An introduction to species distribution modelling in R

This tutorial provides an introduction to Generalized Dissimilarity Modelling (GDM) in R, focusing on data preparation, environmental gradient analysis, and community dissimilarity assessment using core geospatial and ecological modelling libraries.

On this page

Citation: MacFadyen S (2024). An introduction to species distribution modelling in R. https://docs.b-cubed.eu/sdm-r/

Introduction

Species Distribution Modelling (SDM) is a powerful method for predicting the potential distribution of species across different landscapes by analysing relationships between known species occurrences and environmental variables. SDMs are essential in conservation planning, allowing researchers and practitioners to identify critical habitats, model species response to environmental changes, and predict impacts of climate change. This tutorial makes extensive use of data from the Global Biodiversity Information Facility (GBIF), an international network and data platform providing open access to biodiversity occurrence records contributed by museums, scientific institutions, and citizen scientists worldwide. GBIF serves as a vital resource for biodiversity research, enabling users to access large datasets on species occurrences and environmental attributes.

Generalized Dissimilarity Modelling (GDM) is an advanced method for examining patterns of ecological turnover across environmental gradients. GDM is especially useful for mapping and visualising dissimilarity in species composition and diversity across spatial scales. It helps identify the role of environmental variables and geographic distance in driving species distribution patterns. Multi-Site Generalized Dissimilarity Modelling (MS-GDM) is an extension of GDM designed for analysing multiple sites simultaneously, providing deeper insights into how communities differ across locations and allowing comparisons across varied environmental conditions.

Zeta Diversity is an approach for examining biodiversity across multiple sites, quantifying the commonness and rarity of species shared between them. Zeta diversity goes beyond traditional pairwise comparisons, allowing researchers to analyse the turnover and nestedness of species composition across multiple locations, offering a robust tool for biodiversity assessment in complex landscapes.

In this tutorial, participants will learn to use these techniques in R, combining occurrence and environmental data for SDM, MS-GDM, and Zeta Diversity analyses to support species conservation and ecosystem management.

Load libraries

# See https://rsh249.github.io/bioinformatics/spatial.html

# install.packages("ENMeval", INSTALL_opts="--no-multiarch")
library(dismo) # Species Distribution Modelling tools
library(raster) # R tools for GIS raster files >>> USE terra{} instead
library(spocc) # Access species occurrence data from GBIF/iNaturalist
library(ENMeval) # Tuning SDMs
# library(mapr) # Quick mapping tools
library(sf)
library(ggplot2)
library(geodata)
library(sfheaders)

library(dplyr)
library(tidyr)        ## spread()
library(reshape2)     ## dcast(), melt()

# library(gdm)
library(zetadiv)

Download WorldClim climate data

# Download the WorldClim Bioclimatic variables for the world at a 10 arc-minute resolution
bio_10m = getData('worldclim', var='bio', res=10, path='D:/Workshops/Zeta_MSGDM/Data') # Set your own path directory

# Use geodata package instead
bio_10m_rsa = worldclim_country("South Africa", var="tmin", path='D:/Workshops/Zeta_MSGDM/Data')

summary(bio_10m)
bio_10m[[1]] #BIO1 = Annual Mean Temperature

Read descriptions/definitions of Bioclimatic variables.

Crop to your area of interest

# Define 'extent' of boundary
# South Africa
rsa_ext = extent(16, 33, -35, -22)

# Crop Bioclimatic variables to extent of South African boundary
rsaExt_bio_10m = crop(bio_10m, rsa_ext)
plot(rsaExt_bio_10m[[1]]) # Basic plotting

Or crop to country borders

# Using GADM (gives a SpatialPolygonsDataFrame)
rsa = getData('GADM', country='South Africa', level=0, path='D:/Workshops/Zeta_MSGDM/Data')
# rsa = sf::st_as_sf(rsa)

# OR use geodata instead
# rsa = gadm(country="ZA", level=1, path='D:/Workshops/Zeta_MSGDM/Data') # or path=tempdir()
# Use country_codes() to find correct code
# cc = country_codes()
# View(cc)

# Basic National map
ggplot()+
  geom_sf(data=sf::st_as_sf(rsa))+
  ggtitle("South Africa")

# Crop Bioclimatic variables to South African border
rsa_bio_10m = crop(bio_10m, rsa)
plot(rsa_bio_10m[[1]]) # See result

