Dynamical interpolation of surface ocean chlorophyll fields via data assimilation with an iterative ensemble smoother

Inference of the sea surface chlorophyll field from incomplete satellite coverage is posed as a formal inverse problem using a Monte Carlo approach to Bayesian estimation. We introduce a new method, the strong constraint iterative ensemble smoother, for solving the general coupled physical–biological parameter estimation problem where model nonlinearities may be relevant. The forward model is posed in four ways: (1) advection–diffusion, (2) linear advection–diffusion-reaction, (3) nonlinear advection–diffusion-reaction, and (4) a nonlinear nutrient-phytoplankton model. Hindcast skill is demonstrated through analysis of the fit to independent data in a series of experiments utilizing MODIS chlorophyll imagery from the Middle Atlantic Bight during summer of 2006. The data assimilative model demonstrates skill over a range of presumed observational error. Both the purely physical model (advection–diffusion only) and the coupled physical–biological models exhibit skill fitting unassimilated data. The skill of the coupled physical–biological models is greater than the skill of the advection–diffusion model, owing at least in part to greater degrees of freedom in those inversions.

Research highlights
► The ItEnS has skill over a range of presumed observational error for a suite of dynamical models.
► Physics-only model and physical–biological models exhibit skill fitting unassimilated data.
► Skill of the coupled physical–biological models is greater than the physics-only model.