[Biomod-commits] BIOMOD_Projection

Wilfried Thuiller wilfried.thuiller at ujf-grenoble.fr
Thu Oct 6 16:09:18 CEST 2011


Dear Jesus,

It has to do that the data you used for calibrating the models are not representative of the data of the overall study area. This is certainly a problem of extrapolation out of the climatic range.
Perhaps you should also remove  ANN from the projections because it is very bad. Note that GAM, GLM and RF are not very good either. 

Wilfried



Le 6 oct. 2011 à 15:56, Jesus a écrit :

> Dear Wilfried,
> 
> I made models for a species with diferent algorithms and 10 runs per algorithm. However, my results are quite odd when I project my models to the complete study area.  For some of the algorithms some of the runs look ok but other runs appear all as 0. I explain more below.
> 
> a)     I make my models and then check my binary predictions and the continuous values; they look OK.
> 
> The Evaluation results of one run (as example) are:
> 
> 
>    Cross.validation indepdt.data total.score Cutoff Sensitivity Specificity
> 
> ANN            0.657         none        0.83  0.999      66.667      98.644
> 
> GAM            0.788         none       0.912  1.978      91.667      83.122
> 
> GBM            0.939         none       0.986  1.896      91.667      96.398
> 
> GLM             0.78         none       0.913  1.995      83.333      86.969
> 
> RF             0.795         none       0.952  0.665      91.667      91.216
> 
> 
> b)    Then I project the models to the complete study area because in order to train the model I just used 9000 cells of a separated file. However, I get the runs all with values that do not correspond to the above mentioned prediction and all the map runs seem to be the same. Besides that, my values in the colot graph go from “-1e-03” to “1e-03”, but the maps are all of a continuous colour with a value of around 0.
> 
> c)     If I also check the binary predictions of these projections I get exactly the same than above.
> 
> d)    When I extract the projections directly without applying the BinRoc  for example, and then I apply the Cutoff value that appears in the Evaluation.Results using for example ArcGIS for the algorithm and specific run then my binary map is a bit different than when I just call the projection with the binary predicition BinRoc directly from BIOMOD.
> 
> Do you have any idea of what could be going wrong with the projections? This is very weird because as I mentioned before the predictions and the binary predictions look fine but the problems start with the projections in the complete area. 
> 
> Thank you in advance.
> 
> Jesus
> 
> _______________________________________________
> Biomod-commits mailing list
> Biomod-commits at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/biomod-commits

--------------------------
Dr. Wilfried Thuiller
Laboratoire d'Ecologie Alpine, UMR CNRS 5553
Université Joseph Fourier
BP53, 38041 Grenoble cedex 9, France
tel: +33 (0)4 76 51 44 97
fax: +33 (0)4 76 51 42 79

Email: wilfried.thuiller at ujf-grenoble.fr
Personal website: http://www.will.chez-alice.fr
Team website: http://www-leca.ujf-grenoble.fr/equipes/emabio.htm

FP6 European MACIS project: http://www.macis-project.net
FP6 European EcoChange project: http://www.ecochange-project.eu






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20111006/c659f880/attachment.htm>


More information about the Biomod-commits mailing list