A framework for modeling uncertain reasoning in ecosystem management. II. Bayesian belief networks

Figure 3: The interface main panel. project will comprise the design and implementation of a module for supporting the user in the selection of a treatment in accord with the basic tenets of integrated control. name and the damages caused, i.e., a set of pointers to the objects belonging to the damage discrimination tree. At present 35 phytophagouses have been described. The damage discrimination tree. As already described it consists of a tree having a generic damage description at the root and the \most" speciic damage description at the leaves. It has been implemented as a damage object hierarchy, see Figure 2. Each damage is an object deened by an extended description of the damage itself and for each time period by the pointers to the pests which can cause the damage in that period. The damage hierarchy consists of 200 objects. The system interface. In this phase of the project the user of the system is the agricultural technician who can be asked by the farmer to make a diagnosis explaining a set of observed damages. The system interface design rests on the choice to make both the kb information and the information resulting from the interaction accessible at each point during the user-system interaction by clicking with the mouse on the appropriate buttons of the interface (see Figure 3). The navigation along the damage discrimination tree is realized by performing the right choice from a menu activable at each node of the tree. Finally, the diagnostic module has been implemented as a collection of LISP functions. 6 Conclusions and Future Work This paper describes a pest management expert system for apple orchards. The system focuses on two main problems: the classiication of the damage manifestations and their diagnosis. The main problems faced concerning the classiication are related to the identiication of a precise language terminology in the description of the damages. Actually the same damage can be deened using diierent descriptions and terminology and this language is far from being easily formalizable. This point has been found to be the main obstacle in an early implementation of the system, written in LOOM Gre88], a hybrid knowledge representation language with an automatic classiier based on the concept of \term subsumption". The current implementation of the classiication module is based on a \Discrimination Tree". In a future work diierent descriptions of the same damage manifestation will be available allowing the …