Emulating present and future simulations of melt rates at the base of Antarctic ice shelves with neural networks

This dataset contains the data and scripts for the publication "Emulating present and future simulations of melt rates at the base of Antarctic ice shelves with neural networks" in Journal of Advances in Modeling Earth Systems.

Before going into details, here is a reminder that the NEMO runs for the training dataset are called 'OPM+number'. These are the corresponding names given in the manuscript: OPM006=HIGHGETZ, OPM016=WARMROSS, OPM018=COLDAMU and OPM021=REALISTIC. For the testing dataset: 'bf663' is the REPEAT1970 run and 'bi646' is the 4xCO2 run.

Most of the formatting and preprocessing of the training data has been made for Burgard et al. 2022. The raw and to-some-degree processed data can therefore be found here: https://doi.org/10.5281/zenodo.7308352. The raw data for the testing dataset is from Smith et al. 2021, you can find it here: https://doi.org/10.5281/zenodo.7886986

The following folders and files can be found here:

===============raw/

Some geometrical files needed for initial data formatting and masking.

===============interim/

ANTARCTICA_IS_MASKS/ (from INTERIM_ANTARCTICA_IS_MASKS.zip): contains

masks and geometric information for the testing dataset to be included in the input file of the neural network and for the classic parameterisations.

local bedrock and ice meridional and zonal slopes

BOXES/ (from INTERIM_BOXES.zip): contains variables needed to apply the box parameterisation for the testing dataset

PLUMES/ (from INTERIM_PLUMES.zip): contains the variables needed to apply the plume parameterisation for the testing dataset

SMITH_bf663/ and SMITH_bi646/ (from INTERIM_SMITH*.zip): for testing dataset,

corrected_draft_bathy_isf.nc: file containing ice draft and bathymetry corrected by ice draft concentration to account for the biased draft and bathymetry at the grounding line resulting from the interpolation from the native NEMO grid to the stereographic grid (values under ice shelf and NaNs over land

custom_lsmask_Ant_stereo_clean.nc: land-sea mask giving 0 = ocean, 1 = shelf, 2 = land

isfdraft_conc_Ant_stereo.nc: ice-shelf concentration resulting from the interpolation from the native NEMO grid to the stereographic grid

other_mask_vars_Ant_stereo.nc: contains other variables used for the masks

the reference melt: 1D containing integrated melt, 2D containing melt fields, box1 containing melt near the grounding line

T_S_PROF/ (from INTERIM_T_S_PROF.zip)

Mean profiles used as input for traditional parameterisations

T and S 2D fields, extrapolated from the mean profiles to the local ice draft depth (needed as input for the neural network)

Fields of mean and standard deviation T and S for all points (needed as input for the neural network)

NN_MODELS/ (from INTERIM_NN_MODELS.zip) contains all neural networks trained for this paper (for the cross validation and over the whole dataset for testing)

INPUT_DATA/ (from INTERIM_INPUT_DATA.zip) contains all input csv files containing the input datasets for the different training and testing iterations. Also contains the metrics to normalise the input. For the cross-validation, the input csv files are not included because they are too large. However, they can be reconstructed from the individual files for ice shelves and time blocks. The metrics to normalise the data during the cross-validation are included in EXTRAPOLATED_ISFDRAFT_CHUNKS_CV!

===============processed/MELT_RATE/

Contains resulting melt rates

CV_ISF : Cross-validation results over ice shelves

CV_TBLOCKS : Cross-validation results over time

SMITH_bf663 : Neural network results for REPEAT1970

SMITH_bf663_CLASSIC  : "Traditional" parameterisation results for REPEAT1970

SMITH_bi646 : Neural network results for 4xCO2

SMITH_bi646_CLASSIC: "Traditional" parameterisation results for 4xCO2

=====================

The explanation around the scripts can be found in README.rst with the scripts in scripts_paper_simpleNN_basal_melt.zip.Note that these are the scripts needed to produce the results in the paper. You can also find them on Github: https://github.com/ClimateClara/scripts_paper_simpleNN_basal_melt, find the most up-to-date version of the package 'multimelt' here: https://github.com/ClimateClara/multimelt and a version you can install via pip here: https://pypi.org/project/multimelt/

Finally, if anything is unclear, check out the "Methods" section of the paper: https://doi.org/10.1029/2023MS003829