The application of expert systems to significant earth science disciplines such as geology, geophysics and pedology has been hitherto well established and documented (e.g. PROSPECTOR for mineral exploration (Katz, 1991), KBGIS-II (Smith et al., 1987). Development of an expert system for coastal zone management can be seen to be in its infancy; the need for such an entity is clearly indicated.
Figure 1: The configuration of COAMES
COAMES (COAstal Management Expert System) is an expert system currently under development at Plymouth Marine Laboratory. It is being constructed as part of the Land-Ocean Interaction Study (LOIS), a multi-disciplinary multi-institutional initiative. LOIS aims to study the coastal zone in an holistic manner, by a thorough analysis of biological, chemical and physical processes and elements, relating to terrestrial, riverine, marine and atmospheric systems. These systems considered together will give marine environmental scientists a fuller understanding of the constituents and processes operating at the coast. From a pragmatic point of view this fuller appreciation will facilitate improved environmental management.
Anticipated users include private and public environmental agencies, and marine scientists of varying backgrounds, who could potentially offer scenarios and submit queries on an array of coastal issues: biological, chemical and physical. A typical application could be the impact of an offshore oil slick on nutrients in coastal seas. From an initial query worded in natural language, the interface extracts the operative words and passes them to the inference engine, which performs logical processes (e.g. induction, deduction) to select data, knowledge and models appropriately.
Understandably, data from such a breadth of applications are likely to occur in multifarious formats, and true to the nature of the LOIS objectives, the integration of this data is essential The data is stored in a number of raster (e.g. remotely sensed images and interpolated or modelled data) and vector (e.g. point data derived from static moorings or cruises) geographical datasets associated with attribute data. The integration of these disparate data sets is enabled through referencing to an object-oriented structure via geographical co-ordinates, programmed in C++. The structure itself is divided into several classes, categorised according to thematic content. Acting on the operative words drawn from the user's input, the relevant datasets are extracted from the structure; topology is then performed on them only, giving substantial saving on computing time. This arrangement is reflected in other undertakings, notably ESRl's Spatial Database Engine (Maguire, 1995).
Inference engine and knowledge base
Traditionally, the knowledge base (the formal encoding of the knowledge of, for example, marine experts) has manifested itself as a long series of IF-THEN statements where action is taken if a certain condition is met. This exhaustive approach results in the knowledge base and inference engine being closely entwined (i.e. the action is the task of the inference engine). The knowledge base should not be so 'hard-wired' into the system, as it may need to be modified to meet specific demands. This is best done as a separate entity.
Alternatively, rules can be arranged as a hierarchy of objects. The knowledge base is called upon by the inference engine 'Does this rule apply?' 'This one?' etc. until a rule is found that satisfies the operative words and the derived data. This is then repeated for the next tier in the hierarchical object structure. In much the same way, the pertinent models are also invoked (see next section). At this stage no action is taken on the rules - appropriate action is implemented by the inference engine once the levels in the hierarchy have been traversed.
A significant role of COAMES is to link via the inference engine, with various models being developed within LOIS - biological, chemical (e.g. modelling the flux of nutrients associated with the ecosystem present in offshore waters) and physical (e.g. hydrodynamic models). Models themselves encapsulate knowledge, and can be viewed as a knowledge base of processes as opposed to the largely fact-based object-oriented knowledge base.
A further role of the inference engine within COAMES is to select an appropriate method for visualising the inferred results of the initial query.
This paper describes initial developments of the COAMES expert system, as outlined above. In common with the LOIS project objectives, it is hoped that through use of such a system a better understanding of the coastal environment as a complete entity will be gained.
Katz, S.S. 1991. "Emulating the Prospector Expert System with a Raster GIS", Computers and Geosciences, 17, 7, 1033-1050.
Maguire D.J. 1995. "A High Performance, Object-oriented Spatial Database Management System", AGI '95 Conference Proceedings, 1.22.1 - 1.22.4.
Smith, T., Peuquet, D., Menon, S. and Agarwal, P. 1987. "KBGIS-II - A Knowledge-based Geographical Information System", International Journal of Geographical Information systems, 1 (2), 149-172.