One of the crucial choices when using Species Distribution Models (SDM) based on pseudo-absences approaches is the delineation of the background area to fit the model. Defining its extent, however, remains a challenge and often studies are based on partial SDMs (i.e. models build with only part of the species geographic distribution)
MinBAR is an R package that aims at (1) defining the minimum background extent necessary to fit SDMs reliable enough to extract ecologically relevant conclusions from them and (2) optimizing the modelling process in terms of computation demands. In this vignette we show an example of MinBAR usage.
Please see this paper for further details on the package MinBAR as well as for the references of other packages and works mentioned in this vignette.
We need species occurrences, which can be downloaded from different public repositories. In this case we will use the Bioatles (http://bioatles.caib.es/), a data base of species on the Balearic Islands. The occurrences will be downloaded with the R package PreSPickR (https://github.com/xavi-rp/PreSPickR).
library(devtools) #install_github("xavi-rp/PreSPickR") library(PreSPickR) PreSPickR:::bioatles(sp_list = c("Asphodelus aestivus"), out_name = "sp_records_bioatles")
Then, we will download environmental data from WorldClim (https://worldclim.org/data/index.html) and will crop them to the extent of the archipelago.
The idea behind MinBAR is to sequentially fit several concentric SDMs, each with an increased diameter, from the centre of the species geographical distribution to the periphery, until a model which satisfies user’s needs is reached. We call “buffers” to these concentric SDMs.
In order to evaluate the predictive performance of the models, MinBAR includes two metrics based on the Boyce Index, implemented in the R package ecospat. On the one hand, Boyce Index Partial (BI_part) evaluates the accuracy of predictions within the buffer (i.e. training area). On the other hand, Boyce Index Total (BI_tot) assesses predictions beyond the training area, across the whole distribution of the species (i.e. model transferability).
The user can choose either (1) to run the models for all the buffers to see if the selected background area is accurate and how the quality of the models evolves, or (2) to stop the process when it reaches certain conditions, which can be defined by the user as well. For this, the main function minba() has different arguments:
If all the four arguments are NULL, all buffers are preocessed. If one or the two BIs have a value, this or these are the maximum limit to be reached. If one or the two SD_BIs have a value, this or these are the minimum limit to be reached. In this example we will set all of them as NULL.
minba() is implemented for MaxEnt models and the user can choose to use either the R package maxnet (default) or the original java program, if installed. We will run it using maxnet with default parameters, except for the number of background points, which is set to 50% of pixels in the study area.
Now we can run MinBAR to fit the models.
#install_github("xavi-rp/MinBAR") library(MinBAR) MinBAR:::minba(occ = "sp_records_bioatles.csv", varbles = bioclim, wd = getwd(), prj = 4326, num_bands = 10, n_rep = 3, maxent_tool = "maxnet")
As results of the calculations, we can check both a csv file and a plot. While the former shows the best and second best buffer, both with and without execution time, the latter presents a plot with the evolution of the two Boyce Index with the increase of the buffer diameter in kilometres, as well as the execution time.