1. Introduction
Arctic sea ice is important for shipping, oil and gas, fisheries, wildlife, and indigenous communities. For each of these stakeholders, predicting sea ice concentration (SIC) and thickness (SIT) can be beneficial. Recent declines in Arctic SIC and SIT can be attributed to two major categories of causes. The first of these is natural and anthropogenic forcings, such as emissions of greenhouse gases (Kay et al. 2011; Notz and Stroeve 2016; Polvani et al. 2020; Zhang 2010); the second is internal variability in the climate system (Swart 2017; Ding et al. 2019; England et al. 2019). If the effects of internal variability in the climate system on sea ice can be constrained and separated from those resulting from anthropogenic forcing, one can develop a clearer understanding of the anthropogenically forced impacts on sea ice and future trends.
El Niño–Southern Oscillation (ENSO) and its teleconnections form the leading mode of interannual variability in the global climate system (Deser et al. 2010). ENSO cycles between two phases: El Niño, a phase in which sea surface temperatures (SSTs) in the eastern and central Pacific are anomalously warm and La Niña in which they are anomalously cold. The impacts of ENSO extend far beyond the tropical Pacific through a range of mechanisms. Rossby waves are triggered by upper-level divergence generated by deep convection in the tropics (Sardeshmukh and Hoskins 1988; Hoskins and Karoly 1981). Atmospheric deep convection occurs when SSTs exceed roughly 27.5°C and is therefore sensitive to El Niño events where SSTs rise in the equatorial central and eastern Pacific (Yeh et al. 2018). Shifts in the Walker circulation, jet stream changes, anomalous atmospheric circulation and heat fluxes and altered transient eddy activity also provide mechanisms by which ENSO may affect higher latitudes (Yuan et al. 2018). ENSO has been linked to several patterns of Northern Hemisphere atmospheric variability that may influence Arctic sea ice. These connections are typically strongest in boreal winter, when ENSO amplitude peaks, as does sea level pressure (SLP) variability in the Northern Hemisphere.
Anomalously high SLP over the Arctic basin is a key feature of the negative phase of the Arctic Oscillation (AO; otherwise known as the northern annular mode). A negative winter AO response to ENSO (more negative during El Niño events than La Niña) has been found in both observations and models (Li et al. 2014; Zhu and Wang 2016; Domeisen et al. 2019), however, the robustness of this response is questionable (Deser et al. 2017). Observational studies have shown that a positive AO drives anomalous cyclonic sea ice motion, an increased export of thick ice through the Fram Strait and thinning of sea ice in the central Arctic (Rigor et al. 2002; Rigor and Wallace 2004). However, the response is not spatially homogeneous, as ice is thickened north of Greenland and the Canadian Archipelago but thinned north of Russia (Lindsay and Zhang 2005; Park et al. 2018). While the direct impacts of the AO on sea ice occur in winter, these changes to the sea ice volume may persist or possibly initiate positive feedbacks, resulting in reduced sea ice extent in the following summer (Rigor et al. 2002; Lindsay and Zhang 2005; Williams et al. 2016; Park et al. 2018). Some studies suggest that the reduced sea ice associated with the positive AO may not be robust depending on the time period sampled (Stroeve et al. 2011; Day et al. 2012), particularly as its pattern is not consistently reproduced in CMIP3 (Day et al. 2012) or CMIP5 (Cai et al. 2020) models.
A robust link between ENSO and negative SLP anomalies in the North Pacific has been found in studies using both observations (Bjerknes 1969; Horel and Wallace 1981; Trenberth et al. 1998) and models (Hurwitz et al. 2014; Deser et al. 2017). The result is a deepened Aleutian low (AL), often considered to be part of the positive Pacific–North American pattern. In both models and observations, a deep Aleutian low has been shown to drive warm, moist air into the Arctic, particularly in the Pacific Ocean sector (Trenberth and Hurrell 1994; Tokinaga et al. 2017; Svendsen et al. 2018) and result in sea ice loss primarily in the North Pacific but extending into the central Arctic Ocean (Screen and Deser 2019; England et al. 2020).
While there is extensive literature on the links between ENSO and atmospheric modes of variability, and to a degree on the links between these modes of variability and Arctic sea ice, only a handful of studies have attempted to connect these concepts to identify and explain the Arctic sea ice response to ENSO. An observational study by Liu et al. (2004) regressed sea ice concentration data on an ENSO index from 1978 to 2002. They found a spatially heterogeneous response, however, did not address their small sample size of ENSO events and how this might impact the significance of their results. Arctic cooling and reduced sea ice melting was linked with central Pacific El Niños, while the opposite has been linked with eastern Pacific El Niños by Hu et al. (2016), based on results from observations and a single global climate model. The short time period analyzed (1979–2013) and subsampling of El Niño events by type again results in a small sample size of events, limiting confidence in these conclusions. On longer time scales, observations and models indicate that the rate of sea ice decline is linked to El Niño–like trends in tropical Pacific SSTs (Ding et al. 2019).
Observational studies on ENSO’s effect on the Arctic suffer from the small sample size of El Niños/La Niñas during reliable Arctic climate records. The signal of ENSO in Arctic SLP, temperature and SIC is small relative to the background internal variability; therefore, a large sample size is needed to test for significance (Deser et al. 2017). Both observational (O’Reilly 2018) and model-based (Bonan and Blanchard-Wrigglesworth 2020) evidence suggests that the tropical Pacific to Arctic teleconnection may be nonstationarity, as it has been shown to interact with other phenomena (Domeisen et al. 2019). As a result, many observed relationships between ENSO and the Arctic are not statistically significant, and those that are may only be capturing a snapshot of nonstationary behavior, both of which may lead to either misleading or contradictory conclusions.
General circulation models (GCMs) provide one means of supplementing observations in studying the teleconnections of ENSO with the Arctic; however, their fidelity in simulating these processes must be considered. Specifically, a discrepancy between modeled and observed links between tropical Pacific SSTs and Arctic sea ice has been noted a number of studies (Baxter et al. 2019; Bonan and Blanchard-Wrigglesworth 2020; Ding et al. 2019; Li et al. 2019). Two possible interpretations of such discrepancies are available, each of which is true to some degree. The first is that the representation of tropical–Arctic teleconnections and/or sea ice physics is inaccurate in GCMs. The second, as discussed in the previous paragraph, is that the small sample size in observations prevents identification of a robust sea ice response.
Model large ensembles and multimodel studies may be used to mitigate some of the issues of small sample sizes associated with observations and of model fidelity associated with single model studies. In this study we use both observations and six models from the Multi-Model Large Ensemble Archive (MMLEA; Deser et al. 2020). Details on these datasets and our methods are provided in section 2 of this paper. In section 3 we investigate the Arctic sea ice patterns associated with ENSO and to reveal the inability of observational studies alone to confidently constrain these patterns. In section 4 we reveal the atmospheric response to ENSO and explain how this drives the sea ice response seen in models. We proceed to identify a model-dependent mechanism by which ENSO may have a delayed influence on summer Arctic sea ice. In section 5 we highlight the importance of the AO and AL as intermediaries between ENSO and Arctic sea ice. Our summary and conclusions are provided in section 6.
2. Data and methods
a. Datasets
From the MMLEA we analyzed monthly data from the Community Earth System Model, version 1, Large Ensemble (CESM1-LE; 40 ensemble members; Kay et al. 2015); the Second Generation Canadian Earth System Model (CanESM2; 50 ensemble members; Kirchmeier-Young et al. 2017); the Geophysical Fluid Dynamics Laboratory Earth System Model, version 2 (GFDL-ESM2M; 30 ensemble members; Dunne et al. 2012; Rodgers et al. 2015), and Coupled Model, version 3 (GFDL CM3; 20 ensemble members; Sun et al. 2018); the Max Planck Institute Grand Ensemble (MPI-GE; first 40 ensemble members; Maher et al. 2019); and the Commonwealth Scientific and Industrial Research Organisation Mark 3.6 global climate model (CSIRO Mk3.6; 30 ensemble members; Jeffrey et al. 2013). We do not use the EC-Earth Consortium EC-EARTH ensemble because sea ice data were not available from the MMLEA. Each large ensemble is forced with greenhouse gases and aerosol emissions from the CMIP5. Historical forcing is used up to and including 2005, with RCP8.5 forcing used for years 2006–19. Further details about these models are given in Table 1, and an evaluation of the representation of Arctic sea ice in these models is provided in the online supplemental material. Results from 1950 to 2019 (HIST70) are used to analyze atmospheric variables in the models, whereas 1979–2019 (HIST41) is used to analyze sea ice variables, because of the shorter duration of sea ice observations for comparison. Our analysis is typically split by season, with winter defined as December–February (DJF), spring as March–May (MAM), summer as June–August (JJA), and autumn as September–November (SON).
