Ensemble optimal interpolation for adjoint-free biogeochemical data assimilation.

Area of Science:

• Oceanography
• Marine Ecosystem Modeling
• Data Assimilation

Background:

• Advanced marine ecosystem models involve numerous biogeochemical variables, complicating data assimilation.
• Traditional 4dVar methods require tangent linear and adjoint code, difficult for complex models.
• Ensemble-variational methods offer an alternative by using model forecast ensembles for statistics.

Purpose of the Study:

• Introduce and evaluate a novel four-dimensional ensemble optimal interpolation (4dEnOI) technique.
• Assess the suitability of 4dEnOI for coupled physical-ecosystem models.
• Compare 4dEnOI performance against traditional 4dVar methods.

Main Methods:

• Implemented a 4dEnOI technique using small ensembles and covariance localization (spatial and variable).
• Tested the 4dEnOI implementation with a coupled physical-ecosystem model of the California Current System.
• Compared 4dEnOI against a 4dVar benchmark for a model with 4 biogeochemical variables.

Main Results:

• The 4dEnOI implementation effectively reduces model observation misfit, performing similarly to 4dVar.
• Covariance localization, especially variable localization, significantly impacts 4dEnOI results.
• Variable localization proved beneficial when reducing coupling between physical and biogeochemical components.

Conclusions:

• The presented 4dEnOI technique is a viable, easier-to-implement alternative for data assimilation in complex marine ecosystem models.
• The method is computationally suitable and avoids the need for tangent linear/adjoint code.
• Further improvements to 4dEnOI performance are possible with algorithmic modifications like multiple iterations.