International Environmental Modelling and Software Society (iEMSs) 7th Intl. Congress on Env. Modelling and Software, San Diego, CA, USA, Daniel P. Ames, Nigel W.T. Quinn and Andrea E. Rizzoli (Eds.) http://www.iemss.org/society/index.php/iemss-2014-proceedings Interoperable Intelligent Environmental Decision Support Systems: a Framework Proposal Miquel Sànchez-Marrè Knowledge Engineeering & Machine Learning Group (KEMLG), Computer Science Dept., Universitat Politècnica de Catalunya-BarcelonaTech, Jordi Girona 1-3, 08034 Barcelona, Catalonia (miquel@lsi.upc.edu) Abstract: In this paper, an approach for the development of Interoperable Intelligent Environmental Decision Support Systems (IEDSS) is proposed. The framework is based upon the cognitive-oriented approach for the development of IEDSS proposed in (Sànchez-Marrè et al., 2008), where three kind of tasks must be built: analysis tasks, synthesis tasks and prognosis tasks. Now, a fourth level will be proposed: the model construction layer, which is normally an off-line task. At each level, interoperability should be possible and inter-level interoperability must be also achieved. This interoperability is proposed to be obtained using data interchange protocols like Predictive Model Markup Language (PMML), which is a model interchange protocol based on XML language, using an ontology of data and AI models to characterize data types and AI models and to set-up a common terminology, and using workflows of the whole interoperation scheme. In the future, a Multi-Agent System will be used to implement the software components. An example of use of the proposed methodology applied to the supervision of a Wastewater Treatment Plant is provided. This Interoperable IEDSS framework will be the first step to an actual interoperability of AI models which will make IEDSS more reliable and accurate to solve complex environmental problems. Keywords: Model Interoperability, Intelligent Environmental Decision Support Systems. 1. INTRODUCTION Environmental systems/problems are real-world complex processes very difficult to manage and supervise. Most of them are dynamic processes with many decisions involved within them. For instance, the management of a basin river in a country taking into account all the environmental features: water level at different points, water quality of the river and its tributary rivers, wastewater treatment plants operation along the basin river, water use along the basin river, several stakeholders involved, politic and scientific strategies, etc. Another example could be air pollution control and management systems in charge of supervising the air quality of a concrete zone (region, city, etc.) taking into account all the features: car traffic, meteorological condition, chemical variables, physical conditions of the region, wind strength, etc. Environmental Decision Support Systems (EDSS) field has been trying to use some models of the real world being analysed to try to get an insight of the behaviour and evolution of the real system. A model is a description of a system, usually a simplified description less complex than the actual system, designed to help an observer to understand how it works and to predict its behaviour. Typically, models could be divided into mechanistic models and empirical models. Mechanistic models are based on an understanding of the behaviour of a system's components, analysing the system from its first- principles. Usually these mechanistic models are expressed a as set of mathematical formulas and equations (differential equations, etc.). Historically, the first EDSS were using only these kind of models. Notwithstanding, taking into account that usually huge amount of data gathered from the system’s being managed were available, some new empirical models were started to be used. Empirical models are based on direct observation, measurement and extensive data records. The first empirical models used were mathematical and statistical methods like Multiple Linear Regression (MLR) models, Principal Component Analysis (PCA) models, Discriminant Analysis (DA) models, Logistic Regression (LR) Page 501