Details of the six MMLEA ensembles analyzed. Additional model expansions can be found at https://www.ametsoc.org/PubsAcronymList.
We performed more comprehensive analysis using CESM1-LE, because this model has a more sophisticated representation of sea ice thermodynamics and an expanded set of sea ice variables available relative to the other models. Examples of such variables are a decomposition of the rates of change of sea ice concentration and thickness into a “thermodynamic tendency” that represents ice melting and freezing rates, and a “dynamic tendency” that represents ice transport and deformation rates. A detailed evaluation of the simulation of Arctic sea ice in CESM1 is provided in Jahn et al. (2012).
We defined pan-Arctic sea ice area as the sum of gridcell areas in the Northern Hemisphere weighted by each grid cell’s SIC. Pan-Arctic sea ice volume is given by the same sum but weighted by each grid cell’s mean SIT (which includes zero thickness over the open water fraction in the mean) instead of by SIC. For the GFDL models, SIT is given for only ice-covered areas of each grid cell, so we convert this to gridcell mean SIT by multiplying it with each grid cell’s SIC.
We used observations of Arctic SIC derived from satellite passive microwave data from the NASA bootstrap algorithm (Comiso 1986) for January 1979–June 1987 and from the NOAA/NSIDC Climate Data Record from July 1987 to December 2019 (Peng et al. 2013). Data for SLP and atmospheric temperature at 850 hPa (T850) were derived from monthly means of ERA5 reanalysis from 1950 to 2019 (Hersbach et al. 2020). SSTs from 1950 to 2019 were taken from NOAA ERSST, version 3b. Our combined dataset of SIC, SST, SLP, and T850 observations and reanalysis is henceforth denoted as OBS. As the response of the AO to ENSO in reanalysis is of questionable robustness (Deser et al. 2017), we also use SLP data from NCEP–NCAR reanalysis (Kalnay et al. 1996) to evaluate the dependence of our results upon choice of reanalysis product.
b. ENSO compositing
Anomalies for each variable were calculated by removing the ensemble mean or climatological mean for model and OBS data, respectively. A seventh-order, high-pass Butterworth filter with a cutoff of 15 years was then applied to these anomalies. We find filtering to be preferable to the more typical linear detrending of anomalies, as variables such as local sea ice concentration can respond very nonlinearly to climate forcing, which itself displays some nonlinearity in the Arctic during the last century (England 2021). Additionally, filtering reduces the influence of low-frequency internal variability such as the Pacific decadal oscillation on our results.
The Niño-3.4 index (N3.4), the spatial mean SST anomaly between −5°N, 5°S, 170°, and 120°W, was used to represent ENSO amplitude. We selected years when N3.4 in January is above the 90th percentile to represent El Niño years and below the 10th percentile to represent La Niña years. For OBS this gave 7 years of each type in HIST70 and 4 years of each type in HIST41, which can be compared with a much larger sample of 160 of each type of event for CESM1-LE in HIST41. In observations and models, Arctic atmosphere and sea ice quantities were composited on El Niño and La Niña years to identify ENSO’s influence on the Arctic. We commonly refer to El Niño composites minus La Niña composites as the “net ENSO” effect. The significance of results was determined using Monte Carlo testing at a 95% significance level. Two-tailed tests were generally applied, except for in circumstances where a separate analysis had already indicated that a specific sign of response was expected, such as confirming if observational and modeled responses match. We test for field significance in maps of our composites using the false discovery rate method (Wilks 2006, 2016; see the online supplemental material for a full description).
Composites of N3.4 on the El Niño and La Niña events in MMLEA models demonstrate that while each model is capable of capturing some aspects of the observed ENSO cycle, there are notable differences (Fig. S1 in the online supplemental material). For example, CSIRO Mk3.6 has too low an amplitude for both El Niño and La Niña events. However, differences with the ENSO cycle in observations do not necessarily indicate biases in each case. In CESM1-LE, El Niño events are larger than in OBS for the ensemble mean, yet there are several individual ensemble members that closely match the OBS amplitude (Fig. S2 in the online supplemental material). While we note that other biases exist in the representation of ENSO in MMLEA models, we primarily evaluate the teleconnections resulting from ENSO rather than the ENSO cycle itself, as these more directly determine the Arctic sea ice response. The ability of ability of GCMs to simulate ENSO is thoroughly documented, such as in Deser et al. (2012) for CESM1.
c. Role of the Arctic Oscillation and Aleutian low
The major modes of variability of Northern Hemisphere SLP were defined through empirical orthogonal function (EOF) analysis. Largely following the methods of Thompson and Wallace (1998), an estimate of anomalies in 1000-hPa geopotential height (Z1000) north of 20°N as Z1000 = 8(SLP) was generated using filtered SLP anomalies from climatology expressed in hPa. Singular value decomposition with a weighting by latitude was then applied to this Z1000 estimate. Unlike in Thompson and Wallace’s work, all months of the year were incorporated, however, given that SLP variability is highest during winter months, the results remain similar. The leading EOF captures the AO, while the second captures the AL, with well separated variance based on the criteria of North et al. (1982). The EOFs generated from the first ensemble member of each MMLEA model generally agree well with those from ERA5 reanalysis in both their pattern (Fig. S3 in the online supplemental material) and the proportion of total SLP variance (Fig. S4 in the online supplemental material) they explain. Model differences from OBS include exaggerated high SLP in the North Pacific for the AO, and a slight overestimation of the fraction of variance explained by the AO and AL.
The standardized principal components of the EOFs from the ERA5 reanalysis were used as indices for the AO and AL. Indices for the AO and AL for each model were calculated by regressing Z1000 estimates from each month on the associated ERA5 reanalysis loading patterns. The resulting indices were then scaled according to their standard deviations such that their magnitudes were directly comparable to those derived from the reanalysis data.
Regression coefficients were calculated excluding years with El Niño or La Niña events so that there was no overlap in the years used to compute the reconstructed net ENSO response
3. Response of Arctic sea ice to ENSO
The observed response of pan-Arctic sea ice area to El Niños or La Niñas in January is not statistically significant for any month (red/blue dashed lines do not exceed green lines in Fig. 1a), nor is the observed net ENSO sea ice area response (Fig. 1b). There are too few ENSO events during the 41-yr sea ice satellite record to separate the observed sea ice area response to ENSO from internal variability, which motivates us to investigate the response in model large ensembles.
Total pan-Arctic sea ice (a),(b) area and (c),(d) volume anomalies (left) composited on El Niño and La Niña events in OBS and CESM1-LE and (right) showing the net ENSO effects for area and volume for all MMLEA models and area for OBS. Note that the y-axis range in (b) and (d) is double that in (a) and (c). Shaded patches represent the 10th–90th-percentile spread among model ensemble members, and solid lines represent the ensemble mean. Months for which the difference between El Niño and La Niña in CESM1-LE is significant are circled in (a) and (c). In (a), dashed red and blue lines represent the sea ice area response from OBS and dashed green lines indicate the threshold for significance for anomalies associated with El Niño or La Niña in OBS as based on a Monte Carlo test.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Total pan-Arctic sea ice (a),(b) area and (c),(d) volume anomalies (left) composited on El Niño and La Niña events in OBS and CESM1-LE and (right) showing the net ENSO effects for area and volume for all MMLEA models and area for OBS. Note that the y-axis range in (b) and (d) is double that in (a) and (c). Shaded patches represent the 10th–90th-percentile spread among model ensemble members, and solid lines represent the ensemble mean. Months for which the difference between El Niño and La Niña in CESM1-LE is significant are circled in (a) and (c). In (a), dashed red and blue lines represent the sea ice area response from OBS and dashed green lines indicate the threshold for significance for anomalies associated with El Niño or La Niña in OBS as based on a Monte Carlo test.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Total pan-Arctic sea ice (a),(b) area and (c),(d) volume anomalies (left) composited on El Niño and La Niña events in OBS and CESM1-LE and (right) showing the net ENSO effects for area and volume for all MMLEA models and area for OBS. Note that the y-axis range in (b) and (d) is double that in (a) and (c). Shaded patches represent the 10th–90th-percentile spread among model ensemble members, and solid lines represent the ensemble mean. Months for which the difference between El Niño and La Niña in CESM1-LE is significant are circled in (a) and (c). In (a), dashed red and blue lines represent the sea ice area response from OBS and dashed green lines indicate the threshold for significance for anomalies associated with El Niño or La Niña in OBS as based on a Monte Carlo test.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
A detailed examination first of the CESM1-LE indicates that rarely does a single ensemble member have a statistically significant pan-Arctic sea ice area response in the satellite era (shaded regions rarely exceed green lines for either El Niño or La Niña events in Fig. 1a). By adjusting the number of samples taken in our Monte Carlo test we can estimate the number of years of data typically required for a statistically significant pan-Arctic sea ice area response to be determined. From this we find that the CESM1-LE ensemble mean El Niño response could be classified as statistically significant in more than one month if it came from a minimum of 5 ensemble members, which equates to 20 events or 205 years. For La Niña, 12 ensemble members are needed, equating to 48 events or 492 years. These time periods far exceed the length of the observational record. In the ensemble mean of the complete set of 40 members of CESM1-LE, the net ENSO sea ice area response (the response to El Niño events minus the response to La Niña events) is statistically significant in 10 of the 12 months in the following year. When sea ice area anomalies are the same sign for both El Niños and La Niñas, as in CESM1-LE for January–May, the net ENSO effect tends to be weak. In contrast, sea ice area anomalies from June–December are negative for El Niños and positive for La Niñas, resulting in a relatively large negative net ENSO effect (Figs. 1a,b).
