Assessment of Bias and Precision Associated with Roving Creel Survey Designs to Estimate Chinook Salmon Harvest in Idaho

     Recreational fisheries for Chinook salmon Oncorhynchus tshawytscha in Idaho typically consist of relatively short seasons with one to three fish bag limits, high angler effort, and low catch rates.  The fisheries are closely monitored due to concerns associated with mortality of caught and released wild Chinook salmon.  Currently, a roving creel survey is the primary technique used to estimate the dynamics of the fisheries.  We used observations of anglers and historic creel survey data to create profiles of angler use and harvest.  Re-sampling simulations were then used to assess the precision and accuracy of harvest estimates associated with a traditional two-stage cluster design, as well as simple random sampling, random and non-random systematic sampling, and non-uniform probability sampling designs.  Two-stage sampling designs with days as the primary sampling unit and morning and afternoon shifts as the secondary sampling unit consistently produced biased results while simple random sampling and non-uniform probability sampling produced results that were relatively unbiased.  Most sampling designs underestimated harvest over the length of the season, likely due to low catch rates and a high proportion of anglers who did not catch any fish.  Bias can be reduced by sampling the entire length of the fishing day or randomizing the start times of the secondary sampling unit.