Speeding by: Modeling animal movement speed does not overcome limitations of a popular camera trap density estimator

Viewshed density estimators offer a noninvasive alternative to capture-mark-recapture methods, using camera trap images and auxiliary information to estimate density of unmarked animal populations. One well-known estimator, the random encounter model (REM), can yield useful density estimates but requires a travel speed input that can be challenging to derive from telemetry data. A recent continuous time movement model approach to obtain speed estimates could provide a way forward, but this has yet to be rigorously evaluated. We aimed to test important study considerations (data quality, individual behavior, and tagging bias) with this approach using simulations informed by real telemetry data, allowing simulated individuals to move through a virtual camera trap grid. We then combined virtual detection data with modeled movement speeds in an REM framework, evaluating performance across a range of scenarios. We found that declining data quality (i.e., sparser telemetry data) increased bias of REM density estimates and reduced coverage of confidence intervals. Poor data quality also precluded the estimation of key movement model parameters required to yield travel speeds. When movement behavior varied widely, REMs performed even worse, with higher bias across scenarios. We also found that the REM consistently and severely underestimated reference densities for a real-life case study regardless of the speed input we used, suggesting that additional factors related to behavior and imperfect detection may additionally limit model performance. Ultimately, design-based approaches like the REM make strong sampling assumptions and may not produce unbiased density estimates without sufficient incorporation of ecological realism.