Refining Entomological Models: Strengthening Data Foundations for Accurate Predictions

Abstract

“All models are wrong, but some are useful” (Box, 1976) has been frequently cited as a guiding principle for modelers. One of the most significant limitations for model accuracys is bias in the training and validation datasets, which results in models that do not reflect reality. Accuracy and reliability are critical for the adoption of model-based decision tools (Rossi et al., 2019). Consequently, poor-quality training and validation data limit their utility and adoption. The adoption and usage of these tools are critical for the implementation of Integrated Pest Management (IPM) and for facilitating improvements in large-scale crop protection. Therefore, it is critical to improve methodologies used for data collection to produce adequate training and validating model datasets. The development of new and robust datasets will foster the creation of more effective and timely models. The development of degree day models, which describe the temperature-mediated development of many pests and are used to forecast seasonal phenology, is a timely and onerous process. The delays inherent to the conventional process of collecting the necessary data used as model inputs limits a rapid response to invasive species and the updating frequency when regionally adapted biotypes no longer fit the models’ parameters. In Chapter 1, a novel usage of a differential scanning calorimeter is utilized to derive a degree day-based phenological model leveraging the life tables developed by (Neven 2024; Kistner 2017). This novel method of generating a dataset for degree day-based phenology models hastens the development of the prerequisite life tables from months to weeks. This will empower modelers with the ability to train and validate models much more quickly and enable the rapid development of tools tailored to novel invasive pests or regionally adapted biotypes. Gaps in datasets produce biases that reduce accuracy and utility of models. Pest phenology

models used for supporting decision-making require extensive validation of the models’ outputs through ground truthing (Graf et al., 2002). Building the requisite datasets for ground truthing extensive regional models is an expensive and time-consuming process. Therefore, the efficient allocation of resources to maximize coverage and observations in areas at risk from pests is another critical objective for improving the quality of datasets for training regional pest phenology models. In Chapter 2, a methodology is developed to evaluate existing and potentially new monitoring locations by creating a “site score metric.” This metric evaluates the intensity of corn growth at a location, past pest pressure, and the distance to the closest existing monitoring location to quantitatively assess the utility of these locations. Quantitatively assessing these locations will empower those running regional monitoring networks to allocate resources more efficiently toward areas with the most value. Useful models should include variables relevant to biological processes to achieve accurate forecasts. Range expansion models are used to inform long-term decision-making by forecasting the climate change-mediated changes to pests’ distributions. While these models leverage available long-term climate change estimates and current thermal tolerance of insects, they have failed to incorporate the full suite of relevant biological factors such as insects’ capacity for rapid evolution (Garnas, 2018). In Chapter 3, a method is developed to estimate the rate of evolution for cold tolerance for the corn earworm (Helicoverpa zea). This rate is then incorporated into a range expansion model, the output of which is compared to a conventional range expansion model developed by (Lawton et al. 2022) to demonstrate the potentially greater poleward expansion this pest may achieve. The gap between the two models establishes the importance of incorporating novel variables, such as an evolutionary rate, when making long-term forecasting models. Model utility is limited by the quality and size of available datasets for training and validation. This work addresses common limitations in entomological modelling by introducing novel methods for generating life table data more efficiently, optimizing monitoring site selection through a quantitative scoring system, and incorporating evolutionary dynamics into long-term range expansion models. Together, these advancements contribute to the creation of more timely, adaptable, and biologically relevant models that can better inform pest management decisions in an era of rapid environmental change.