change defaultprevalence to get estimates of occupancy probability?

[Megan]

Hi All, 

Will changing the defaultprevalence in Maxent to the true prevalence from PA data (= # of detections/total number of sites sampled) adjust the logistic output to get estimates of occupancy probabilities? 

Thanks! 

[Jamie M. Kass]

Where are you finding this parameter that you can edit? I wasn't aware you could.

[Megan]

If you look in the help, defaultprevalence is one of the flags. It is set at 0.5 but can be changed. 

[welc...@gmail.com]

Hi Megan,

Please see Guillera-Arroita 2014 (attached). They caution that adjusting the defaultprevalence argument does not address prevalence across the landscape, however they provide a post-processing method in the appendices to adjust the logistic output to obtain estimates of occupancy probability.

-Heather

[Jamie M. Kass]

Megan, thanks for posting this -- I realize I'd overlooked this paper, which makes pretty important points. First off, it does appear you can alter the defaultprevalance parameter, along with some other post-processing transformations they suggest, but they still advocate for using GLMs to model PA data with appropriate model selection methods to avoid overfitting (they suggest regularization via the glmnet package in R). Thus, if you have PA data that you're using to inform the selection of this Maxent parameter, you should probably use a GLM to model occupancy, or a hierarchical method that accounts for detection probability (see papers by Royle). However, if you have a small PA dataset not sufficient for modeling but informative as to what the prevalence per site may be, changing defaultprevalence and running Maxent might br the way to go. Just understand, as the paper says, Maxent will only give you relative suitability, not probability of occupancy.

Jamie Kass
PhD student
City College, NYC