Maxent
Maxent and sample size
12-09-2008 19:05
From earlier posts you may know that I’m a fan of
Maxent. Recently a new paper* appeared in which
different software on species modeling were compared.
This time there was special emphasis on the
performance with the use of a small number of
occurrences. This is usually problematic for many
modeling programs. Here is the result: Maxent has
about “the best predictive power across all samples
sizes”.
This is certainly good news, but it doesn’t solve a problem what I have with the modeling of endemic, range-restricted species. These are usually known from very few localities (up to 4). So far, I haven’t seen solutions proposed for working with these small numbers. But if anybody has an idea for an approach that could solve this nifty problem, please keep me posted! You would certainly make my day.
Reference:
Wisz, M. S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A. & NCEAS Working Group (2008). Effects of sample size on the performance of species distribution models. Diversity and Distributions, 14, 763-773.
This is certainly good news, but it doesn’t solve a problem what I have with the modeling of endemic, range-restricted species. These are usually known from very few localities (up to 4). So far, I haven’t seen solutions proposed for working with these small numbers. But if anybody has an idea for an approach that could solve this nifty problem, please keep me posted! You would certainly make my day.
Reference:
Wisz, M. S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A. & NCEAS Working Group (2008). Effects of sample size on the performance of species distribution models. Diversity and Distributions, 14, 763-773.
With some help and inspiration
13-10-2007 19:55
The Maxent package I
previously wrote about has a - like most freeware
software - a discussion group
which in my case was
helpful to solve some initial problems. One
problem - the grid output file that should be
suitable for input in DIVA-GIS - was a bug in the
beta version that I ran and was fixed very swiftly
by the Steven Phillips, the main author. After
this I could use the software to do useful
analyses on the Ecuadorian data. The result is a
series of distribution maps on (sub)generic level.
Although several studies (e.g. Elith et al., 2006;
Pearson et al., 2007) have shown that Maxent is
robust enough to handle datasets with only a few
occurrences per species, I preferred to do the
analyses on aggregated data to get a more
comprehensive result. Here is one of the maps,
obtained by processing the grid output file of
Maxent in DIVA-GIS.
Ecological niche model for Plekocheilus. The blue dots are localities where the genus is present. Colours represent logistic values of likelihood for occurrence: from 0-0.2 (dark green) to 0.8-1.0 (red).
When I out of curiousity tried to feed Maxent with occurrence data (also Plekocheilus) from the Guayana Shield, the resulting picture of the distribution showed a remarkable disjunct pattern. An inspirational indication that my hypothesis about linkages between the Pantepui region in Venezuela and the Andes in Colombia is worth further investigation!
Ecological niche model for Plekocheilus from Venezuela, Pantepui, as depicted by Maxent.
This even becomes more clear when all the known localities of Plekocheilus from Venezuela are taken into account.
Ecological niche model for all Plekocheilus known from Venezuela, as depicted by Maxent.
See also this blog for comments on Maxent.
References:
Elith, J. e. a. (2006). Novel methods improve prediction of species' distributions from occurrence data. Ecography, 26, 129-151.
Pearson, R. G., C.J. Raxworthy, M. Nakamura & A. Townsend Peterson. (2007). Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos from Madagascar. J. Biogeogr, 34, 102-117.
Ecological niche model for Plekocheilus. The blue dots are localities where the genus is present. Colours represent logistic values of likelihood for occurrence: from 0-0.2 (dark green) to 0.8-1.0 (red).
When I out of curiousity tried to feed Maxent with occurrence data (also Plekocheilus) from the Guayana Shield, the resulting picture of the distribution showed a remarkable disjunct pattern. An inspirational indication that my hypothesis about linkages between the Pantepui region in Venezuela and the Andes in Colombia is worth further investigation!
Ecological niche model for Plekocheilus from Venezuela, Pantepui, as depicted by Maxent.
This even becomes more clear when all the known localities of Plekocheilus from Venezuela are taken into account.
Ecological niche model for all Plekocheilus known from Venezuela, as depicted by Maxent.
See also this blog for comments on Maxent.
References:
Elith, J. e. a. (2006). Novel methods improve prediction of species' distributions from occurrence data. Ecography, 26, 129-151.
Pearson, R. G., C.J. Raxworthy, M. Nakamura & A. Townsend Peterson. (2007). Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos from Madagascar. J. Biogeogr, 34, 102-117.