[R-sig-eco] Error in Maxent maps

[Bede-Fazekas Ákos]

Dear Irene,
Yes, you're right, the values should be between 0 and 1. And all the 
values are between 0 and 1 (even if they do not span through the whole 
interval), so the prediction is OK.
I guess you used complementary log-log link function (this is the 
default) to get the predicted values. This results in correct 
probability values if we assume that typical presences have a 1/point 
(1/cell) expected abundance (Phillips et al. 2017), but you have no data 
to reliably accept or reject this assumption. So you can state only 
that, for a certain species, predicted value 0.05 is larger than 0.02, 
which means that the environment is more suitable for the species in the 
first location than in the second one.
I have no publication about this theoretical issue, sorry.
Have a nice weekend,
Ákos


2021.01.15. 10:46 keltezéssel, Irene Rojo írta:
> Dear Ákos,
>
> Thank you very much for your reply. I have been thinking about it but 
> I still don't understand it very well. As I am having the logistic 
> output from Maxent, as far as I know, the values should be between 0 
> and 1 because Maxent does a transformation in order to get a 
> probability of finding the species referred to my study area (as you 
> say it is not a probability of occurrence in absolute terms). I don't 
> want to compare among species, but for the rest of the species the 
> predicted values obtained fall within that range of 0-1, while for a 
> couple of them the values are between 0-0.05. That is why I am 
> surprised. Maybe you have any publication which could help me to 
> understand it better?
>
> Thank you so much,
>
> Irene
>
> El mar, 12 ene 2021 a las 15:47, Bede-Fazekas Ákos 
> (<bfalevlist using gmail.com <mailto:bfalevlist using gmail.com>>) escribió:
>
>     Dear Irene,
>
>     Since your background points are not real occurrences (so you do not
>     know the prevalence of the species), and MaxEnt is a presence-only
>     method, the predicted value can not directly treated as
>     probability of
>     occurrence. Also you should never compare the raw predicted values
>     between species (or between different models), even if you use
>     presence-absence data/method. 0.005 for Species1 is not better/worse
>     than 0.95 for Species2. So I would say that the predicted values
>     are OK.
>     If you really want to compare predictions between species, I
>     recommend
>     you to rescale the raw values to a 5-level ordinal scale using
>     specific
>     thresholds that account for observed presences. Please refer to
>     Somodi
>     et al. (2017):
>     https://www.researchgate.net/publication/318561746_Implementation_and_application_of_multiple_potential_natural_vegetation_models_-_a_case_study_of_Hungary
>     <https://www.researchgate.net/publication/318561746_Implementation_and_application_of_multiple_potential_natural_vegetation_models_-_a_case_study_of_Hungary>
>
>     Have a nice week,
>     Ákos Bede-Fazekas
>     Hungarian Academy of Sciences
>
>     2021.01.12. 13:47 keltezéssel, Irene Rojo írta:
>     > Dear all,
>     >
>     > My name is Irene and I am performing Maxent SDM in R with the dismo
>     > package.
>     >
>     > I am working with several species, performing one analysis for each
>     > species, and all are working fine, i.e. the models have sense,
>     except two
>     > of them. For those species, I can run Maxent models and I get a
>     very high
>     > value of AUC (0.99) but the resulting maps show values between 0
>     and 0.005,
>     > when for the rest of the species they are between 0 and (almost)
>     1. So maps
>     > have no sense. The predictor variables are the same, and those
>     two species
>     > are not the ones with the lowest number of occurrences (this is
>     the first
>     > reason I thought that could be affecting the results).
>     >
>     > I know it is not easy to know what is going on, but I am doing
>     exactly the
>     > same with all species. If anyone can guess the error or know
>     which part of
>     > the analysis is more sensitive so I can check it again I would
>     be very
>     > grateful.
>     >
>     > Thanks a lot in advance,
>     >
>     > Irene
>     >
>     >       [[alternative HTML version deleted]]
>     >
>     > _______________________________________________
>     > R-sig-ecology mailing list
>     > R-sig-ecology using r-project.org <mailto:R-sig-ecology using r-project.org>
>     > https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>     <https://stat.ethz.ch/mailman/listinfo/r-sig-ecology>
>
>     _______________________________________________
>     R-sig-ecology mailing list
>     R-sig-ecology using r-project.org <mailto:R-sig-ecology using r-project.org>
>     https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>     <https://stat.ethz.ch/mailman/listinfo/r-sig-ecology>
>


	[[alternative HTML version deleted]]