Evaluation of a model for predicting onset of Septoria nodorum blotch in winter wheat

Authors: ADHIKARI, URMILA - North Carolina State University; Cowger, Christina; OJIAMBO, PETER - North Carolina State University

Submitted to: Plant Disease
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/19/2022
Publication Date: 4/24/2023
Citation: Adhikari, U., Cowger, C., Ojiambo, P. 2023. Evaluation of a model for predicting onset of Septoria nodorum blotch in winter wheat. Plant Disease. 107:1122-1130. https://doi.org/10.1094/PDIS-06-22-1469-RE.
DOI: https://doi.org/10.1094/PDIS-06-22-1469-RE

Interpretive Summary: Computer models can help farmers make decisions about applying fungicides to control diseases of wheat. One disease of wheat is Septoria nodorum blotch (SNB, caused by the fungus Parastagonospora nodorum). The start or onset of an SNB epidemic in winter wheat is influenced by location, amount of wheat residue in the field, and certain weather factors during mid- and late March. A model previously developed based on these variables was evaluated for its ability to predict disease onset. The research to develop that model had shown there was a significant association between SNB epidemics that started “early” – before April 11 or 12 – and reduced wheat yields. In the current study, a field experiment was conducted at three locations in North Carolina in 2018, 2019 and 2020. Plots either received wheat straw covering the surface >20%, or received no straw. Plots were monitored for SNB symptoms, and disease onset was defined as the date when an average of 50% of plants per plot that were examined showed SNB symptoms. Of the 298 plots in the dataset, disease onset occurred early (i.e., prior to the April 11-12 cutoff) in 257 cases, while onset was late (after that April date) in 41 cases. The accuracy of the model in predicting whether disease onset would be before or after the April date ranged from 0.67 to 0.95, with a mean of 0.87 across the study period. The model was highly successful in predicting as “early onset” the plots that actually had early SNB onset (rates of 0.88 to 1.0 with a mean of 0.98 across all years). However, the model had a low success rate at predicting as “late onset” the plots that actually had a late onset of SNB (i.e., after April 11/12), with an average rate of 0.15 across the study period. Overall, there was no significant difference in the model’s prediction of plots as “early” or “late” and the actual outcome of “early” or “late” in those plots. The timing of disease onset was significantly correlated with grain yield, as had been observed when the model was developed, with earlier epidemics associated with lower yields. Thus, the study performed well in predicting early disease onset, and can help guide decisions on when to scout fields for SNB and whether to apply fungicides for reduction of SNB in wheat.

Technical Abstract: Prediction models that aid growers in making decisions on timing of fungicide application are important components of integrated management programs for several foliar diseases of wheat. The risk of Septoria nodorum blotch (caused by Parastagonospora nodorum) onset in winter wheat has been reported to be influenced by location, amount of wheat residue in the field, and cumulative daily infection values 2 weeks prior to day of year (DOY) 102. A model previously developed based on these predictor variables was evaluated for its ability to predict disease onset under field conditions. An experiment was conducted at three locations in North Carolina in 2018, 2019, and 2020, where plots were either treated with >20% wheat residue or received no residue treatment. Plots were monitored for disease symptoms, and disease onset was defined to have occurred when mean disease incidence in a plot was 50%. Of the 298 disease cases recorded, disease onset occurred early (i.e., prior to DOY 102) in 257 cases, while onset was late (i.e., on or after DOY 102) in 41 cases. Model accuracy based on correct classification ranged from 0.67 to 0.95, with a mean of 0.87 across the study period. Similarly, sensitivity rates of the model ranged from 0.88 to 1.0 with a mean of 0.98 across all years. However, the model had low specificity, with a mean rate of 0.15 across the study period. Overall, there was no significant difference in the frequency of observed and predicted cases in the study ('2 = 0.50, P = 0.7788, df = 2). Time to disease onset was significantly correlated with grain yield and explained 26% of variation in yield (P < 0.0001). Results indicated that the disease onset model performs well in predicting early disease onset but requires further evaluation and improvement, particularly in the Piedmont, where it over-predicted early onset in 2 successive years.