Supplementary Software 1

A meta-analysis of global fungal distribution reveals climate-driven patterns

Tomáš Větrovský, Petr Kohout, Martin Kopecký, Antonín Machač, Matěj Man, Barbara Doreen Bahnmann, Vendula Brabcová, Jinlyung Choi, Lenka Meszárošová, Zander Rainier Human, Clémentine Lepinay, Salvador Lladó, Rubén López-Mondéjar, Tijana Martinović, Tereza Mašínová, Daniel Morais, Diana Navrátilová, Iñaki Odriozola, Martina Štursová, Karel Švec, Vojtěch Tláskal, Michaela Urbanová, Joe Wan, Lucia Žifčáková, Adina Chuang Howe, Joshua Ladau, Kabir Gabriel Peay, David Storch, Jan Wild, Petr Baldrian

### Analyses of "environmental drivers" part of the paper by Vetrovsky et al. "" in Nature Communications. 
### R script written by Martin Kopecký (ma.kopecky@gmail.com) and works in R 

library(HH) # ver. 3.1-35
library(vegan) # ver. 2.5-2
library(extendedForest) # ver. 1.6.1
library(gradientForest) #ver. 0.1-17

#set WD
setwd("c:/MK/Dokumenty/DATA/Rdata/globalFungi")

### GRADIENT FOREST

# load envi data for 742 aggregated sites
envi.dta <- read.table("env_ClimSoilNdvi.csv", row.names=, header=T, sep=";",dec=".")

# select final envi data for models
envi <- envi.dta[,c("CH_2","CH_18","CH_9","CH_19","CH_8","CH_15","GPP", "BulkD", "pH")]

# check VIF (all variables with VIF<5)
sort(HH::vif(chelsa), decreasing = T)

# SH occurence data in 742 aggreegated sites
species <- read.table("2019_top496_agreg_sites.csv", row.names=1, header=T, sep=";",dec=".")

# remove site identification
species2 <- species[,3:471]

# transform to P/A data
species.pa <- decostand(species2, "pa")

nSites <- dim(species2)[1]
nSites
nSpecs <- dim(species2)[2]
nSpecs

# set levels for conditional permutation
lev <- floor(log2(nSites * 0.368/2))
lev

# fit gradient forest
gf <- gradientForest(cbind(envi, species.pa),
                     predictor.vars = colnames(envi), response.vars = colnames(species.pa),
                     ntree = 500, transform = NULL, compact = T,
                     nbin = 201, maxLevel = lev, corr.threshold = 0.5, trace=T)

# sort outputs from random forest model for each SH with R2>0
sorted_models <- gf$res.u[order(gf$res.u$spec, -gf$res.u$rsq.var),]

sorted_models



  ### RICHNESS ACROSS LATITUDES AND REGIONS

  ### REGRESSION ANALYSES

  ### INPUT
  ### x ... dataframe with cells, their richness, latitude, longitude
  ### ndata ... dataframe with climatic data for all grid cells (from chelsa)


  library(SDMTools)
  library(maptools)
  library(raster)


  ### EXPLORE GEOGRAPHIC PATTERNS IN THE RICHNESS DATA

  ### upload map template from directory
  blank  <- raster.from.asc(read.asc ("C:\\blank_map.asc")) ### template map
  blank -> blank0
  blank[] <- ndata$bio1

  ### plot richness on the map showing world climate
  plot (blank, main= "richness across the world - against bio1")
  points (x$longitude, x$latitude, pch=20, cex = x$sprich,col="darkblue")

  ### loess fitting
  plot (x$sprich ~ x$latitude, xlab = "Latitude",ylab="richness",font.lab=2,pch=19, col="darkgrey", main="Latitudinal gradient (LDG)")
  mo1 <- cbind(x$latitude,predict (loess (x$sprich ~ x$latitude, span=0.75, degree=2)))
  lines( mo1[order(mo1[,1]),],col="darkblue",lwd=4)

