Linking Climate Variables to Large-Scale Spatial Pattern and Risk of Citrus Huanglongbing: A Hierarchical Bayesian Modeling Approach

Abstract

Huanglongbing (HLB) is one of the most important diseases affecting citriculture in the world. Knowledge of climatic factors linked to HLB risk at large spatial scales is limited. We gathered HLB presence and absence data from official surveys conducted in the state of Minas Gerais, Brazil, over 13 years. The total count of orange and mandarin orchards, and mean orchard area, normalized to a spatial grid of 60 cells (55 × 55 km), were derived from the same database. Monthly climate normals (1984 to 2013) of rainfall, mean temperature, and wind speed split into rainy (September to April) and dry (May to August) seasons (annual summary was retained) were obtained for each grid cell. Two hierarchical Bayesian modeling approaches were evaluated, both based on the integrated nested Laplace approximation method. The first, the climate covariates model (CC model), used orchard, climate, and the spatial effect as covariates. The second, principal components (PC model), used the first three components from a principal component analysis of all variables and the spatial effect as covariates. Both models showed an inverse relationship between posterior prevalence and grid cell mean temperature during the dry season. Annual wind speed, as well as annual and rainy season rainfall, contributed to HLB risk in the CC and PC models, respectively. A partial influence of neighboring regions on HLB risk was observed. The results should assist policymakers in defining regions at HLB risk and guide monitoring strategies to mitigate further spread of HLB in the state of Minas Gerais.