Dear Zafar,

If I understand your question correctly, you can train the models using the biomod2 package of R and you can also make predictions, but you are lack of only one quick step that categorize/classify your continuous (0-1) prediction output. In this case, let's say that the predictions are in a numeric vector called 'predictions', then you can use

berryFunctions::classify(x = predictions, method = "custom", breaks = c(0.2, 0.4, 0.6, 0.8, 1)

But I recommend you to

1) do the classification of the several model outputs separately and combine them afterwards;

2) do not use pre-defined breaks. Use breaks that depend on your data (e.g classInt::classIntervals())

For a more sophisticated method (that make classified predictions comparable between the studied species) see the following publication and the R script in its Supplementary Material.

Article Implementation and application of multiple potential natural...

HTH,

Ákos

Our bear distribution and ASF forecast articles include a suitable GIS method (smoothing followed by classification in three classes) of your model output.

Relabeling the model/classification output providing probabilities of presence as habitat or habitat suitability is another topic.

Somewhat deviating from your original question... but I would follow another approach and abstain from lumping together all models' continuous predictions. Consider to threshold all individual models, for example, using maxSSS (Liu et al. 2013), and to report the proportion of models predicting presence for each cell in your raster. In this case one obvious class would be 0% which would represent "almost certainly inhospitable" or something similar.

You could refer to Araujo & New 2007 for justification.

Dear Zafar,

You can categorise your ensemble model out put into suitability classes by following theses steps (If your output data from biomod2 is in grd format)

rm(list=ls())

setwd("C:/Users/Desktop/biomod")

library(rgdal)

library(raster)

##

files