Avoid extrapolation

[FernandoBiganzoli]

Hi all

is there a way to avoid extrapolation during model projection in R?

I'm using the function maxnet.predictRaster from the ENMeval package

Thanks in advance

Fernando

[Bede-Fazekas Ákos]

Hi Fernando,
do you mean spatial extrapolation or environmental extrapolation? Spatial extrapolation is not a problem, but if you want, you can easily mask/crop your prediction raster to show only the region that you used for model training. Environmental extrapolation is a much more interesting and relevant issue. Maybe there are more established methods to remove environmental extrapolation, but my simple idea is the following:
1) save the min and max values of all your predictors within the training dataset (i.e., presence points + background points)
2) train and evaluate the model, and make prediction in the same way as previously
3) mask your prediction, i.e. set those cells to NA that are outside the learnt environmental hypercube. So delete all the cells that are lower than the saved minimum or higher than the saved maximum for any of the predictors.
HTH,
Ákos Bede-Fazekas

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maxent+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/888b07ae-02d8-4463-803d-e6e04376b5c7n%40googlegroups.com.

[Belay Ejigu]

Dear all,

I need your help to fill my knowledge gap on maxent.

Please how I can know how the maxent model calculates the probability of similarity of current and future environmental conditions at a given point of space, then determine the probability of area suitability for the existence of the concerned species?.

I am running GUI mode Maxent model version 3.4.4. Another, is the logistic regression model is the main component of maxent model?. How it correlate species occurrence and environmental variables?. Is is use pearson correlation coefficient values to determine the probability to be similar?

Thanks a lot in advance!

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/95e9c0da-d0ff-8285-ab54-349aaaf02282%40gmail.com.

[Bede-Fazekas Ákos]

Hello,

Please read this paper:
https://onlinelibrary.wiley.com/doi/full/10.1111/j.1472-4642.2010.00725.x

I hope it will answer your questions,
Ákos

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/CAL8t7yASGrgzW554bTqdH_%2B56ZHbHMuajK-YhRgN5DKibKqmNg%40mail.gmail.com.

[Belay Ejigu]

ok. thanks a lot. i will check it out.be blessed ever

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/53a0a411-a77e-327f-287b-fab147b6fa48%40gmail.com.

[Jamie M. Kass]

For those interested in avoiding extrapolation, just turn on clamping in the Maxent GUI or set doClamp = TRUE when running ENMeval 2.0.0. When making predictions with a model, clamping sets all environmental variable values that are outside the range of your training data to the max or min of the training data. Additionally, in ENMeval, you can control which parts of the environmental variable distribution to clamp (left or right or both).

Jamie