We evaluate the robustness of the net ENSO pan-Arctic sea ice area response by examining six model ensembles that participate in the MMLEA (Fig. 1b). We find that the response of a single ensemble member in any model is generally not significant (see the online supplemental material for further discussion). The pattern of more negative pan-Arctic sea ice area anomalies in response to El Niños relative to La Niñas during summer and autumn is consistent across MMLEA models, as the net ENSO effect is negative in each, but varies from near zero for CSIRO Mk3.6 to −6 × 105 km2 for CanESM2 in June–July.
The net ENSO pan-Arctic sea ice volume response is nearly always negative in the ensemble mean of each MMLEA model, indicating that sea ice volume is typically lower following El Niños than following La Niñas (Fig. 1d); however, the timing and magnitude of the volume loss varies from model to model. For CanESM2 the largest negative volume anomaly occurs in January, coincident with peak ENSO amplitude. Local minima of volume anomalies occur in June for GFDL CM3, MPI-GE, and CanESM2. For CESM1-LE and CSIRO Mk3.6, sea ice volume anomalies decrease through summer and into autumn. The El Niño and La Niña volume responses for each model are not necessarily opposite. For example, in CESM1-LE the January–June response is negative for both El Niños and La Niñas, and therefore the net ENSO effect in these months is statistically insignificant (Fig. 1c), which we attribute partly to the persistence of reduced sea ice caused by El Niño–like conditions the year before El Niños and La Niñas in CESM1-LE (see Fig. S1 in the online supplemental material). After June, the sign of the sea ice volume responses in Fig. 1c diverges, leading to the largest net ENSO volume response of any model in autumn (−800 km3). While we note that nonlinearities exist in the response to ENSO, we proceed to focus primarily on the net ENSO effects–those associated with El Niño minus those associated with La Niña. Doing so facilitates easier comparison across observations and each MMLEA model and improves the signal-to-noise ratio in comparison with the response to El Niño or La Niña separately.
One reason that the pan-Arctic sea ice anomalies associated with ENSO are small is that the SIC and SIT anomalies are spatially heterogeneous, with regions of positive and negative SIC (Fig. 2) and SIT (Fig. 3) anomalies associated with each ENSO phase, causing much of the signal to cancel out in pan-Arctic totals. Henceforth, we focus on the spatial patterns of sea ice anomalies associated with ENSO, as opposed to pan-Arctic totals, as the spatial heterogeneity is useful in diagnosing important mechanisms.
Net ENSO Arctic SIC anomalies for OBS and MMLEA models in each season. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (the green line indicates the climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S5 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO Arctic SIC anomalies for OBS and MMLEA models in each season. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (the green line indicates the climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S5 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO Arctic SIC anomalies for OBS and MMLEA models in each season. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (the green line indicates the climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S5 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO Arctic SIT anomalies for MMLEA models in each season. SIT data are not available for OBS. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (green line indicates climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S6 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO Arctic SIT anomalies for MMLEA models in each season. SIT data are not available for OBS. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (green line indicates climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S6 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO Arctic SIT anomalies for MMLEA models in each season. SIT data are not available for OBS. Plots for winter and spring span a wider latitude range to account for the shifting sea ice edge (green line indicates climatology 15% SIC contour). For plots excluding grid cells that fail a local 95% significance test, see Fig. S6 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
The magnitude of SIC anomalies in OBS for the net ENSO effect exceeds 15% in many regions (Fig. 2), however, only approximately 5% of grid points pass a local significance test at a 95% confidence interval in any season (Fig. S5 in the online supplemental material). No grid points pass a two-tailed field significance test (see the online supplemental material); therefore, we cannot rule out the possibility that the response seen in OBS is simply the result of random chance. This finding is insensitive to the method chosen, as doubling the sample size of ENSO events by reducing the required N3.4 threshold or using regression instead of compositing gives a comparable lack of statistical significance (not shown).
The net effects of ENSO on SIC patterns in MMLEA models (Fig. 2) are largely similar to those for SIT (Fig. 3). The SIT effects tend to extend farther into the central Arctic than the SIC effects, particularly in winter and spring when SIC is typically close to 100% in the central Arctic. With the exception of CSIRO Mk3.6, there are many features of the net ENSO SIC and SIT response to ENSO that are common across models. In spring, SIC and SIT are increased around the coast of Alaska, with decreases to the west of this, particularly in the East Siberian Sea for SIT. In summer SIC and SIT anomalies are negative in most models in the Kara, East Greenland, and Beaufort Seas, but positive in the East Siberian sea. The autumn SIC and SIT patterns are generally dampened versions of the respective summer patterns. The CESM1-LE SIT pattern is an exception to this, with large negative thicknesses persisting through autumn. The largest anomalies, exceeding −15% for SIC and −30 cm for SIT, typically occur in the Pacific sector of the Arctic, particularly the Beaufort Sea, which is unsurprising given that ENSO occurs in the Pacific Ocean, and a deepened AL causing reduced sea ice in the Beaufort Sea is consistent with previous studies (Trenberth and Hurrell 1994; Tokinaga et al. 2017; Svendsen et al. 2018; Screen and Deser 2019; England et al. 2020).
The response of Arctic sea ice to ENSO in CESM1-LE in summer is possibly unexpected as ENSO SSTs have dissipated substantially by summer (Fig. S2 in the online supplemental material), yet this is when the largest response occurs in this model both spatially (Figs. 2, 3) and for pan-Arctic totals (Fig. 1). We decompose the spatial SIT response in CESM1-LE into changes through thermodynamic (Figs. 4a,b) and dynamic (Fig. 4c) processes. The dynamic SIT tendency response is much greater than the thermodynamic SIT tendency response in winter and spring. These dynamic changes result from anomalous anticyclonic motion (Fig. 4c), which transports sea ice from the western to the eastern Arctic, causing thinning in the west and thickening in the east. By summer, the dynamic response is reduced and the thermodynamic response dominates. Increased melt rates are seen across the Arctic, with exceptions only where SIC is near zero. Although in absolute terms, the ENSO effect on thermodynamics is largest in summer; when normalized by the standard deviation, the anomalies are more uniform in magnitude across seasons. Integrated over the year, anomalies in the dynamic and thermodynamic tendencies are roughly equal in magnitude (not shown); however, it is thermodynamics that drive the delayed summer sea ice response to ENSO in CESM1-LE.
