High‐stakes steeplechase: a behavior‐based model to predict individual travel times through diverse migration segments

Many migratory species traverse highly heterogeneous landscapes, often including habitats that have been altered by human activities. Modeling migration dynamics is challenging because individual variability in behavior at multiple spatial and temporal scales can produce complex, multi-modal distributions in migration travel times. Moreover, behavioral responses to conditions encountered en route can affect habitat-specific migration rates which then influence bioenergetic costs and mortality risk over the entire migration. To quantify impacts of conditions within migration corridors, refined analyses of behavior are needed. In this study, we developed a behavior-based simulation model that predicts individual adult salmon migration duration over 24 distinct river reaches totaling 922 km, including eight hydropower dams. The study population, threatened Snake River spring/summer Chinook salmon (Oncorhynchus tshawytscha), had observed migration durations ranging from 23 to 108 d. In a novel application of N-dimensional mixture models, which can account for subpopulations that behave differently, we simulated “fast” vs. “slow” travel through migration reaches. The proportion of migrants in each category was determined by diel, seasonal, and proximate river conditions, which captured the temporally shifting bimodal patterns in the data. We fit reach-specific models with data from 2188 tagged salmon migrating in 2000–2013 and validated the cumulative model with additional data through 2015. By accounting for multiple behaviors in this way, the model successfully recreated the breadth and variability in total travel times to within 3% of observed durations throughout the 5th–95th quantiles. En route mortality appeared to account for the loss of the slowest fish that encountered record-breaking high temperatures in 2015. For Chinook salmon, this combined reach and cumulative travel-time model provides an opportunity to link high-resolution behavioral data to individual fitness and population-level impacts on viability. More generally, the N-dimensional modeling approach offers a framework for assessing the cumulative impacts of alternative behaviors at small spatial and temporal scales. Improved accounting of changes in migration rate in response to local conditions will aid recovery efforts for species of concern traversing complex migration corridors.