A function at applies a bayesian network to SDMs. Based on Staniczenko et al. (2017), doi:10.1111/ele.12770

For optimal performance this should be run in parallel
More Cores = More Better
Currently uses multisession because I'm writing this on Windows but will change to forking later

devtools::source_url("https://github.com/davidyshen/bayesian_network_sdm/blob/main/bnSDM.R?raw=TRUE") or git clone the repository or download just bnSDM.R

bnSDM(in_dir, out_dir = "BN_out/", focal, direction, method = "or", ncores = "auto")
Outputs a tif in the output directory named focal.species_bnSDM.tif

• in_dir = (required) character string of directory of folder ending in "/" containing .tif input raster SDMs of all species e.g. "folder/"
• out_dir = character string ending in "/" for output directory, defaults to "BN_out/"
• focal = (required) character string file name of focal species inside in_dir folder including file extension e.g. "Gonimbrasia_belina.tif"
• direction = (required) vector of 1s (positive interaction) and -1s (negative interactions) that denote the direction of interaction of interacting species on focal species in alphabetical order e.g. c(1, -1, 1)
• method = character string "or" (default) or "and" specifies state table filling method OR vs AND (box 2 of Staniczenko et al)
• ncores = integer value or character string auto, specifies number of cores/threads to run in multithreaded mode, defaults to auto using all threads on all cores

SDM rasters are loaded and used alphabetically, so the direction vector must be in the same order

Networks with more than ~12 dependencies are not computationally possible (I tried running it over 30 cores and its still not possible 😢 - size of matrix increases by 2^n)

• parallel
• raster
• stringr

smush function compresses many species into a single raster (or more depending on how many you want). Takes the highest probability value in each cell over all input rasters, and selects n of them. Used to reduce many species down into a functional group and reduce to a single or few interactors.

smush(in_dir, out_dir, out_name, n, ncores = "auto") Outputs n tifs in the output directory with name out_name.tif

• in_dir = (required) character string of path to folder containing the input data
• out_dir = (required) character string of output directory
• out_name = (required) character string of output tif filename
• n = integer of how many files to output. If more than 1, takes n of the highest probabilities in each cell over all the input rasters. Defaults to 1
• ncores = number of cores to use. Defaults to auto and uses all cores available