Understanding Trade-Offs Between Modeling Approaches to Forecast Plant Disease Spread and Risk

Modeling disease spread and risk is crucial for disease management, with numerous options for models. Multiple factors affect model selection, including the model's spatial and temporal dimensions, the quantity and quality of infection data, management objectives, and the research questions being addressed. We developed a workflow to filter model types based on pathogen biology and infection data availability, then select the appropriate model for the research question and management objectives. We applied the workflow to two case study pathosystems (late blight and sudden oak death) and evaluated three modeling approaches representing different spatial and temporal dimensions: a temporally focused weather-risk model (WRM), a spatially focused species distribution model (SDM), and a spatiotemporal spread model, the Pest or Pathogen Spread (PoPS) model. We qualitatively compared how each model's spatiotemporal outputs affected predicted disease risk or spread. Additionally, we conducted a simulation study in which infection data were reduced to determine the minimum data density required to obtain reliable parameter estimates with PoPS. Our results suggest that the temporal WRM model is suitable for disease management in agricultural pathosystems, where stakeholders possess sufficient spatial knowledge. In the natural pathosystem, the spatiotemporal model's greater accuracy reduced the total high-risk infection area by >99% compared with the other models. We found that although a spatiotemporal model is more accurate and helps with resource allocation, performance dropped significantly when data density fell below 50%. The effect of decreasing data density on SDM performance varied with pathosystem. Our work underscores the need for scalable disease-detection methods.