Building phenological models from presence/absence data for a butterfly fauna.

Species phenology is increasingly being used to explore the effects of climate change and other environmental stressors. Long-term monitoring data sets are essential for understanding both patterns manifest by individual species and more complex patterns evident at the community level. This study used records of 78 butterfly species observed on 626 days across 27 years at a site in northern California, USA, to build quadratic logistic regression models of the observation probability of each species for each day of the year. Daily species probabilities were summed to develop a potential aggregate species richness (PASR) model, indicating expected daily species richness. Daily positive and negative contributions to PASR were calculated, which can be used to target optimum sampling time frames. Residuals to PASR indicate a rate of decline of 0.12 species per year over the course of the study. When PASR was calculated for wet and dry years, wet years were found to delay group phenology by up to 17 days and reduce the maximum annual expected species from 32.36 to 30. Three tests to determine how well the PASR model reflected the butterfly fauna dynamics were all positive: We correlated probabilities developed with species presence/absence data to observed abundance by species, tested species' predicted phenological patterns against known biological characteristics, and compared the PASR curve to a spline-fitted curve calculated from the original species richness observations. Modeling individual species' flight windows was possible from presence/absence data, an approach that could be used on other similar records for butterfly communities with seasonal phenologies, and for common species with far fewer dates than used here. It also provided a method to assess sample frequency guidelines for other butterfly monitoring programs.