Outputs uncertainty analysis of four crop protection models relative to agrometeorological inputs measurement errors.

The use of computer models in crop protection increases our management and forecast capacities as well as it allows to reduce the increasing pressure of agricultural activity on natural resources by the optimization of phytos anitary products use. The CIPRA system (Centre Informatique de Prevision de Ravageurs en Agriculture), a modeling tool gathering, under a common compu ter frame, several forecasting models using standard weather data (temperat ure, wind, precipitation, relative humidity), is one of the first Canadian operational decision support systems in crop protection. Since models are j ust a simplification more or less representative of the corresponding bioph ysics system, it is of primary importance to study the implications and the limitations of their application by determining the uncertainty level on m odel outputs. The purpose of the present paper was to evaluate the impact o f uncertainty associated with weather inputs measurement on the outputs of four models within CIPRA system. These models are the Cercospora blight of carrots (Cercospora carotae (Pass.) Solheim), the onion leaf blight (Botryt is squamosa J.C. Walker), the carrot weevil (Listronotus oregonensis (LeCon te)) and the European corn borer (Ostrinia nubilalis (Hubner)). This object ive was achieved using the relative sensitivity and the uncertainty propaga tion. Results show that the cercospora blight model is more sensitive to te mperature than with relative humidity. In this case uncertainty on the rela tive humidity is a significant source of error in the outputs. The onion le af blight model is primarily sensitive to the relative humidity fluctuation s. The relative humidity is, therefore, the principal source of error in th e model outputs. Finally, the two insects models (weevil and borer) are mor e sensitive to the maximum temperature than to the minimum temperature. Unc ertainties on the two models outputs are low. In addition to the evaluation of the reliability and level of significance of the results provided by th e four models, this study allows to identify the most significant meteorolo gical variables for the models.