Help

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Help

R-sig-geo mailing list
Bonjour , j'aimerais utiliser  maxent pour modéliser la distribution potentielle du niébé sur la base des données de présence seuls. En effet, jai acquis un certains nombre de variables environnementales et bioclimatiques concernant ma zone d'étude.  Mais pour choisir les variables les plus contributives dans le modèle; j'aimerai faire une analyse de correlation de celles-ci. Sur ce, pourriez vous m'expliquer etape par etape les procedures à suivre sous R ? J'aimerais dire par là le scripts pour:  -    compiler et appeler toutes les variables environnementales et les données d'occurence; -    executer le tester de correlation;-    pour faire une analyse discriminante?

Merci par avance














SADDA Abou-Soufianou











        [[alternative HTML version deleted]]

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Reply | Threaded
Open this post in threaded view
|

Re: [FORGED] Help

Rolf Turner

On 14/05/18 20:17, Soufianou Abou via R-sig-Geo wrote:

> Bonjour , j'aimerais utiliser  maxent pour modéliser la distribution
> potentielle du niébé sur la base des données de présence seuls. En
> effet, jai acquis un certains nombre de variables environnementales
> et bioclimatiques concernant ma zone d'étude.  Mais pour choisir les
> variables les plus contributives dans le modèle; j'aimerai faire une
> analyse de correlation de celles-ci. Sur ce, pourriez vous
> m'expliquer etape par etape les procedures à suivre sous R ?
> J'aimerais dire par là le scripts pour:  -    compiler et appeler
> toutes les variables environnementales et les données d'occurence; -
> executer le tester de correlation;-    pour faire une analyse
> discriminante?
>
> Merci par avance.
La langue de cette liste est l'anglais.
S'il vous plaît exprimer votre question en anglais.

I'm afraid that my French is insufficient to follow your question
properly, but I gather that you have presence-only data (for some
phenomenon) and a number of environmental variables from which you hope
to predict occurrences of this phenomenon.

You also express an interest in undertaking a correlation analysis of
your predictors and performing "tests of correlation".

Given that I am understanding you correctly, I would advise against
this.  The proper strategy (IMHO) is to *fit a model* using your
predictors and then assess their predictive power in this model,in some way.

If the "presence only" data, that you have, can be considered to be
point locations, and if the values of your predictors are available at
all points of your study region, then you may be able to effect the
required model fitting using the facilities of the spatstat package.

Anyway, please re-post your question en anglais, if you can.  You are
much more likely to get a useful answer if you do.  Bon chance.

Cordialement,

Rolf Turner

--
Technical Editor ANZJS
Department of Statistics
University of Auckland
Phone: +64-9-373-7599 ext. 88276

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Reply | Threaded
Open this post in threaded view
|

Help

R-sig-geo mailing list
Dear Rolf Turner,

 I have points of presence of cowpea in Niger in CSV format; in addition to other variables (soil texture, soil pH, altitude, I downloaded from worldclim archives, the 19 environmental variables, I cut them all at the Niger scale and I converted them under ASCUI format. The idea for me is to choose the best variables to include in the model.
 NB. I'm using Maxent model, but I'm not good in R software.

Merci














SADDA Abou-Soufianou

--------------------------------------

Doctorant

Université Dan Dicko Dankoulodo deMaradi-Niger

BP 465 120, avenue MamanKoraou- ADS

                   &

Institut d’Ecologie et des Sciencesde l’Environnement de Paris (iEES-Paris)

Centre IRD France Nord-(iEES Paris)-32,av.Henri Varangnat 93143 BONDY cedex.

|  
Lien: https://ieesparis.ufr918.upmc.fr/index.php?page=fiche&id=378&droit=1


 [hidden email]

GSM : Niger : (+227) 96-26-99-87/91-56-35-19 ; France (+ 33)  07-55-79-39-93

 
  |  
 
  |










    Le lundi 14 mai 2018 à 12:05:40 UTC+2, Rolf Turner <[hidden email]> a écrit :  



Please keep your posts "on-list".  You are much more likely to get a
useful answer that way.  There are many others on the list whose
knowledge and insight are far greater than mine.

I have therefore cc-ed the list in this reply.

On 14/05/18 21:48, Soufianou Abou wrote:

> Thank you for advice, Rolf Turner
>
> My question is as follows:
>
> I'd use maxent to model the potential distribution of cowpea on the
> basis of the only presence data. Indeed, I have acquired a number of
> environmental variables and bioclimatic regarding my area of study. But
> to choose the most contributive variables in the model; I would like to
> make a correlation analysis of these. On this, could you explain to me
> the step by step procedures to follow in R? I would like to say scripts
> for:- compile and call all environmental variables;- run the correlation
> test to select the least correlated ones.

As I said before, I don't think this is the right approach, but I can't
be sure without knowing more about your data.  I find your description
to be vague.

