easterly (westerly) wind stress anomalies (Fig. 9a; vectors) associated with anticyclonic (cyclonic) anomalies (Fig. 9a; contours) were generated in the northern Coral Sea, leading to increased (decreased) transports entering the northern Coral Sea and the East Australian Current off the east coast of Australia. As a poleward flowing western boundary current system, the East Australian Current and its Extension thus transport more (less) warm water from the Tropics to Australia’s southeast region, resulting in positive (negative) SST anomalies along the current and in the TSbox. However, at the same time, westerly (easterly) wind stress anomalies (Fig. 9a; vectors) are generated in the Tasman Sea, leading to increased (decreased) transports in the separating East Australian Current along the Tasman Front. The resultant positive (negative) SST anomalies in the TSbox are then partly cancelled out. Our results were supported by several previous studies, which have shown that transport variations in the East Australian Current, together with sea level variations at Australia’s east coast, could be modulated by westward propagating oceanic Rossby waves on ENSO to decadal timescales caused by wind stress forcing in the interior South Pacific^46,50,58,59.

a, b Lagged regression coefficients of sea level pressure (blue (red) contours for positive (negative) values), wind stress (vectors), wind speed (shading in a) and downward net surface heat flux (shading in b) anomalies onto the Niño 3.4 index in the Coral and Tasman Sea. The contour interval is 0.15 hPa for sea level pressure anomalies. The negative (positive) numbers in the titles indicate the months that the Niño 3.4 index lags (leads). Stippling in a and b indicate the areas where the regression coefficients of wind speed and heat flux anomaly fields are statistically insignificant at the 95% level, while only significant sea level pressure and wind stress anomalies are shown. c Standard deviations of downward latent and sensible heat flux, shortwave and longwave radiation components of net surface heat flux anomalies, while d shows the correlation coefficients between these components and the net surface heat flux anomalies. e Standard deviations of the reconstructed downward latent heat flux anomalies and their components (see equation (3) in “Methods”), while f shows their correlation coefficients with the downward latent heat flux anomalies. Stippling in d and f indicate the areas where the correlation coefficients are statistically insignificant at the 95% level.

Full size image

The contribution of ENSO-driven buoyancy forcing from the extratropical South Pacific was also found to mainly result from heat flux forcing in the Coral and Tasman Sea, as shown by the close standard deviations and high correlation coefficients of TSbox SST anomalies between the PACE-EM-SPBuoy and PACE-EM-CTHeat experiments (see “Methods” and Supplementary Fig. 1c). The decomposition of \({Q}_{O}^{{\prime} }\) (see “Methods”) shows that \({Q}_{{LH}}^{{\prime} }\) is the primary contributor over the Coral and Tasman Sea, as this component has the largest standard deviations and highest correlation coefficients among all components (Fig. 9c, d). The decomposition of \({Q}_{{LH}}^{{\prime} }\) (see “Methods”) further reveals that around the TSbox off southeast Australia, \({Q}_{{LH}}^{{\prime} }\) is largely due to \({Q}_{{LH}-\triangle q}^{{\prime} }\) driven by air-sea humidity gradient anomalies, while \({Q}_{{LH}-W}^{{\prime} }\) due to wind speed anomalies plays a smaller role (Fig. 9e, f). The dominant role of \({Q}_{{LH}-\triangle q}^{{\prime} }\) demonstrates that the local ENSO-driven heat flux forcing is mainly generated by changes in SST anomalies which are driven by wind stress forcing remotely from the tropical Pacific (via oceanic teleconnections) as well as from the Coral and Tasman Sea. During El Niño (La Niña), positive (negative) SST anomalies increase (decrease) the humidity gradient and evaporation, enhancing (suppressing) the heat loss from ocean to the atmosphere and in turn damping the positive (negative) SST anomalies (Fig. 9b; shadings). This can be considered as an interesting but less discussed case of ocean-driven heat flux variability as indicated by ref. ⁵³.

