Presenter: Jonathan Izett (University of California, Santa Cruz)

Description:
There are many challenges to overcome when assimilating biogeochemical (BGC) data into ocean models. For example, it can be prohibitively difficult to convert the nonlinear BGC equations into the tangent linear (TLM) and adjoint (ADJ) models needed for traditional 4D-Var data assimilation. It is therefore desirable to identify simpler, yet still accurate, alternative approaches in order to improve the accessibility of BGC data assimilation. Model-reduced 4D-Var (MR4D-Var) was proposed to mitigate the challenge of translating the BGC equations by using a linear combination of underlying orthogonal modes to create the TLM and ADJ. Whereas the method has been demonstrated in idealized settings, it must also be tested in real-world applications before adoption as a viable alternative. To test the method’s viability in complex settings, we perform a series of numerical experiments in which we assimilate real chl-a observations into a three-dimensional model of the California Current System (CCS). We use a simple NPZD BGC model constructed in the Regional Ocean Modeling System (ROMS), with the experiments designed to compare the performance of MR4D-Var to the traditional 4D-Var as well as solutions without any data assimilation. Our results indicate that the MR4D-Var approach can indeed significantly reduce the model-observation misfit, even within the complex CCS domain, providing similar error statistics to traditional 4D-Var and performing much better than a non-assimilative run. The corrections are, however, limited by the structure of the orthogonal modes, which are dominated by coastal processes and seasonal dynamics. We also test a second alternative, a 'passive-tracer' approach, where biogeochemical dynamics are ignored in the linearized models and corrections are only made through advection and diffusion. Surpisingly, the passive-tracer approach provides very similar error statistics to the full 4D-Var and avoids some of the challenges associated with the BIOEOFs approach, while further simplifying the BGC assimilation.