inst/examples/devel.cvmeasure.R

In inlabru: Bayesian Latent Gaussian Modelling using INLA and Extensions

\donttest{
if (bru_safe_inla() &&
    require(ggplot2, quietly = TRUE) &&
    bru_safe_sp() &&
    require("sp")) {

  # Load Gorilla data

  data("gorillas", package = "inlabru")

  # Use RColorBrewer

  library(RColorBrewer)

  # Fit a model with two components:
  # 1) A spatial smooth SPDE
  # 2) A spatial covariate effect (vegetation)

  pcmatern <- INLA::inla.spde2.pcmatern(gorillas$mesh,
    prior.sigma = c(0.1, 0.01),
    prior.range = c(0.01, 0.01)
  )

  cmp <- coordinates ~ vegetation(gorillas$gcov$vegetation, model = "factor_contrast") +
    spde(coordinates, model = pcmatern) -
    Intercept(1)

  fit <- lgcp(cmp, gorillas$nests,
    samplers = gorillas$boundary,
    domain = list(coordinates = gorillas$mesh),
    options = list(control.inla = list(int.strategy = "eb"))
  )

  # Predict SPDE and vegetation at the mesh vertex locations

  vrt <- fm_vertices(gorillas$mesh, format = "sp")
  pred <- predict(
    fit,
    vrt,
    ~ list(
      joint = spde + vegetation,
      field = spde,
      veg = vegetation
    )
  )

  # Plot component mean

  multiplot(ggplot() +
    gg(gorillas$mesh, color = pred$joint$mean) +
    coord_equal() +
    theme(legend.position = "bottom"),
  ggplot() +
    gg(gorillas$mesh, color = pred$field$mean) +
    coord_equal() +
    theme(legend.position = "bottom"),
  ggplot() +
    gg(gorillas$mesh, color = pred$veg$mean) +
    coord_equal() +
    theme(legend.position = "bottom"),
  cols = 3
  )

  # Plot component variance

  multiplot(ggplot() +
    gg(gorillas$mesh, color = pred$joint$var) +
    coord_equal() +
    theme(legend.position = "bottom"),
  ggplot() +
    gg(gorillas$mesh, color = pred$field$var) +
    coord_equal() +
    theme(legend.position = "bottom"),
  ggplot() +
    gg(gorillas$mesh, color = pred$veg$var) +
    coord_equal() +
    theme(legend.position = "bottom"),
  cols = 3
  )

  # Calculate variance and correlation measure

  vm <- devel.cvmeasure(pred$joint, pred$field, pred$veg)
  lprange <- range(vm$var.joint, vm$var1, vm$var2)

  # Variance contribution of the components

  csc <- scale_fill_gradientn(colours = brewer.pal(9, "YlOrRd"), limits = lprange)
  boundary <- gorillas$boundary

  plot.1 <- ggplot() +
    gg(gorillas$mesh, color = vm$var.joint, mask = boundary) +
    csc +
    coord_equal() +
    ggtitle("joint") +
    theme(legend.position = "bottom")
  plot.2 <- ggplot() +
    gg(gorillas$mesh, color = vm$var1, mask = boundary) +
    csc +
    coord_equal() +
    ggtitle("SPDE") +
    theme(legend.position = "bottom")
  plot.3 <- ggplot() +
    gg(gorillas$mesh, color = vm$var2, mask = boundary) +
    csc +
    coord_equal() +
    ggtitle("vegetation") +
    theme(legend.position = "bottom")

  multiplot(plot.1, plot.2, plot.3, cols = 3)

  # Covariance of SPDE field and vegetation

  ggplot() +
    gg(gorillas$mesh, color = vm$cov)

  # Correlation between field and vegetation

  ggplot() +
    gg(gorillas$mesh, color = vm$cor)

  # Variance and correlation integrated over space

  vm.int <- devel.cvmeasure(pred$joint, pred$field, pred$veg,
    samplers = fm_int(gorillas$mesh, gorillas$boundary),
    mesh = gorillas$mesh
  )
  vm.int
}
}