The project plan is to develop the initial prototypes and integrate these in the first “wave”. In the second “wave”, these prototypes are iteratively refined to include more realistic features. In the third “wave” we will focus on performance and GPU acceleration, before we finally focus on the non-linear data assimilation.

Project start: April 1, 2016.

Work package 1:

Simplified ocean current model This work package consists of developing an efficient two-layer ocean current model capable of utilizing GPU resources. The work package can again be broken down into several milestones:

Milestone A: The first milestone will create a simulator (simplified ocean model) based on existing one-layer codes, but extended to two layers.

Milestone B: The second milestone will verify and validate the simulator against know test cases and existing codes to ensure validity of results.

Milestone C: Milestone C will extend the simulator to utilize the GPU efficiently through OpenCL.

Work package 2:

Drift trajectory model The second work package will focus on extending and the ocean circulation model from work package 1 to include a drift component. Our initial version will be a decoupled passive drift component, which will be refined over the project.

Milestone A: The first milestone will use a passive Lagrangian particle model to represent e.g., a floating object. This prototype will be verified and validated against existing simulations to ensure validity of results.

Milestone B: The second milestone will extend the first prototype to also support transport of ice-bergs, including air and water side drag and Coriolis force. This will be performed by extending the ocean circulation model with a new component.

Work package 3:

Ensemble generation Work package 3 will be a major undertaking in the project, and concerns the initialization of ensemble members that sample the probability density function. We can break this task down into several milestones:

Milestone A: The first milestone in this work package will use classical Monte Carlo sampling to generate a set of ensemble members. This prototype will then represent the first fully integrated simulation system with uncertainty quantification.

Milestone B: The second milestone will move from Monte Carlo to multi-level Monte Carlo to sample the probability density function of our uncertain input parameters.

Work package 4:

Non-linear data assimilation Work package 4 concerns non-linear data assimilation, whereby observations are used to correct the simulation. The key objective is here to make optimal use of the few observations we are expected to have.

Milestone A: The first milestone will implement a particle filter without any special handling with respect to the probability density function. It is therefore expected that the ensemble will gradually collapse into only a few uncorrelated samples.

Milestone B: The second milestone in this work package will take special care to avoid the problem of a collapsing ensemble when reinitializing ensemble members.

Project end: March 31, 2020.