As for the smaller role of \({Q}_{{LH}-W}^{{\prime} }\) in the southern Tasman Sea including the TSbox, it suggests that during El Niño (La Niña), increased (decreased) westerly wind speeds in this region can lead to increased (decreased) evaporation and thus SST cooling (warming) off the coast. This can be linked to a recent study which shows that the local SST variations in the Tasman Sea could be modulated by SAM while SAM itself was also impacted by ENSO⁶⁰. However, this effect of \({Q}_{{LH}-W}^{{\prime} }\) was found to be generally weak and likely to be masked by the effect of \({Q}_{{LH}-\triangle q}^{{\prime} }\).

In addition, it is notable that \({Q}_{{LH}-W}^{{\prime} }\) due to wind speed anomalies primarily dominates \({Q}_{{LH}}^{{\prime} }\) in the northern Coral Sea (Fig. 9d, e). During El Niño (La Niña), while increased (decreased) easterly wind stress in the northern Coral Sea (Fig. 9a; vectors and shadings) can lead to strengthened (weakened) southward advection of warm water of the East Australian Current system and thus positive (negative) TSbox SST anomalies, the associated increased (decreased) wind speeds can also enhance (suppress) evaporation, leading to southward advection of colder (warmer) than normal water along the current system and finally into the TSbox. In other words, the effects of wind stress and heat flux forcings in the northern Coral Sea can partly cancel out each other. The considerable contribution of \({Q}_{{LH}-W}^{{\prime} }\) to \({Q}_{{LH}}^{{\prime} }\) in the northern Tasman Sea is expected to be in the similar situation.

The superposition of non-ENSO climate variability on ENSO modulations of SST off southeast Australia

The TSbox SST anomalies are shown to be modulated by both ENSO and non-ENSO climate variability (Fig. 2b). Similar to the WAbox, the relative importance of ENSO and non-ENSO climate variability on the SST field in the TSbox was further examined by comparison between the PACE-M13 group and PACE-M14 group experiments. Driven by ENSO and non-ENSO forcings simultaneously, the TSbox SST anomalies from the PACE-M13-Full and PACE-M14-Full experiments are poorly correlated (\(r=0.33\), \(p < 0.05\)) and have much larger standard deviations (0.33 °C and 0.34 °C) than those of the PACE-EM-Full experiment (standard deviation of 0.17 °C; Fig. 5e; Table 1). These results indicate that non-ENSO climate variability plays an important role in modulating the TSbox SST variations.

The important role of non-ENSO climate variability on the TSbox SST variations was found to mainly result from forcing in the extratropical South Pacific. The TSbox SST anomalies from PACE-M13-Full is nearly equal to the sum of the PACE-M13-TPWind, PACE-M13-SPWind and PACE-M13-SPBuoy experiments, and the same finding also applies to the PACE-M14 group experiments (Supplementary Fig. 2c, d). In contrast to the high correspondence between the PACE-M13-TPWind and PACE-M14-TPWind experiments (\(r=0.93\), \(p < 0.05\); Fig. 5f), low correlations were found between the TSbox SST anomalies from the PACE-M13-SPWind and PACE-M14-SPWind experiments, and between the TSbox SST anomalies from the PACE-M13-SPBuoy and PACE-M14-SPBuoy experiments (\(r=-\mathrm{0.12,0.20}\) respectively; Fig. 5g, h). These low correlations can be explained by the low ENSO-driven variance ratio for the wind stress and buoyancy fluxes over the South Pacific Ocean. For wind stresses, heat and freshwater flux, the extratropical South Pacific including the Coral and Tasman Sea exhibits a small variance ratio (<0.3 for most ocean areas), indicating the dominance of non-ENSO climate variability over the region (Fig. 6). The weak SST response to ENSO in the region is therefore further weakened. While the Tasman Sea is known as an eddy-active region, the results suggest that in addition to the effect from chaotic mesoscale eddies, the SST response to ENSO in the region is also weakened by forced deterministic variability beyond ENSO.

For some ocean regions, it can be difficult to cleanly separate out the contributions of different mechanisms from each other on the oceanic responses. Specifically, separating the contribution to regional sea surface temperature variations from different sources and modes of climate variability using empirical methods is challenging based on relatively short available observational records. To address this, we have set up and applied a novel modelling approach—which combines a large ensemble of tropical Pacific pacemaker simulations and regional OGCM perturbation experiments—to explore ENSO modulations of these regions and identify underlying processes. Here, we used this modelling approach to identify the contribution and mechanisms of ENSO to SST variations in two regions (the WAbox and TSbox) on opposite sides of Australia.

