Dynamic time warping analysis of accelerometry data: a tool for interpreting fine-scale movement patterns during fish...

Post-release survival in catch-and-release fisheries is highly variable and context dependent, yet management agencies often apply uniform survival estimates. A significant source of this uncertainty is our limited understanding of how the capture event itself drives physiological stress and recovery. Fight dynamics reflect how much effort a fish expends and the exhaustion it experiences. Non-invasive accelerometers attached to the fishing line to record the capture event provide a powerful yet underused tool for quantifying these fight dynamics. Traditional analytical approaches reduce accelerometry data to summary statistics, obscuring fine-scale temporal patterns. Here, we introduce a toolbox based on dynamic time warping (DTW) to preserve the temporal structure of capture events and link fight behaviour to physiological disturbance. We attached tri-axial jerk accelerometers ('jerk' tags) to fishing lines and captured Chinook salmon (Oncorhynchus tshawytscha) and coho salmon (O. kisutch) using rod-and-reel angling off the Pacific coast of Vancouver Island in British Columbia, Canada. We quantified physiological disturbance by sampling blood pH and plasma lactate 1 h post-capture. To determine whether biologging tags can predict physiological outcomes, we systematically evaluated 13 analytical pipelines, including jerk summary metrics (fight duration, burst frequency, intensity patterns) and DTW-based approaches using raw, filtered and differentiated jerk data. These complementary approaches captured different behavioural dimensions linking fight to physiology. Summary metrics described broad fight patterns and clustered fish with temporally similar fight signatures, while DTW detected Y-axis dimensions linked to individual recovery (plasma lactate). This toolbox is transferable across species and logger types, requiring only accelerometers attached to angling gear. By systematically evaluating processing pipelines rather than defaulting to conventional metrics only, researchers can optimize inference and identify behavioural signatures that predict physiological disturbance. This approach provides a scalable tool for developing evidence-based best practices to improve conservation outcomes.