multivariate environmental similarity surface (MESS) analysis

[Workualem]

Dear all,

I'm using Maxent and GBM to model an endangered species distribution with land cover type (as a categorical variable) and other topographic variables as predictor variables. However, categorical variables are frequently problematic and can significantly impact model results when the presence/absence points are not evenly distributed across the categories within the layer.

So, how do I convert this layer to multiple continuous layers and perform multivariate environmental similarity surface (MESS) analysis? Is it possible to run MESS analysis if one of the predictors is categorical?

I appreciate your consideration.

[Jamie M. Kass]

It's not possible to run MESS with categorical variables because they need to have a minimum and maximum that signify low and high values. You could calculate the proportion of the land cover classes within a buffer zone around each point and use these numbers as continuous variables. I would caution against using all of them (so that summing the proportions add up to 1), because that will lead to very high correlation between two or more of them. What Maxent usually does when variables are highly correlated is choose one to do most of the explaining, but the one it chooses can be random, and so it's hard to trust the variable importance values. If you are just interested in getting an accurate prediction, this is not something to worry about too much, but if you plan on interpreting the model response curves, etc., you should make sure collinearity is not too high.

Jamie