(a),(b) Thermodynamic and (c) dynamic net ENSO SIT tendency anomalies from CESM1-LE in each season. In (b), anomalies are expressed in standard deviations away from the climatological mean. Red arrows indicate the net ENSO volumetric ice flux direction and relative magnitude.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
(a),(b) Thermodynamic and (c) dynamic net ENSO SIT tendency anomalies from CESM1-LE in each season. In (b), anomalies are expressed in standard deviations away from the climatological mean. Red arrows indicate the net ENSO volumetric ice flux direction and relative magnitude.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
(a),(b) Thermodynamic and (c) dynamic net ENSO SIT tendency anomalies from CESM1-LE in each season. In (b), anomalies are expressed in standard deviations away from the climatological mean. Red arrows indicate the net ENSO volumetric ice flux direction and relative magnitude.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
4. Response of Arctic atmosphere to ENSO
The atmosphere acts as a bridge, allowing ENSO to influence higher latitudes (Lau and Nath 1996; Graf and Zanchettin 2012; Hu et al. 2016). We investigate the ENSO effect on the Arctic atmosphere in order to understand the link between ENSO and Arctic sea ice, as well as to evaluate the accuracy of MMLEA models in representing this link. The teleconnection bridge from the tropics to the Arctic is visible in the net ENSO SLP response shown in Fig. 5 (see Fig. S7 in the online supplemental material for maps with a local significance test applied). In OBS and MMLEA models, the net ENSO effect includes a deepened AL with low pressure anomalies in the North Pacific, and high pressure anomalies over the Arctic similar to the negative phase of the AO (Fig. S3 in the online supplemental material). The SLP response is largest in winter, which is expected as Northern Hemisphere SLP variability peaks in winter and the propagation time for Rossby waves between the tropics and the North Pacific is approximately two weeks (Alexander et al. 2002). The observed low SLP winter response in the North Pacific passes a field significance test, yet the high SLP response in the central Arctic does not. In spring, the SLP response is smaller and generally not significant for OBS at high latitudes, but still statistically significant in the large ensemble datasets, which still display a deepened AL and high pressure over the Arctic. The SLP response is larger in most models than in OBS, potentially indicating a bias in teleconnection strength and persistence. In summer and autumn the SLP response is even further diminished in both OBS and models. The net ENSO SLP effect is qualitatively consistent with the anomalous sea ice volume fluxes shown in Fig. 4c, with high pressure over the Arctic driving anticyclonic sea ice drift.
Net ENSO SLP anomalies for OBS and MMLEA models for each season. For plots excluding grid cells that fail a local 95% significance test, see Fig. S7 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO SLP anomalies for OBS and MMLEA models for each season. For plots excluding grid cells that fail a local 95% significance test, see Fig. S7 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO SLP anomalies for OBS and MMLEA models for each season. For plots excluding grid cells that fail a local 95% significance test, see Fig. S7 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
The Arctic T850 response to ENSO in winter and spring is relatively consistent across OBS and MMLEA models (Fig. 6). Warming of around 2°C occurs over much of northern North America except for in CSIRO Mk3.6, while northern Russia is simultaneously cooled by around 1°C. In OBS this cooling extends into the Barents and Kara Seas and across half of the central Arctic, unlike in MMLEA models in which warming tends to dominate over the whole basin, as noted by McCrystall and Screen (2021). This is somewhat at odds with the conclusions of Lee (2012), who found that El Niños and La Niñas result in decreased and increased Arctic surface air temperatures respectively in winter in observations; however, our overall net ENSO pattern of warming over North America and cooling over Siberia retains many similarities with theirs. In our results, only small regions of the T850 response are significant for OBS, but notably well over 5% of the grid points pass a local significance test, which is more than would be expected by chance (Fig. S8 in the online supplemental material). The winter and autumn observed T850 responses pass field significance tests, although the region crossing the threshold for field significance in winter is confined to the North Pacific. However, the good agreement between OBS and MMLEA in winter and spring supports the conclusion that the observed patterns of T850 across the Arctic are more than just noise. Further, the Arctic warming, particularly in the Beaufort region, is consistent with the effects of a deepened AL identified in other studies (e.g., see Tokinaga et al. 2017). During summer, T850 anomalies in MMLEA models are typically positive over the Arctic basin, but only reach magnitudes of approximately 0.5°C. The equivalent OBS response is slightly negative but not significant. During autumn, the T850 response in OBS passes a local and field significance test; however, given that this pattern is not reflected in MMLEA models it is unclear if it is noise or if there is an ENSO response not captured in the models.
Net ENSO T850 anomalies for OBS and five MMLEA models for each season. T850 data were not available for MPI-GE. For plots excluding grid cells that fail a local 95% significance test, see Fig. S8 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO T850 anomalies for OBS and five MMLEA models for each season. T850 data were not available for MPI-GE. For plots excluding grid cells that fail a local 95% significance test, see Fig. S8 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO T850 anomalies for OBS and five MMLEA models for each season. T850 data were not available for MPI-GE. For plots excluding grid cells that fail a local 95% significance test, see Fig. S8 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
The net ENSO surface downwelling longwave radiation response over the Arctic Ocean is generally positive in MMLEA models (Fig. 7) and resembles the aforementioned T850 patterns over the ocean, as expected since this flux is largely controlled by lower-tropospheric temperatures (Vargas Zeppetello et al. 2019). The surface downwelling longwave radiation anomaly in Fig. 7 is greatest in winter and spring around the coast of Alaska, often exceeding 15 W m−2. By summer the net ENSO surface downwelling longwave flux response is weakly positive, consistent with an increase in cloudiness (not shown). The majority of these longwave flux anomalies are absorbed due to the high surface emissivity (e.g., ε = 0.95 in CESM1-LE for snow and sea ice), which contributes to increased melt rates (Fig. 4b).
Net ENSO anomalies in surface downwelling longwave radiation in winter, spring, and summer; downwelling shortwave radiation in summer; and absorbed shortwave radiation in summer in MMLEA models. Downwelling longwave data were not available for GFDL CM3, and none of the surface energy flux terms were available for MIP-GE. The full evolution of each of these terms in each season in CESM1-LE is shown in Fig. S9 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO anomalies in surface downwelling longwave radiation in winter, spring, and summer; downwelling shortwave radiation in summer; and absorbed shortwave radiation in summer in MMLEA models. Downwelling longwave data were not available for GFDL CM3, and none of the surface energy flux terms were available for MIP-GE. The full evolution of each of these terms in each season in CESM1-LE is shown in Fig. S9 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Net ENSO anomalies in surface downwelling longwave radiation in winter, spring, and summer; downwelling shortwave radiation in summer; and absorbed shortwave radiation in summer in MMLEA models. Downwelling longwave data were not available for GFDL CM3, and none of the surface energy flux terms were available for MIP-GE. The full evolution of each of these terms in each season in CESM1-LE is shown in Fig. S9 in the online supplemental material.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
The net ENSO surface downwelling shortwave flux over the Arctic Ocean is weak in most seasons except in summer (Fig. 7), when it is mostly negative in the central Arctic in MMLEA models. We attribute the negative anomalies to increased cloudiness, consistent with the increase in surface downwelling longwave flux in summer. In turn, most models absorb less shortwave flux at the surface, except in the Beaufort Sea. Each model has negative sea ice concentration anomalies in the Beaufort Sea in summer (Fig. 2), which implies an increased fraction of open water and melt ponds, both of which lower surface albedo, causing an increase in shortwave absorption. In CESM1-LE, the net ENSO absorbed shortwave flux is positive throughout the Arctic, with values up to 10 W m−2, which explains the large summertime loss of sea ice through thermodynamics (Fig. 4a) and large summertime reduction in pan-Arctic sea ice area and volume (Fig. 1) in CESM1-LE. The increased shortwave absorption is confined to summer when there is sufficient downwelling shortwave flux and lower surface albedos (see the online supplemental material and supplemental Fig. S9).
The availability of additional surface variables for CESM1-LE allows us to further investigate the causes of its net ENSO reduced summer albedo across the Arctic ocean. The first factor is reduced sea ice concentrations (Fig. 2b), exposing open water with much lower albedo in the marginal Arctic seas, but with little influence on the central Arctic. Negative net ENSO cumulative snowfall (Fig. 8a) and positive cumulative snowmelt (Fig. 8b) until summer in the central Arctic result in decreased summer snow concentrations of up to 10% (Fig. 8c). Summer melt pond fraction is also increased by up to 5% in the central Arctic (Fig. 8d). Fresh snow has a higher albedo than old snow, which has an albedo higher than that of sea ice, while melt ponds have a much lower albedo (Perovich et al. 2002). These patterns of reduced snow concentration and increased melt pond fraction closely match those of increased absorbed shortwave radiation (Fig. 7a).
CESM1-LE net ENSO anomalies in cumulative (a) snowfall and (b) snowmelt from winter through to halfway through summer (the sum of the winter and spring patterns and one-half of the summer pattern) and net ENSO (c) snow concentration and (d) melt pond fraction in summer.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
CESM1-LE net ENSO anomalies in cumulative (a) snowfall and (b) snowmelt from winter through to halfway through summer (the sum of the winter and spring patterns and one-half of the summer pattern) and net ENSO (c) snow concentration and (d) melt pond fraction in summer.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
CESM1-LE net ENSO anomalies in cumulative (a) snowfall and (b) snowmelt from winter through to halfway through summer (the sum of the winter and spring patterns and one-half of the summer pattern) and net ENSO (c) snow concentration and (d) melt pond fraction in summer.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Ultimately, much of the mechanism through which Arctic sea ice responds to ENSO can be traced back to SLP anomalies (Fig. 5). Sea ice motion is controlled primarily by surface winds, which are themselves related to the SLP. Anticyclonic sea ice motion (Fig. 4c) occurs under high pressure in the central Arctic (negative AO), redistributing SIC and SIT. A deepened AL drives advection of warm, moist air into the Arctic (as in e.g., Trenberth and Hurrell 1994; Tokinaga et al. 2017), melting sea ice and snow and lowering surface albedo. Increased shortwave absorption drives further melt and increased albedo reduction, forming a positive feedback in summer in CESM1-LE.
