A framework based on generalised linear mixed models for analysing pest and disease surveys

Lucie Michel (1) , François Brun (2, 3) , David Makowski (4)

1 Agronomie - Agronomie
2 AGIR - AGroécologie, Innovations, teRritoires
3 Acta - Les instituts techniques agricoles
4 Agronomie

In several countries, regional surveys are carried out to detect the presence of pests and diseases in crops. During these surveys, the incidence of major diseases and the presence of pests are recorded on various dates during the growing season. In this study, we aim to develop a framework to make better use of these regional surveys to estimate pest and disease dynamics, to analyse their variability across sites and years, and to assess uncertainty. Our framework is illustrated in four case studies: Septoria leaf blotch on wheat, downy mildew on grapevine, yellow sigatoka on banana and weevils on sweet potato. We showed that frequentist and Bayesian generalised linear mixed models gave similar results. This type of models is flexible enough to handle different types of data. They can be used to estimate disease and pest dynamics from observations collected in regional surveys and could help regional extension services evaluate risk levels at the regional scale.

Mots clés: alert system; bayesian model; disease survey; generalised linear mixed model; plant pest

Domaines: Applications [stat.AP]; Sciences agricoles

Lucie Michel, François Brun, David Makowski. A framework based on generalised linear mixed models for analysing pest and disease surveys. European Journal of Plant Pathology, 2017, 94, pp.1-12. ⟨10.1016/j.cropro.2016.12.013⟩. ⟨hal-01530793⟩