Bayesian models describing microhabitat associations of infrequently captured small mammals sampled under a complex hierarchical design

Abstract

Knowledge about the relationship between habitat structure and abundance of a target species facilitates biodiversity conservation in managed forests. However, modeling the relationship for infrequent small mammal species in silvicultural experiments introduces the challenge of excessive zero counts and complex hierarchical sampling. A common solution has been to ignore infrequent species. The goal of this study was to model microhabitat associations of infrequently captured forest floor small mammal species with Bayesian models that accounted for subsampling and the blocking design of a large-scale variable-retention harvest experiment. Poisson, negative binomial and overdispersed Poisson Generalized Linear Mixed Models (GLMMs) were fitted to data for three small mammal species with different rates of capture. Shrew–mole (Neurotrichus gibbsii) and Keen’s deer mouse (Peromyscus keeni) were the two infrequent species and southern red-backed vole (Myodes gapperi) was the frequent species selected for modeling. Capture rate was predicted from variables representing vegetation structure, and results were compared to corresponding Generalized Linear Models (GLMs). GLMMs predicted stronger and sometimes contrary effects of vegetation structure with wider confidence intervals compared to GLMs. The overdispersed Poisson GLMM provided the most consistent and adequate fit to infrequent species. Capture rate of the shrew–mole was found to be negatively associated with tall shrub cover and coarse woody debris volume. Similarly, capture rate of Keen’s deer mouse was negatively associated with herb cover and coarse woody debris volume. Finally, captures of southern red-backed vole was associated negatively with herb cover and coarse woody debris volume but positively associated with vertical complexity of overstory vegetation. With correct GLMM specification, statistical inferences of habitat predictors were more reliable as autocorrelation between samples was properly accounted for and valid standard errors were estimated. Furthermore, the GLMMs in this study fitted capture rates of infrequent species well and produced admissible results on the association of these species to microhabitat features. Infrequent species need not be excluded from analysis; in fact, inclusion of these species is crucial to conservation of species diversity by designing silvicultural treatments that produce or protect suitable habitat.