Aggressive Space Mapping for the Optimization of a Marine Ecosystem Model
In this paper we apply the Aggressive Space Mapping (ASM) algorithm by Bandler et. al. to the parameter optimization of a one-dimensional marine ecosystem model of NPZD type. We show that this approach leads to a very satisfactory solution while yielding a significant reduction in the total optimization cost. The ecosystem model, developed by Oschlies and Garcon, simulates the distribution of nitrogen, phytoplankton, zooplankton and detritus in a water column and is driven by ocean circulation data. A key issue is to optimize model parameters in order to minimize the misfit between the model output and given observational data. In the ASM approach, reducing the overall optimization cost by avoiding expensive function and derivative evaluations is achieved by using a surrogate model that replaces the original one. Furthermore the ASM algorithm solves a nonlinear system of equations which is conditionally equivalent to use this surrogate in the optimization run. We use a coarser time discretization for obtaining a suitable low-fidelity model. This is then corrected to create a physically-based surrogate, where the correction is obtained through a parameter mapping which provides the minimizer of the distance between the fine and the coarse model output. We show that this surrogate provides a good approximation of the fine model. The applicability of the ASM technique to the problem at hand is verified by using synthetic target data. Results are compared to those of the direct fine model optimization. We show that a very reasonable fit of the target data can be obtained with an average reduction in the computational cost of about 65 %.