Knowledge representation in the domain of environmental dynamics

Resumen

The management of environmental areas, with a particular focus in coastal and marine zones, has become as an essential mechanism in ensuring sustainable development. These areas has proved to be not only an invaluable ecological and economic asset but also especially fragile because of the artificial and human activities taking place in them. The degeneration of these areas has increased dramatically in recent years. These changes had created the need to establish effective new forms of organization to ensure the sustainability of these areas. Artificial intelligence techniques could contribute significantly to the development of a novel approach of integral management within environmental dynamic field. Some of these methods had been applied successfully in other scientific fields. Nevertheless, their application is significantly lower in environmental works. In particular, knowledge representation may exhibit an extraordinary and valuable resource to structure complex systems. Knowledge representation is a subfield of artificial intelligence which main aim is mod- eling a domain of discourse in an efficient and computer understandable manner. Knowledge representation field emerged from the need of enhancing tradicional procedural methods, which had been proven insufficient to solve certain complex problems using computers. Although logi- cians was already confronted with the problem of formalizing knowledge declaratively before the inception of computers, their efforts were focused on mathematical knowledge. Non-mathematical knowledge representation is a scientific field based on more expressive logics. With the emergence of computers and the evolution of the non-mathematical knowledge representation, another artificial intelligence subfield called automatic reasoning began to assume greater importance. Automatic reasoning allows inferring non-explicit knowledge, make new assertions or verify the consistency of declared facts. At the same time, knowledge bases size was reduced because it was no longer needed to represent the whole knowledge explicitly. Nevertheless, many works within dynamic environment field start frequently from previous research, or their data source is not public available. For these scenarios it is not possible to either formalize knowledge nor apply automatic reasoning. Thus, knowledge discovery methods need to be employed. Knowledge discovery usually involves several steps. The first one is selecting and pre-processing data. After that, the core step is taken place, which is generally called data mining. Eventually, results are validated and interpreted. In other circumstances, data has to be obtained from non-numerical sources. Techniques widely applied in environmental dynamics like remote sensing or video-monitoring frequently need an additional feature extraction step. In these situations, another subfield of the artificial intelligence, the computer vision, becomes relevant to acquire new kind of data. While the application of these methods can be very beneficial for environmental dynamic field, the selection of the suitable technique and its implementation is out of the work scope of most environmental engineers. It is essential to carry out an interdisciplinary work to overcome these difficulties and make artificial intelligence tools easily accesible for environmental researchers. The main contribution of this Thesis is to provide a framework to environmental researchers who want to apply artificial intelligence techniques to their works. This Thesis was conducted in collaboration with several research groups. The work was mainly done at the multidisciplinary research group of Environmental Fluid Dynamics of the University of Granada. The main field of expertise of the group is the integral management of natural resources and the related infrastructures for their exploitation, with specific focusing on coastal areas, ports and river basins. Those topics are investigated through a combined methodology of theoretical analysis, numerical modeling, measurements both at the field and at the laboratory, and data from different sources. This group cooperated with the LexiCon Research Group to produce a lexical environmental resource called EcoLexicon, which is the basis of some of the chapters in the Thesis. Finally, as a result of a collaboration with the Terrestrial Ecology Research Group, an ecological indicator system in Sierra Nevada was also developed. In summary, the data used in this Thesis come from a variety of projects and works developed in collaboration with theses research groups.