5. Role of the AO and AL in connecting ENSO and Arctic sea ice
In this section we quantify the importance of the AO and the AL in connecting ENSO with Arctic sea ice by establishing the extent to which this sea ice response may be reconstructed solely from the response of the AO and AL to ENSO, and typical sea ice responses to the AO and AL.
To begin, we calculate the AO and AL response to El Niños and La Niñas in each month for OBS and MMLEA models by compositing the AO and AL indices on ENSO events (shown for OBS and CESM1-LE in Figs. 9a,c). The net ENSO AO and AL index (for El Niño events minus La Niña events) for all MMLEA models and OBS are shown in Figs. 9b and 9d. Because the AO and AL are orthogonal in space and time, the net ENSO AO and AL indices multiplied by their respective EOF patterns (Fig. S3 in the online supplemental material) can be summed to give a good approximation to the full net ENSO SLP response. The square of the pattern correlation between this approximation and the full net ENSO response R2 indicates the fraction of the variance in the full response that can be explained by the AO and AL response. In January–March the majority of the net ENSO SLP response is captured by the AO and AL response in models (R2 = 0.63), while a smaller portion is captured by the AO and AL response in OBS (R2 = 0.43), in part due to the smaller sample size and lower signal-to-noise ratio in OBS.
Anomalies in the (a),(b) AO and (c),(d) AL associated with El Niño and La Niña in (left) OBS and CESM1-LE and (right) net ENSO in OBS and MMLEA models. Shaded patches represent the 10th–90th-percentile spread among model ensemble members. Dashed lines represent results from ERA5 reanalysis, with solid circles indicating significant differences between El Niño and La Niña anomalies. Dotted lines represent results from NCEP–NCAR reanalysis.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Anomalies in the (a),(b) AO and (c),(d) AL associated with El Niño and La Niña in (left) OBS and CESM1-LE and (right) net ENSO in OBS and MMLEA models. Shaded patches represent the 10th–90th-percentile spread among model ensemble members. Dashed lines represent results from ERA5 reanalysis, with solid circles indicating significant differences between El Niño and La Niña anomalies. Dotted lines represent results from NCEP–NCAR reanalysis.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Anomalies in the (a),(b) AO and (c),(d) AL associated with El Niño and La Niña in (left) OBS and CESM1-LE and (right) net ENSO in OBS and MMLEA models. Shaded patches represent the 10th–90th-percentile spread among model ensemble members. Dashed lines represent results from ERA5 reanalysis, with solid circles indicating significant differences between El Niño and La Niña anomalies. Dotted lines represent results from NCEP–NCAR reanalysis.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
In each MMLEA model and in OBS, the net ENSO SLP response includes a negative AO index (Fig. 9b) and a positive AL index (Fig. 9d) from January to April (henceforth we refer to these quantities as the net ENSO AO and AL responses). The net ENSO AO and AL responses then decline to near zero until the next winter. The OBS net ENSO AO and AL responses generally fall within the 10th–90th percentile ensemble spread of MMLEA models, although there are indications that the AL response might persist into spring too long in models. The AO and AL responses to El Niño and La Niña events in ERA5 OBS and NCEP–NCAR OBS are nearly identical (Figs. 9a,c). In OBS and CESM1-LE the AO and AL responses to El Niño and La Niña events are nearly always opposite in sign (Figs. 9a,c). The net ENSO AO response in OBS is statistically significant in February and April, while the net ENSO AL response in OBS is significant in January and February (solid circles in Figs. 9a and 9c), which combined with matching behavior in MMLEA models leads us to conclude that both the negative AO and positive AL net ENSO responses are likely robust.
Having established that there are robust net ENSO AO and AL responses in MMLEA models and OBS, we are motivated to reconstruct the SIC and SIT responses to ENSO [see Eq. (1)] by combining the net ENSO AO and AL responses and the typical sea ice patterns associated with the AO and AL and compare them with the net ENSO SIC and SIT responses found through compositing (Figs. 2 and 3).
An example of reconstructing the net ENSO SIC and SIT responses is shown for CESM1-LE in Fig. 10. In this example, the June patterns of SIT associated with the AO and AL (Figs. 10a,b) are multiplied by the net ENSO AO (−1.6) and AL (2.4) index responses, respectively, and summed together to give the reconstructed net ENSO SIT response (Fig. 10c). The similarity of this reconstruction with composites of the net ENSO SIT response in June (Fig. 10d) in CESM1-LE is reflected by a high pattern correlation coefficient between the two (R = 0.80), implying that much of the net ENSO sea ice response in this month can be attributed to ENSO’s influence on the AO and AL. However, an important caveat is that the sea ice patterns associated with the AO and AL (Figs. 10a,b) are not necessarily driven by the AO and AL, merely correlated with these indices.
CESM1-LE June SIT regressed on January–March (a) AO index, (b) AL index, and (c) CESM1-LE reconstructed June SIT based on patterns in (a) and (b) and the net ENSO response of the AO and AL index (see methods). (d) CESM1-LE June net ENSO SIT composite. (e) Pattern correlation coefficients between the reconstructed and composited net ENSO SIT (dashed lines) and SIC (solid lines) response in OBS and MMLEA models. Blue-shaded region represents the 10th–90th-percentile spread of pattern correlations for single ensemble members of CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
CESM1-LE June SIT regressed on January–March (a) AO index, (b) AL index, and (c) CESM1-LE reconstructed June SIT based on patterns in (a) and (b) and the net ENSO response of the AO and AL index (see methods). (d) CESM1-LE June net ENSO SIT composite. (e) Pattern correlation coefficients between the reconstructed and composited net ENSO SIT (dashed lines) and SIC (solid lines) response in OBS and MMLEA models. Blue-shaded region represents the 10th–90th-percentile spread of pattern correlations for single ensemble members of CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
CESM1-LE June SIT regressed on January–March (a) AO index, (b) AL index, and (c) CESM1-LE reconstructed June SIT based on patterns in (a) and (b) and the net ENSO response of the AO and AL index (see methods). (d) CESM1-LE June net ENSO SIT composite. (e) Pattern correlation coefficients between the reconstructed and composited net ENSO SIT (dashed lines) and SIC (solid lines) response in OBS and MMLEA models. Blue-shaded region represents the 10th–90th-percentile spread of pattern correlations for single ensemble members of CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
The MMLEA models produce a range of pattern correlation coefficients between the reconstructed and composited net ENSO SIC and SIT responses for each month (Fig. 10e). For example, the correlation coefficients for SIT in CanESM2 exceed 0.9 from February to July, indicating that the response of SIT to ENSO is almost exactly what we would expect given the effects of ENSO on the AO and AL. In contrast, the correlation coefficients for CSIRO Mk3.6 are negative in several months for both SIC and SIT, which is perhaps to be expected given that the response of the AO and AL to ENSO is small in CSIRO Mk3.6 (Figs. 9b,d) and the issues in Arctic sea ice representation in this model (Fig. S10 in the online supplemental material). Indeed, MPI-GE displays the second lowest correlation coefficients and is the model with the second weakest AO and AL response (Figs. 9b,d). In these models other modes of variability may drive much of the sea ice response to ENSO. For many models the correlation coefficients rise from January through to late spring and early summer when they are at their highest, which is intuitive as the AO and AL response to ENSO is confined to winter and spring (Figs. 9b,d), and their influence on Arctic sea ice integrates over that time. Correlation coefficients for SIC in OBS are near zero, which is generally consistent with results from individual members of the CESM1-LE (Fig. 10e; black lines fall within blue shading for 9 of 12 months).
