Bull trout streamflow and temperature linear SCR model

Abstract In the Pacific Northwestern United States, climate change is increasing air temperatures, decreasing warm season (April–September), and increasing cool season (October–March) streamflow. Warmer water temperatures produced by both reduced streamflow and warmer air temperatures may alter conditions for migratory, cold-water fishes like bull trout (Salvelinus confluentus). Consequently, understanding bull trout migration and survival is critical for species conservation and restoration in an uncertain future. We evaluated pre- and post- spawning migrations and survival of fluvial bull trout radiotagged and tracked in the Salmon River basin, Idaho from 1992–1994. Both 1992 and 1993 recorded two of the most extreme warm season streamflows during the last three decades. These extremes provided a unique opportunity to retrospectively compare bull trout survival and migration under potential climate change scenarios. We used a Cormack Jolly-Seber linear spatial capture-recapture model to simultaneously model migration and survival of radio-tagged pre-spawning (n = 63) and post-spawning (n = 23) bull trout among weeks and river reaches with streamflow, water temperature, and habitat covariates. Most individual pre-spawning migrations (May 26–September 28) were similar among tagged fish, whereas post-spawning fish (August 12–May 12) adopted multiple migration and overwintering strategies. Movements of pre-spawning bull trout were larger when between-weekly changes in streamflow decreased, weekly average daily maximum streamflow increased, and weekly average daily maximum water temperature increased. More than 50% of spawners died and mean weekly pre-spawning apparent survival was higher in the low streamflow year (x̄ = 0.97, CI 0.93–1), compared to the higher and more variable streamflow year (x̄ = 0.91, CI 0.76–0.98). Survival during the 38-week post-spawning period was lowest (x̄ = 0.95, CI 0.90–0.98) when weekly maximum average daily water temperatures were coldest. Bull trout detections (n = 880 detections) were generally higher in sites with more complex habitats, less large woody debris, and fewer undercut banks. Our results increase knowledge of bull trout migration and survival and offer insights into changes that might be expected under future climate. Methods We adapted the Raabe et al. (2014) linear spatial capture-recapture (SCR) model for use with the bull trout radio-telemetry data where reaches correspond to arrays used in other fish migration studies (Gardner et al. 2010). The SCR model uses the basic framework of the Jolly-Seber open spatial capture-recapture model (Gardner et al. 2010) with a Cormack Jolly-Seber formulation that is conditional on first capture. SCR models are an efficient, precise, and unbiased method for estimating the spatial scale of detection rates, and effects of environmental covariates on detection rates, movements, and survival simultaneously (Gardner et al. 2010; Raabe et al. 2014; Harris et al. 2020), even for species with low detection rates (Blanc et al. 2013; Royle et al. 2014; Leuenberger et al. 2019). The Raabe et al. (2014) linear SCR model is an extension of the Gardner et al. (2010) SCR model for migrating stream fishes that requires detection coordinates, capture-recapture histories of marked individuals, and covariates (Leuenberger et al. 2019). The main components of the SCR model are an observation model based on detections (λijt), a state model (zit), based on whether the fish was alive and in the river system, dead, or had emigrated, and latent individual activity (or home range) centers (Si; Gardner et al. 2010; Raabe et al. 2014). We used Raabe et al.’s (2014) SCR model in a Bayesian framework to evaluate the relationship between environmental and habitat covariates and bull trout detection rates, survival, and movement distances over time. The SCR model requires three standard open capture-recapture model assumptions: tagging did not influence survival, relocation probability was similar among tagged individuals, and tags were not lost or missed (Williams et al. 2002). We believe we have met these assumptions because fish survived tagging and were later detected near spawning sites, we detected fish throughout Rapid River and the Salmon River, and mortalities were removed from the dataset and included fish that stopped moving.