Input and output data for: Climate-informed cryospheric reanalysis via hierarchical Bayesian data assimilation

This record contains all input data (inputs.zip) needed to run the experiments described in the manuscript "Climate-informed cryospheric reanalysis via hierarchical Bayesian data assimilation", as well as all output files generated by these experiments (results.zip). Using the associated code repository, crane, available on GitHub, the input data can be used to rerun the experiments and subsequently reproduce all results, tables, and most figures in the manuscript, or to run completely new experiments. In addition, the output data can be used to directly reproduce the figures and tables without rerunning any experiments. The input data in inputs.zip has subdirectories structured according to the type of data, either seasonal snow data in snow or glacier data in glacier, and further subdirectory splits according to study sites. In the seasonal snow subdirectory, there are four snow study sites (in separate subdirectories): Abisko (AK), Filefjell (FF), Hornsund (HS), and Weissfluhjoch (WF). For each snow site, we provide daily topographically-downscaled ERA5 forcing data, in-situ SWE data, and MODIS fractional snow-covered area (FSCA) data from the ESA Snow_cci for the encompassing 0.01° (approximately 1 km) pixel. In the glacier subdirectory, there are four glaciers (in separate subdirectories): Austre Brøggerbreen (AB), Claridenfirn (CD), Storbrean (SB), and Storglaciären (SG). For each glacier, we provide daily topographically-downscaled ERA5 forcing data, glacier-wide mass balance data, and ancillary data from RGI7. The output data in results.zip contains results for each experiment and site that we conducted. Three experiments were conducted: in-situ SWE data assimilation at snow sites, MODIS FSCA data assimilation at snow sites, and glacier-wide mass balance data assimilation for the glaciers. Results include reanalyses from each experiment and site. For example, Snow_AK_D1_F0_LO1.mat contains prior and posterior snow reanalyses in a structure (r) for Abisko (AK) assimilating in-situ SWE (D1) with a withholding data (a leave out LO1) period, obtained via complete pooling (cp), no pooling (np), partial pooling with nested particle smoothers (pp), approximate partial pooling via type-II MAP (map), and intensive partial pooling via PMCMC (mcmc) which are all stored in substructures together with the parameter settings (p), and observations (obs) used. Similarly, Snow_FF_D0_F1_LO0.mat contains prior and posterior snow reanalyses in a structure (r) for Filefjell (FF) assimilating MODIS FSCA (F1) with no artificially imposed withholding of data (no leave out LO0) period, obtained in the same way as the previous example with the same substructures. Likewise, Glacier_CD_LO1.mat contains prior and posterior glacier reanalyses in a structure (r) for Claridenfirn (CD) assimilating glacier-wide mass balance data with a withholding data (a leave out LO1) period, obtained in the same way and with the same substructures as above. For the in-situ SWE and glacier data assimilation experiments we ran experiments both with and without a withholding (leave out) period to test the effects of climate-informed reanalysis in an unseen validation period by seeing if hierarchical inference can help learn the parameter climatology. For the FSCA data assimilation experiment this was not done, as assimilation data was only available for the 21st century MODIS-era which created a natural validation/calibration split. More details are provided in the paper, and examples of how to use these output files are provided in the associated code repository. The output also includes the full PMCMC results for each experiment and study site stored in the corresponding chains*.mat files. We do not recommend rerunning these Markov chains unless you have a lot of time on your hands or an idling server at your disposal; each site takes approximately 12 hours running 10 chains in parallel on 10 cores. When all methods but PMCMC are used (so the PMCMC switch is off), all experiments take only a few minutes per site to run on a standard laptop. All input and output files are provided as binary .mat files (v5) containing structures produced in MATLAB. These can be read natively in MATLAB, but also in programming languages such as Python via the routine scipy.io.loadmat. We gratefully acknowledge all data providers that we relied on when producing the input data. The FSCA data, specifically ESA Snow_cci MODIS SCFG v4.0 satellite retrievals, are available via https://archive.ceda.ac.uk/. In-situ data from the seasonal snow sites were provided by: Institute of Geophysics Polish Academy of Sciences for Hornsund on request, SMHI for Abisko via https://opendata.smhi.se, NVE for Filefjell on request, and SLF for Weissfluhjoch on request. Glacier-wide mass balance data were provided by: NPI for Austre Brøggerbreen via https://mosj.no/en/, Stockholm University for Storglaciären via https://bolin.su.se/data/v7/, NVE for Storbrean via https://glacier.nve.no/Glacier/viewer/CI/en/, GLAMOS for Claridenfirn via https://www.glamos.ch, and the WGMS which synthesized data for all these glaciers through the FoG database via https://doi.org/10.5904/wgms-fog-2025-02b. All experiments were forced by topographically downscaling ERA5 reanalysis data (1940-2024) obtained from the Copernicus Climate Change Service (C3S) Climate Data Store https://cds.climate.copernicus.eu/. These data were generated using modified Copernicus Climate Change Service information. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains. Please get in touch with [e-mail] in case of any questions.