Evaluating probability sampling strategies for estimating redd counts: an example with Chinook salmon (Oncorhynchus tshawytscha)

Abstract

Precise, unbiased estimates of population size are an essential tool for fisheries management. For a wide variety of salmonid fishes, redd counts from a sample of reaches are commonly used to monitor annual trends in abundance. Using a 9-year time series of georeferenced censuses of Chinook salmon (Oncorhynchus tshawytscha) redds from central Idaho, USA, we evaluated a wide range of common sampling strategies for estimating the total abundance of redds. We evaluated two sampling-unit sizes (200 and 1000 m reaches), three sample proportions (0.05, 0.10, and 0.29), and six sampling strategies (index sampling, simple random sampling, systematic sampling, stratified sampling, adaptive cluster sampling, and a spatially balanced design). We evaluated the strategies based on their accuracy (confidence interval coverage), precision (relative standard error), and cost (based on travel time). Accuracy increased with increasing number of redds, increasing sample size, and smaller sampling units. The total number of redds in the watershed and budgetary constraints influenced which strategies were most precise and effective. For years with very few redds (<0.15 reddsÁkm -1 ), a stratified sampling strategy and inexpensive strategies were most efficient, whereas for years with more redds (0.15-2.9 reddsÁkm -1 ), either of two more expensive systematic strategies were most precise.Résumé : La gestion des pêches requiert comme outils essentiels des estimations précises et non faussées de la taille des populations. Chez une grande diversité de poissons salmonidés, le décompte des frayères dans un échantillon de sections de cours d'eau sert couramment à suivre les tendances annuelles d'abondance. En utilisant une série chronologique de 9 années d'inventaires géoréférencés de frayères de saumons chinook (Oncorhynchus tshawytscha) dans le centre de l'Idaho, É .-U., nous testons une gamme étendue de stratégies courantes d'inventaire pour l'estimation de l'abondance totale des frayères. Nous évaluons deux tailles d'unités d'échantillonnage (sections de 200 et de 1000 m), trois proportions échantillonnées (0,05, 0,10 et 0,29), ainsi que six stratégies d'échantillonnage (échantillonnage par indices, échantillonnage aléatoire simple, échantillonnage systématique, échantillonnage stratifié, échantillonnage par groupements adaptatifs et un plan équilibré spatialement). Nous évaluons les stratégies d'après leur exactitude (couverture de l'intervalle de confiance), leur précision (erreur type relative) et leur coût (temps de déplacement). L'exactitude augmente en fonction directe du nombre de frayères et de la taille des échantillons et en fonction inverse de la taille des unités d'échantillonnage. Le nombre total de frayères dans le bassin versant et les contraintes budgétaires déterminent quelles stratégies sont les plus précises et les plus efficaces. Les années avec très peu de frayères (<0,15 frayèreÁkm -1 ), la stratégie d'échantillonnage stratifié et les stratégies moins chères s'avèrent être les plus efficaces, alors que les années a...

Trade-offs between monitoring objectives and monitoring effort when classifying regional conservation status of Pacific salmon (Oncorhynchusspp.) populations

“…In the field though, adult coho salmon density is linked to factors such as wetland occurrence, local geology, stream gradient, fishery harvest, and land-use (Pess et al 2002), while trends in abundance at individual sites are sometimes linked to changes in harvest and landuse patterns (Bradford and Irvine 2000;Bilby and Mollot 2008). When spatial patterns such as these exist, sampling performance is usually improved by using more sophisticated spatial designs, such as stratified sampling, systematic sampling, and spatially balanced stratified sampling (Stevens and Olsen 2004;Courbois et al 2008).…”

“…In the former, sites are selected for primarily logistical reasons (i.e., nonprobabilistic sampling) while in the latter, sites are grouped into clusters based on location and the clusters themselves are randomly selected for sampling. Realistic cost functions for sample design and methodology have been included in evaluations of small-scale monitoring programs (Courbois et al 2008;AlChokhachy et al 2009). A useful extension of our study would be to develop a habitat-based model of adult salmon distribution and then examine how spatial complexity and realistic cost functions affect the relative performance of alternative sampling designs at a regional scale.…”

Proceedings of the coast redwood forests in a changing California: A symposium for scientists and managers

“…Redd density (an index of abundance) in LCS was lower between 2009 and 2011 than observed since 2000 and was outside the range of data we used earlier (Gallagher et al 2010b) to develop sample size estimates. Courbios et al (2008) found that a larger sampling fraction and higher redd abundance resulted in better accuracy for GRTS. At low redd abundance none of their sampling designs were accurate.…”

The spatial distribution of salmon and steelhead redds and optimal sampling design

“…Redd surveys occurred weekly from mid-September through early November for fall Chinook salmon, and every 2 weeks from early March through late May for winter steelhead. During each foot survey, newly identified redds were flagged and georeferenced using recreationalgrade global positioning system (GPS) units (Courbois et al 2008;Jacobs et al 2009). The reported accuracies from the GPS units generally ranged from 2 to 15 m. In subsequent surveys, previously flagged redds were inspected to determine whether they should be classified as "still visible" or "not visible".…”

“…While probability sampling can guarantee unbiased estimates, variance can be high owing to the patchy distributions common in salmonid spawning locations (e.g., Irvine et al 1992;Isaak and Thurow 2006). Stratified designs that concentrate sampling effort in areas with higher redd density and spatially balanced designs that ensure an even spatial distribution of sample reaches have been shown to substantially reduce estimator variance (Courbois et al 2008;Gallagher et al 2010;Jacobs and Nickelson 1998;Jacobs et al 2009). These results align with general sampling theory that predicts increased accuracy for designs that sample units with higher predicted variance more intensively (e.g., Cochran 1977) and constrain samples to be representative with respect to auxiliary variables that are related to the variable of interest (e.g., Tillé 2011).…”