Comparing and interpreting AUC values from Maxent runs with environmental variables at different resolutions.

[Timothy Mayer]

Hello All.

I conducted Maxent modeling with 1 meter resolution Lidar and land cover data as environmental variables. I then repeated the same modeling effort; however, I replaced the slope and land cover data with 30 meter products resampled to 1 meter. I was interested in how useful (for ecological modeling) coarse 30 meter products resampled to 1 m are when fine resolution data is not available.  I.e. how useful is 30 meter data when 1m  Lidar isn't available for lower mobility small mammal species?

Now for my results, my highest resolution (originally 1 m products) data model run resulted in a AUC of 0.738, while my lower resolution (30m resampled to 1m) data model runs resulted in an AUC of 0.843.

Please correct me if my interpenetration is misguided.

From these result, I believe I should convey first that these are two distinct models and cross comparison is tricky. But essentially the lower resolution (30m resampled to 1m) model performed at a higher rate than the higher resolution model (originally 1 m products) at accurately distinguishing between presence-only points and background/pseudo-absence points to construct the ROC and subsequently the AUC. So in a sense the lower resolution model run performed better, but that does not inherently mean that it is ecologically more useful for identifying potential suitable habitat than the higher resolution (originally 1 m products) model.

Obviously I need some direction in my thinking as I mull these results over.

I can easily supply further details if necessary.

Any thoughts on this matter of how these results should most effectively and accurately be interpreted would be greatly appreciated.

Thanks so much,

Tim

[Милош Поповић]

Hello Tim,

I have similar results with slope variable using 30m and 1km resolution, where
the contribution of slope was much lower using 30 m layer. In contrary, land
cover improved the prediction (as expected) when resolution increased.

It is quite logical that 1 km, 30 m and 1 m data give different results for
topographic variables. I.e. slope constructed over 1 km data shows how steep the
terrain is in general, slope from 30 m layer could detect differences over small
cliffs or hills and finally 1 m layer could detect large rocks as steep terrain.
Final choice depends on the species you study, but I guess 1 m resolution will
not give you very useful data regarding slope. We’ll see what the others think.




Cheers,
Miloš


У нед, 11. 06 2017. у 16:32 -0700, Timothy Mayer пише:

> -- 
> You received this message because you are subscribed to the Google Groups
> "Maxent" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to maxent+un...@googlegroups.com.
> To post to this group, send email to max...@googlegroups.com.
> Visit this group at https://groups.google.com/group/maxent.
> For more options, visit https://groups.google.com/d/optout.

[Jamie M. Kass]

Tim,

How did you do this resampling? Did you simply change the cell size, so that large cells were broken into smaller ones? Or did you do interpolation and predict what the finer resolution cell values should be based on something?

Jamie

> email to maxent+unsubscribe@googlegroups.com.