Hello, can anybody please tell an alternative to AUC for assessing performance of species distribution models?
You could also use Cohen's Kappa or the True Skill Statistic (deals with prevalence better than Kappa) which are dependent on thresholds. Mainali et al. compares these different statistics in the attached publication. There are plenty of other papers as well that discuss these statistics.
Hello
You can find some interesting information on next link:
http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2664.2006.01214.x/pdf
It will actually depends a lot on your research question. Why are you producing a model? without knowing this piece of information it's basically impossible to reply (as model evaluation always depends on the aim). are you interested in discrimination power? the AUC is the way to go (or kappa or TSS, which are basically all the same thing). are your interested in calibration power? Boyce is the way to go. are you interested more in commission error than omission?
I suggest a null-model test that actually tests the significance of an SDM. Most SDM measures of accuracy are flawed when applied to presence-only data! http://onlinelibrary.wiley.com/doi/10.1111/j.2007.0906-7590.05041.x/abstract. I also provided a R-script on my RG page.
Pietra and the scaled Brier scores to tackle the insensitivity issue of AUC measure when adding strong features to the model
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs