Weather-based logistic regression models for predicting the risk of wheat blast epidemics

• Ano de defesa: 2023
• Autor(a) principal: Cól, Monalisa Cristina de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa; Fitopatologia
• Palavras-chave em Português: Trigo - Doenças e pragas - Previsão; Magnaporthe oryzae; Triticum aestivum; Fitopatologia
• Link de acesso: https://locus.ufv.br//handle/123456789/31930; https://doi.org/10.47328/ufvbbt.2023.529

Wheat blast, caused by Pyricularia oryzae Triticum lineage, is an important yield- limiting disease in the tropics of Brazil. This study aimed to develop models for predicting the within-season risk of wheat blast outbreaks. Data sets used in this study were obtained from field trials conducted in Patos de Minas (n = 103 epidemics) as well as in 10 other locations (n = 40 epidemics); the latter as part of the cooperative fungicide trial network. The trials were conducted across six states of Brazil over a nine-year period (2012-2020). A binary response variable was created based on disease incidence being ≥20% or <20%. Daily meteorological variables including minimum (Tmin), maximum (Tmax), and mean temperature (Tmean), relative humidity (RH) and precipitation (P) were obtained from the NASA POWER. The wheat heading date (WHD) was used to define four time windows, consisting of two intervals of seven days each, before and after the WHD. These windows combined with the weather variables resulted in 36 prediction variables (9 weather variables × 4 windows). Logistic regression models were fitted with selection of variables using LASSO followed by best subset selection. Four accuracy measures were computed and the model performance was evaluated using leave-one-out cross validation (LOOCV). The models were further run for six sites using a historical series of 23 years of weather data. The variables that best explained outbreak risk were RH, days with Tmean < 22ºC and Tmean x RH two weeks before WHD, Tmean x RH one week before WHD, and RH and Psum one week after WHD. The selected models included 2 to 5 predictors, with accuracies ranging from 0.80 to 0.85. Sensitivities ranged from 0.80 to 0.91, specificities from 0.72 to 0.86, and AUC values from 0.89 to 0.91. The accuracy values obtained for LOOCV ranged from 0.77 to 0.81. The model predictions generally agreed with most historical reports in the tropical region of Brazil. This study enhanced our understanding of the complex relationship between weather variables and wheat blast, contributing valuable insights for disease management. Keywords: Magnaporthe oryzae. Triticum aestivum. Prediction models.