We firstly examined the individual contributions of ENSO-driven atmospheric forcings from different regions to the ENSO responses in the WAbox and TSbox, thus inferring and identifying the ENSO teleconnection mechanisms on regional SST anomalies. Our results show that off Australia’s west coast, the SST response to ENSO in the WAbox is dominantly driven by remote tropical Pacific ENSO-driven wind stress forcing via oceanic teleconnections, and moderately enhanced by ENSO-driven wind stress forcing locally off the west coast of Australia. The wind stress forcings positively contributed to SST anomalies by modulating the variations of southward advection of warm water along the Leeuwin Current. The local ENSO-driven latent heat flux forcing is then induced by local SST anomalies driven by wind stress forcings, acting to damp the wind stress-induced SST anomalies.

Compared to the mechanisms underpinning the strong ENSO response in the WAbox SST along Australia’s west coast, the mechanisms behind the weak ENSO signature in SST in the TSbox off southeast Australia are much less understood. Here, we found that tropical Pacific ENSO-driven wind stress forcing contributes to the TSbox SST via coastal wave propagation anticlockwise around Australia. Compared to the WAbox, the coastal waves “deliver” the effect of tropical Pacific wind stress forcing to the TSbox through a longer and more complicated propagation pathway. This results in smaller contribution of tropical Pacific wind stress forcing to the TSbox than the WAbox. ENSO-driven wind stress forcing in the Coral and Tasman Sea plays a secondary role on the TSbox ENSO response by modulating the variations of southward advection of warm water along the East Australian Current system. We found that there is a large offset in the TSbox SST response due to ENSO-driven buoyancy forcing in the Coral and Tasman Sea. Such buoyancy forcing can be generated by local SST anomalies responding to ENSO-driven wind stress forcings, and act against the contributions arising from the wind stress forcings, damping the wind stress-induced SST anomalies. This offset leads to the weak ENSO response in the TSbox.

With the modelling approach, we also quantitively identified and compared the relative importances of non-ENSO climate variability and ENSO on SST variations in the WAbox and TSbox. Our results show that SST variations in the WAbox off Australia’s west coast are dominated by the contribution from ENSO relative to non-ENSO climate variability. In contrast, SST variations in the TSbox off Australia’s southeast coast are strongly modulated by non-ENSO climate variability from the South Pacific, which further weakens the weak SST response to ENSO in the region.

By undertaking this contrast study, we have gained an improved understanding of why the ENSO modulation is weak off Australia’s southeast coast (Fig. 10). Specifically, we found that the much weaker east coast SST response to ENSO compared with Australia’s west coast is due to: (1) the smaller positive contribution from ENSO-driven wind stress forcing as well as the larger offset among different ENSO-driven forcings which lead to a relatively small net effect of ENSO on Australia’s southeast region SST variability; and (2) the effect of ENSO being further weakened by the significant contribution from non-ENSO climate variability.

Contributions from different ENSO wind stress and buoyancy forcings, together with the superposition effects from non-ENSO variability on the regional SST variations of the a WAbox and b TSbox. The reservoir boxes represent the corresponding SST anomalies in the WAbox and TSbox. The arrows pointing towards (away from) the reservoir boxes together with plus (minus) signs indicate positive (negative) contributions of the forcings to the SST anomalies, with the numbers representing specific contribution values as listed in Table 1. Please note that the relative sizes of the boxes are approximate and not drawn to scale.

Full size image

There are several limitations in this study. First, it should be noted that our study focused on the regional SST variations and identified the ENSO mechanisms by investigating the responses of SST to their direct atmospheric drivers: wind stress and buoyancy flux forcings. Using ocean model experiments and associated analyses, our study did not completely eliminate the role of, for example, changes in SST driving atmospheric wind changes and thus SST-induced wind changes driving further SST changes through wind stress and heat flux forcings. It would be worthwhile for future studies to conduct a more in-depth investigation into the feedback mechanisms involved in air-sea interactions in the regions, e.g., via coupled ocean-atmosphere model setup.

