This repository contains codes and sample notebooks for downloading and processing the SSH mapping data challenge.

The quickstart can be run online by clicking here:

The goal is to investigate how to best reconstruct sequences of Sea Surface Height (SSH) maps from partial satellite altimetry observations. This data challenge follows an Observation System Simulation Experiment framework: "Real" full SSH are from a numerical simulation with a realistic, high-resolution ocean circulation model: the reference simulation. Satellite observations are simulated by sampling the reference simulation based on realistic orbits of past, existing or future altimetry satellites. A baseline reconstruction method is provided (see below) and the practical goal of the challenge is to beat this baseline according to scores also described below and in Jupyter notebooks.

The reference simulation is the NATL60 simulation based on the NEMO model (Ajayi et al. 2020 doi:10.1029/2019JC015827). The simulation is run without tidal forcing.

The SSH observations include simulations of Topex-Poseidon, Jason 1, Geosat Follow-On, Envisat, and SWOT altimeter data. This nadir altimeters constellation was operating during the 2003-2005 period and is still considered as a historical optimal constellation in terms of spatio-temporal coverage. The data challenge simulates the addition of SWOT to this reference constellation. No observation error is considered in this challenge.

The SSH reconstructions are assessed over the period from 2012-10-22 to 2012-12-02: 42 days, which is equivalent to two SWOT cycles in the SWOT science phase orbit. For reconstruction methods that need a spin-up, the observations can be used from 2012-10-01 until the beginning of the evaluation period (21 days). This spin-up period is not included in the evaluation. For reconstruction methods that need learning from full fields, the reference data can be used from 2013-01-02 to 2013-09-30. The reference data between 2012-12-02 and 2013-01-02 should never be used so that any learning period or other method-related-training period can be considered uncorrelated to the evaluation period.

µ(RMSE): average RMSE score.
σ(RMSE): standard deviation of the RMSE score.
λx: minimum spatial scale resolved.
λt: minimum time scale resolved.

You can follow the quickstart guide in this notebook or launch it directly from binder.

The data are hosted on the AVISO+ website and tagged with DOI: 10.24400/527896/a01-2020.002. The website also provides a data handbook. This is the recommended access. This wiki can help you create an AVISO account to access the data. The data are also temporarily available here. They are presented with the following directory structure:

. 
|-- dc_obs
|   |-- 2020a_SSH_mapping_NATL60_topex-poseidon_interleaved.nc
|   |-- 2020a_SSH_mapping_NATL60_nadir_swot.nc 
|   |-- 2020a_SSH_mapping_NATL60_karin_swot.nc
|   |-- 2020a_SSH_mapping_NATL60_jason1.nc
|   |-- 2020a_SSH_mapping_NATL60_geosat2.nc
|   |-- 2020a_SSH_mapping_NATL60_envisat.nc

|-- dc_ref
|   |-- NATL60-CJM165_GULFSTREAM_y****m**d**.1h_SSH.nc

To start out download the observation dataset (dc_obs, 285M) from the temporary data server, use:

wget https://ige-meom-opendap.univ-grenoble-alpes.fr/thredds/fileServer/meomopendap/extract/MEOM/OCEAN_DATA_CHALLENGES/2020a_SSH_mapping_NATL60/dc_obs.tar.gz

and the reference dataset (dc_ref, 11G) using (this step may take several minutes):

wget https://ige-meom-opendap.univ-grenoble-alpes.fr/thredds/fileServer/meomopendap/extract/MEOM/OCEAN_DATA_CHALLENGES/2020a_SSH_mapping_NATL60/dc_ref.tar.gz

and then uncompress the files using tar -xvf <file>.tar.gz. You may also use ftp, rsync or curlto donwload the data.

The baseline mapping method is optimal interpolation (OI), in the spirit of the present-day standard for DUACS products provided by AVISO. OI is implemented in the baseline_oi Jupyter notebook. The SSH reconstructions are saved as a NetCDF file in the results directory. The content of this directory is git-ignored.

The evaluation of the mapping methods is based on the comparison of the SSH reconstructions with the reference dataset. It includes two scores, one based on the Root-Mean-Square Error (RMSE), the other based on Fourier wavenumber spectra. The evaluation notebook example_data_eval implements the computation of these two scores as they could appear in the leaderboard. The notebook also provides additional, graphical diagnostics based on RMSE and spectra.

Cross-functional modules are gathered in the src directory. They include tools for regridding, plots, evaluation, writing and reading NetCDF files. The directory also contains a module that implements the baseline method.

The structure of this data challenge was to a large extent inspired by WeatherBench.