KNOWLEDGE USED BY FARMERS IN THE PLANNING AND USE OF CULTIVATION DEVICES

This paper describes a study an the knowledge farmers use to manage th eir crops and discusses its consequences for the design of clop manage ment supports. Crop management is considered to be the control of an o pen dynamic environment. Thus farmer must design and perform a set of coordinated actions to reach a specific stare (generally defined by th e quality and quantity of the harvested pans of the plants) and to lim it as far as possible the negative effects of irreversible events on y ield and on the physical, chemical and biological components of the so il. Crop management also implies the coordination of actions on differ ent scales (the plot, the crop area, the farm) and in time (one task c an influence other tasks due to lack of manpower or equipment at a giv en time. Each task must be considered in relation to all the others do ne on the same plot during the cultivation period). The farmer's task may be represented as a loop including information gathering, prognosi s and planning, decision making, performance and checking. The problem is to understand how knowledge of the biophysical processes and actio ns are related to each other and the way they are used to complete thi s loop. The research examined a specific cultivation task: tilling and sowing sugar beet. A description of the task showed the components of the biophysical processes and the actions involved in designing and p erforming this task. Similarities in the way farmers managed sugar bee t setting up were identified. We postulated that the semantics of acti on networks can be used to identify these similarities and we tested t he extent to which such a knowledge representation was useful for repr esenting farmers' knowledge. Data were collected from 8 farmers whose farms differed in their organisation and agronomic constraints. One ex periment was based on classifying pictures of soil. A second involved asking farmers to design a coordinated set of actions for each soil pi cture used in the first experiment. The way the task teas carried out peas then analysed. Representing farmer's knowledge as a semantic netw ork of action helped to understand the role of procedures in informati on gathering and prognosis. But this representation is not enough to s upport the inferences made by the farmers to describe soil states and changes. In real world situations, the task performance is also the re sult of a compromise between applying an optimal set of actions on the plot (the one accessed through the semantics of actions network) and minimizing the consequences of negative events which might occur at th e crop area level.