Parameter estimates of a zero-dimensional ecosystem model applying the adjoint method

Schartau, M., Oschlies, A. and Willebrand, J. (2001) Parameter estimates of a zero-dimensional ecosystem model applying the adjoint method. Deep Sea Research Part II: Topical Studies in Oceanography, 48 (8-9), 1769-1800. (doi:10.1016/S0967-0645(00)00161-2).

Abstract

Assimilation experiments with data from the Bermuda Atlantic Time-series Study (BATS, 1989–1993) were performed with a simple mixed-layer ecosystem model of dissolved inorganic nitrogen (N), phytoplankton (P) and herbivorous zooplankton (H). Our aim is to optimize the biological model parameters, such that the misfits between model results and observations are minimized. The utilized assimilation method is the variational adjoint technique, starting from a wide range of first-parameter guesses. A twin experiment displayed two kinds of solutions, when Gaussian noise was added to the model-generated data. The expected solution refers to the global minimum of the misfit model-data function, whereas the other solution is biologically implausible and is associated with a local minimum. Experiments with real data showed either bottom-up or top-down controlled ecosystem dynamics, depending on the deep nutrient availability. To confine the solutions, an additional constraint on zooplankton biomass was added to the optimization procedure. This inclusion did not produce optimal model results that were consistent with observations. The modelled zooplankton biomass still exceeded the observations. From the model-data discrepancies systematic model errors could be determined, in particular when the chlorophyll concentration started to decline before primary production reached its maximum. A direct comparison of measured 14C-production data with modelled phytoplankton production rates is inadequate at BATS, at least when a constant carbon to nitrogen C : N ratio is assumed for data assimilation.

This record has no associated files available for download.

Contributors

Download statistics