Predicting the behavior of individuals acting under their own motivation is a challenge shared across multiple scientific fields, from economic to ecological systems. In rivers, fish frequently change their orientation even when stimuli are unchanged, which makes understanding and predicting their movement in time-varying environments near built infrastructure particularly challenging. Cognition is central to fish movement, and our lack of understanding is costly in terms of time and resources needed to design and manage water operations infrastructure that is able to meet the multiple needs of human society while preserving valuable living resources. An open question is how best to cognitively account for the multi-modal, -attribute, -alternative, and context-dependent decision-making of fish near infrastructure. Here, we leverage agent- and individual-based modeling techniques to encode a cognitive approach to mechanistic fish movement behavior that operates at the scale in which water operations river infrastructure is engineered and managed. Our cognitive approach to mechanistic behavior modeling uses a Eulerian-Lagrangian-agent method (ELAM) to interpret and quantitatively predict fish movement and passage/entrainment near infrastructure across different and time-varying river conditions. A goal of our methodology is to leverage theory and equations that can provide an interpretable version of animal movement behavior in complex environments that requires a minimal number of parameters in order to facilitate the application to new data in real-world engineering and management design projects. We first describe concepts, theory, and mathematics applicable to animals across aquatic, terrestrial, avian, and subterranean domains. Then, we detail our application to juvenile Pacific salmonids in the Bay-Delta of California. We reproduce observations of salmon movement and passage/entrainment with one field season of measurements, year 2009, using five simulated behavior responses to 3-D hydrodynamics. Then, using the ELAM model calibrated from year 2009 data, we predict the movement and passage/entrainment of salmon for a later field season, year 2014, which included a novel engineered fish guidance boom not present in 2009. Central to the fish behavior model’s performance is the notion that individuals are attuned to more than one hydrodynamic signal and more than one timescale. We find that multi-timescale perception can disentangle multiplex hydrodynamic signals and inform the context-based behavioral choice of a fish. Simulated fish make movement decisions within a rapidly changing environment without global information, knowledge of which direction is downriver/upriver, or path integration. The key hydrodynamic stimuli are water speed, the spatial gradient in water speed, water acceleration, and fish swim bladder pressure. We find that selective tidal stream transport in the Bay-Delta is a superset of the fish-hydrodynamic behavior repertoire that reproduces salmon movement and passage in dam reservoir environments. From a cognitive movement ecology perspective, we describe how a behavior can emerge from a repertoire of multiple fish-hydrodynamic responses that are each tailored to suit the animal’s recent past experience (localized environmental context). From a movement behavior perspective, we describe how different fish swim paths can emerge from the same local hydrodynamic stimuli. Our findings demonstrate that a cognitive approach to mechanistic fish movement behavior modeling does not always require the maximum possible spatiotemporal resolution for representing the river environmental stimuli although there are concomitant tradeoffs in resolving features at different scales. From a water operations perspective, we show that a decision-support tool can successfully operate outside the calibration conditions, which is a necessary attribute for tools informing future engineering design and management actions in a world that will invariably look different than the past.