We use the dynamic and thermodynamic tendency terms from CESM1-LE and Eq. (2) to decompose the response of summer SIT to ENSO into that which can be attributed to thermodynamics and dynamics associated with the AO and AL (Fig. 11). The dynamic components of both the AO and AL response are roughly equal in magnitude and share many similarities with the net ENSO dynamic SIT patterns in winter and spring (Fig. 4c). The smaller thermodynamic components indicate that the AO and AL primarily influence SIT patterns through dynamics in winter and spring. Sea ice anomalies in winter and spring persist for several months generally across the Arctic (Blanchard-Wrigglesworth et al. 2011) and specifically in relation to the AO (Rigor et al. 2002; Park et al. 2018). A negative SIT anomaly in the central Arctic is associated with AL thermodynamics (Fig. 11d), which suggests that the delayed summer melt in the central Arctic associated with ENSO is primarily the result of changes to the AL, consistent with the connection between the AL and increased heat flux into the Arctic. Conversely, the thermodynamic component associated with the AL in the Beaufort Sea is positive, indicating increased ice growth or reduced melt rates despite warmer temperatures over that region (Fig. 6). We explain this as a response to the dynamic components of the AL, which results in a SIT reduction of over 0.6 m in the Beaufort Sea (Fig. 11b), causing a regional absence of sea ice available to be melted in summer, and the presence of open water or thin ice that freezes quickly in winter. We note that the delayed summer melting in response to ENSO in CESM1 is unusual among MMLEA models (Fig. 3); therefore, while we do not perform this decomposition for the other MMLEA models, we hypothesize that they would display even smaller thermodynamic tendency anomalies than in CESM1, and that the greater role of dynamics relative to thermodynamics in driving the sea ice response to ENSO is likely robust to model selection.
Contribution of ice (a),(b) dynamic and (c),(d) thermodynamic processes associated with the (left) AO and (right) AL toward the net ENSO SIT pattern in summer in CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Contribution of ice (a),(b) dynamic and (c),(d) thermodynamic processes associated with the (left) AO and (right) AL toward the net ENSO SIT pattern in summer in CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Contribution of ice (a),(b) dynamic and (c),(d) thermodynamic processes associated with the (left) AO and (right) AL toward the net ENSO SIT pattern in summer in CESM1-LE.
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Given our conclusions on the importance of the AO and AL in linking ENSO and Arctic sea ice, it follows that an accurate representation of the sea ice response to ENSO in models relies upon accurate representation of the effects of AO and AL on Arctic sea ice. We evaluate this representation in CESM1-LE by comparing the SIC regression coefficients
Summer SIC regressed on January–March (a),(b) AO index and (c),(d) AL index for (left) CESM1-LE and (right) OBS. (e) Pattern correlation coefficients for
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Summer SIC regressed on January–March (a),(b) AO index and (c),(d) AL index for (left) CESM1-LE and (right) OBS. (e) Pattern correlation coefficients for
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
Summer SIC regressed on January–March (a),(b) AO index and (c),(d) AL index for (left) CESM1-LE and (right) OBS. (e) Pattern correlation coefficients for
Citation: Journal of Climate 34, 24; 10.1175/JCLI-D-20-0958.1
We quantify the agreement between OBS and CESM1-LE patterns by calculating pattern correlation coefficients Robs between them in each season (Fig. 12e, dashed lines). Internal variability and a small sample size mean we expect the OBS patterns to have a low signal-to-noise ratio, resulting in a mismatch with the CESM1-LE ensemble mean patterns. To provide an appropriate benchmark we also calculate pattern correlation coefficients between individual CESM1-LE ensemble members and the CESM1-LE ensemble mean (RCESM; Fig. 12e, solid lines and shading). Robs does not fall within the range of RCESM for the AO in spring, and the AL in winter, and spring, indicating that the patterns of SIC associated with the AO and AL in CESM1-LE in these seasons are inconsistent with observations. While part of this disagreement may be due to biases in the location of the ice edge in CESM1-LE (Fig. S10 in the online supplemental material), we still highlight that model biases in the response of Arctic sea ice to the AO and is a key source of uncertainty in our conclusions.
6. Summary and conclusions
The influence of the teleconnections between tropical Pacific SST patterns, including ENSO, and the Arctic on Arctic sea ice is challenging to decipher. The short time period of satellite sea ice observations results in a small sample size of observed ENSO events to analyze, which when combined with the internal variability of Arctic sea ice prevents identification of a statistically significant ENSO signal in Arctic sea ice observations. The use of model large ensembles provides a means of addressing this shortfall.
In each MMLEA model we find that El Niño events in January are associated with a reduction in total Arctic sea ice area and volume in the following summer and autumn when compared with La Niñas. These models are mostly consistent with each other in their spatial patterns of Arctic SIC and SIT responses to ENSO, each indicating large regional sea ice responses to ENSO, which may be a source of predictability for sea ice forecasts.
The sea ice response to ENSO can be explained by the coupling of Arctic sea ice to the Arctic atmosphere. Sea level pressure anomalies, specifically a negative AO and positive AL, drive surface wind anomalies, which result in dynamic redistribution of sea ice, leading to a spatially heterogeneous pattern of sea ice response in winter and spring. In most MMLEA models, the summer and autumn sea ice response to ENSO is mostly determined by the persistence of SIT anomalies from earlier in the year, which are shown in to be dynamically induced in CESM1-LE. However, unlike most other MMLEA models, CESM1-LE also displays delayed thermodynamic thinning of sea ice in the central Arctic in summer. This occurs due to reduced snowfall and increased melt rates earlier in the year, which precondition the ice to have a lower albedo, resulting in increased shortwave absorption and sea ice melt during summer when the downwelling shortwave flux in climatology is greatest. We tie this delayed melting mechanism to the deepened AL, which other studies have identified as a mechanism for advection of warm air into the Arctic (e.g., Svendsen et al. 2018).
Much of the previous work on teleconnections from the tropics to the Arctic has focused on coincident analysis of summer SSTs and summer SIC (Baxter et al. 2019; Ding et al. 2019; Bonan and Blanchard-Wrigglesworth 2020). Our findings suggest that a lagged response from winter SST patterns such as ENSO should also be considered for a full understanding of the effects of tropical Pacific SSTs on summer Arctic sea ice. That the sea ice response to ENSO can be reconstructed from the AO and AL response to ENSO in January–March underscores the importance of winter teleconnections. An additional reason for the focus on winter teleconnections is that most summer sea ice area forecast error growth is generated locally, bringing into question the impact of remote teleconnections on the Arctic summer atmosphere at seasonal and subseasonal time scales (Blanchard-Wrigglesworth and Ding 2019).
While our use of large model ensembles overcomes sample size issues, it leads to two main sources of uncertainty in our results. The first source is inaccuracies in the representation of ENSO tropical dynamics and its teleconnections to the north. The second source is inaccuracies in the response of sea ice to these teleconnections in models. Our results suggest that the first source is minimal since we find that the SLP and T850 response in most MMLEA models is largely consistent with that in OBS. With regard to the second source, we show that the response of Arctic sea ice to the AO and AL in CESM1-LE is inconsistent with OBS in some seasons. We believe improvement of the Arctic sea ice response to modes of atmospheric variability in models would greatly aid our understanding of the influence of tropical Pacific SSTs on Arctic sea ice.
Analysis of the differences in the Arctic sea ice response to eastern and central Pacific ENSO events, as in Hu et al. (2016), would be a natural extension of the work done in this study; however, this is not without further challenges. The sample size of eastern and central Pacific ENSO events in observations is smaller than for both event types combined, while GCMs such as CESM1 often fail to realistically distinguish between eastern and central Pacific events. The impact of ENSO on western Pacific SSTs is likely very important in determining the resulting teleconnection pattern to the Arctic (Dong et al. 2019), and differs greatly between eastern and central Pacific ENSO events (Hu et al. 2016). Differences in western Pacific SSTs associated with ENSO is therefore a potential source of much of the difference in teleconnection patterns among MMLEA models.
Acknowledgments
We are grateful for support from Office of Naval Research N00014-15-1-2873 and N000141812175. We thank three anonymous reviewers for their valuable feedback.
Data availability statement
Satellite sea ice concentration data were provided by the NOAA/OAR/ESRL PSD (https://www.esrl.noaa.gov/psd/). Sea surface temperature data, NOAA_ERSST_V3b, were provided by the NOAA/OAR/ESRL PSL (https://psl.noaa.gov/). ERA5 data are freely available online (https://cds.climate.copernicus.eu/cdsapp#!/home). Climate model data are available from the U.S. CLIVAR Working Group on Large Ensembles Multi-Model Large Ensemble Archive (https://www.cesm.ucar.edu/projects/community-projects/MMLEA/). Analysis code is available from the corresponding author on request.
