Late last year, Dan Warren, Michael Turelli, and I wrote a paper about niche evolution in which we developed new metrics and statistical tests for comparative studies of environmental niche models (ENMs). Our basic metrics permit quantification of similarity between ENM model projections generated by two or more populations. These metrics may either be explored in a phyogenetic context, or used in association with pseudoreplicated datasets to test two null hypotheses at opposite ends of the niche similarity continuum : (1) ENMs are identical and, (2) ENMs no more similar than expected by chance. Although our methods could work with several types of niche modeling algorithms, they are best suited to output generated by the maximum entropy method implemented in the program Maxent. We have now written a program of our own called ENMTools that interacts extensively with Maxent to implement the analyses discussed in our paper (more accurately, Dan Warren wrote a program and Michael and I beta tested it). Dan has done a masterful job with this Perl application, which presents as a simple GUI interface on any platform capable of running ActivePerl (including Linux, Mac OSX, and Windows). In addition to performing the methods we've already introduced, new functionality is being added constantly (although some of the coolest stuff is purposely left unexplained so we can publish the methods before they start getting used by others). Dan has set up a website and a blog to keep people informed about the latest developments, and hopes you appreciate his retro-internet stylings.
Saturday, April 18, 2009
New Program for Studies of Environmental Niche Evolution
Late last year, Dan Warren, Michael Turelli, and I wrote a paper about niche evolution in which we developed new metrics and statistical tests for comparative studies of environmental niche models (ENMs). Our basic metrics permit quantification of similarity between ENM model projections generated by two or more populations. These metrics may either be explored in a phyogenetic context, or used in association with pseudoreplicated datasets to test two null hypotheses at opposite ends of the niche similarity continuum : (1) ENMs are identical and, (2) ENMs no more similar than expected by chance. Although our methods could work with several types of niche modeling algorithms, they are best suited to output generated by the maximum entropy method implemented in the program Maxent. We have now written a program of our own called ENMTools that interacts extensively with Maxent to implement the analyses discussed in our paper (more accurately, Dan Warren wrote a program and Michael and I beta tested it). Dan has done a masterful job with this Perl application, which presents as a simple GUI interface on any platform capable of running ActivePerl (including Linux, Mac OSX, and Windows). In addition to performing the methods we've already introduced, new functionality is being added constantly (although some of the coolest stuff is purposely left unexplained so we can publish the methods before they start getting used by others). Dan has set up a website and a blog to keep people informed about the latest developments, and hopes you appreciate his retro-internet stylings.
Subscribe to:
Post Comments (Atom)
4 comments:
PS - This blog isn't intended as a platform for discussion of programs that it's authors are directly involved with. We'd like to post on any new programs in our field, so please let us know if you've got new stuff that Dechronization readers are likely to be interested in.
I should add that we're very much in a beta phase right now - the program is rapidly improving due to the feedback I'm getting from people who use it. If anyone out there runs into something they don't like, or has a feature they'd like to add, I would really like to hear from you. That extends to new tests and metrics as well; I've already added a few new features thanks to suggestions from Town Peterson and his crew at KU, in addition to all the cool ideas that Rich has been coming up with.
I still think the ENMtools website needs to play a MIDI of "Rock You Like a Hurricane" like we discussed some time ago.
Dear Glor,
coming across blogs too infrequently to be efficient in contributing, I would still like to leave a comment on the approach you used in the Warren et al. (2008). I have a real problem with using the geographic overlap of environmental niche models (or species distribution models) as measure of climate niche overlap. If, as is common, a specific climate combination (e.g. moderately warm and moist) is very common in the study region, then the occurrence probability of the two species for this particular climate combination will dominate the entire overlap.
I find it much more intuitive to predict, based on the species distribution model, to a climate hyperspace with equidistant sampling points in parameter space.
The example you show in the appendix to your paper, where this approach failed, was (in my view) more to do with the way MaxEnt extrapolates beyond the range of parameters (in contrast to tree-based methods), not of the parameter-space-based niche overlap calculations.
It's becoming a bit technical here, but I mainly wanted to express my concern about the geographic overlap as measure of niche equivalence (or similarity; I find that distinction too fuzzy).
Cheers,
Carsten (Dormann)
Post a Comment