Choosing environmental variables

[Megan S]

Hi all,

I am planning to run many SDMs for lots of different species. I am trying to figure out the most efficient approach. I am wondering if model outputs of current distributions (ignoring fit statistics and variable contributions) are very sensitive to the choice of environmental variables. From my reading it seems like collinearity and spurious variables are more of an issue when projecting models to different scenarios such as under climate change and when examining variable contributions. If all I am interested in is a general species distribution, do I need to worry so much about the choice of environmental variables?

Any ideas and/or references to read would be greatly appreciated.

Thanks,

Megan

[Nyasha Mwendera]

Hi Megan

These  many species of yours, are they similar in some respect? are they animals or plants? because each species is affected by its own environmental variables. However for some species such as plants the environmental variables could be similar.

[Megan S]

Hi Nyasha,

In the end I will be modeling species of all different taxa - animals, plants, etc. - but I could do some groupings of environmental variables specific to each taxa. And I will also be adding in ones that are specific to certain species, but was planning to have several that I would throw into almost every model - 19 bioclim, elevation, slope, and for plants - soil type, aspect, etc. My main question though is whether collinearity and spurious variables negatively affect models of current distribution when examining variable importance is not important - the purpose is just to see the map of the current distribution.

[HeatherD]

Hi Megan,

This paper (located here) was a huge help for me in understanding how collinearity might affect my model. It's important to assess collinearity while keeping in mind which variables are biologically important for your species, like what you said about groupings of environmental variables. In reading several papers on the topic, it seems that too many intercorrelated variables can influence the outcome of the model, which would include the map of the current distribution. I used ENMTools to determine which of my WorldClim variables were highly correlated.

[Megan Sebasky]

Hi Heather,

Thanks so much for that paper! That is really helpful. In terms of your correlation analysis, for variables correlated at R>0.7, how did you choose which to keep? Some people use variable importance in MaxEnt or a PCA, in combination with biological importance to the species (when it is known).

Thanks,

Megan

On Mon, Feb 23, 2015 at 10:17 AM, HeatherD <heath...@gmail.com> wrote:

Hi Megan,

This paper (located here) was a huge help for me in understanding how collinearity might affect my model. It's important to assess collinearity while keeping in mind which variables are biologically important for your species, like what you said about groupings of environmental variables. In reading several papers on the topic, it seems that too many intercorrelated variables can influence the outcome of the model, which would include the map of the current distribution. I used ENMTools to determine which of my WorldClim variables were highly correlated.

--
You received this message because you are subscribed to a topic in the Google Groups "Maxent" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/maxent/a6g5vWb_wp0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to maxent+un...@googlegroups.com.
To post to this group, send email to max...@googlegroups.com.
Visit this group at http://groups.google.com/group/maxent.
For more options, visit https://groups.google.com/d/optout.

[Dimitris Poursanidis]

After the 0.7 criterion you have to see the biological importance for the studied specirs

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.

[Megan Sebasky]

What if you have 2 variables that you know are biologically important and have to choose one?

[Dimitris Poursanidis]

Hmmm.... try 3 models. One with both and 2 with each one.. the trial n error has to be used