Skip to content
m-pfeil edited this page Dec 27, 2021 · 2 revisions

bgc-ann

Marine ecosystem models are important to identify the processes that affects for example the global carbon cycle. Computation of an annual periodic solution (i.e., a steady annual cycle) for these models requires a high computational effort.

To reduce this effort, we use two different strategies. Firstly, we apply larger time steps for the spin-up calculation (i.e., a long-time integration for the calcuation of a steady annual cycle). Using larger time steps shortens the runtime of the spin-up obviously. As an application of the use of larger time steps, we implemented two algorithms (a step size control algorithm and a decreasing time step algorithm) that automatically adapt the time steps during the spin-up calculation in order to use the time steps always as large as possible (see python package timesteps and scripts TimestepsSimulation). Secondly, we approximate an exemplary marine ecosystem model by different artificial neural networks (ANNs). We used a fully connected network, then applied the sparse evolutionary training (SET) procedure, and finally applied a genetic algorithm (GA) to optimize both the network topology (see Python package ann and scripts ArtificialNeuralNetwork).

The parameter identification is a challenging task for marine ecosystem models. Therefore, we implemented a surrogate-based optimization (SBO) to identify optimal model parameters for marine ecosystem models (see Python package sbo and scripts SurrogateBasedOptimization).

For many marine ecosystem models, the existence and uniqueness of periodic solutions has not yet been analytically proven. Thus, we implemented a generator for different initial concentrations and started simulations with these different initial concentrations (see Python package initialValue and scripts InitialValueSimulation).

Clone this wiki locally