Minimizing the influence of sampling variation in ordination and in the field for lake fish communities

Sampling variation was studied from two perspectives. First, using simulated communities, I examined the effects of sampling error and gradient length on five ordination methods. I found that correspondence analysis and principal component analysis using presence-absence data were most robust to the influence of sampling error and gradient length. Second, I examined how variance is apportioned in lake fish communities, using four intensively sampled lakes selected from an extensive dataset. I compared how accurately five gears measured the proportional abundance of individual fish species, and I resampled the dataset to determine the combination of gears that most accurately sampled each community (i.e. sample richness was closest to actual richness). I found that gear choice has important effects on variance, gear or gear by lake contributes most to variance estimates, and recommend the use of 5 trapnets and 25 plastic traps to obtain the best estimate of richness while minimizing effort.