Comparing relative abundance models from different indices, a study case on the red fox

Abstract

The correct interpretation of relative abundance indices provided by different sampling methods is essential to correctly estimate population size. Although multiple indices and models have been proposed, their ability to estimate relative abundances and their performance in models explaining abundance trends remains unclear. We used the red fox (Vulpes vulpes) as a model species to compare the relationship and derived models of relative abundance between three indices of relative abundance: RAI (number of captures/total occasions); NI (number of photo-identified individuals) obtained by camera-trapping, and NSE (number of segments with scats) obtained by the scat census sampling method. In addition, we modelled the relationship between a set of habitat predictors and fox relative abundance for each of the three estimated relative abundance indices. We compared the relative abundance models explained for each index against N-mixture models that estimate abundance controlled for variation in detection. Results showed a positive correlation between the RAI and NI indices, while both indices showed a negative relationship with the NSE index. Relative abundance models and N-Mixture models showed a different selection of predictors to explain abundance trends. NSE and RAI indices selected predictors that could explain variability in fox detection rather than fox abundance. In contrast, the NI index and N-Mixture models selected the same predictors to explain fox abundance. Our results suggest the use of the NI index for abundance models without the need to control for variation in detection. Relative abundance indices based on scats and captures per occasion are suboptimal indices for species abundance studies due to possible bias caused by animal behaviour. If count-based methods on captures per occasion (RAI) are selected, we suggest using session-based data processing to incorporate detectability variation in N-mixture models.