How are your data stored?  What information do you have about the
"distribution of cowpea".  Do you have *points* where cowpea is present
or more extensive *regions* where it is present?  (And could these
regions be "considered to be points" on the scale of interest?) How are
your predictors stored?  Are the values of these predictors known at
every point of your study area?  Can you show us a bit of your data (use
the function dput() to include *a small sample* of your data in the body
of your email).

If you insist on mucking about with correlation and testing, perhaps the
function cor.test() will give you what you want.  I reiterate however
that this seems to me to be a wrong approach.

cheers,

Rolf Turner

--
Technical Editor ANZJS
Department of Statistics
University of Auckland
Phone: +64-9-373-7599 ext. 88276
 
        [[alternative HTML version deleted]]

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Reply | Threaded
Open this post in threaded view
|

Re: Help

Bede-Fazekas Ákos
Dear Soufianou,
this is just a framework. Let's say that you have a vector ('variables')
containing the name of the environmental variables.

library(raster)
library(dismo)
library(corrplot)

for (variable in variables) {
     assign(variable, raster(paste0(variable, ".asc"))
}
environment <- brick(variables)
environment_standardized <- data.frame(scale(x =
as.data.frame(environment), center = TRUE, scale = TRUE))

correlation_matrix <- cor(environment_standardized, use = "na.or.complete")
corrplot(corr = correlation_matrix)
VIF <- vif(environment_standardized)
CN <- kappa(na.omit(environment_standardized), exact = TRUE)
# You can select variables that fulfill your criteria about correlation
structure
selected_variables <- variables[c()] # subsetting

maxent(x = environment[[selected_variables]] , p = presence_points)

HTH,
Ákos Bede-Fazekas
Hungarian Academy of Sciences

2018.05.14. 12:28 keltezéssel, Soufianou Abou via R-sig-Geo írta:

> Dear Rolf Turner,
>
>   I have points of presence of cowpea in Niger in CSV format; in addition to other variables (soil texture, soil pH, altitude, I downloaded from worldclim archives, the 19 environmental variables, I cut them all at the Niger scale and I converted them under ASCUI format. The idea for me is to choose the best variables to include in the model.
>   NB. I'm using Maxent model, but I'm not good in R software.
>
> Merci
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> SADDA Abou-Soufianou
>
> --------------------------------------
>
> Doctorant
>
> Université Dan Dicko Dankoulodo deMaradi-Niger
>
> BP 465 120, avenue MamanKoraou- ADS
>
>                     &
>
> Institut d’Ecologie et des Sciencesde l’Environnement de Paris (iEES-Paris)
>
> Centre IRD France Nord-(iEES Paris)-32,av.Henri Varangnat 93143 BONDY cedex.
>
> |
> Lien: https://ieesparis.ufr918.upmc.fr/index.php?page=fiche&id=378&droit=1
>
>
>   [hidden email]
>
> GSM : Niger : (+227) 96-26-99-87/91-56-35-19 ; France (+ 33)  07-55-79-39-93
>
>  
>    |
>  
>    |
>
>
>
>
>
>
>
>
>
>
>      Le lundi 14 mai 2018 à 12:05:40 UTC+2, Rolf Turner <[hidden email]> a écrit :
>
>
>
> Please keep your posts "on-list".  You are much more likely to get a
> useful answer that way.  There are many others on the list whose
> knowledge and insight are far greater than mine.
>
> I have therefore cc-ed the list in this reply.
>
> On 14/05/18 21:48, Soufianou Abou wrote:
>
>> Thank you for advice, Rolf Turner
>>
>> My question is as follows:
>>
>> I'd use maxent to model the potential distribution of cowpea on the
>> basis of the only presence data. Indeed, I have acquired a number of
>> environmental variables and bioclimatic regarding my area of study. But
>> to choose the most contributive variables in the model; I would like to
>> make a correlation analysis of these. On this, could you explain to me
>> the step by step procedures to follow in R? I would like to say scripts
>> for:- compile and call all environmental variables;- run the correlation
>> test to select the least correlated ones.
> As I said before, I don't think this is the right approach, but I can't
> be sure without knowing more about your data.  I find your description
> to be vague.
>
> How are your data stored?  What information do you have about the
> "distribution of cowpea".  Do you have *points* where cowpea is present
> or more extensive *regions* where it is present?  (And could these
> regions be "considered to be points" on the scale of interest?) How are
> your predictors stored?  Are the values of these predictors known at
> every point of your study area?  Can you show us a bit of your data (use
> the function dput() to include *a small sample* of your data in the body
> of your email).
>
> If you insist on mucking about with correlation and testing, perhaps the
> function cor.test() will give you what you want.  I reiterate however
> that this seems to me to be a wrong approach.
>
> cheers,
>
> Rolf Turner
>

_______________________________________________
R-sig-Geo mailing list
[hidden email]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo