The indirect effect of winter Arctic Oscillation (AO) events on the following summer Arctic sea ice extent suggests an inherent winter-to-summer mechanism for sea ice predictability. On the other hand, operational regional summer sea ice forecasts in a large number of coupled climate models show a considerable drop in predictive skill for forecasts initialised prior to the date of melt onset in spring, suggesting that some drivers of sea ice variability on longer timescales may not be well represented in these models. To this end, we introduce an unsupervised learning approach based on cluster analysis and complex networks to establish how well the latest generation of coupled climate models participating in phase 6 of the World Climate Research Programme Coupled Model Intercomparison Project (CMIP6) are able to reflect the spatio-temporal patterns of variability in Northern Hemisphere winter sea-level pressure and Arctic summer sea ice concentration over the period 1979–2020, relative to ERA5 atmospheric reanalysis and satellite-derived sea ice observations, respectively. Two specific global metrics are introduced as ways to compare patterns of variability between models and observations/reanalysis: the adjusted Rand index – a method for comparing spatial patterns of variability – and a network distance metric – a method for comparing the degree of connectivity between two geographic regions. We find that CMIP6 models generally reflect the spatial pattern of variability in the AO relatively well, although they overestimate the magnitude of sea-level pressure variability over the north-western Pacific Ocean and underestimate the variability over northern Africa and southern Europe. They also underestimate the importance of regions such as the Beaufort, East Siberian, and Laptev seas in explaining pan-Arctic summer sea ice area variability, which we hypothesise is due to regional biases in sea ice thickness. Finally, observations show that historically, winter AO events (negatively) covary strongly with summer sea ice concentration in the eastern Pacific sector of the Arctic, although now under a thinning ice regime, both the eastern and western Pacific sectors exhibit similar behaviour. CMIP6 models however do not show this transition on average, which may hinder their ability to make skilful seasonal to inter-annual predictions of summer sea ice.
Arctic sea ice is a key component of the polar climate system, acting as a barrier which both reflects incoming solar radiation and regulates the rate of energy exchange between the atmosphere and ocean. Over the past four decades, however, it can be seen as a direct barometer for climate change, having suffered significant losses in areal extent across all seasons
For analysing summer sea ice concentration variability in the observations, we compute the average of monthly mean June, July, August, and September (JJAS) sea ice concentration fields between 1979 and 2020 from three separate observational data sets based on the series of multi-frequency passive microwave satellite observations since October 1978. These include the National Snow and Ice Data Center (NSIDC) NASA Team
The AO is typically defined as the leading mode of variability in mean sea-level pressure data north of 20
CMIP6 models used in this study.
We assess the seasonal patterns of variability in sea ice concentration and mean sea-level pressure outputs from 31 different GCMs participating in CMIP6. In order to compare with recent observations, we combine monthly averaged model outputs from historical runs (1979–2014) with ScenarioMIP run SSP5-8.5
Generating complex networks here follows the methodology from previous studies (
Complex network of DJFM mean sea-level pressure from ERA5, computed between 1979 and 2020. The covariance-based link weights are computed from Eq. (
The standardised (Std.) temporal component of DJFM mean sea-level pressure (SLP) variability from ERA5 (dashed curve). Temporal components are computed via Eq. (
In general, we can consider a network as a group of vertices, or
Figure
Complex networks of JJAS sea ice concentration for
In Fig.
Before we introduce the two metrics which are used to compare similarities between complex networks, it is worth saying a few words about what information we can expect to obtain when comparing models and observations/reanalysis. Due to the fact that any CMIP6 model ensemble member is in its own phase of internal variability (e.g.
Synthetic example of cell clusters to illustrate the concept of the Rand index (see Sect.
The adjusted Rand index (ARI;
ARI and
Winter sea-level pressure networks from
The network distance metric (
For every available ensemble member from each of the CMIP6 models outlined in Table
In Fig.
In this section we compute individual complex networks of JJAS sea ice concentration over the period 1979–2020 for every available ensemble member from each of the CMIP6 models and then compute ARI and
ARI and
Summer sea ice concentration networks from
In Fig.
Network link weight between the DJFM ERA5 “AO node” (dashed time series from Fig.
We now turn to an investigation of the winter AO to summer sea ice teleconnection. We begin by illustrating how we can use the network framework to exploit this relationship in the observational and reanalysis data by effectively considering both winter sea-level pressure and summer sea ice concentration networks as individual layers within a
In this section we use the temporal component of variability associated with the “AO node” of the ERA5 sea-level pressure network to define the time series corresponding to the winter AO (i.e. the dashed time series in Fig.
For each CMIP6 model ensemble member we extract the temporal component from the node with the highest strength of each winter sea-level pressure network and compute the covariance-based link weight with each node of its respective summer sea ice concentration network. In Fig.
ARI and
Covariance-based link weights between the winter AO node time series and each node of the summer sea ice concentration network between 1979 and 2020 for
The average covariance-based link weights between the winter AO node time series and each node of the summer sea ice concentration networks across both the observations
In Fig.
The average summer sea ice thickness from 25 CMIP6 models (49 realisations) and PIOMAS in the Laptev Sea
Average winter AO to summer sea ice teleconnection for 15 CMIP6 model ensemble members with the lowest average root mean square error (RMSE) in mean sea ice thickness relative to PIOMAS in the East Siberian, Laptev, and Beaufort seas.
In this study we used a combination of cluster analysis and complex networks to derive spatio-temporal patterns of variability in Northern Hemisphere winter sea-level pressure and Arctic summer sea ice concentration over the period 1979–2020 and to subsequently understand the spatio-temporal network connectivity between the winter Arctic Oscillation (AO) and summer sea ice cover over the same period. We analysed these patterns in both satellite observational data sets and ERA5 atmospheric reanalysis and also from 31 of the latest generation general circulation models (GCMs) participating in the most recent phase of the Coupled Model Intercomparison Project (CMIP6). We also introduced two global metrics for comparing patterns of variability between two networks: the adjusted Rand index
The following repository contains Python code written by William Gregory, which can be used to access and download CMIP6 data volumes, as well as to perform the complex networks analysis of all data types:
The satellite-derived sea ice concentration data are available from NSIDC and OSI-SAF at the following locations: NASA Team (
The ERA5 atmospheric reanalysis data are available from
PIOMAS sea ice thickness data are available from
CMIP6 model ensemble members used in this study were hosted on the JASMIN UK supercomputer; however these same files can be downloaded directly from
The supplement related to this article is available online at:
WG developed the networks code and subsequently ran the analysis for all the observational, reanalysis, and model data. JS originally suggested the idea of applying the complex networks framework to analysing sea ice teleconnections in CMIP6 models and also provided technical support and input for this paper. MT provided technical support and input for this paper. All co-authors contributed to all sections of the paper.
At least one of the (co-)authors is a member of the editorial board of
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Michel Tsamados acknowledges support from the NERC “PRE-MELT” (grant NE/T000546/1) and “EXPRO+ Snow” (ESA AO/1-10061/19/I-EF) projects.
William Gregory received funding from the UK Natural Environment Research Council (NERC) (grant NE/L002485/1).
This paper was edited by Yevgeny Aksenov and reviewed by Céline Gieße and Robin Clancy.