How to Develop More Efficient Sampling Designs When Estimating Salmon and Steelhead Adult Abundance

Different sampling designs can produce dramatic differences in the effort required to achieve a given accuracy.  This is especially true for resources that are both highly variable and spatially aggregated such as fish or redd density along a stream network. Here we use known redd locations for three populations over several years to compare five different probability sampling designs through simulation. The coefficient of variation (CV) for estimates based on simple random sampling was high with values well over 15% when sampling a third of the reaches. Moving to a spatially balanced sampling design (generalized random tessellation stratified, GRTS) produced improvements in two of the three watersheds (16-22% reduction in CV). Estimates based on a stratified GRTS design and a GRTS design that included a census of all reaches close to the peak count had higher accuracy with approximate CV of one half to a third of GRTS alone.  We show how these improvements are predicted by theory and under which conditions the different approaches are likely to perform well.