Using Formal Meta-Data Descriptions for Automated Ecological Modeling

Abstract

System dynamics is a mathematical modeling ap-proach widely used in environmental studies for repres-enting and simulating ecological systems, giving sup-port to prediction and decision making. The know-ledge sources for model design are, essentially, human expertise and ecological data at various levels of ab-straction. We observe that property descriptions of eco-logical data (ecological meta-data), such as functional, temporal and spatial relations between variables, seem to be cognitively close to the concepts and reasoning used by model designers. Can the process of linking ecological meta-data to model design be automated? We aim to answer this question by: (1) develop-ing a formal language for expressing ecological meta-data; (2) building a logic-based formalisation of con-nections between the meta-data descriptions and model design; and (3) semi-automating model design endorsed by meta-data descriptions. Logic-based approaches for ecological modeling have been proposed in (Robertson et al. 1991). Emphasis is placed on the use of domain knowledge to support modeling automation, making model assumptions ex-plicit to enable more informed model analysis. Our work evolves from these ideas, adding to them by in-vestigating how ecological meta-data (which play a part in domain knowledge) can be conducive to model con-struction. Automated modeling based on meta-data can only be attractive if the mechanisms supporting it are not restrictive to specic datasets. Pre-dening a detailed knowledge representation system still general enough to express properties of every ecological dataset is in-feasible. Thus, what is needed is a general framework for meta-data description which can be instantiated to speci c datasets and model purposes. We have a proto-type ontology, named Ecolingua, for such a framework, providing a vocabulary and axioms for ecological meta-data description. Ecolingua has been designed by trying and describ-ing a diverse dataset generated by a tropical forest log-ging experiment in the Amazon, Brazil (Biot 1995). As design tool we have been using the www-based Ontolin-gua Server (Farquhar, Fikes, & Rice 1996) for ontology