Probabilistic Machine Learning Estimation of Ocean Mixed Layer Depth from Dense Satellite and Sparse In-Situ Observations: Preprocessed Satellite and In-situ observation datasets

This record includes all of the prepared data used in the manuscript, "Probabilistic Machine Learning Estimation of Ocean Mixed Layer Depth from Dense Satellite and Sparse In-Situ Observations" (citation information forthcoming). As a part of this manuscript, we analyzed the ability for machine learning models to extract sea surface information (from salinity, temperature, sea height anomaly) to predict mixed layer depth. In this manuscript there are two experimental datasets: (1) info derived from CESM POP2 ocean model dataset (1989-1998), and (2) info derived from a combination of satellite sources and MLD from Argo profiles. More details below. 

All of these data files are preprocessed and organized to be used with the ml-ocean-bl github code found at https://github.com/NCAR/ml-ocean-bl/mloceanbl/.

• CESM POP2 Ocean model dataset

Preprocessed sea surface salinity (SSS), temperature (SST), sea surface height anomalies (SSH), and ocean mixed layer depth (MLD, or HMXL) derived from the CESM POP2 Ocean model. Specifically, CESM POP2 model in a hindcast forced by JRA55do atmospheric reanalysis from 1958 to present and initialized with an oceanic climatology as in e.g. Deppenmeier et al. (2021). The model outputs include the ocean mixed layer depth (MLD), sea surface salinity (SSS), sea surface temperature (SST), and sea height anomaly (SSH) at a temporal frequency of 5-days and an approximate latitude and longitude resolution of 0.1 degrees.

Relevant files:

1. full_EPO.nc, full_SIO.nc
   • NetCDF4 containing SSS, SST, SSH, MLD for the equatorial Pacific Ocean (EPO) and southern Indian Ocean (SIO) (see manuscript for details). Data is regridded onto a 1/2 degree lat/lon 5 day grid to correspond with data used for Argo datasets (see below).
2. clim_EPO.nc, clim_SIO.nc, clim_std_EPO.nc, std_clim_EPO.nc, std_clim_SIO.nc
   • NetCDF4 containing mean and standard deviation climatologies of SSS, SST, SSH, and MLD for EPO and SIO.
3. std_anomalies_EPO.nc, std_anomalies_SIO.nc
   • NetCDF4 containing SSS, SST, SSH, and MLD standardized anomalies for EPO and SIO. This is the dataset directly used for training in aforementioned manuscript. Use with ml-ocean-bl/ml-ocean-test/data. 

• Satellite and Argo datasets

Preprocessed satellite sea surface salinity (SSS), temperature (SST), and sea surface height anomalies (SSH) and Argo-based mixed layer depth (MLD) profiles. Original data can be found at:

(SST): Remote Sensing Systems. 2017. MW optimum interpolated SST data set. Ver. 5.0. PO.DAAC, CA, USA.  Further information available at at https://doi.org/10.5067/GHMWO-4FR05. Data can be accessed at https://podaac-tools.jpl.nasa.gov/drive/files/allData/ghrsst/data/GDS2/L4/GLOB/REMSS/mw_OI/v5.0/.

(SSS): Oleg Melnichenko. 2018. Aquarius L4 Optimally Interpolated Sea Surface Salinity. Ver. 5.0. PO.DAAC, CA, USA. Further information at https://doi.org/10.5067/AQR50-4U7CS. Data can be accessed at https://podaac-tools.jpl.nasa.gov/drive/files/SalinityDensity/aquarius/L4/IPRC/v5/7day. 

(SSH): Zlotnicki, Victor; Qu, Zheng; Willis, Joshua. 2019. SEA_SURFACE_HEIGHT_ALT_GRIDS_L4_2SATS_5DAY_6THDEG_V_JPL1609. Ver. 1812. PO.DAAC, CA, USA. Information available at https://doi.org/10.5067/SLREF-CDRV2. Data can be accessed at https://podaac-tools.jpl.nasa.gov/drive/files/SeaSurfaceTopography/merged_alt/L4/cdr_grid

(MLD) Argo-based ocean surface mixed layer depths using the buoyancy gradient definition of Whitt Nicholson and Carranza (2019) processed dataset available at https://doi.org/10.5281/zenodo.4291175.

Relevant files:

1. https://github.com/NCAR/ml-ocean-bl/mloceanbl/preprocess_mld.py and .../preprocess_sss_sst_ssh.py.
   • Preprocessing code
2. sss_sst_ssh_anomalies.nc.
   • Regridded and resampled SSS, SST, SSH onto a 1/2 degree lat/lon 7day grid. Contains preprocessed seasonal data along with anomalies.
3.  mldb_climatology_climatologystd_binned.nc
   • Smoothed argo-based mixed layer depths are used to calculate climatologies and standardized climatologies. 4 degree lat/lon gridded climatologies.
4. mldb_full_anomalies_stdanomalies_climatology_stdclimatology.nc
   • Contains the Argo profile-derived MLD, anomalies, standard anomalies, climatologies, and standardized climatologies with corresponding argo locations, times, and corresponding weeks. 
5. equatorial_pacific_model_oi_re.nc,  southern_indian_model_oi_re.nc
   • Model outputs for the Equatorial Pacific Ocean and Southern Indian Ocean. These gridded files contain the model outputs (vlcnn, vlcnn variance, OI, OI variance, reanalysis, and reanalysis variance - see manuscript for nomenclature details) at each of the 200 weeks available. It should be noted that, in the equatorial Pacific Ocean, the lat/lon location of (-138.75, -9.75) is masked during the training and filled with a NaN in the .nc files. 

Contact D. Foster with any questions.