Future projection

[bipana sadadev]

Hi Maxent group, 

I have a couple of questions associated with the future projection. I hope the group will be able to clarify me. 

1. To my understanding, while projecting to the future we put the same variables, we use in the present situation, in the projection layer of the maxent. The variables for the present situation are selected after doing the multi-collinearity test. Do we need to do the collinearity test for the future variables too? Or simply we can add it as per our present collinearity test? 
2. While selecting the random test percentage, if we make it 25. Can't we use cross-validate, I got an error message when the random test percentage was set to 25 and replication type to cross-validate and Maxent popped me a message to reset it on Subsample. Can you please clarify this for me? 
3. Also, if we plan to project for the future using climatic variables, however, we do have anthropogenic variables such as population density, human settlement, and distance to the road in our present environmental variables which will change in the future, but we don't know in what ratio. Then in this situation how do we do the future projection? Do we do it by putting only the climatic variables in the projection layer of Maxent or are there other ways to do it? 
4. For model evaluation, TSS seems to be best over AUC. Can we calculate the TSS of 10 models (after running the Maxent selecting 10 Replicates and cross-validate replication type) using background predictions and sample predictions of each model?  I have selected the best features based on delta AIC = 0 using the ENMevaluate function in R. Can I use just those features to run the models or Can I use other features too if that delta AIC = 0 donot  give me good models or can I select other variables which were removed during collinearity test? 

Thankful for your time and consideration. 

Bipana

[Bede-Fazekas Ákos]

Dear Bipana,

I have answers for your questions no. (1) and (3).
(1) Collinearity is typically calculated for the reference period ("present"). It is not a problem, if non-correlated/non-collinear variables became correlated/collinear in the future. Therefore, there is no need to study future multicollinearity.
(3) There are two options:
- you use only those variables for the model training that are updated for the future period or are known to be static/unchanged
- you use all variables that are available for the reference period, you update some of them for the future (e.g. climate), and write some sentences in your manuscript's Discussion or Methods section about the limited availability of future data for the other variables (eg. population density) and discuss the uncertainty of your future predictions caused by this.

Have a nice week,
Ákos
___________
Ákos Bede-Fazekas
Centre for Ecological Research, Hungary

--
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/6d8fa862-ee4d-4b58-8bca-3922360d0b82n%40googlegroups.com.

Vírusmentes.www.avast.com

[bipana sadadev]

Hi Bede, 

Thank you so much for the response. I projected for the future as you said however I am getting the blue maps. What do you think might be the reason behind it? The blue maps are in the .html file but when I transfer them to the arc gis, it shows very low probability. How can we solve this problem? My study area is small. When we get the .avg.asc file, that is in grid format, and when we convert ascii to raster and reclassify them based on the selected threshold the final map is also in grid format. How can we fit the grid map to the boundary of our study area? I may have wrong-typed that but how can we create a smooth classified map covering the full boundary of our study area? 

Thank you so much.

[Bede-Fazekas Ákos]

Dear Bipana,
I'm not familiar with the stand-alone MaxEnt software (I use it from R), so I cannot answer your question.

Have a nice week,
Ákos

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/150bd7e5-522d-4a03-921f-fdc22c1bda7bn%40googlegroups.com.

[tdonald]

Hello

I would like to have a tutorial on how to run sdm with R ( variable collinearity, models, distribution map, etc.).

Thanks in advance 

[Beatriz]

I'm having the same issue projecting future scenarios when I place the future variable folders in the 'projection layers directory/file'. It ends up generating blue maps in the HTML and the future map ASC files have nothing. I'm going to try to project using R, which seems to be a bit more complicated. I need tutorials.

Não contém vírus.www.avast.com

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/150bd7e5-522d-4a03-921f-fdc22c1bda7bn%40googlegroups.com.