[Biomod-commits] Biomod-commits Digest, Vol 34, Issue 15

[Brenna Forester]

Hi Andreas - you are on the right track, except for the prediction part:

To predict:
1) For future predictions: I load a file with the same coordinates as the current file and with the future values of the variables/ For projections in other areas: I load a file with different coordinates and with the values of the variables observed on the new coordinates;

**Yes.

2) I use Ensemble.Forecasting() setting as Proj.name my new dataset;

**No - first use Projection() to project your models onto your current conditions, future conditions, new area, etc.  Name your projections with "Proj.name" within the Projection() call.  Then you can use those Proj.names in Ensemble.Forecasting().  You can also just use Projection() and then create the ensemble/consensus model yourself if you like (in R or a GIS).

Hope that helps!
Brenna

PS - another good paper, in addition to the ones listed by Wilfried, is:

Grenouillet G, Buisson L, Casajus N, Lek S (2011) Ensemble modelling of species distribution: the effects of geographical and environmental ranges. Ecography 34:9–17

In this paper they find that the consensus (average) model has better predictive accuracy (based on AUC) on current data than any of the single models separately.  I have also found this to be the case in my use of ensemble modeling so far.


________________________________
From: biomod-commits-bounces at r-forge.wu-wien.ac.at [biomod-commits-bounces at r-forge.wu-wien.ac.at] on behalf of Andreas Soteriades [andreassot10 at yahoo.com]
Sent: Monday, February 20, 2012 9:38 AM
To: biomod-commits at lists.r-forge.r-project.org
Subject: Re: [Biomod-commits] Biomod-commits Digest, Vol 34, Issue 15

Hi all,

Thank you so much for your willingness to help! Here are some additional questions, as derived by your useful comments:

>This is not necessarily the case, and (in my opinion) is one of the main arguments for using an ensemble approach to >forecasting species distributions onto new time period or locations.

>However, as recommended by Brenna, this is not something really recommended anymore as ensemble forecasts are >much more robust than the best model.

>We let the function BestModel for people who wants to quickly extract which is the best model for a given data but >this is something which should be applied when extrapolating the models into new areas or time.

> Ensemble.Forcasting try to give you a consensus modeling projection, that means that every selected model will be >contribute (more or less depending on arguments given) to a consensus and supposed better projection.

>From the comments above I can see that what you suggest is to avoid using the model that predicted best my already known localities for projections onto other areas or into the future; I should use ensemble forecasting instead, right?

Can you please confirm that I have understood all steps of the methodology:

To initialise:
1) I load the file with the coordinates, variables and presence/absence table of my species;
2) I use Initial.State() to initialise my dataset;
3) I run the Models() function by choosing the models I want and by setting different parameters;

To predict:
1) For future predictions: I load a file with the same coordinates as the current file and with the future values of the variables/ For projections in other areas: I load a file with different coordinates and with the values of the variables observed on the new coordinates;
2) I use Ensemble.Forecasting() setting as Proj.name my new dataset;
3) I analyse the outputs;

Is the above procedure correct?

Cheers,

Andreas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20120220/260495a3/attachment-0001.html>