Managing and Analysing Camera Trapping Data: An Advanced Toolbox
Jürgen Niedballa · Refubium (Universitätsbibliothek der Freien Universität Berlin) · 2018
Camera trapping has become a prime source of information about wild terrestrial mammals over the recent years, particularly for rare and elusive species and in challenging habitats. Key inference from camera-trapping encompasses species habitat associations, density and abundance estimations, or species interactions, amongst others. The rapid development of those methods and the large amount of data collected entail new challenges in terms of data management and analysis. The aim of this thesis was to contribute to the development of new methods for managing (Chapter 2) and analysing (Chapter 3 and 4) camera trapping data and to thus increase the efficiency and effectiveness of the use of camera-trapping data for practitioners both in academia and conservation. Camera-trapping can generate vast volumes of data over short periods of time, making efficient yet flexible data management imperative. In my first manuscript (Chapter 2), I developed a free and open-source R package for camera trap data management, camtrapR. It is the first such toolbox in the popular R language and was designed to offer a comprehensive and flexible workflow from data acquisition to creating input for well-developed downstream analytical tools, e.g. in occupancy or spatial capture-recapture frameworks. The package has quickly gained popularity and is now being used worldwide in scientific and conservation work, while it is still being actively maintained and developed. Species occurrence data from camera-traps can be combined with habitat information at camera traps via occupancy models in order to identify habitat associations of species while explicitly accounting for imperfect detection. The spatial scale at which habitat information are collected (grain and extent) will influence results heavily. In my second manuscript (Chapter 3), I assessed the influence of spatial scale on estimates of species-habitat relationships by varying the spatial resolution and extent of habitat covariates used in single-species occupancy models for six mammal species from Sabah, Malaysian Borneo. Habitat data from high-resolution (5-m RapidEye) satellite imagery had considerably higher model support than lower resolution data (≥30 m). Likewise, habitat data from patches of 50 meters around camera traps had higher model support than smaller (10 m) or larger (100 – 500 m) habitat patches. This study was the first to use 5-m RapidEye imagery in occupancy models and demonstrated the potential of such high-resolution satellite imagery for obtaining more realistic species-habitat associations in occupancy modelling, particularly in heterogeneous landscapes. The flexibility high-resolution satellite imagery offer in defining suitable spatial scales further add to their value. Species distributions in space and time are not only shaped by habitat preferences, but also by interactions between species, such as predator-prey relationships or various forms of competition. Discovering such spatiotemporal interactions in camera trapping data is challenging due to the sparseness and randomness of camera trapping data and further exacerbated by a lack of systematic comparisons of methods to assess such interactions. Therefore, in my third manuscript (Chapter 4), I developed a method to flexibly simulate camera trapping records of two interacting species. These simulated data are used for the first comparative assessment of the statistical power and robustness of a suite of statistical tests for spatiotemporal interactions. Linear models were the most powerful and flexible method to discover such interactions. Nevertheless, only strong interactions could be detected reliably with any of the methods tested. This novel simulation approach and the recommendations given can serve as guidelines for practitioners wishing to assess interactions between or within species from camera trapping data.