Integrating environmental component models. Development of a software framework

Abstract

Integrated models consist of interacting component models that represent various natural and social systems. They are important tools to improve our understanding of environmental systems, to evaluate cause–effect relationships of human–natural interactions, and to forecast the behaviour of environmental systems. The construction of these models is a conceptual and technical challenge, as it requires integrating various environmental processes, often occurring at different spatial or temporal scales. Several software tools are available supporting the development of integrated models but their diversity and IT–orientation hampers domain specialists like hydrologists in the construction and assessment of integrated models. This dissertation contributes to the development of one modelling framework allowing for the construction of component models, their multi–scale coupling to integrated models, and the analysis thereof. We consider different aspects including the model syntax and execution, spatial and temporal scaling, integration of legacy models, support for model analysis and semantic interoperability. We propose solutions to these issues tailored to domain specialists. Model construction and evaluation becomes difficult for domain specialists when technical barriers are present. We show that operations implementing domain specific concepts can be used by modellers to express spatio–temporal processes in component models, and they can be used to express couplings and therefore feedback effects between component models. To execute models described with domain specific concepts, we analyse with a client–server and a request–reply approach two control flow mechanisms for component models and integrated models. Both approaches allow for the sound execution of multi–scale integrated models. The client–server approach allows for better optimisation of the model execution due to the overall knowledge of the model. By introducing the accumulator as a generic model building block, we provide the modeller a means to program scale transfer operations on space and time with operations representing domain concepts similar to map algebra operations. The accumulators allow for a coupling of multi–scale components without the need to modify process implementations and spatial or temporal characteristics of individual component models. We present two ways to integrate existing models into the modelling frameworks, allowing therefore to reuse functionality from available legacy models. The integration of external models done by software engineers on the level of APIs provides modellers functionality in the same modelling language as components are constructed. The integration of applications by system calls is easier to realise for domain specialists. The numerical implementation representing environmental processes does not necessarily expose the scientific meaning of component models. We develop a formal description allowing for a semantically enhanced description of spatial and temporal characteristics of the main model building blocks, and evaluate whether an implementation–independent formalisation improves the scientific interoperability. We demonstrate that the presented concepts for component construction and multi–scale coupling can be merged into one modelling platform. Software framework prototypes are used in illustrative case studies, where we construct component models with spatio–temporal processes simulating amongst others land use change and hydrological processes, and couple these to integrated models.