I am thinking of using Maxent for global species distribution model. However while I am reading papers relating to the issue, papers don't seem to provide a clue on how their ensemble was done. If I picked like five GCM outputs from worldclim or Chelsa, how do I get a single ensemble output from the five GCM outputs that I downloaded from worldclim or Chelsa?
Have a look at our ensemble SDM model including maxent from Andalusia.
Our bear article will inform you where CHELSA might be preferred over WorldClim and why.
In SDM studies, many factors can cause uncertainties, like choice of SDMs, GCMs, and extent of study area. I would suggest to use multiple SDMs and multiple GCMs. You can project habitat suitability under each GCM and then make an ensemble output. Please check biomod2 R package. Please feel free to contact me if you need further help.
James R, Hatten Joaquín Duque Lazo Zhixin Zhang Hein Van Gils
Thank you all of you for taking time for answering my question.
Can I check with you all for the GCM ensemble, if I interpreted the method of creating the ensemble correctly? In the event that I decided to work with 19 biovariables from 5 GCM models, the steps I will take will be as follow:
1) I will first download the biovariables Geotiff for the 5 GCMs I chose from databases like Worldclim or CHELSA at the same resolution.
2) I will use the raster calculator on a GIS software to sum the 5 GCMs outputs I got from the database and average them out to get a Geotiff for each of the biovariable that is the mean of the 5 GCMs (if I decided to not weigh each of the GCMs), which I will call it mean model
3) I will then continue to run my SDM base on mean model as climate projection input to get my output
Is my interpretation correct? If my interpretation is correct, for the next step, if I choose 2 or more SDMs for my projection, do I run each SDM once to get output for each of the SDM, then similarly use the raster calculator to sum the different output and average them out if I only want to present one output in my study?
Thank you for your attention.
You can use forecast package R and dismo another package. I prefer WORLDCLIM for terrestial layer and BIORACLE y MARSPEC for benthic layer. I recomend you araujo y new 2007
Chan Wee How , a reasonable workflow would be to
1) download climate data (and data on species' occurrences)
2) fit an ensemble of SDMs for each of the five GCMs (do not average them). So if you for example run an ensemble of 10 SDM algorithms, you would end up with 50 projections.
3) combine the ensemble projections into a consensus projection
Of course one way to proceed would be to average the GCMs as you suggested, but then you would lose information that the different GCMs contain. If you would fit SDMs separately on the different GCMs, this will allow you to see which part of the variability is related to GCMs, which to SDMs, as noted by Zhixin Zhang above.
Note also that the choice of climatic variables to be used in the models is not a trivial task. Often a large array of variables are used (as you mentioned 19), but preferably one should choose variables that based on literature determine the geographic distribution of the species in question. From my perspective this yields biologically more realistic models and the results are easier to interpret.
- Badges
- Science topic
- Similar topics
- Papers