  ################################################################

  ###  CREATE DATASET WITH SPECIES RICHNESS, CHAO, AND CHELSA CLIMATIC VARIABLES

  xdataset <- data.frame(x,matrix(ncol=dim(ndata)[[2]]-2,nrow=dim(x)[[1]],data=NA))

  for (i in 1:19){

       blank [] <- ndata[,2+i]
       xdataset[,10+i]  <- extract (blank, xdataset[,c("longitude","latitude")])
       names(xdataset)[10+i] <- names(ndata)[2+i]

       }


  ### save the created dataset, including environmental predictors
  write.table(xdataset,file="Fungi dataset OTU.xls",sep="\t",row.names=F)



  ### run best-subset regression on the environmental data
  library (leaps)

  xdataset2  <- xdataset [complete.cases(xdataset),]
  rg1 <- regsubsets (xdataset2[,c(11:29)],xdataset2$sprich)

  which.max(summary(rg1)$adjr2)
  which.min(summary(rg1)$cp)
  which.min(summary(rg1)$bic)


  ### identify the best model and predict geographic patterns in richness based on the model
  ### (here shown an example of a linear model with eight predictors)
  
  coef(rg1,8)
  ft1 <- lm (sprich ~ bio2 + bio3 + bio4 + bio5 + bio6 + bio12 + bio13 + bio17, data= xdataset2)

  pr1 <- predict (ft1,ndata[,c("bio2","bio3","bio4","bio5","bio6","bio12","bio13","bio17")],interval="predict")
  blank[] <- pr1[,"fit"] ### fit
  plot(blank, main = "richness R2=17.31%")

  ### plotting

  library(rasterVis)

  dev.new(width=25, height=15)
  blank[] <- pr1[,"fit"] ### fit
  myTheme <- rasterTheme(region=(brewer.pal('Reds', n=9)))
  levelplot(blank, margin = list (FUN = "sum"), par.settings=myTheme, contour=F, main = "richness R2=17.31%")

  dev.new(width=25, height=15)
  blank[] <- pr1[,"lwr"] ### lower prediction interval
  myTheme <- rasterTheme(region=(brewer.pal('PuBu', n=9)))
  levelplot(blank, margin = list (FUN = "sum"), par.settings=myTheme, contour=F, main = "richness LOWER BOUND R2=17.31%")

  dev.new(width=25, height=15)
  blank[] <- pr1[,"upr"] ### upper prediction interval
  myTheme <- rasterTheme(region=(brewer.pal('PuBu', n=9)))
  levelplot(blank, margin = list (FUN = "sum"), par.settings=myTheme, contour=F, main = "richness UPPER BOUND R2=17.31%")



  ### RANDOM FOREST
  ### predicting species richness, based on climate, environment, etc.

  ### input
  ### x ... dataframe with richness, chao, chelsa climate, environment, etc.
  ### ndata ... dataframe with chelsa climate, soil and vegetation for the entire world


  library(randomForest)


  xcomp  <- x[complete.cases(x),]


  rdf    <- randomForest (sprich ~   CH_01 + CH_02 + CH_03 + CH_04 +
                                     CH_05 + CH_06 + CH_07 + CH_08 +
                                     CH_09 + CH_10 + CH_11 + CH_12 +
                                     CH_13 + CH_14 + CH_15 + CH_16 +
                                     CH_17 + CH_18 + CH_19 +

                                     BLDFIE_M_sl1_ + CECSOL_M_sl1_ + EVMMOD3a_ +
                                     OCSTHA_M_sd1_ + PHIHOX_M_sl1_ + veg_gpp_mod17a2h.annual_d_,

                                     ntrees = 500,
                                     data = xcomp)

  rdf

  plot (rdf$predicted ~ xcomp$sprich, col="darkblue",pch=19)



  ### PLOT ON THE MAP

  library(SDMTools);library(maptools);library(raster);library(rasterVis)
  blank  <- raster.from.asc(read.asc ("C:\\blank_map.asc")) ### template map uploaded from directory
   
  dev.new(width=25, height=15)
  blank[] <- predict (rdf,newdata=ndata[,-(1:2)]) ### fit
  myTheme <- rasterTheme(region=(brewer.pal('Reds', n=9)))
  levelplot(blank, margin = list (FUN = "sum"), par.settings=myTheme, contour=F, main = "Random forest")