SDAP data assimilation platform

SDAP (a recursive acronym for “SDAP Data Assimilation Platform”) is a data assimilation framework developed by Pierre De Mey-Frémaux and collaborators since 2001 and distributed under the GNU General Public License (GPL) v3 and CeCill v2 licence.

The main objectives of SDAP are:

1. to provide a flexible platform to carry out multivariate assimilation of geophysical data in a numerical model;
2. to provide Ensemble-based explainability metrics to facilitate the interpretation of results;
3. to provide other tools such as objective analysis tools and array design tools.

The program is multi-geometry (finite differences or finite elements), multi-algebra (plug-in analysis kernels), multi-model (simple standardized interfaces), multi-subdomain (MPI groups and communicators), and multi-grid (e.g., parent grid + child grid). The program supports almost uniquely Ensemble approaches, such as the Stochastic Ensemble Kalman Filter (S-EnKF) for assimilation, and Ensemble-based metrics such as probabilistic scores.

As an “advanced” data assimilation method, S-EnKF, implemented in SDAP via the BELUGA analysis kernel, is well adapted to coastal ocean data assimilation, as it includes built-in nonlinear dynamical error propagation and does not make any simplifying hypotheses for the uncertainties.

SDAP/BELUGA is a stochastic Ensemble filter in the sense that it uses perturbed observations to correct the too low spread of the analysis ensemble in the original EnKF. It is a four-dimensional filter in the sense that it compares observations with the background at the observation time. It is localized in the sense that it uses observations (“predictors”) in the vicinity (“influence bubble”) of the model grid point where an analysis increment is sought. It is suboptimal, in the sense that an algorithmic predictor selection operates within the influence bubble.

SDAP can be integrated in three types of workflows:

• the classic in-core workflow (SEQUOIA), which is maintained but not much used anymore;
• the scripted (bash, Python) workflow;
• the agent-based SPRUCE workflow for parallel execution.

A few years ago, SDAP has been interfaced with several models, such as:

• The MOG2D barotropic finite-element model, precursor to the T-UGOm model ( https://sirocco.obs-mip.fr/ocean-models/tugo/ ).
• The NEMO community ocean model ( https://nemo-ocean.eu/ ).
• The SYMPHONIE coastal ocean model ( https://sirocco.obs-mip.fr/ocean-models/s-model/ ).