SIMULATING EXTREMES IN PESTICIDE MISAPPLICATION FROM BACKPACK SPRAYERS

This problem of how to quantify the worst-case or extreme misapplicati on that could occur for a particular backpack sprayer scenario was stu died by considering two disparate approaches. First, a highly conserva tive position is taken by assuming that all errors are dependent on on e another and that no single error can cancel another. In this case, e rrors were represented by triangular fuzzy numbers. Second, a less con servative position is taken by assuming that individual errors occur i ndependently, and any one error can cancel out another of similar valu e. In this case, errors were represented by random samples from unifor m probability distributions. Simulations with errors represented by tr iangular fuzzy numbers resulted in worst-case misapplication levels al ways greater than that found in probabilistic simulations. When measur ement errors were 10%, ratios of worst-case simulation results from fu zzy number analysis to probabilistic analysis ranged from 0 .-9-10 . 2 times less pesticide delivered to 1 . 1-12 . 7 times more pesticide d elivered; simulations with error represented by fuzzy numbers were abo ut 3 . 5 times more conservative overall than probabilistic simulation s. This study shows that when little information is available, a model of a pesticide spray process can be used to simulate potential extrem es in misapplication. Under conditions where only the range of errors can be determined, highly conservative estimates of worst-case misappl ication can be generated using possibilistic techniques, including int erval analysis or fuzzy arithmetic. If the statistical distributions o f errors can be estimated or determined from experiments, Monte-Carlo simulations can be used to generate useful risk-based elements of misa pplication.