The use of multiple regression via principal components in forecasting early season aphid (Homoptera: Aphididae) flight

Abstract

AbstractThe Rothamsted Insect Survey has kept records of aerial aphid activity using suction traps since 1965. Previous work has shown that, for certain species, there is a linear relationship between the date of first record in the trap each year and the mean temperature during the preceding winter. This paper describes and evaluates a more complex technique which relates the date of first record to a range of weather variables over 28 winter time periods. The technique uses principle components analysis to remove correlations between weather variables before these are regressed on the date of first record. Data from 1966 to 1988 inclusive are used to generate models to predict the date of first record of Myzus persicae (Sulzer) at Rothamsted, and the predictive values of models using both simple and multiple regression are assessed using data from 1989 to 1992. The accuracy of the multiple regression models was no greater than that of the simple regression models during these years. However, the multiple regression approach identified relationships with other variables for time periods when the correlation with mean temperature was weaker and may therefore be more widely applicable.