[Biomod-commits] BIOMOD EXAMPLE

[wilfried.thuiller at ujf-grenoble.fr]

Dear Pep,

>   I just started running the example in the manual and there is
> something I do not understand. In the example NbRepPa=2 AND
> nb.absences=1000 (page 13,14,15). Although the data contains enough
> absences, we created for the porpouse of example 2 runs of pseudo
> absences. By setting this parameters, what is the behavior of BIOMOD?
>
>   That means that 2 runs of the models are made using pseudo absences
> (2=NbRepPA) and 3 (NbRunEval) are made using raw data with real
> absences (?) or...will BIOMOD feed data on real absences with PA in
> order to achieve data for calibration/evaluation (80 % / 20% in our
> example) and hence keeping prevalence constant?


You did understand the pseudo-absence part. When you set the NbRepPA,  
two selections of pseudo-absence will be carried out. This means than  
1000 pseudo absences will be drawn from the original dataset to build  
the presence-absence matrix. That will done twice with the algorithm  
you selected (random, or other).

Then, NbRunEval is another thing. It is a step after.
NbRunEval ismade for the evaluation step. A random training part will  
be extracted from the dataset for training (e.g. 70%) and the  
remaining, let's say 30%, will be used for testing. This will be  
repeated 3 times if you select NbRunEval = 3. It allows to account for  
the uncertainty linked to this step.

So the question from you, it is which dataset will be used for this  
training/testing test.
The original data if you did not select pseudo-absence selection  
(NbRepetPA=0).
The presence/pseudo absence matrix if you did select pseudo-absence selection.

So how build a matrix if you do not trust the absences? In arcgis or R  
(preliminary GIS step) create a presence / absence matrix by feeding  
the matrix with the true presence, and put 0 elsewhere (let's call  
them 'No value'). It could be the entire area of interest or already a  
subselection (random or expert). Then let BIOMOD with the different  
algorithms select the most plausible absences from this 'No value'  
data, as shown in the manual (even if we do have real absence in our  
case).

Does it make sense?

Best
Wilfried












>   To sum up: I would like to know if running this PA is worth when
> not enough data on real absences is available and whether BIOMOD does
> feed automatically our original dataset wiht PAs or we have to
> reconstruct the dataset.
>
>   Thank you,
>
>   PS: and thank you Wilfried Thuiller for the latest release of
> BIOMOD where finding the manual is much more easy!
>
> Pep Serra Díaz
>
> Despatx C1/211
> Unit of Botany.
> Department of Animal Biology, Plant Biology and Ecology. C Builiding
>
>   Bioscience Faculty. Autonomous University of Barcelona.
>
>   08193 Cerdanyola del Vallès
> Spain
>