Results of running maxent with repeated layer

[ٍElham]

I run maxent with one repeated layer (just the layer name is different). In contribution results one of them show high value (e.g. 80%) and the other low value (e.g.5%)!!! What is the reason? They have exactly the same content! Why this happen??

And in jackknife plot , as expected, both of them show the same value.

[Jamie M. Kass]

Maxent is a machine learning method. It will use all the data you give it and fit a response based on the flexibility for complexity you allow it. Why did Maxent choose to use 20% of one and 80% of the other? I don't know. Maxent does similar things with variables that are highly correlated -- it sometimes chooses to include both at different levels of contribution because it's just trying to fit the best response it can. But since your variables are exactly the same, the true reason for these exact proportions is a mystery. Can I ask why you are doing this in the first place?

Jamie

[ٍElham]

I did it because I wanted to test the correction of my aligned layers (following my previous question in google group:)). Although I checked the content, I wanted to be surer.

First I used the aligned reference layer (that I used as reference for aligning layer with different geographic dimension) and the original references layer in maxent; and the result was one with low and the other with high contribution. In the next step I just copy the original reference layer and just change the name and run these two layers (the original and the copy of original!) in maxent and again the result was one with low and the other with high contribution; and this was so amazing!!  

[Jamie M. Kass]

I see now, but fhe simplest way to check for equivalence of rasters is to simply do r1 == r2 in R, or in python in ArcGIS or QGIS. A better way to detect how similar rasters are is with Schoener's D, available in dismo through the nicheOverlap() function.

Jamie

I