Nonstationarity in wildlife disease dynamics: insights from the prairie dog–plague system

Ecological relationships often vary in strength and direction across ecosystems, a pattern that has become increasingly evident as ecology expands to broader spatial and temporal scales. Although such nonstationarity – the changing effect of predictor variables across spatiotemporal contexts – is widely acknowledged in ecological research, it has received limited attention in the study of wildlife diseases. Here, we evaluated nonstationarity in sylvatic plague epizootics among black‐tailed prairie dog Cynomys ludovicianus colonies across nine federally managed grasslands in the Great Plains, USA. Using a 29‐year dataset (~ 1992–2020) and a mixed‐effects machine learning approach, we assessed how the effects of climate and colony characteristics on plague varied in strength and direction at each grassland. Across sites, colony size, clustering, and structural connectedness showed generally stationary, positive associations with outbreak probability. By contrast, weather effects were strongly nonstationary: temperature and precipitation alternately increased, decreased, or showed little association with outbreaks depending on the grassland. Predictor importance also shifted across space and within epizootics; at some sites precipitation signaled outbreak onset, whereas host spatial structure governed subsequent spread and persistence. These results indicate that host aggregation provides a consistent scaffold for transmission, while environmental drivers act locally and vary through time. Recognizing this contrast can improve forecasting and mitigation, as interventions such as vaccination or vector control are likely to perform differently across environmental contexts. Methodologically, explicitly modeling slope heterogeneity identified where and when covariate effects differ, offering a transferable approach for disentangling local from broad‐scale drivers of wildlife disease across heterogeneous landscapes.