Hi,
I used Maxent to create a model identifying the best habitats for the Short-eared Owl. I used a 800mX800m grid (0,5 by 0,5 miles). I calculated, for each environmental variable, the percentage of each pixel occupied by that variable. Those were my Environmental layers. My problem is with the response curves: when the value of a single variable is changed and the others are kept to their mean values, my response curves are very flat and all close to 1 (see image "Capure1"). But when each variable is used without the others, the response curves make more sense (see "Capture 2"). Anybody has an explanation?
Thank you very much.
"But when each variable is used without the others, the response curves make more sense."
When modeling with maxent (I recommend using Ensemble forecasting), what you are looking for is to see the area that a species can occupy according to the variables that you consider to influence it (what you are going to predict is the climate niche of the species), But what interests you is to see how the response curves are, the problem is that despite the fact that I assume that the variables you used are not correlated or unrelated, they do not act by themselves. Look at the jackniffe values to know how each variable influences the model by itself and hence the explanation you give to those response curves.
Hello..
May be variables are correlated, if this happen then the combine graph of variable response curve is hard to interpret. You must have to do some kind of correlation test to select the best variable.
Regards.
Dear Marie-Helene,
"each variable is used without the others": do you mean you train separate models using one variable per model?
I agree with Asiel Cabrera Guerrero and Souvik Barik : the issue might be related to the correlation between the variables.
Regards,
Ákos
I recommend before modeling, do a correlation test spearman to discard the variables that are more correlated and obtain a suitable set of variables, in addition, I also advise you to calculate the VIF. What resolution are you modeling? You should worry about species with a small number of presence records,
Hello all,
I have 8 environmental variables (percentage of a habitat type in a in a 0,5X0,5 miles pixel). I had problems with the default output's response curves (cloglog), the problem being that some response curves were absolutely flat and very close to 1. This morning I ran Maxent again, testing each of the 4 outputs possible. Raw, logistic and cumulative output gave good results (the response curves were similar wheter each environmental layer was used with the value of the other layers kept at their mean value (1st set of response curves in Maxent results), or each environmental value was used by itself (2nd set of response curves in Maxent results). My question is: why is the cloglog output's response gives "flat lines" and not the other outputs?
- Badges
- Science topic