Modeling specific behaviors using Maxent

[Ari Whiteman]

Hi All,

I'm working on using Maxent to determine where high-quality bear denning habitat is in Eastern Europe. For my sample data, I'm using 62 confirmed locations of bear dens and as an output, I'm hoping to find where bears are likely to den. As far as I'm aware, using Maxent to model the presence liklihood of specific behaviors rather than entire ranges is pretty rare in the literature. I was wondering if any of you have come across examples of it in the past? 

Best,

Ari

[Jamie M. Kass]

First off, it sounds like you can get good absence data for denning sites. If you have good absence data, a GLM approach (logistic regression) would be preferable to Maxent, as the results would likely be more accurate. If you have no absence data and only presence, Maxent would be a good technique for you to try. One potential problem for you if you use a presence/background approach like Maxent is that the background points may include areas with denning sites you have no explored, and this might bias your results. Further, denning selection is likely largely driven by microclimate preference, and thus your predictor variables must reflect these, maybe in addition to large-scale climate variables too. There are more things to consider, obviously, but I don't want to delve too far into it before you consider other techniques. Have you looked into resource selection functions? I think the literature there may have more behavioural studies. Just a thought. But I don't see any big, red warning flags for using Maxent to predict suitable areas for bear denning sites.

Jamie Kass
PhD Student
City College, NY