Environmental DNA (eDNA) analysis is a technique for detecting organisms based on genetic material in environments such as air, water, or soil. Observed eDNA concentrations vary in space and time due to biological and environmental processes. Here we investigate variability in eDNA production and loss by sampling water adjacent to a managed population of non-native cetaceans on a near-hourly timescale for 48 hours. We used diverse sampling approaches and modeling methods to describe time variability in observed eDNA concentrations and then compare the magnitude of production and loss mechanisms. We parsed production and loss in a Bayesian process model, and compared biological and physical loss rates using a decay experiment and a physical transport-and-diffusion tracer model. We then evaluated eDNA concentrations along a transect away from the animal enclosure in light of model parameter estimates. We conclude that eDNA production is best conceptualized using a time-varying mixed-state model and biological losses are small relative to physical losses in the marine environment. Because physical loss is unsteady and nonlinear, tracer models are especially helpful tools to estimate it accurately.