Model-Driven Software Engineering for Computational Science
Applied to a Marine Ecosystem Model
- Zugl.: Kiel, Univ., Diss. 2015
|Seitenbereich||xxiii, 404 S.
This interdisciplinary thesis contributes to both software engineering and ecological modeling. In the discipline of software engineering, we introduce Sprat, which is a model-driven software engineering approach for computational science.
In the field of ecological modeling, we present the Sprat Marine Ecosystem Model—a spatially-explicit fish stock model for marine end-to-end modeling. The ever-increasing complexity of in silico experiments in computational science is reflected in the growing complexity of the simulation software enabling these experiments.
This development results in a need for collaboration between scientists from different disciplines in the development of
such elaborate scientific software. However, in this process, state-of-the-art software engineering methods are rarely employed, which negatively affects the maintainability and performance of the software as well as the reliability
of its results. To tackle this challenge, we introduce the Sprat Approach, which hierarchically integrates multiple domain-specific languages to facilitate the cooperation of scientists from different disciplines and to support them in creating well-engineered software without extensive software engineering training.
In order to evaluate the Sprat Approach, we apply it to the implementation of the Sprat Marine Ecosystem Model in an exploratory case study.
The Sprat Marine Ecosystem Model is a fish stock model that is coupled with biogeochemical ocean models to simulate all trophic levels of a marine ecosystem. The model utilizes a novel modeling approach based on population balance equations that combines the advantages of existing end-to-end modeling frameworks while preventing their main drawbacks.
For solving the partial differential equations that constitute the Sprat Model, we develop a flux-corrected transport finite element scheme that uses explicit multi-step methods to integrate in time. In order to evaluate the Sprat Model, we apply it to the eastern Scotian Shelf ecosystem with its intertwined direct and indirect fish stock interactions, which previously could not be modeled satisfactorily. Our simulation results provide new insights into the main drivers of regime shifts in marine ecosystems.