Methodology:

Model data

The tables contain the model names for all models included in the analysis, and the headers reference the CMIP experiment or forcing scenario from which data was taken. The table also lists additional details of the runs used. The details of the experiment protocol for CMIP5 and CMIP6 are given in Taylor et al. (2012) and O'Neill et al (2016) respectively:

Taylor, K. E., Stouffer, R. J., & Meehl, G. A. (2012). An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society, 93(4), 485-498. https://doi.org/10.1175/BAMS-D-11-00094.1

O'Neill, B. C., Tebaldi, C., Vuuren, D. P. V., Eyring, V., Friedlingstein, P., Hurtt, G., et al. (2016). The scenario model intercomparison project (ScenarioMIP) for CMIP6. Geoscientific Model Development, 9(9), 3461-3482. https://doi.org/10.5194/gmd-9-3461-2016

Sea ice Concentration data from satellites

'NASA-Team v1.1': Cavalieri, D. J., Parkinson, C. L., Gloersen, P., & Zwally, H. J. (1996), updated yearly. Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS Passive Microwave Data, Version 1.1 [Dataset]. NASA National Snow and Ice Data Center Distributed Active Archive Center. https://doi.org/10.5067/8GQ8LZQVL0VL. [Accessed Apr 1 2021].

"Bootstrap v3.1": Comiso, J. C. (2017). Bootstrap Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS, Version 3.1 [Dataset]. NASA National Snow and Ice Data Center Distributed Active Archive Center. https://doi.org/10.5067/7Q8HCCWS4I0R. [Accessed Apr 1 2021].

Processing

CMIP data was accessed from the CEDA archives of CMIP5 and CMIP6 data on the UK computing resource JASMIN. CMIP5 data was accessed in July 2020; CMIP6 data was accessed in November 2020.

SIA was calculated from the CMIP5 variable 'sic' or CMIP6 variable 'siconc', or, where available, is the CMIP6 variable 'siareas'. Global mean surface air temperature was calculated from CMIP variable 'tas'.

We define an 'exceedance year' as the year in which a model's 5-year trailing average February SIA drops below 0.1Mkm2 . Exceedance year is calculated separately for historical and scenario runs, and the exceedance year taken as the minimum of these values. Due to the use of the trailing average, there are no exceedances in the first four years of the historical (1950-1953) or future (2015-2018) analysis.

Historical climatologies (SIAhist) use the period 1979-2014; CMIP5 historical runs end in 2005, so climatologies also use data from rcp45 or, where this is unavailable, rcp85. End-of-the-century climatologies, to calculate the change in SIA or TAS (dSIA and dTAS in data table), use the period 2081-2100.

Ensemble linear regressions were performed (using python's statsmodels ordinary least squares method). Constrained projections were then produced using these linear regression models and one of the observation-derived SIA datasets as the predictor.

Other data

Equilibrium Climate Sensitivity (ECS) and Transient Climate Response (TCR) data from Meehl et al. (2020):

Meehl, G. A., Senior, C. A., Eyring, V., Flato, G., Lamarque, J. F., Stouffer, R. J., et al. (2020). Context for interpreting equilibrium climate sensitivity and transient climate response from the CMIP6 Earth system models. Science Advances, 6(26). https://doi.org/10.1126/sciadv.aba1981

Data description: regression statistics

For each combination of forcing scenario, dataset (i.e. CMIP5 or CMIP6 or special subsets therein) and season (February or September) and for each pair of variables named ''''''xname' and 'yname', the datafile regression_statistics.csv contains full regression statistics:

* ensemble minimum, mean and max (as in Holmes et al. (2022) Table 1) and standard deviation of each variable

* r2 and its p-value, regression slope and intercept

* Where relevant, the constrained pr...(8)

Data collection:

The data is aggregated over the Southern Ocean, derived from CMIP models at various spatial resolution.

Data quality:

Missing values in one of the future scenario run columns indicate that the given model was not run for that scenario, or, that data was not available to the authors.