BioMod and Ensemble Forecasting

[cicada]

HI all--

As part of a manuscript revision, we have been using both MaxEnt and
the various methods within BIOMOD. We have gotten everything to work
correctly, and in general, the MaxEnt and BIOMOD ensemble predictions
for our dataset mostly agree. Fiddling with various parameters or
variables in any of the methods has only minor effects on the
predictions, so everything seems robust and believable.

We have been criticized for the separate presentations of MaxEnt and
BIOMOD methods and results. However, recent papers I've dredged up
(pub. dates in 2011 and 2012) all have this same problem, as there
seems to be no really good way combine MaxEnt and BIOMOD results.
Methodologically, it also seems as though we have to treat them at
least a bit separately; while the presence records used in both MaxEnt
and BIOMOD analyses are the same, we also have observed absence
records, which we make full use of in the BIOMOD analysis.

Does anybody have suggestions or exemplar papers that handle this
problem in a way you find particularly helpful?

Many thanks!

[lilibamej]

Hi,
There is a package that allows to run maxent in R, it is called
dismo.

http://cran.r-project.org/web/packages/dismo/dismo.pdf

It might be worthy to run your models there and argued that all the
methods have been ran under the same platform.

I hope it helps,

Liliana.

[Matthew B]

I would recommend comparing models by their broader categories ie regression based, machine learning & finally pres only maxent. It is possible to run maxent in the swd format. you could compare the swd predictions that make use of your absence data to the default maxent which uses background points as absence data. Finally I would take caution before added mAxent predictions to a score weighted ensemble. If the background extent is too large maxent have a strong leverage on the ensemble

[Matthew B]

[Iman Momeni]

Hi
Could you please tell me what are the advantages of ensemble approach over MaxEnt?