Wednesday, 17 April 2013

Jane Elith et al: range-shifting species (Friday 19th at 2)

The first paper in our Eco-Stats Symposium reading group is first-authored by the first speaker to present at the Symposium – Jane Elith.  It is called “The art of modelling range-shifting species”
and was published in Methods in Ecology & Evolution in 2010, the journal's first volume.  This paper has attracted a lot of attention, and is currently the second most highly cited paper in the journal, and by my estimation is solely responsible for about 11% of the journal’s 2011 impact factor!  We'll discuss this paper at UNSW on Friday 19th, 2-3pm and would be interested to hear what you make of it... feel free to post here any questions you have about it, comments, ideas...

Feel free to post any time, but if you aim for Friday 2-3 we might be able to get a discussion happening.


  1. Here's a few questions that have come up so far in discussions here at UNSW...
    - why were pres/abs weighted as they were, and how might that effect conclusions?
    - we know a lot about cane toad biology (e.g. they have been evolving in Austraila and have different traits at different parts of their range) so how can you include that into a model?

  2. I would be interested to hear what suggestions people came up with for how might weighted pres/abs effect the conclusions?

  3. A couple more:
    - things shift range for different reasons (e.g. invasion vs climate change)... would you use a different approach for these different cases?
    - could you evaluate effectiveness predicting range shift retrospectively, e.g. use historical range to predict to current range?

  4. Applying weights to presences and absences will surely change the results - it's hard to qualitatively say what may change without having a look at the data, but the models will place greater emphasis on the environments corresponding to sites with higher weights.

    Related to this issue is the reduction of data to one observation per 5x5 km grid cell as required by MAXENT but not by other modelling approaches such as GLM. Such a loss of information will reduce the environmental signal in those areas.