Retrospective Sampling Strategies Using Video Recordings to Estimate Fish Passage at Fishways

Abstract

AbstractReliable estimates of population size are critical for fisheries management and for testing ecological hypotheses but can be expensive and time consuming to obtain. Sampling methodologies have been developed to obviate complete enumeration, but their effectiveness can be limited by logistical constraints. A posteriori sampling from digital video recordings, however, permits the application of otherwise impractical sampling schemes. We evaluated the time savings of estimating total run size by a posteriori sampling of video recordings (assisted by motion detection software) in comparison with on‐site counting methods. The evaluation was based on 9 years of complete counts enumerated either on site or by a posteriori counts from video recordings of alewives Alosa pseudoharengus migrating through a fishway in Nova Scotia. We compared results obtained using analytical estimation methods with results from simulation‐based methods and found them to be the same for simple random sampling and daily stratified random sampling. We tested the application of using motion detection software to automatically omit sample units with zero counts from sampling strategies; large reductions in sampling requirements were obtained for data sets with large proportions of zero counts. We also evaluated the effect of sample unit size on sampling effort requirements; use of short but frequent sample units allowed for large reductions in sampling effort. Use of motion detection software in combination with shorter sample units achieved highly significant aggregate reductions in sampling effort. For example, at the shortest tested sample unit size of 1 min, it was possible to reduce sampling requirements to 4 min/d based on daily stratified random sampling to achieve an estimate of the true population within 20% and with 95% confidence. Finally, we evaluated linear interpolation to estimate fish passage for missed days; although average bias was small, bias was substantial when a peak run day was missed.