טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentCohen Avihu
SubjectKnowledge Based Spatial Decision Support System for Pest
Management Applied for Mediterranean Fruit Fly
(Medfly)
DepartmentDepartment of Civil and Environmental Engineering
Supervisors Professor David Broday
Dr. Yafit Cohen
Full Thesis textFull thesis text - English Version


Abstract


The Mediterranean fruit fly (Medfly) is a major pest of fruit crops. In Israel it is a key pest in citrus. The spraying decision making against it is based on domain experts. Due to time constraints and fear of Medfly population outbreak excessive sprays are carried-out resulting in environmental damages and extra costs. In addition, customer awareness and global market regulations impose the usage of advanced decision making tools.

The general objective of this study was to develop a genuine and comprehensive approach to improve the decision making process in pest management. The suggested approach, Knowledge Based Spatial Decision Support System (KB-SDSS), was evaluated by: Developing a KB-SDSS for Medfly control in citrus in Israel (MedCila); Assessing its performance; Developing a learning mechanism to improve its performance.

Development of the MedCila involved: acquisition of expert and domain knowledge; identification, modeling and integration of the relevant criteria into a Geographic Information System (GIS). The assessment of the MedCila performance involved verification, evaluation and assessment of the MedCila's acceptance potential and its ability to improve the experts' decisions. Two expert-panels were carried out to evaluate the MedCila as well as its knowledge based approach. For the learning mechanism the MedCila recommendations (MR) and the Expert's decisions (ED) were used. A k means clustering was used to locate groups of cases in which MR-ED conflicts exist. The learning mechanism analyzes the data and identifies the MR-ED conflicts roots. Then, if an action is needed it can be directed to either the MedCila, its database, or to the expert.

The MedCila produces for each plot a grade between -1 and 1 (the strongest certainty against/ for spraying respectively). The grades are divided into four recommendations: Spray, Spraying is recommended, Spraying is not recommended, and Do not spray. The recommendations are shown on a symbolized map accompanied by a visual explanation. The results from the experts’ panels showed that generally the MedCila recommendations were accepted by the experts. The MedCila recommendations are divided between trivial and complex recommendations. As such, the expert can easily and quickly decide upon the trivial cases and dedicate more time to complex plots. Moreover, the MedCila enables using more information than normally used by the experts. Another conclusion was that the MedCila may reduce spraying actions by at least 8%. In summary, the KB-SDSS approach can be appropriate for many pest management practices where data are missing but experts’ knowledge is available.