I appreciate that perspective—the 'fixed-point' nature of cameras definitely creates spatial bias that GPS collars don't have.

To make this data more 'movement-model ready,' we are focusing on Individual Re-Identification (Re-ID). If our AI can consistently tag unique individuals across a distributed network, we can generate a 'sparse trajectory' (x,y,t).

Regarding your point on 'no spatial gaps,' how do you feel about Probabilistic Occupancy? If we can't have 100% coverage, would providing 'Detection Probability' metadata alongside the coordinates help your models account for the sampling bias?

Also, for tools like MoveApps, would you prefer the output in a flattened CSV (Standard Event Format) or a specific MoveBank-aligned schema?? Whats your take?

I think the success of a probabilistic occupancy model would depend entirely upon the target species and how homogeneous the environment is in the context of that species. In more diverse environments ensuring that you have your sample locations well stratified might help solve some of the problems, however, there will always be movement related questions that cannot be confidently answered.

It should be noted that GPS based data can also be strongly biased by heterogeneous habitats, particularly because of loss of GPS signal under heavy cover/underwater/underground or even environments where the animal changes its posture. Unfortunately, it is rare to see these limitations/biases acknowledged in the literature.