Bowling for bees: optimal sample number for “bee bowl” sampling transects

Abstract

The use of bowl traps to sample bee diversity has become common in recent years. To provide guidance in optimizing bee bowl sampling effort, we used several independent datasets from North America (Pacific Northwest golf courses, suburban Maryland, and Chihuahuan Desert) to estimate the number of bowls in a sampling event at which additional bowls added little to estimates of species richness or to cumulative species lists. We examined changes in the value of a nonparametric richness estimator (Chao2) as a function of bowl number by randomly resampling (without replacement) the data for each sampled transect. We also carried out rarefaction (interpolation of expected species richness for all numbers of bowls up to the actual number). Using as our criterion of “optimal” number of bowl samples per transect the number of samples at which five additional bowls added <1 additional species to the Chao2 estimate, we find that 30-bowl transects would have met this standard for nearly three quarters or more of the sampling events in any of our three datasets (for the suburban Maryland and Chihuahuan Desert data sets, at least three quarters of sampling events met this criterion with just 20 and 25 bowls, respectively). Results from rarefaction analysis for all sampling events indicate that 30-bowl transects would have met our “five additional bowls” standard for more than 80 % of our suburban Maryland and Chihuahuan Desert sampling events, but for only about 61 % of our Pacific Northwest sampling events. For the Pacific Northwest sampling events, there was notable among-site heterogeneity in required sampling intensity. Thus, although 30-bowl transects appear to be adequate for assessing local bee species richness in most instances, investigators may wish to increase (or decrease) this number based on their particular goals and specific knowledge about local bee diversity.