# OR use geodata package instead
# rsa_bio_10m = worldclim_country("South Africa", var="tmin", path='D:/Workshops/Zeta_MSGDM/Data')

Extract raster values to points

# Create 100 random points across South Africa
random_pts = spsample(rsa, n=100, type="random")    

# See randomly generate points on RSA map
plot(rsa_bio_10m[[1]])
plot(random_pts, add=T)

# Extract data to points
bio_values_1 = extract(rsa_bio_10m, random_pts)

# Use cbind.data.frame to get results as data.frame
bio_values_df = cbind.data.frame(coordinates(random_pts), bio_values_1)
head(bio_values_df)

Use your own tabular data

# Read directly from csv file
all_lepidoptera.sf = st_as_sf(read.csv('D:/Workshops/Zeta_MSGDM/Data/0097482-230224095556074.csv'), coords = c("x", "y"), crs = 4326)  
plot(all_lepidoptera.sf['family'])

###############################################################################
# EXTRACT RASTER DATA TO POINT LOCALITIES
# Extract raster value by points
lepidop_enviro = raster::extract(rsa_bio_10m, all_lepidoptera.sf)
lepidop_enviro
# Combine raster values with point and save as a CSV file.
lepidop_enviro.ptData = cbind(all_lepidoptera.sf, lepidop_enviro)
# Check output
names(lepidop_enviro.ptData)
head(lepidop_enviro.ptData)

Get data from GBIF

See https://poldham.github.io/abs/gbif.html
First sign-up for a free account.

library(dplyr)
library(readr)  
library(rgbif) # for occ_download

# Fill in your gbif.org credentials 
user = "xxxxxx" # your gbif.org username 
pwd = "xxxxxx" # your gbif.org password
email = "xxxx@xxxxx" # your email

The main functions related to downloads are:
occ_download(): start a download on GBIF servers.
occ_download_prep(): preview a download request before sending to GBIF.
occ_download_get(): retrieve a download from GBIF to your computer.
occ_download_import(): load a download from your computer to R.

gbif_download = occ_data(scientificName='Lepidoptera',
                         country='ZA',
                         hasCoordinate=TRUE,
                         hasGeospatialIssue=FALSE,
                         limit=500)
lepidoptera.df = as.data.frame(gbif_download$data)
head(lepidoptera.df)

lepidop.df = lepidoptera.df[,c(1,3,4,16,31:33,42:46)]
# head(lepidop.df)

lepidop.sf = st_as_sf(lepidop.df, coords = c("decimalLongitude", "decimalLatitude"), crs = 4326)
plot(lepidop.sf['stateProvince'])

Citing your download

If you end up using your download in a research paper, you will want to cite the download’s DOI. Please see these citation guidelines for properly citing your download. When using this dataset please use the following citation:

GBIF.org (16 March 2023) GBIF Occurrence Download https://doi.org/10.15468/dl.uvu2qm

Working with species occurence and environmental data tables

# CONVERT TO DATAFRAME
# all_lepidoptera.df = as.data.frame(all_lepidoptera.sf)
all_lepidoptera.df = as.data.frame(cbind(all_lepidoptera.sf, st_coordinates(all_lepidoptera.sf)))[,-11]
head(all_lepidoptera.df) 

# lepidop_enviro.df = as.data.frame(lepidop_enviro.ptData)
lepidop_enviro.df = as.data.frame(cbind(lepidop_enviro.ptData, st_coordinates(lepidop_enviro.ptData)))[,-30]
head(lepidop_enviro.df)

# OR use sfheaders{}
# lepidop_enviro.df = sf_to_df(lepidop_enviro.ptData, fill=TRUE)
# head(lepidop_enviro.df)
# str(lepidop_enviro.df)
# summary(lepidop_enviro.df)

# SET COLUMN FORMATS
lepidop_enviro.df$nameF = as.factor(lepidop_enviro.df$scientific)
lepidop_enviro.df$dateT = as.Date(lepidop_enviro.df$date)
head(lepidop_enviro.df)

# Create unique SITE IDs
lepidop_enviro.df$unqIDF = as.factor(as.integer(as.factor(paste(lepidop_enviro.df$X,lepidop_enviro.df$Y, sep='_'))))
head(lepidop_enviro.df)
str(lepidop_enviro.df)

