23_Vulnerability.R

################################################################################
#
#  RiskExtrapolation
#
#  Composite indices of vulnerability
#
################################################################################

#---------------------
# Reduce dimension
#---------------------

#----- Compute PLS

# Create formula
coefvars <- grep("coef", names(metadf), value = T) |> sort()
plsform <- sprintf("cbind(%s) ~ %s", paste(coefvars, collapse = ", "), 
  paste(metaprednames, collapse = " + "))

# Compute PLSR (basic PLS computed as a linear model)
plsres <- plsr(as.formula(plsform), data = metadf, scale = T)

# Extract scores for all locations and age
comps <- predict(plsres, newdata = metadf, type = "scores")
colnames(comps) <- sprintf("Comp%i", 1:ncol(comps))
metadf <- bind_cols(metadf, comps, .name_repair = "check_unique")

# Correlation with original metapredictors
coords <- plsres$loadings

#------------------
# Plot correlation with initial variables
#------------------

# Select right coordinates
coord_plot <- coords[, 1:npc, drop = F]
rownames(coord_plot) <- names(metaprednames)

# Color scale
pal <- scico(200, palette = "cork")

# Plot coordinates
corrplot(t(coord_plot), method = "square", is.corr = T, col.lim = c(-1, 1), 
  tl.srt = 45, tl.col = "black", cl.cex = .7, cl.align.text = "l",
  col = pal, cl.length = 5) 

# Save
dev.print(pdf, file = "figures/Fig2_compcor.pdf")