Novel method of determining parameters for the effective accumulated temperature model by using seasonal pest occurrence data

To improve the prediction accuracy of pest forecasting, we developed a machine-learning-based method of estimating parameters of the effective accumulated temperature (EAT) model on the basis of numerous historical occurrence data on target pests and meteorological data in the field. The parameters were estimated by using past occurrence data of the white-backed planthopper (WBPH), Sogatella furcifera, collected in light traps at 20 sites on Shikoku Island from 1980 to 2000. When the accuracy of the estimated parameters was compared with that of those derived from traditional approaches using field data from 2001 to 2022, the parameters estimated by using the novel method had the best accuracy, with a predictive lower threshold temperature (Tp) = 12.29 (°C) and an effective accumulated temperature (KTp) = 372.23 (degree-days) for the first generation of WBPH and Tp = –5.00 and KTp= 800.31 for the second generation. Use of these parameters improved the prediction accuracy by approximately 3 days compared with the traditional approach. We also found that prediction of parameters on the basis of the coefficient of variation of the predicted EATs resulted in better forecasting accuracy than prediction based on the standard deviation. Our novel method of estimating parameters for the EAT model will contribute to better forecasting of insect pests.