A ‘How-to’ Guide for Interpreting Parameters in Habitat-Selection Analyses

Abstract

Resource-selection and step-selection analyses allow researchers to link animals to their environment and are commonly used to address questions related to wildlife management and conservation efforts. Step-selection analyses that incorporate movement characteristics, referred to as integrated step-selection analyses, are particularly appealing because they allow modeling of both movement and habitat-selection processes.Despite their popularity, many users struggle with interpreting parameters in resource-selection and step-selection functions. Integrated step-selection analyses also require several additional steps to translate model parameters into a full-fledged movement model, and the mathematics supporting this approach can be challenging for biologists to understand.Using simple examples, we demonstrate how weighted distribution theory and the inhomogeneous Poisson point-process model can facilitate parameter interpretation in resource-selection and step-selection analyses. Further, we provide a “how to” guide illustrating the steps required to implement integrated step-selection analyses using the amt package.By providing clear examples with open-source code, we hope to make resource-selection and integrated step-selection analyses more understandable and accessible to end users.

Analysis of local habitat selection and large-scale attraction/avoidance based on animal tracking data: is there a single best method?

“…SSM approaches showed slightly inflated type I error rates, which were probably explained by the fact that these methods parameterise the movement kernel without adjusting for habitat selection, which can lead to biased estimators of habitat-selection parameters [ 73 ]. iSSMs in contrast use movement characteristics in the linear predictor to reduce this bias [ 38 , 46 ]. SLRM-based methods showed strongly inflated type I error rates, probably because the spatio-temporal and angular autocorrelations were not considered appropriately within these methods.…”

“…Although ST-PPMs and iSSMs showed similar statistical powers under some circumstances (e.g., habitat selection in continuous habitats), we recommend using iSSMs rather than ST-PPMs for the following reasons: (1) the power of iSSMs is more robust compared with ST-PPMs (the latter performed worse in categorical habitats and for the detection of large-scale attraction); (2) iSSMs do not need the time-consuming initial empirical determination of the optimal spatial extent of the dummy point grid; which is related to the fact (3) that computation times for iSSMs are much shorter than for ST-PPMs, which is especially important in the context of large data sets; and (4) iSSM model implementation is user-friendly and well-documented, using the provided R-package amt and/or the worked examples in the Supplemental Appendices of Ref. [ 38 ], in contrast to spatio-temporal PPMs, for which there is currently (to the best of our knowledge) no available R-package. In addition, Avgar et al [ 46 ] noted additional advantages of iSSMs over previous methods, including their predictive capacity (e.g., for landscapes different from the landscape used for the model fit), and the ability to derive and parametrize a mechanistic movement model.…”

“…Dummy points were generated based on step lengths and turning angles, assuming a gamma and a von Mises distribution [ 46 , 47 ]. The iSSM approach differs from the SSM approach by additionally considering fixed-effect predictors related to different movement characteristics [ 38 , 47 ]. For iSSMs, we therefore performed a model selection step based on the Akaike information criterion (AIC) [ 58 ] comparing five different models reflecting all existing combinations of main effects and interactions in relation to the different autocorrelation terms, namely the logarithm of the spatial distance to the forgoing tracking point l o g ( d x ), and cosine of the turning angle c o s ( t a ) [ 47 ] (due to equidistant time steps, the temporal distance to the foregoing step was not considered).…”

“…First, point process models (PPMs) naturally and automatically resolve many of the questions and pitfalls arising from previous techniques [ 8 , 26 , 27 , 37 ]. In particular, PPMs allow the role and number of dummy points to be deduced purely mathematically by aiming at an efficient estimation of an integral as a part of the PPM likelihood [ 26 , 37 ] leading to parameter convergence for sufficiently large dummy point numbers [ 38 ]. Additionally, PPMs represent a generalisation of many other frequently used methods [ 26 , 39 , 40 ], and the PPM likelihood can be approximated using various standard generalised linear modelling regression software [ 27 , 41 , 42 ] (possibly including mixed and/or additive modelling [ 43 ]).…”

“…It is therefore not surprising that both methods can lead to divergent conclusions [ 45 ]. Finally, all four methods (ST-PPMs, SSMs, SLRMs, and iSSMs) can be used not only to test hypotheses, but also to estimate utilization distributions respectively for predictions [ 8 , 38 , 46 – 48 ].…”