Scott, C. B.; Cleary, S.; Halliday, F. W.; Joyner, B.; O'Keeffe, K.; Stiver, I.; Mitchell, C. E.

Forecasting disease epidemics may require considering both abiotic and biotic conditions. Abiotic conditions can influence pathogen dispersal, survival, and infection, while prior infection by one pathogen species can alter host susceptibility to subsequent pathogens, creating historical contingencies. Yet, the relative importance of environmental conditions, within-host pathogen interactions, and their potential interplay in predicting seasonal disease dynamics remain underexplored. To better understand these interactions and improve ecological forecasts of disease risk, we analyzed more than 41,000 plant-level observations of three foliar fungal diseases on the grass species tall fescue in North Carolina, collected from 2017--2024. We built temporally explicit random forest models which for two of the three focal diseases accurately (>80%) predicted future disease dynamics from a complex abiotic and biotic predictor space. From these models, we identified key environmental thresholds that intensified annual epidemics. We then supplemented those machine-learning models with Bayesian hierarchical models and survival analyses, finding that pathogen-pathogen interactions can be as important as environmental conditions in predicting disease risk. Furthermore, for two of the three focal diseases, prior infection by a different pathogen facilitated the subsequent infection and the strength of this facilitation was modulated by environmental context. From this study we draw two major conclusions, one ecological and one methodological. First, knowledge of a host's current disease state may be as important as local environmental conditions for predicting disease dynamics. Second, by integrating complementary modeling approaches, we can develop both predictive forecasts and mechanistic insight into the biotic and abiotic drivers of infectious disease.