Decision analysis as a tool for integrating simulation with expert systems when risk and uncertainty are important

Abstract

Risk and uncertainty are important components of agricultural decision making. The methodology of applied decision analysis is especially useful in addressing such problems, but has not been widely integrated with expert systems. One problem has been the difficulty of handling uncertainty within the expert system framework in a way which is logically consistent with rational decision criteria. An additional problem in agriculture is the need to combine uncertain or incomplete information from simulations and statistical studies with the subjective knowledge of one or several experts. In this paper, it is argued that Bayesian probability theory provides a natural approach, and a methodology is developed for combining diverse sources of information within the framework of an expert system. The methodology is developed within the context of an expert system for protection of soybeans against corn earworm, using information from HELSIM, a heliothis population model (R. Stinner) and from the SOYGRO soybean crop model (G. Wilkerson).