Point d'avancement

[AndreaGarciaJuan]

• Résumer le travail qui a était fait et définir nos besoins pour discuter avec le reste de partenaires.

Je vais préparer un résumé de 5-10 lignes sur ce qui a été fait et on peut réfléchir ensemble sur nos besoin jeudi prochain, si vous êtes d'accord.

[AndreaGarciaJuan]

Done

PCM (Profile Classification Model) algorithms have been applied to 3D gridded products (GLOBAL-ANALYSIS-FORECAST-PHY-001-024, GLOBAL_REANALYSIS_PHY_001_030, GLOBAL_REANALYSIS_PHY_001_031 and GLOBAL_REANALYSIS_BIO_001_029). A Jupyter notebook has been created, where we can select a dataset, choose model parameters, train the model and predict classes labels for each profile. Results are plotted in some figures, including vertical structure of each class (median profiles) and spatial and temporal distribution of classes (see notebook ). Plot functions are organized in a new python class called Plotter.

This initial notebook is now been derived in to 2 notebooks :

1. a model development notebook, where we can optimize the model (with the help of new intermediate figures), fit and save it,
2. and a prediction notebook, where we can load a model, classify any dataset and plot the results.

We expect users to trigger a workflow made of notebooks 1+2 or only 2.

PCM algorithms and plotting functions have also been tested with a dataset of argo profiles (created using argopy library) and biological variables from GLOBAL_REANALYSIS_BIO_001_029 product, with satisfactory results. Ocean patterns indicator is almost ready for 3D gridded datasets.

Needs (to talk with partners)

• [ ] Format in-situ product INSITU_GLO_NRT_OBSERVATIONS_013_030 to apply pyxpcm algorithms easily
• [ ] Create a cache system where we can stock fitted models to avoid the development of the same model twice. We need a system to name models uniquely.
• [ ] Define with partners the level of choice the user will have in dataset, model, and plot options?