A comparison of abundance and distribution model outputs using camera traps and sign surveys for feral pigs

Abstract

Species distribution models play a central role in informing wildlife management. For models to be useful, they must be based on data that best represent the presence or abundance of the species. Data used as inputs in the development of these models can be obtained through numerous methods, each subject to different biases and limitations but, to date, few studies have examined whether these biases result in different predictive spatial models, potentially influencing conservation decisions. In this study, we compare distribution model predictions of feral pig (Sus scrofa) relative abundance using the two most common monitoring methods: detections from camera traps and visual surveys of pig sign. These data were collected during the same period using standardised methods at survey sites generated using a random stratified sampling design. We found that although site-level observed sign data were only loosely correlated with observed camera detections (R2 = 0.32-0.45), predicted sign and camera counts from zero-inflated models were well correlated (R2 = 0.78-0.88). In this study we show one example in which fitting two different forms of abundance data using environmental covariates explains most of the variance between datasets. We conclude that, as long as outputs are produced through appropriate modelling techniques, these two common methods of obtaining abundance data may be used interchangeably to produce comparable distribution maps for decision-making purposes. However, for monitoring purposes, sign and camera trap data may not be used interchangeably at the site level.