Articles | Volume 9, issue 10
Geosci. Model Dev., 9, 3729–3750, 2016
Geosci. Model Dev., 9, 3729–3750, 2016

Development and technical paper 19 Oct 2016

Development and technical paper | 19 Oct 2016

Metos3D: the Marine Ecosystem Toolkit for Optimization and Simulation in 3-D – Part 1: Simulation Package v0.3.2

Jaroslaw Piwonski and Thomas Slawig Jaroslaw Piwonski and Thomas Slawig
  • Institute for Computer Science and Kiel Marine Science – Centre for Interdisciplinary Marine Science, Cluster The Future Ocean, Kiel University, 24098 Kiel, Germany

Abstract. We designed and implemented a modular software framework for the offline simulation of steady cycles of 3-D marine ecosystem models based on the transport matrix approach. It is intended for parameter optimization and model assessment experiments. We defined a software interface for the coupling of a general class of water column-based biogeochemical models, with six models being part of the package. The framework offers both spin-up/fixed-point iteration and a Jacobian-free Newton method for the computation of steady states.

The simulation package has been tested with all six models. The Newton method converged for four models when using standard settings, and for two more complex models after alteration of a solver parameter or the initial guess. Both methods delivered the same steady states (within a reasonable precision) on convergence for all models employed, with the Newton iteration generally operating 6 times faster. The effects on performance of both the biogeochemical and the Newton solver parameters were investigated for one model. A profiling analysis was performed for all models used in this work, demonstrating that the number of tracers had a dominant impact on overall performance. We also implemented a geometry-adapted load balancing procedure which showed close to optimal scalability up to a high number of parallel processors.

Short summary
In order to fundamentally tackle the problem of parameter identification for marine ecosystem models in 3-D, we introduced a general biogeochemical programming interface that fits into the optimization context. Moreover, we implemented a comprehensive parallel solver software for periodic steady states that uses the interface to couple marine ecosystem models to a transport matrix driver. We validated the new implementation using a hierarchy of biogeochemical models.