Monthly Sea Surface Temperature, Sea Ice, and Sea Level Pressure over 1850–2023 from Coupled Data Assimilation

When using this dataset, please cite:Cooper, V. T., G. J. Hakim, and K. C. Armour, 2025: Monthly Sea-Surface Temperature, Sea Ice, and Sea-Level Pressure over 1850–2023 from Coupled Data Assimilation. Journal of Climate. https://doi.org/10.1175/JCLI-D-25-0021.1. Brief dataset description: Reconstruction of fully gridded, monthly mean sea-surface temperature (SST), sea-ice concentration (SIC), sea-level pressure (SLP), and near-surface air temperature (T) over 1850-2023. The reconstruction method is "strongly coupled online data assimilation" with eight different forecast models used as model priors. Each forecast model is a "linear inverse model" trained on one of the CMIP6-class coupled climate models. The assimilation step uses observations of SST from HadSST4, marine SLP from ICOADS, terrestrial T from CRUTEM5, and satellite-era SIC from the NSIDC CDR. The dataset provided here includes 200 (of the 1600) individual ensemble members, the ensemble means using each of the eight forecast models, and the grand mean across all 1600 ensemble members. All .nc files are on a 96x144 lat-lon grid (nominal 2 degree).All variables are anomalies from the 1961-1990 climatological mean. The SST component of the reconstruction could be compared with other fully gridded datasets, such as NOAA ERSST, Met Office Hadley Centre HadISST, COBE-SST2, etc. File descriptions Files titled 'grandmean' and 'ensmeans': 1) The 'grand mean' across all eight model priors (i.e., the mean of the eight ensemble means). The grand mean is the best estimate of the state at any time, but it should not be used to assess variability over time; the variability of the mean will be damped (see the ens_members files for assessing temporal variability). The grand mean corresponds to the mean of all 1600 ensemble members. 2) The 'ensemble mean' for each of the eight model priors. These demonstrate the spread arising from differences in model physics across the eight forecast models. The ensemble means correspond to the mean of the 200 ensemble members associated with each prior. Files titled 'ensmembers': 3) A subset of 200 'ensemble members' (of the 1600 total ensemble members), equally weighted across the eight model priors. There are 25 ensemble members from each prior instead of the full 200 members because of limits on the size of the repository. The ensemble members account for uncertainty in SST bias corrections, as provided by the HadSST4 ensemble. When assessing temporal variability (e.g., computing the ENSO power spectrum) or uncertainty in trends, the ensemble members should be used rather than the ensemble/grand mean. Additional file added with version 1.0.14) 'ensmembers-subset_siconc-actuals-extent_1850-2023.nc' contains actuals (rather than anomalies) of monthly gridded ice concentration for 14 ensemble members. These 14 members have been selected for "amip-type" simulations because they approximately span the uncertainty range in the full ensemble. This file has been included to facilitate calculation of ice extent. However, as noted in the accompanying manuscript, caution is advised when comparing ice extent across different datasets due to the strong influence of the grid/masking. Version tracker: 1.0.0: Original submission. 1.0.1: Added 'ensmembers-subset_siconc-actuals-extent_1850-2023.nc' file with subset of ice concentration actuals. No changes to previously uploaded files. Note: A version of the SST and sea ice dataset designed for use as the lower boundary condition in atmospheric general circulation models will be released at a later date. Please contact [e-mail] if you are interested and that dataset has not yet been posted online.