Second, while the study focused on the atmosphere and ocean surface, mixed layer temperature evolution beneath the surface may also provide additional information of different ENSO modulation mechanisms and their contributions to the regional SST variations. Future research including a mixed layer heat budget analysis may be beneficial here. Third, small-scale chaotic intrinsic ocean variability influences were not examined. For example, with poleward transports in the East Australian Current Extension contributed by the procession of warm-core mesoscale eddies, temperature changes off southeast Australia are strongly influenced by this shorter time variability^61,62,63. It would be worthwhile in future analyses of SST variations in Australia’s west and southeast regions to explore the influence of mesoscale activity (e.g., mesoscale thermal feedback⁶⁴) on the larger scale ENSO modulations, with a high-resolution regional ocean model set up properly representing mesoscale eddies. In addition, despite good ENSO performance supported by evaluation studies (e.g., refs. ^24,65,66,67), the CESM is a climate model and known to exist model bias—which can inevitably affect the estimated ENSO responses in the pacemaker experiments. A valuable future study would be to extend the current study to use a set of tropical Pacific pacemaker ensemble simulations performed by multiple climate models.

The novel modelling approach applied in our study—which combines an ENSO pacemaker experiment with regional MITgcm experiments—has proven to be an effective tool to explore the important contributions and dynamical mechanisms that explain how ENSO modulates regional SST variations. Many previous empirical studies have discussed estimates of global or regional responses to ENSO and explored the associated teleconnections mechanisms by applying statistical methods including composite and regression analyses to a particular period of observation record (e.g., refs. ^{68,69,70,71,72}). This usually requires the period to be sufficiently long to remove the noise from climate variability beyond ENSO, especially for those regions with weak ENSO responses like the southeast coasts of Australia. However, currently the reliable observational records are typically less than 50 years. Within our modelling approach, the CESM Pacific pacemaker experiment with multiple ensemble members efficiently minimizes the noise due to non-ENSO climate variability and provides a robust dynamically oriented estimate of global responses to ENSO. For a specific region of interest, the designed MITgcm perturbation experiments further enables us to quantify and identify the contributions of different ENSO forcings from different regions together with their related underpinning mechanisms. The relative importance of ENSO and non-ENSO climate variability for the region is also able to be identified.

In the face of global warming, MHWs have been increasing in intensity, frequency and duration⁷², and are expected to continue in the future⁷³. Both the west and southeast coasts of Australia studied here have been significantly impacted by substantive MHWs in recent years^29,30,31,32. While many previous studies focused on the overall SST variations in these regions (e.g., refs. ^42,50,60,74), we have teased out the specific ENSO contributions and identified the associated underpinning teleconnection mechanisms with the use of the developed modelling approach. With well-developed ENSO prediction systems^{8,9,10,11,12,13,14,15}, a good understanding of ENSO modulation mechanisms of regional SST variations can benefit MHW predictions in the future. With ENSO significantly contributing to MHWs globally^{11,15,34,75,76}, we expect this flexible and informative modelling approach will be useful to examine ENSO modulations on SSTs, including ocean temperature extremes, in other regions around the world. This can benefit marine ecosystem management, as well as fisheries and aquaculture industries, with improved knowledge of the potential predictability of MHWs. We expect that by using different pacemaker ensembles, this modelling approach will also be useful to explore the role of other climate modes and their teleconnection mechanisms underpinning SST variations and MHW predictability in regions of interest around the world.

The Community Earth System Model1 Pacific Pacemaker Ensemble Experiment

In this study, we used the CESM1 Pacific Pacemaker Ensemble Experiment (hereafter, CESM Pacific pacemaker experiment)²⁴ to isolate and examine the influence of ENSO in specific regions outside of the tropical Pacific. This dataset was selected due to its good performance in simulating ENSO teleconnections, which can be supported by evaluations of atmospheric circulation and SST responses to ENSO in the previous studies^24,65,67,77.

