Maxent results, jackknife test, problems in aligning rasters

348 views
Skip to first unread message

Rollyna Domingo

unread,
Jun 19, 2021, 3:44:47 PM6/19/21
to Maxent
Hello everyone,

I am fairly new to Maxent and QGIS and I am currently making a predictive model for binturongs in Palawan. I am using the climatic variables from WorldClim and non-climatic variables such as elevation, human population, percent tree cover, and land cover. I"ve had many failed attempts in aligning these variables through QGIS but I have recently succeeded. However, it seems suspicious to me that all 19 climatic variables have equal values in the jackknife test. They have varying values in the permutation importance table but this is bothering me and Im not sure if I should push through with the model. Attached are some photos for reference

I hope someone can see and help me with this. 


Thank you!
Screen Shot 2021-06-19 at 1.27.55 PM.png
“arctictis_binturong”_jacknife.png
“arctictis_binturong”.png

Bede-Fazekas Ákos

unread,
Jun 20, 2021, 3:05:17 AM6/20/21
to max...@googlegroups.com
Hello,
the jacknife result is really strange. Although I don't know the reason of it, but I have a suggestion for you that may also connect to this issue. Your model uses all the 19 bioclimatic variables, that are surely multicollinear. It is highly recommended that you select some uncorrelated variables (before training your model!) considering correlation/multicollinearity criteria, ecology of the species, interpretability and some other aspects (e.g. avoiding the use of combined variables, https://doi.org/10.1111/2041-210X.13488).MaxEnt is just an algorithm... you are the one who should know more about the ecology of the species, so do not let the algorithm choose from the correlated variable set.
Have a nice weekend,
Ákos
--
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/2a2ab6b7-9939-4203-9927-d4582429602fn%40googlegroups.com.

Rollyna Domingo

unread,
Jun 20, 2021, 7:50:49 AM6/20/21
to Maxent
Thank you for this recommendation, Ákos! I have figured that the problems in the bioclimatic variables may be due to the realignment of the rasters through QGIS. 

Anyway, do you have any guides on how to check for multicollinearity among variables? I also saw papers that recommend this but I cannot find a step-by-step guide on how to do it. I am also not quite familiar with R but maybe if there is a guide I can do it. I also saw some people say it can be done through excel. 

Thank you so much for your reply! Have a great weekend too!

Bede-Fazekas Ákos

unread,
Jun 20, 2021, 8:00:56 AM6/20/21
to max...@googlegroups.com
Hello,
I cannot recommend any tutorials/guides, but I'm sure that there are a plenty of them in the internet. I always use R, and there are several alternative ways of multicollineraity/correlation calculation in that software. I usually check
1) the pairwise Pearson correlation coefficients (|r|<0.7) - functions cor() or Hmisc::rcorr() or raster::layerStats();
2) the variance inflation factor (VIF) of each variable (VIF < 10) - function usdm::vif();
3) the condition number (CN) of the dataset (CN < 30) - function kappa().
HTH,
Ákos

Rollyna Domingo

unread,
Jun 20, 2021, 10:27:25 AM6/20/21
to Maxent
Thank you, again! I will check this out.
Reply all
Reply to author
Forward
0 new messages