Hierarchical forecasting models of stink bug population dynamics for pest management

Abstract

In recent decades, the intensification of agricultural production, accompanied by an increasing pressure from pests in various crops, has resulted in a substantial increase in the use of synthetic pesticides. Integrated pest management (IPM) provides a framework for the development and use of sustainable control strategies, which include the monitoring of the crop pest complex and the use of decision tools such as predictive models. In this study, an empirical modeling approach based on a hierarchical Bayesian model with a state-space structure was developed to perform stink bug (Pentatomidae) density forecasts to assist in deciding when to carry out pest control interventions, thus increasing the efficacy and efficiency of IPM. Using stink bug abundance and crop phenology data, along with meteorological data from eight different sites in Argentina, we made 1-week forecasts of population density, evaluated the predictive capacity of different models using Leave-One-Out-Cross-Validation, and analyzed how the uncertainty in the predictions vary as a function of the number of vertical beat sheet samples. The forecasts made with our best model showed a reasonable degree of accuracy. We found that i) the observation error of the vertical beat sheet method was much larger than expected, and ii) the uncertainty analysis suggested a sample size of 40 to obtain a good balance between precision and sampling effort, which is in stark contrast to the average sample size usually taken by advisors. Our approach provides advisors with a tool to make better-informed decisions about when and if to carry out pest control interventions.