The CESM Pacific pacemaker experiment is a 20-member ensemble based on the fully coupled Community Earth System Model (CESM) version 1.1⁷⁸. In this ensemble, SST anomalies in the central and eastern tropical Pacific (10°S-10°N, 160°-90°W, with a linearly tapering buffer zone extending to 20°S and 20°N, 180°W to the American coast), were nudged towards the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstruction SST version 3b data (ERSST v3b; ref. ⁷⁹) over the period from 1920 to 2013²⁴. Consequently, the observed time-varying SST variability was maintained in the central and eastern Pacific. For each ensemble member, the remainder of the model was fully coupled and free to evolve after an initial atmospheric temperature perturbation applied on the first day of the simulation²⁴. In this way, the ensemble mean of the CESM Pacific pacemaker experiment provides a dynamically based estimate of the climate system’s response to time-varying tropical Pacific SSTs (e.g., refs. ^24,27). Tropical Pacific SST variability is largely explained by the first two empirical orthogonal functions which together describe the nonlinear evolution of ENSO (e.g., ref. ⁸⁰; Supplementary Fig. 3). Hence, the ensemble mean can be considered as a dynamically based estimate of the global climate system’s response to ENSO on interannual timescales. Each ensemble member can be interpreted as the response to ENSO plus non-ENSO climate variability outside of the tropical Pacific.

Monthly anomalies derived from the ensemble mean and ensemble members of the CESM Pacific pacemaker experiment over the period from 1994 to 2013 were used in the study. The ensemble-mean SST anomalies were used to identify the SST responses to ENSO in two regional case study boxes—(a) the WAbox off Australia’s west coast (bounded by (110°-116°E, 22°-32°S); Fig. 1b) and (b) the TSbox off Australia’s southeast coast (bounded by (147°-156°E, 37°-46°S); Fig. 1b). The wind stress, heat flux and freshwater flux anomalies from the ensemble mean and individual ensemble members were used to force a suite of MITgcm perturbation experiments. The ensemble-mean sea level pressure, shortwave solar radiation, longwave radiation, latent heat flux, sensible heat flux, 2-m air temperature and wind speed anomalies were also derived and used for lagged regressions and/or decompositions of net surface heat flux as well as latent heat flux anomalies in the study.

In this study, we focused on the mechanisms generating interannual (specifically here, ENSO) SST variability over the 20-year period from 1994 to 2013. All monthly anomalies analysed were calculated by subtracting the monthly climatology (calculated over the baseline period from 1994 to 2013) from the corresponding month of each year and then linearly detrending. A 5-month running mean filter was further used to smooth the monthly anomalies prior to analysing the interannual variability, which is a common practice for ENSO-related studies on interannual time scale (e.g., ref. ⁸¹). Here, the results were found to be insensitive to other filter choices (e.g., 3-month running mean filter).

Model setup, validation, and experiment design

The OGCM used in this study was the MIT general circulation model (MITgcm) developed at the Massachusetts Institute of Technology²⁸. The configured regional MITgcm ___domain covers the tropical and South Pacific basin together with the Indian Ocean (36°E-68°W, 54°S-26°N; Fig. 1a), with 1/3° resolution in both the zonal and meridional directions. There are 51 vertical layers, with layer thicknesses gradually increasing from 10 m near the surface to 500 m in the deep ocean. The model configuration is mainly based on previous work from refs. ^26,82 with several minor modifications. Notably, both Laplacian and biharmonic schemes were used for the horizontal diffusivity (horizontal viscosity), with coefficients of ~10³ m²s⁻¹ (~10³ m²s⁻¹) and ~10¹¹ m⁴s⁻¹ (~10¹¹ m⁴s⁻¹). In the vertical direction, Laplacian diffusivity and viscosity were set up with coefficients of ~10⁻⁵ m²s⁻¹ and ~10⁻⁴ m²s⁻¹. These diffusivity and viscosity parameters were kept the same for all perturbation experiments. Relatively large horizontal viscosities and diffusivities were chosen to suppress the chaotic intrinsic variability and focus on the forced deterministic ocean variability. The model bathymetry was derived from the Earth topography dataset ETOPO5, a 5-min gridded world bathymetry product. The flux form of the atmospheric surface forcing (i.e., wind stresses, heat flux and freshwater flux) was used for the study.

