Towards model integration and model-based decision support for environmental applications

We argue that scientific results from different fields that can help decision making about environmental problems should be delivered in the form of executable and combinable models. For this purpose, such models have to be stated in a coherent modeling framework. For their integration, it is not essential (and even unnecessary) that they share a mathematical formalism. More fundamentally, they have to be stated at a conceptual level in order to identify and relate the objects and quantities that occur in the various model fragments. In order to be truly compositional, the models have to be formulated as independent model fragments that represent elementary processes, and they have to be stated in a context-independent way to enable their reuse for different purposes. Computer-based decision support should be based on such models and generic algorithms for drawing inferences from these models. The basic steps in such a model-based decision support system are situation assessment, which is the task of generating a model that is compliant with the observations provided. Based on such a situation assessment, the next step is therapy proposal, which amounts to generating a model that combines the model of the current situation and models of human interventions and is compliant with the goals to be achieved by the remedy. Figure 1 displays this basic architecture. These steps can be formalized as instances of model revision in logic and realized by consistency-based diagnosis techniques. As the modeling formalism, we propose what has been developed as process-oriented modeling in Artificial Intelligence. The elementary model fragments, called processes, contain an explicit representation of their preconditions, stated in conceptual terms by reference to objects, their existence, properties, and relations. Their effect part does not only relate quantities, but also create objects and relations. Such model fragments can be formalized as logical formulas of the form StructuralConditions ? QuantityConditions = StructuralEffects ? QuantityEffects. The logical foundation of this modeling formalism allows for the integration with the logic-based model revision algorithm. Since the effects of a process may imply (or negate) the preconditions of other proccesses, the composition of the model for a certain scenario based on a library of elementary processes can be performed by an automated reasoning process and is not dependent on modeling or domain experts. As a consequence, this approach promises multiple benefits, the major ones being • support for the integration of models from different research fields and sources, • re-use of model fragments in different contexts and for different purposes, • availability of expert knowledge (captured by the model library) for non-experts.