REFERENCES
Alexander, M. A., I. Bladé, M. Newman, J. R. Lanzante, N.-C. Lau, and J. D. Scott, 2002: The atmospheric bridge: The influence of ENSO teleconnections on air–sea interaction over the global oceans. J. Climate, 15, 2205–2231, https://doi.org/10.1175/1520-0442(2002)015<2205:TABTIO>2.0.CO;2.
Baxter, I., and Coauthors, 2019: How tropical Pacific surface cooling contributed to accelerated sea ice melt from 2007 to 2012 as ice is thinned by anthropogenic forcing. J. Climate, 32, 8583–8602, https://doi.org/10.1175/JCLI-D-18-0783.1.
Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163–172, https://doi.org/10.1175/1520-0493(1969)097<0163:ATFTEP>2.3.CO;2.
Blanchard-Wrigglesworth, E., and Q. Ding, 2019: Tropical and midlatitude impact on seasonal polar predictability in the Community Earth System Model. J. Climate, 32, 5997–6014, https://doi.org/10.1175/JCLI-D-19-0088.1.
Blanchard-Wrigglesworth, E., K. C. Armour, C. M. Bitz, and E. DeWeaver, 2011: Persistence and inherent predictability of Arctic sea ice in a GCM ensemble and observations. J. Climate, 24, 231–250, https://doi.org/10.1175/2010JCLI3775.1.
Bonan, D. B., and E. Blanchard-Wrigglesworth, 2020: Nonstationary teleconnection between the Pacific Ocean and Arctic sea ice. Geophys. Res. Lett., 47, e2019GL085666, https://doi.org/10.1029/2019GL085666.
Cai, L., V. A. Alexeev, and J. E. Walsh, 2020: Arctic sea ice growth in response to synoptic- and large-scale atmospheric forcing from CMIP5 models. J. Climate, 33, 6083–6099, https://doi.org/10.1175/JCLI-D-19-0326.1.
Comiso, J. C., 1986: Characteristics of Arctic winter sea ice from satellite multispectral microwave observations. J. Geophys. Res., 91, 975–994, https://doi.org/10.1029/JC091iC01p00975.
Day, J. J., J. C. Hargreaves, J. D. Annan, and A. Abe-Ouchi, 2012: Sources of multi-decadal variability in Arctic sea ice extent. Environ. Res. Lett., 7, 034011, https://doi.org/10.1088/1748-9326/7/3/034011.
Deser, C., M. A. Alexander, S.-P. Xie, and A. S. Phillips, 2010: Sea surface temperature variability: Patterns and mechanisms. Annu. Rev. Mar. Sci., 2, 115–143, https://doi.org/10.1146/annurev-marine-120408-151453.
Deser, C., and Coauthors, 2012: ENSO and Pacific decadal variability in the community climate system model version 4. J. Climate, 25, 2622–2651, https://doi.org/10.1175/JCLI-D-11-00301.1.
Deser, C., I. R. Simpson, K. A. McKinnon, and A. S. Phillips, 2017: The Northern Hemisphere extratropical atmospheric circulation response to ENSO: How well do we know it and how do we evaluate models accordingly? J. Climate, 30, 5059–5082, https://doi.org/10.1175/JCLI-D-16-0844.1.
Deser, C., and Coauthors, 2020: Publisher correction: Insights from Earth system model initial-condition large ensembles and future prospects. Nat. Climate Change, 10, 791, https://doi.org/10.1038/s41558-020-0854-5.
Ding, Q., and Coauthors, 2019: Fingerprints of internal drivers of Arctic sea ice loss in observations and model simulations. Nat. Geosci., 12, 28–33, https://doi.org/10.1038/s41561-018-0256-8.
Domeisen, D. I. V., C. I. Garfinkel, and A. H. Butler, 2019: The teleconnection of El Niño Southern Oscillation to the stratosphere. Rev. Geophys., 57, 5–47, https://doi.org/10.1029/2018RG000596.
Dong, Y., C. Proistosescu, K. C. Armour, and D. S. Battisti, 2019: Attributing historical and future evolution of radiative feedbacks to regional warming patterns using a Green’s function approach: The preeminence of the western Pacific. J. Climate, 32, 5471–5491, https://doi.org/10.1175/JCLI-D-18-0843.1.
Dunne, J. P., and Coauthors, 2012: GFDL’s ESM2 global coupled climate–carbon Earth System Models. Part I: Physical formulation and baseline simulation characteristics. J. Climate, 25, 6646–6665, https://doi.org/10.1175/JCLI-D-11-00560.1.
England, M. R., 2021: Are multi-decadal fluctuations in Arctic and Antarctic surface temperatures a forced response to anthropogenic emissions or part of internal climate variability? Geophys. Res. Lett., 48, e2020GL090631, https://doi.org/10.1029/2020GL090631.
England, M. R., A. Jahn, and L. Polvani, 2019: Nonuniform contribution of internal variability to recent Arctic sea ice loss. J. Climate, 32, 4039–4053, https://doi.org/10.1175/JCLI-D-18-0864.1.
England, M. R., L. M. Polvani, and L. Sun, 2020: Robust Arctic warming caused by projected Antarctic sea ice loss. Environ. Res. Lett., 15, 104005, https://doi.org/10.1088/1748-9326/abaada.
Flato, G. M., and W. D. Hibler, 1992: Modeling pack ice as a cavitating fluid. J. Phys. Oceanogr., 22, 626–651, https://doi.org/10.1175/1520-0485(1992)022<0626:MPIAAC>2.0.CO;2.
Graf, H.-F., and D. Zanchettin, 2012: Central Pacific El Niño, the “subtropical bridge,” and Eurasian climate. J. Geophys. Res., 117, D01102, https://doi.org/10.1029/2011JD016493.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with the Southern Oscillation. Mon. Wea. Rev., 109, 813–829, https://doi.org/10.1175/1520-0493(1981)109<0813:PSAPAW>2.0.CO;2.
Hoskins, B. J., and D. J. Karoly, 1981: The steady linear response of a spherical atmosphere to thermal and orographic forcing. J. Atmos. Sci., 38, 1179–1196, https://doi.org/10.1175/1520-0469(1981)038<1179:TSLROA>2.0.CO;2.
Hu, C., S. Yang, Q. Wu, Z. Li, J. Chen, K. Deng, T. Zhang, and C. Zhang, 2016: Shifting El Niño inhibits summer Arctic warming and Arctic sea-ice melting over the Canada Basin. Nat. Commun., 7, 11721, https://doi.org/10.1038/ncomms11721.
Hurwitz, M. M., N. Calvo, C. I. Garfinkel, A. H. Butler, S. Ineson, C. Cagnazzo, E. Manzini, and C. Peña-Ortiz, 2014: Extra-tropical atmospheric response to ENSO in the CMIP5 models. Climate Dyn., 43, 3367–3376, https://doi.org/10.1007/s00382-014-2110-z.
Jahn, A., and Coauthors, 2012: Late-twentieth-century simulation of Arctic sea ice and ocean properties in the CCSM4. J. Climate, 25, 1431–1452, https://doi.org/10.1175/JCLI-D-11-00201.1.
Jeffrey, S., L. Rotstayn, M. Collier, S. Dravitzki, C. Hamalainen, C. Moeseneder, K. Wong, and J. Skytus, 2013: Australia’s CMIP5 submission using the CSIRO-Mk3.6 model. Aust. Meteor. Oceanogr. J., 63, 1–13, https://doi.org/10.22499/2.6301.001.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, https://doi.org/10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kay, J. E., M. M. Holland, and A. Jahn, 2011: Inter-annual to multi-decadal Arctic sea ice extent trends in a warming world. Geophys. Res. Lett., 38, L15708, https://doi.org/10.1029/2011GL048008.
Kay, J. E., and Coauthors, 2015: The Community Earth System Model (CESM) large ensemble project: A community resource for studying climate change in the presence of internal climate variability. Bull. Amer. Meteor. Soc., 96, 1333–1349, https://doi.org/10.1175/BAMS-D-13-00255.1.
Kirchmeier-Young, M. C., F. W. Zwiers, and N. P. Gillett, 2017: Attribution of extreme events in Arctic sea ice extent. J. Climate, 30, 553–571, https://doi.org/10.1175/JCLI-D-16-0412.1.
Lau, N.-C., and M. J. Nath, 1996: The role of the “atmospheric bridge” in linking tropical Pacific ENSO events to extratropical SST anomalies. J. Climate, 9, 2036–2057, https://doi.org/10.1175/1520-0442(1996)009<2036:TROTBI>2.0.CO;2.
Lee, S., 2012: Testing of the Tropically Excited Arctic Warming Mechanism (TEAM) with traditional El Niño and La Niña. J. Climate, 25, 4015–4022, https://doi.org/10.1175/JCLI-D-12-00055.1.