For model evaluation, a Hindcast experiment was designed and performed (Supplementary Table 1). This experiment was forced with daily atmospheric forcing fields from the European Centre for Medium-Range Weather Forecasts (ECMWF) ReAnalysis—Interim (ERA-Interim)⁸³. Open boundary conditions (temperature, salinity, horizontal velocities) were derived from the Estimating the Circulation and Climate of the Ocean Version 4 Release 4 (ECCO v4r4)⁸⁴. The Hindcast experiment was commenced after a 40-year spin-up integration. The spin-up started from the ocean state of ECCO v4r4 on the first day of 1993 driven by ECMWF ERA-Interim atmospheric fluxes as well as ECCO v4r4 ocean boundary conditions from the year 1993. This idea of applying a repeating normal-year type forcing over a long-enough period to stabilize the ocean has been commonly used in many previous studies^51,85,86. For our study, the year 1993 was carefully selected as the “normal year”, because both IOD and ENSO phases are close to near-neutral. Once the model reached quasi-equilibrium after this 40-year spin-up, the Hindcast experiment ran from January 1994 to December 2016. Relaxations of SST towards the NOAA Optimum Interpolation (OI) V2 data⁸⁷ and sea surface salinity (SSS) towards climatology of Aquarius Level 3 SSS dataset⁸⁸ were applied with timescales of 30 days in both the spin-up and Hindcast experiments.

The current MITgcm configuration was mainly based on previous modelling works^26,82, in which model evaluation against observations were already done. We further evaluated the performance of the Hindcast experiment by comparing with the latest ocean observations and reanalyses. One evaluation is the comparison of the simulated SSH with observations and the ECCO v4r4 output. As an ocean variable integrating information over the whole water column⁸⁹, SSH is an effective parameter for examining model performance. We compared the simulated SSH anomaly standard deviation and long-term means against the ECCO v4r4 for the 1994–2016 period (Fig. 1a, b). In general, the magnitude and spatial patterns of SSH simulated by the Hindcast experiment compares well with those from the ECCO v4r4. The time series of interannual SSH anomalies averaged over the WAbox and TSbox were also found to exhibit high correspondence with those from the ECCO v4r4 and observations from the Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) dataset (Fig. 1c, d). A similar evaluation was also performed for SST and good agreements are found while comparing the simulated SST with those from the ECCO v4r4 and NOAA OI SST V2 datasets (Supplementary Fig. 4). The high correspondence of SSH and SST anomalies indicated that the model was suitable for our analysis.

With the aim of understanding ENSO modulations of SST variations in the WAbox (off Australia’s west coast) and TSbox (off Australia’s southeast coast) regions, we carried out a set of MITgcm perturbation experiments forced with atmospheric fluxes derived from the CESM Pacific pacemaker experiment (see Supplementary Table 1). All the MITgcm experiments started from the same initial condition as the Hindcast experiment. The Control experiment was repeatedly forced by ECMWF ERA-Interim atmospheric fluxes and ECCO v4r4 ocean boundary conditions of 1993, while SST and SSS relaxations were applied and archived. Since the model had been spun up for 40 years to reach quasi-equilibrium, this Control experiment did not show any significant interannual variations while retaining seasonal cycles with repeated atmospheric and boundary forcing. Similar idea is also used in previous studies (e.g., ref. ⁵¹).

Perturbation experiments were next performed. For all perturbation experiments, SST and SSS relaxations were prescribed from the archived values during the Control experiment, rather than being calculated on the fly (e.g., ref. ⁸¹). This non-adaptive relaxation is an efficient way to keep model state not drifting while allowing the model to respond freely to prescribed flux forcing. For the perturbation experiments with ENSO-driven atmospheric forcings (see Supplementary Table 1), the PACE-EM-Full experiment was first performed to provide model estimates of ENSO and its SST response across the entire model ___domain. In this experiment, ENSO-driven wind stress and buoyancy flux (i.e., heat and freshwater fluxes) anomalies from 1994 to 2013—which were derived from the ensemble mean of the CESM Pacific pacemaker experiment—were imposed on the repeat-year forcing of 1993 (1993-RYF) applied in the Control experiment. To separate the respective contribution of wind stress and buoyancy forcing, in the PACE-EM-Wind experiment, ENSO-driven wind stress anomalies over the period from 1994 to 2013 were imposed on the 1993-RYF, while the buoyancy fluxes were the same as the 1993-RYF. Conversely, in the PACE-EM-Buoy experiment, ENSO-driven buoyancy flux anomalies over the period from 1994 to 2013 were imposed, while wind stresses were the same as the 1993-RYF.