Get your table into the right format

# str(lepidop_enviro.df)
head(lepidop_enviro.df)

lepidop_enviro.pa =lepidop_enviro.df %>%
  group_by(unqIDF, across(28:29), nameF, across(9:27)) %>%
  tally() %>%
  spread(nameF, n, fill=0)
head(lepidop_enviro.pa)
# View(lepidop_enviro.pa)

Multi-site generalised dissimilarity modelling for a set of environmental variables and distances

How to compute compositional turnover using zeta diversity

Using zetadiv: https://rdrr.io/cran/zetadiv

lepidop_enviro.pa_noNA = lepidop_enviro.pa[complete.cases(lepidop_enviro.pa), ] 
Sitexy = as.data.frame(lepidop_enviro.pa_noNA[,1:3])
# xy = st_as_sf(Sitexy, coords = c("X", "Y"), crs = 4326)
xy = as.data.frame(lepidop_enviro.pa_noNA[,2:3])

envRast = rsa_bio_10m
envTab = as.data.frame(lepidop_enviro.pa_noNA[,c(1:22)])
sppTab = as.data.frame(lepidop_enviro.pa_noNA[,c(1:3,23:1109)])

# envRast
# str(envTab) # 5430 obs(sites) of  20 variables(enviro)
# str(sppTab) # 5430 obs(sites) of  1088 variables(species)

# ##site-species, table-table
# exFormat1a = formatsitepair(sppTab, 1, siteColumn="unqIDF", XColumn="X", YColumn="Y", predData=envTab)
# exFormat1a
# 
# ##site-species, table-raster
# exFormat1b = formatsitepair(sppTab, 1, siteColumn="unqIDF", XColumn="X", YColumn="Y", predData=envRast)
# exFormat1b
# 
# # plot(exFormat1b, plot.layout=c(2,3))
# plot(exFormat1b[[1]])

# OR USE tidyr
long = sppTab %>%
  pivot_longer(!c(unqIDF,X,Y), names_to = "nameF", values_to = "value")

# Dealing with biases associated with presence-only data
#--------------------------------------
# weight by site richness
# long[complete.cases(long), ]

gdmTab = formatsitepair(long, bioFormat=2, XColumn="X", YColumn="Y",abundColumn='value',
                         sppColumn="nameF", siteColumn="unqIDF", predData=envTab)

gdmTab.rw = formatsitepair(long, bioFormat=2, XColumn="X", YColumn="Y",
                            sppColumn="nameF", siteColumn="unqIDF",abundColumn='value',
                            predData=envTab, weightType="richness")
# weights based on richness (number of species records)
gdmTab.rw$weights[1:5]

# remove sites with < 10 species records
gdmTab.sf = formatsitepair(long, bioFormat=2, XColumn="X", YColumn="Y",
                            sppColumn="nameF", siteColumn="unqIDF",abundColumn='value',
                            predData=envTab, sppFilter=10)

gdm: generalized dissimilarity modelling

Read more about gdm analysis here

gdm.1 = gdm(gdmTab.sf, geo=T)
# summary(gdm.1)
str(gdm.1)

# gdm plots
#--------------------------------------------------------
length(gdm.1$predictors) # get idea of number of panels
plot(gdm.1, plot.layout=c(4,3))

gdm.1.splineDat = isplineExtract(gdm.1)
str(gdm.1.splineDat)

par(mfrow=c(1,1))
plot(gdm.1.splineDat$x[,"Geographic"], gdm.1.splineDat$y[,"Geographic"], lwd=3,
     type="l", xlab="Geographic distance", ylab="Partial ecological distance")

zeta.ispline = Zeta.msgdm(sppTab[,c(4:1088)], envTab[,c(4:20)], xy, order=2,
                          rescale = TRUE,
                          rescale.pred = TRUE,
                          method = "mean",
                          normalize = "Jaccard",
                          reg.type = "ispline",
                          sam = 100)

# zeta.ispline.r = Return.ispline(zeta.ispline, envTab[,c(4:20)], distance = TRUE)
# zeta.ispline.r

dev.new()
Plot.ispline(msgdm = zeta.ispline, data.env = envTab[c(4:20)], distance = TRUE)