Li, F., H. Wang, and J. Liu, 2014: The strengthening relationship between Arctic Oscillation and ENSO after the mid-1990s. Int. J. Climatol., 34, 2515–2521, https://doi.org/10.1002/joc.3828.
Li, Z., W. Zhang, M. F. Stuecker, H. Xu, F.-F. Jin, and C. Liu, 2019: Different effects of two ENSO types on Arctic surface temperature in boreal winter. J. Climate, 32, 4943–4961, https://doi.org/10.1175/JCLI-D-18-0761.1.
Lindsay, R. W., and J. Zhang, 2005: The thinning of Arctic sea ice, 1988–2003: Have we passed a tipping point? J. Climate, 18, 4879–4894, https://doi.org/10.1175/JCLI3587.1.
Liu, J., J. A. Curry, and Y. Hu, 2004: Recent Arctic sea ice variability: Connections to the Arctic Oscillation and the ENSO. Geophys. Res. Lett., 31, L09211, https://doi.org/10.1029/2004GL019858.
Maher, N., and Coauthors, 2019: The Max Planck Institute Grand Ensemble: Enabling the exploration of climate system variability. J. Adv. Model. Earth Syst., 11, 2050–2069, https://doi.org/10.1029/2019MS001639.
McCrystall, M. R., and J. A. Screen, 2021: Arctic winter temperature variations correlated with ENSO are dependent on coincidental sea ice changes. Geophys. Res. Lett., 48, e2020GL091519, https://doi.org/10.1029/2020GL091519.
McFarlane, N. A., G. J. Boer, J.-P. Blanchet, and M. Lazare, 1992: The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5, 1013–1044, https://doi.org/10.1175/1520-0442(1992)005<1013:TCCCSG>2.0.CO;2.
North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sampling errors in the estimation of empirical orthogonal functions. Mon. Wea. Rev., 110, 699–706, https://doi.org/10.1175/1520-0493(1982)110<0699:SEITEO>2.0.CO;2.
Notz, D., and J. Stroeve, 2016: Observed Arctic sea-ice loss directly follows anthropogenic CO2 emission. Science, 354, 747–750, https://doi.org/10.1126/science.aag2345.
O’Reilly, C. H., 2018: Interdecadal variability of the ENSO teleconnection to the wintertime North Pacific. Climate Dyn., 51, 3333–3350, https://doi.org/10.1007/s00382-018-4081-y.
Park, H.-S., A. L. Stewart, and J.-H. Son, 2018: Dynamic and thermodynamic impacts of the winter Arctic Oscillation on summer sea ice extent. J. Climate, 31, 1483–1497, https://doi.org/10.1175/JCLI-D-17-0067.1.
Peng, G., W. N. Meier, D. J. Scott, and M. H. Savoie, 2013: A long-term and reproducible passive microwave sea ice concentration data record for climate studies and monitoring. Earth Syst. Sci. Data, 5, 311–318, https://doi.org/10.5194/essd-5-311-2013.
Perovich, D. K., T. C. Grenfell, B. Light, and P. V. Hobbs, 2002: Seasonal evolution of the albedo of multiyear Arctic sea ice. J. Geophys. Res., 107, 8044, https://doi.org/10.1029/2000JC000438.
Polvani, L. M., M. Previdi, M. R. England, G. Chiodo, and K. L. Smith, 2020: Substantial twentieth-century Arctic warming caused by ozone-depleting substances. Nat. Climate Change, 10, 130–133, https://doi.org/10.1038/s41558-019-0677-4.
Rigor, I. G., and J. M. Wallace, 2004: Variations in the age of Arctic sea-ice and summer sea-ice extent. Geophys. Res. Lett., 31, L09401, https://doi.org/10.1029/2004GL019492.
Rigor, I. G., J. M. Wallace, and R. L. Colony, 2002: Response of sea ice to the Arctic Oscillation. J. Climate, 15, 2648–2663, https://doi.org/10.1175/1520-0442(2002)015<2648:ROSITT>2.0.CO;2.
Rodgers, K. B., J. Lin, and T. L. Frölicher, 2015: Emergence of multiple ocean ecosystem drivers in a large ensemble suite with an Earth system model. Biogeosciences, 12, 3301–3320, https://doi.org/10.5194/bg-12-3301-2015.
Sardeshmukh, P. D., and B. J. Hoskins, 1988: The generation of global rotational flow by steady idealized tropical divergence. J. Atmos. Sci., 45, 1228–1251, https://doi.org/10.1175/1520-0469(1988)045<1228:TGOGRF>2.0.CO;2.
Screen, J. A., and C. Deser, 2019: Pacific Ocean variability influences the time of emergence of a seasonally ice-free Arctic Ocean. Geophys. Res. Lett., 46, 2222–2231, https://doi.org/10.1029/2018GL081393.
Stroeve, J. C., J. Maslanik, M. C. Serreze, I. Rigor, W. Meier, and C. Fowler, 2011: Sea ice response to an extreme negative phase of the Arctic Oscillation during winter 2009/2010. Geophys. Res. Lett., 38, L02502, https://doi.org/10.1029/2010GL045662.
Sun, L., M. Alexander, and C. Deser, 2018: Evolution of the global coupled climate response to Arctic sea ice loss during 1990–2090 and its contribution to climate change. J. Climate, 31, 7823–7843, https://doi.org/10.1175/JCLI-D-18-0134.1.
Svendsen, L., N. Keenlyside, I. Bethke, Y. Gao, and N.-E. Omrani, 2018: Pacific contribution to the early twentieth-century warming in the Arctic. Nat. Climate Change, 8, 793–797, https://doi.org/10.1038/s41558-018-0247-1.
Swart, N., 2017: Natural causes of Arctic sea-ice loss. Nat. Climate Change, 7, 239–241, https://doi.org/10.1038/nclimate3254.
Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25, 1297–1300, https://doi.org/10.1029/98GL00950.
Tokinaga, H., S.-P. Xie, and H. Mukougawa, 2017: Early 20th-century Arctic warming intensified by Pacific and Atlantic multidecadal variability. Proc. Natl. Acad. Sci. USA, 114, 6227–6232, https://doi.org/10.1073/pnas.1615880114.
Trenberth, K. E., and J. W. Hurrell, 1994: Decadal atmosphere-ocean variations in the Pacific. Climate Dyn., 9, 303–319, https://doi.org/10.1007/BF00204745.
Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N.-C. Lau, and C. Ropelewski, 1998: Progress during TOGA in understanding and modeling global teleconnections associated with tropical sea surface temperatures. J. Geophys. Res., 103, 14 291–14 324, https://doi.org/10.1029/97JC01444.
Vargas Zeppetello, L. R. V., A. Donohoe, and D. S. Battisti, 2019: Does surface temperature respond to or determine downwelling longwave radiation? Geophys. Res. Lett., 46, 2781–2789, https://doi.org/10.1029/2019GL082220.
Wilks, D. S., 2006: On “field significance” and the false discovery rate. J. Appl. Meteor. Climatol., 45, 1181–1189, https://doi.org/10.1175/JAM2404.1.
Wilks, D. S., 2016: “The stippling shows statistically significant grid points”: How research results are routinely overstated and overinterpreted, and what to do about it. Bull. Amer. Meteor. Soc., 97, 2263–2273, https://doi.org/10.1175/BAMS-D-15-00267.1.
Williams, J., B. Tremblay, R. Newton, and R. Allard, 2016: Dynamic preconditioning of the minimum September sea-ice extent. J. Climate, 29, 5879–5891, https://doi.org/10.1175/JCLI-D-15-0515.1.
Yeh, S.-W., and Coauthors, 2018: ENSO atmospheric teleconnections and their response to greenhouse gas forcing. Rev. Geophys., 56, 185–206, https://doi.org/10.1002/2017RG000568.
Yuan, X., M. R. Kaplan, and M. A. Cane, 2018: The interconnected global climate system—A review of tropical–Polar teleconnections. J. Climate, 31, 5765–5792, https://doi.org/10.1175/JCLI-D-16-0637.1.
Zhang, X., 2010: Sensitivity of arctic summer sea ice coverage to global warming forcing: Towards reducing uncertainty in arctic climate change projections. Tellus, 62A, 220–227, https://doi.org/10.1111/j.1600-0870.2010.00441.x.
Zhu, Y., and T. Wang, 2016: The relationship between the Arctic Oscillation and ENSO as simulated by CCSM4. Atmos. Ocean. Sci. Lett., 9, 198–203, https://doi.org/10.1080/16742834.2016.1149287.