For the SST response to ENSO in the WAbox off Australia’s west coast, the PACE-EM-TPWind and PACE-EM-WAWind experiments were performed to further isolate the contributions of ENSO-driven wind stress forcing from the different regions. The PACE-EM-TPWind experiment was the same as the PACE-EM-Wind experiment, except that ENSO-driven wind stress anomalies were only imposed over the tropical Pacific (120°E-70°W, 13.5°S-13.5°N) while the 1993-RYF of wind stress was applied elsewhere. This experiment aims to investigate the role of ENSO-driven wind stress forcing in the tropical Pacific remotely affecting the WAbox region. Similarly, in the PACE-EM-WAWind experiment, ENSO-driven wind stress anomalies were applied locally in the eastern Indian Ocean (105°-125°E, 35°-15°S) with the aim of investigating the influence of local wind stress forcing arising from ENSO teleconnections. The PACE-EM-WABuoy experiments—in which ENSO-driven heat and freshwater flux anomalies were imposed in the eastern Indian Ocean (105°-125°E, 35°-15°S)—were performed to isolate the contribution of local buoyancy forcing to the ENSO response off Australia’s west coast. The PACE-EM-WAHeat experiments—in which ENSO-driven heat flux anomalies were imposed in the eastern Indian Ocean (105°-125°E, 35°-15°S)—were further performed to isolate the contribution of local heat flux forcing.

For the SST response to ENSO in the TSbox off Australia’s southeast coast, the PACE-EM-SPWind experiment was preformed, in which ENSO-driven wind stress anomalies were imposed over the South Pacific region (140°E-70°W, 50°-13.5°S). The PACE-EM-SPBuoy experiment was also performed by imposing ENSO-driven heat and freshwater flux anomalies over the same South Pacific region. These two experiments aimed to isolate the individual contributions of ENSO-driven wind stress and buoyancy forcings in the South Pacific extratropics (poleward of 13.5°S). The PACE-EM-CTWind and PACE-EM-CTHeat experiments were further performed. They were the same as the PACE-EM-SPWind and PACE-EM-SPBuoy experiments respectively, except that ENSO-driven wind stress and heat flux anomalies were only imposed over the Coral and Tasman Sea (148°E-180°, 50°-13.5°S) to examine the effects of local forcings arising from ENSO teleconnections. Based on the PACE-EM-TPWind experiment, the PACE-EM-TPWind-Sponge experiment was additionally performed. This experiment closely followed the PACE-EM-TPWind experiment but included the addition of a sponge wall off southwest Australia (36°-120°E, 54°-20°S; red box in Fig. 8b). Within the sponge wall, temperature, salinity, and horizontal velocities were restored towards the Control run climatology with timescales of one model timestep. In other words, oceanic exchanges between the tropical and extratropical Pacific though the Indian Ocean were virtually blocked. This experiment aimed to identify the teleconnection pathway(s) of ENSO signatures into the southwest Tasman Sea.

In addition to the experiments forced with ENSO-driven atmospheric fluxes, the experiments forced with atmospheric fluxes from the individual ensemble members were also performed to examine the contributions of non-ENSO forcings to SST variations off Australia’s west and southeast coasts (Supplementary Table 1). Showing the highest correspondence with WAbox and TSbox SST anomalies from the ERSST v3b, ensemble members #13 (\(r=0.67\), \(p < 0.05\) for WAbox region) and #14 (\(r=0.47\), \(p < 0.05\) for TSbox region) were selected as representatives of all ensemble members and used for the ensemble-member experiments. The PACE-M13 (PACEmaker-Member#13) group and PACE-M14 group experiments were essentially the same as the PACE-EM group experiments discussed above, except that the monthly anomalies of wind stress and buoyancy forcing derived from the ensemble mean were replaced by those derived from the ensemble members #13 and #14 respectively.

Decomposition of ENSO-driven heat flux

To identify and compare the roles of different ENSO atmospheric teleconnection mechanisms on the WAbox and TSbox, we examined different components of ENSO-driven heat flux which can link to different mechanisms.

