[R-sig-eco] Background points in SDMs: regular or random?

[Carsten Dormann]

Hi Marco,

you may want to check this reference:

Phillips, S.J., Dudik, M., Elith, J., Graham, C., Lehmann, A., 
Leathwick, J., & Ferrier, S. (2009) Sample Selection Bias and 
Presence-Only Models Of Species Distributions: Implications for 
Selection of Background and Pseudo-absences Ecological Applications, 19, 
181-197

Carsten ( no co-author ;-) )

On 10.02.11 12:42, Marco wrote:
> Dear list:
>
> I am learning SDMs recently and got one questions and I am sure I can
> get the right answer here.
>
> For a certain area, the prediction of species distribution might
> dependent on the input of the background points, ie., the points that
> we would like to know if one or more species could reside. Two
> strategies can be seen in literatures providing the input background
> points: regular grids and random points. In the first case, all the
> grids that contain the predictor information of the whole research
> area are included in the analysis as data points, result in a regular
> point matrix, such as in MAXENT.. While in the latter case, random
> number of points were generated within the area and assigned the
> values of the predictor they overlay, such as the algorithm employed
> in BIOMOD. It seems that in BIOMOD, we can also implement such regular
> grid as input background points, but the question for me is that I do
> not know which strategy make sense in SDMs.
>
>
>
> I think random sampling of background points are both included in
> MAXENT and in algorithms that BIOMOD implemented. So, the main
> difference between the two strategies might be:
>
> 1)    Whether more than 1 point fall inside one grid. In regular grid,
> all point fall inside different grid as guaranteed by the generation
> procedure, but in random points, extra steps should be used to exclude
> the closely dispersed points.
>
> 2)    The stability of the model result. In regular grid, each grid
> get it prediction from the model, given the same model, the output
> distribution is always the same. While in random points, all the
> random points get the prediction from the model, and the final
> distribution map might different given the same model when the points
> are arranged differently.
>
>   If these are correct, then regular grid as input points should be
> superior than random. But I  am not confident on this. Can somebody
> help me out?
>
> Best wishes~
>
> Marco
>

-- 
Dr. Carsten F. Dormann
Department of Computational Landscape Ecology
Helmholtz Centre for Environmental Research-UFZ	
(Department Landschaftsökologie)
(Helmholtz Zentrum für Umweltforschung - UFZ)
Permoserstr. 15	
04318 Leipzig				
Germany

Tel: ++49(0)341 2351946
Fax: ++49(0)341 2351939
Email: carsten.dormann at ufz.de
internet: http://www.ufz.de/index.php?de=4205

Registered Office/Sitz der Gesellschaft: Leipzig
Commercial Register Number/Registergericht: Amtsgericht Leipzig, 
Handelsregister Nr. B 4703
Chairman of the Supervisory Board/Vorsitzender des Aufsichtsrats: MinR 
Wilfried Kraus
Scientific Managing Director/Wissenschaftlicher Geschäftsführer: Prof. 
Dr. Georg Teutsch
Administrative Managing Director/Administrativer Geschäftsführer: Dr. 
Andreas Schmidt