Trained model coefficients, validation and testing data, and predictions, for the modeled driving of low-frequency regional Arctic sea ice concentration variability by large-scale climate modes, 1920-2022

This dataset contains data describing the relationship between global climate modes of variability and regional Arctic sea ice concentration anomalies lagged by 2-20 years. Data originates from global climate models from the Coupled Climate Model Intercomparison Project - Phase 6 (CMIP6), which are freely available from the Earth System Grid Federation (ESGF). This data from CMIP6 focuses on the historical simulation period of 1920-2014, but also leverages pre-industrial control simulations. Climate mode of variability data is obtained using the Climate Variability Diagnostics Package datasets (doi:10.1002/2014EO490002). Observational Arctic sea ice concentration data are obtained from the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST1), doi:10.1029/2002JD002670.
Firstly, this dataset firstly provides coefficients for the linear model relating standardized climate modes of variability with regional Arctic sea ice concentration (SIC) anomalies for 42 individual large ensembles and multi-model large ensembles, and is recorded in the data file "Linear_model_weights.nc". These linearly modeled relationships (found via machine learning) are able to be compared against a different method of linking these phenomena using instances of extreme SIC anomalies correlated with standardized climate modes of variability, as is recorded in the file "Binary_correlations_SIC_anomalies.nc".
The performance of the linear model (and three other neural network configurations) is evaluated by correlating the predicted SIC anomalies using climate mode of variability inputs into the linear model against the known SIC anomalies and is recorded in the files "ML_validation_correlation_coefficients.nc" for the separated 15% validation data, "LE_test_ensemble_member_correlations.nc" for the remaining 10% of the data, and "MMLE_3_test_ensemble_member_correlations.nc" for all remaining unseen large ensemble members. The correlations using the linear model can be compared with the correlation obtained from SIC anomaly persistence, as a measure of the linear model's skill, as recorded in the file "Persistence_Pearson_correlations.nc".
Finally, the skillful linear model can be used to make predictions into the future based on observed climate modes of variability, as recorded in the file "Linear_model_predictions_using_observations.nc". The historical performance of the linear model predicting SIC anomalies, based solely on observed climate modes of variability, is evaluated by correlating the observed SIC anomalies with those predicted by the linear model, as recorded in the file "Linear_model_prediction_vs_observations_correlations.nc".
This dataset was created to record the influence of large-scale climate modes of variability on regional Arctic sea ice concentration anomalies and is used in the article C. Wyburn-Powell & Jahn A. (2024), Large-Scale Climate Modes Drive Low-Frequency Regional Arctic Sea Ice Variability, Journal of Climate, https://doi.org/10.1175/JCLI-D-23-0326.1. This work was conducted at the University of Colorado Boulder from 2022-2024. The figures from the Journal of Climate article can be reproduced from the datasets provided here. The code used to create the datasets is archived with Zenodo and can be located at https://doi.org/10.5281/zenodo.12580233