Ocean surface heat flux \({Q}_{O}\) consists of shortwave solar radiation \({Q}_{{SW}}\), longwave radiation \({Q}_{{LW}}\), latent heat flux \({Q}_{{LH}}\) and sensible heat flux \({Q}_{{SH}}\), with all defined positive downward,

In the bulk formula, downward latent heat flux can be further determined by the air-sea specific humidity gradient \(\triangle q\) and surface wind speed \(W\) (e.g., refs. ^90,91):

where \({L}_{v}\) is the latent heat of vaporization (\(2.5\times {10}^{6}{\rm{J}}{{\rm{kg}}}^{-1}\)), \({C}_{e}\) is an aerodynamic transfer coefficient (\({10}^{3}\)), \({\rho }_{a}\) is the air density (\(1.2\times {10}^{3}{\rm{kg}}{{\rm{m}}}^{-3}\)), \({q}_{{sat}}\left({SST}\right)\) and \({q}_{{sat}}\left({T}_{2m}\right)\) are the saturated specific humidity of SST and air temperature at 2 m, and \({RH}\) is the relative humidity. Latent heat flux can be decomposed as:

where overbars indicate climatological means and primes indicate anomalies. \({Q}_{{LH}-\triangle q}^{{\prime} }=-({L}_{v}{C}_{e}{\rho}_{a}{\left(\Delta {q}\right)}^{\prime}\overline{W})^{\prime}\) represents the anomalies driven by air-sea humidity gradient anomalies. Positive SST anomalies can increase humidity gradient and evaporation, in turn damping the SST anomalies. \({Q}_{{LH}-W}^{{\prime} }=-({L}_{v}{C}_{e}{\rho}_{a}\overline\Delta {q}{W}^{\prime})^{\prime}\) represents the anomalies driven by wind speed anomalies. Increased wind speed anomalies can increase evaporation, resulting negative SST anomalies.

In this study, we firstly decomposed the CESM Pacific pacemaker ensemble-mean monthly heat flux anomalies over the eastern Indian Ocean as well as the Coral and Tasman Sea following \((1)\). After identifying the primary roles of latent heat flux anomalies in both regions, we further decomposed the ensemble-mean monthly latent flux anomalies following \((3)\). By examining and comparing the contributions of different heat flux anomaly components, we identified the contributions of different ENSO atmospheric mechanisms to WAbox and TSbox.

ENSO-driven variance ratio

In this study, we used a variance analysis method to quantify the ratio of the variance explained by ENSO to the total variance internally generated within the global climate system. Originally developed by ref. ⁹², this variance analysis method is often used to assess climate model ensemble runs forced by identical external forcing (e.g., Coupled Model Intercomparison Project Phase 6 (CMIP6) models), to separate and quantify the variance forced by external radiative fluxes and internally generated climate variabilities within the global climate system (e.g., refs. ^93,94). Here, we applied this analysis method to the ensemble members of the CESM Pacific pacemaker experiment, aiming to distinguish and quantify the variances explained by ENSO and non-ENSO climate variability. In the CESM Pacific pacemaker experiment, ENSO was considered as the “common” forcing in the form of SST nudging in the central and eastern tropical Pacific (rather than as internal climate variability), analogous to the “common” external radiative forcing applied in CMIP6 large ensemble experiments.

In each ensemble member of the CESM Pacific pacemaker experiment, at each grid point, the total internally generated variance \(v\) of the variable \(X\) can be interpreted as the sum of the ENSO-driven variance \({v}_{E}\) explained by ENSO and the non-ENSO variance \({v}_{{NE}}\) explained by climate variability beyond ENSO:

For a model with a relatively large number of ensemble members, the ENSO and non-ENSO variances (\({v}_{E}\) and \({v}_{{NE}}\)) can be estimated from the ensemble average and deviations from the ensemble average, respectively. Adapting from previous work⁹⁴, the variances can be expressed as

where \(M\) is the time length of the estimations and \(N\) is the specific number of ensemble members (i.e., 20 in this study). The ratio \(R\) of the ENSO-driven variance to total internal variance can be estimated using the unbiased variances and expressed as

By applying this variance analysis method to the wind stress and buoyancy flux anomalies of each CESM Pacific pacemaker experiment ensemble member, we separated and quantified the variances explained by ENSO and non-ENSO climate variability at each grid cell over the 1994–2013 period. We further used the derived variance ratios to investigate the relative importance of ENSO-driven and non-ENSO atmospheric forcings in the specific regions.