Forecasting of grape powdery mildew disease risk in vineyards
Citation
Lu, W., Carisse, O., Atkinson, D. 2018. Forecasting of grape powdery mildew disease risk in vineyards using a Bayesian learning network model. Joint Statistical Meetings (JSM), Vancouver, B.C., Canada, July 28 - August 2
Résumé en langage clair
Vineyards suffer substantial grape yield losses from damaging diseases, such as powdery mildew (Erysiphe necator) - a polycyclic, airborne disease. The severity and transmission of this grapevine disease depends on a complex interaction of genetic (pathogen and host) and environmental factors, including crop management practices. Climate change and variability is also contributing uncertainty in managing this disease by raising temperatures affecting winter chill requirements, lengthening growing seasons in northern climates, producing more days without frost, and more intense heatwave and rainfall events. It remains unclear how to best protect and minimize the impact of this disease in commercial vineyards. We present a Bayesian learning network model for forecasting disease risk in time and space. This approach combines diverse types of data, and complex causal relationship between variables. We present findings from validating this model against for grapevine disease data collected within Quebec vineyards. Key scientific recommendations and challenges associated with its use as a precision viticulture tool are discussed.
Résumé
Vineyards suffer substantial grape yield losses from damaging diseases, such as powdery mildew (Erysiphe necator) - a polycyclic, airborne disease. The severity and transmission of this grapevine disease depends on a complex interaction of genetic (pathogen and host) and environmental factors, including crop management practices. Climate change and variability is also contributing uncertainty in managing this disease by raising temperatures affecting winter chill requirements, lengthening growing seasons in northern climates, producing more days without frost, and more intense heatwave and rainfall events. It remains unclear how to best protect and minimize the impact of this disease in commercial vineyards. We present a Bayesian learning network model for forecasting disease risk in time and space. This approach combines diverse types of data, and complex causal relationship between variables. We present findings from validating this model against for grapevine disease data collected within Quebec vineyards. Key scientific recommendations and challenges associated with its use as a precision viticulture tool are discussed.