Dalton Hance, Russell Perry, Adam Pope, John Plumn, Ken Tiffan, Mike Dodrill

U.S. Geological Survey, 5501A Cook-Underwood Road, Cook, WA 98605

The passive integrated transponder (PIT) systems of the Columbia River basin are a generational investment that remain unrivalled by any other fisheries monitoring system in terms of volume and cohesiveness of data. In any given year, millions of PIT-tagged fish are released, detected, recaptured, or recovered at a multitude of monitoring locations. This wealth of data is unified in PTAGIS, the structure of which reflects shared elements of the data-generating ecological processes. The location and timestamp of each detection or recovery event after release result from common survival, migration-timing, and detection processes. Data on individuals migrating at similar times and locations will be similarly affected by environmental or management factors that influence the underlying processes. PTAGIS is an informational gold mine for fisheries management.

However, the potential of PTAGIS to inform management is not fully realized, in part because commonly used analytical methods discard large amounts of data. One critical management question can be stated: “How do environmental and operational conditions at the time of passage at a given location affect fish survival?” A common approach to answering this question is to aggregate PIT tagged fish into virtual release (VR) cohorts (i.e., a group of fish detected at a location within a given time window). Each cohort is then treated as independent for the purposes of estimating survival and detection probabilities. The length of the grouping interval (e.g. two-weeks) is typically selected to balance the limitations of conventional statistical models with the need to understand the effect of management actions at useful timescales. Because the VR approach only uses information on detected fish smaller timeframes result in more imprecise estimates, particularly as sample sizes decrease due to low detection probabilities. Conversely, as timeframes expand, VR-based analyses become less useful for quantifying factors affecting survival because covariates of interest (e.g., spill, % gas, flow) must be summarized over the same timeframe. This reduces contrast in the covariate data, potentially obscuring important relationships. These shortcomings of the VR approach limit the ability of managers to use information from PIT tags to design more effective and efficient fish passage operations.

We developed the temporally stratified Cormack-Jolly-Seber (TSCJS) model framework to overcome the limitations of the VR approach. TSCJS explicitly models a realistic data generating process of temporally-varying survival, arrival, and detection probabilities – including separate routes of passage where applicable (e.g., spillway vs bypass at Lower Granite Dam). This framework was designed to accommodate multiple releases of fish each with varying release locations and times and their subsequent detection at any number of downstream locations. By jointly modelling arrival timing, the TSCJS framework uses information from all fish released rather than just fish detected. This greatly improves statistical efficiency and allows for precise estimates of survival and detection at time scales relevant to management (e.g., daily). We applied the TSCJS to Snake River Fall Chinook salmon to gain insight into an important population which presented challenges for traditional methods due to a protracted migration timing. For example, we used TSCJS to estimate daily and annual abundance of age-0 fall Chinook salmon passing Lower Granite Dam and to partition abundance by origin (natural or hatchery). In this talk, we will provide an overview of the TSCJS model framework, application to the Snake River fall Chinook population, and highlight advantages of this approach over traditional methods.