Re: Threshold options in Maxent

[John Baumgartner]

Hi James,

It depends which threshold options you're referring to. If you mean the "Apply threshold rule" options in the Advanced tab of settings, then these are binarization thresholds that can be applied to habitat suitability surfaces. If you set this, Maxent will produce the normal output, but in addition it will convert that output to a grid of 1s and 0s, representing suitable sites (those with HS values above the chosen threshold), and unsuitable sites (those below). The HS values associated with the various thresholds are given in Maxent's html output and in maxentResults.csv. 

You could take a look at http://onlinelibrary.wiley.com/doi/10.1111/j.0906-7590.2005.03957.x/abstract (Selecting thresholds of occurrence in the prediction of species distributions) for some more insight into thresholding output.

If you were referring to the threshold options in the Experimental tab, these are occurrence sample size thresholds that must be met before using particular feature types when auto features is selected. (Although Beta threshold refers to the regularization parameter to apply to threshold features.)

Threshold features, on the other hand, are one of the feature types produced by environmental layers, and take on value 1 when the variable has value greater than some threshold, and 0 otherwise.

The convergence threshold, specified in the Advanced tab of settings, determines the point at which model training should stop, in terms of log loss per iteration. 

I assume you mean the "apply threshold rule" settings, and these are a relatively recent addition to Maxent (from memory).

Cheers,

John

On Wednesday, 6 February 2013 14:33:51 UTC+11, James wrote:

I'm currently using Maxent to hypothesize unique lineages for a species. I have been unable to find anything in the literature regarding the thresholds that can be applied in the program. Specifically what they mean and their purpose.

-James