Interpreting clamping and Mess analysis

[Kris]

Dear all,

I have attached some results from a project I am using Maxent for.  To predict distribution changes on plant species under future climate change.  I recieve some interesting results which may be hard to interpret.

The first attachment is the model trained on current conditions, 2nd is the projection onto 2080 grids.  When I project to 2080 the entire area becomes suitable(red), although the attachement is blue.  I only get red when I click on the explain.bat.  Clamping picture is blue, Mess analysis (novel) is red and novel limiting shows 2 bioclim variables outside their training range. 

So my question is: why is the clamping picture blue and the Mess analysis red?  Is it showing that clamping worked to constrain these two bioclim variables and they no longer have an affect on my prediction?

I am wondering if there is something wrong with my projection.
I find it suspect that the entire area would become suitable. 

Any thoughts?

Thanks guys,
Kris 

[John Baumgartner]

Hi Kris,

The clamping grid indicates the effect of "clamping" response curves to their values at the extremities of training values for predictors ("treats variables outside the training range as if they were at the limit of the training range" - from the Tutorial). The grid shows the difference between clamped vs unclamped predictions. In your case, there seems to be very little (if any) effect of clamping, which would suggest that the response curves had already leveled out at the end of the scale to which the environmental change is moving.

MESS is unrelated, and indicates, for each cell, the extent to which predictors are outside of their training range (more precisely, the extent to which the most dissimilar variable is outside its training range). If response curves flatten out towards the extremities (little effect of clamping) then even in very novel environments, clamping will have little effect on predictions.

You still need to be wary of projections into novel environmental space, as the response curves are calculated on training data alone, and we can't be sure that their flattened ends can be reliably extrapolated beyond this range. 

wrt your other issue, I think that the Explain tool stretches the colour ramp to cover the range of your projected probabilities. The prediction for the cell your cursor is hovering over should be given in the status bar at the bottom-left of the window. 

The png files that Maxent produces should give a consistent colour ramp, so best to refer to those (or open the .asc in ArcMap/QGIS/etc.).

Hope that helps,

John

[TLJohnson]

Hi all,

 

I have been trying to follow the recommendations for clamping etc. when projecting to future climates, however, I would like to ask for input on a couple of things.  Below are images produced for one of the GCMs I used.  It does not appear that clamping is having much impact (Iscap_HAD2050_clamping.png), except for some areas in the southern part of the region in what appear to be riparian habitats.  The novel distribution map (Iscap_HAD2050_novel.png) shows that at least one variable is beyond the range of values present in the current climate, and that predictions in these areas, i.e. most of the study area minus the north central and NE regions, may be suspect.  The predicted distribution shows the range now confined to the NE portion of the area, a region (mostly) where variables were not beyond the range of current limits and where there was little impact of clamping.  Is it accurate to use the Iscap_HAD2050.png distribution as the predicted distribution in the face of climate change (for this model anyhow) given the low level of novel values and the negligible effect of clamping.

 

Thanks for your input!

 

tj