Automatic time step adjustment for shortening the runtime of the simulation of marine ecosystem models

Abstract:

In investigating the global carbon cycle, shortening the runtime of the simulation of marine ecosystem models is an important issue. More specifically, steady annual cycles mostly are used to assess and validate the models against
observational data and to identify relevant biogeochemical processes. Offline simulations based on the transport matrix method already reduce the high computational effort significantly. Furthermore, they facilitate the application
of larger time steps in a simple way. In this paper, we present two different methods that automatically adjust the time step during the simulation of a steady state using transport matrices. The algorithms use either an adaptive
step size control or decreasing time steps. Their aim is to apply always the time step as large as possible but without any manual selection. We applied the methods for a variety of ecosystem models of different complexity, using Latin
hypercube samples of size 100 for the model parameters of each model. We showed that both methods computed an approximation of the steady annual cycle that was of the same accuracy as solutions obtained with a fixed time step. Both algorithms lowered the runtime of the steady annual cycle computation significantly. The performance gain depended on the complexity of the models. Moreover, the adaptive method has a certain overhead that might lead to higher computational cost in special cases.

Content:

Tracer concentrations of a reference solution for all parameter vectors and biogeochemical models

SQLite database including the results using the decreasing time steps algorithm

Tracer concentrations of the results using the decreasing time steps algorithm

SQLite database including the results using the step size control algorithm

Tracer concentrations of the results using the step size control algorithm