Transform a mesh object into a mesh with correlation barriers

Usage

add_barrier_mesh(
  spde_obj,
  barrier_sf,
  range_fraction = 0.2,
  proj_scaling = 1,
  plot = FALSE
)

Arguments

spde_obj: Output from make_mesh().
barrier_sf: An sf object with polygons defining the barriers. For example, a coastline dataset for ocean data. Note that this object must have the same projection as the data used to generate the x and y columns in spde_obj.
range_fraction: The fraction of the spatial range that barrier triangles have.
proj_scaling: If spde_obj was created with scaling of the coordinates after the projection (e.g., dividing UTMs by 1000 so the spatial range is on a reasonable scale) the x and y values in spde_obj are multiplied by this scaling factor before applying the projection from barrier_sf.
plot: Logical.

Value

A list similar to make_mesh() but with spde_barrier and a couple other helper list elements added.

If plot = TRUE, then a basic plot will be created as a side effect. Each grey dot represents the center of a "normal" mesh triangle. Each red cross represents the center of a "barrier" mesh triangle.

References

Bakka, H., Vanhatalo, J., Illian, J., Simpson, D., and Rue, H. 2019. Non-stationary Gaussian models with physical barriers. https://arxiv.org/abs/1608.03787

https://sites.google.com/a/r-inla.org/www/barrier-model

https://haakonbakkagit.github.io/btopic107.html

Examples

if (require("sf", quietly = TRUE) &&
  require("ggplot2", quietly = TRUE) &&
  require("dplyr", quietly = TRUE) &&
  require("INLA", quietly = TRUE)) {

# First, download coastline data for our region.
# We will use `bc_coast` from the package data,
# but you can recreate it with the following.

# For applied situations on finer scales, you may with to use scale = "large".
# For that, first: remotes::install_github("ropensci/rnaturalearthhires")
# map_data <- rnaturalearth::ne_countries(
#   scale = "medium",
#   returnclass = "sf", country = "canada")
#
# # Crop the polygon for plotting and efficiency:
# st_bbox(map_data)
# bc_coast <- suppressWarnings(suppressMessages(
#   st_crop(map_data,
#     c(xmin = -134, ymin = 46, xmax = -120, ymax = 57))))

crs_utm9 <- 3156 # Pick a projection, here UTM9

st_crs(bc_coast) <- 4326 # 'WGS84'; necessary on some installs
bc_coast <- st_transform(bc_coast, crs_utm9)

# Project our survey data coordinates:
survey <- pcod %>% select(lon, lat, density) %>%
  st_as_sf(crs = 4326, coords = c("lon", "lat")) %>%
  st_transform(crs_utm9)

# Plot our coast and survey data:
ggplot(bc_coast) +
  geom_sf() +
  geom_sf(data = survey, size = 0.5)

# Note that a barrier mesh won't do much here for this
# example data set, but we nonetheless use it as an example.

# Prepare for making the mesh
# First, we will extract the coordinates:
surv_utm_coords <- st_coordinates(survey)

# Then we will scale coordinates to km so the range parameter
# is on a reasonable scale for estimation:
pcod$X1000 <- surv_utm_coords[,1] / 1000
pcod$Y1000 <- surv_utm_coords[,2] / 1000

spde <- make_mesh(pcod, xy_cols = c("X1000", "Y1000"),
  n_knots = 200, type = "kmeans")
plot(spde)

# Add on the barrier mesh component:
bspde <- add_barrier_mesh(
  spde, bc_coast, range_fraction = 0.1,
  proj_scaling = 1000, plot = TRUE
)

# In the above, the grey dots are the centre of triangles that are in the
# ocean. The red crosses are centres of triangles that are over land. The
# spatial range will be assumed to be 0.1 (`range_fraction`) over land compared
# to over water.

# We can make a more advanced plot if we want:
mesh_df_water <- bspde$mesh_sf[bspde$normal_triangles, ]
mesh_df_land <- bspde$mesh_sf[bspde$barrier_triangles, ]
ggplot(bc_coast) +
  geom_sf() +
  geom_sf(data = mesh_df_water, size = 1, colour = "blue") +
  geom_sf(data = mesh_df_land, size = 1, colour = "green")

# Now, when we fit our model with the new mesh, it will automatically
# include a barrier structure in the spatial correlation:
fit <- sdmTMB(density ~ s(depth, k = 3), data = pcod, mesh = bspde,
  family = tweedie(link = "log"))
fit
}
#> Linking to GEOS 3.10.2, GDAL 3.4.1, PROJ 8.2.1; sf_use_s2() is TRUE
#> 
#> Attaching package: ‘dplyr’
#> The following objects are masked from ‘package:stats’:
#> 
#>     filter, lag
#> The following objects are masked from ‘package:base’:
#> 
#>     intersect, setdiff, setequal, union
#> This is INLA_23.04.24 built 2023-04-24 19:15:35 UTC.
#>  - See www.r-inla.org/contact-us for how to get help.
#>  - To enable PARDISO sparse library; see inla.pardiso()
#> Loading required package: INLAspacetime
#> Warning: there is no package called ‘INLAspacetime’
#> Warning: Please install the `INLAspacetime` package
#>  which contains an implementation that runs faster!

#> Spatial model fit by ML ['sdmTMB']
#> Formula: density ~ s(depth, k = 3)
#> Mesh: bspde (isotropic covariance)
#> Data: pcod
#> Family: tweedie(link = 'log')
#>  
#>             coef.est coef.se
#> (Intercept)     2.06    0.21
#> sdepth          4.09    0.30
#> 
#> Smooth terms:
#>            Std. Dev.
#> sds(depth)      60.2
#> 
#> Dispersion parameter: 12.59
#> Tweedie p: 1.57
#> Matérn range: 16.69
#> Spatial SD: 2.67
#> ML criterion at convergence: 6400.432
#> 
#> See ?tidy.sdmTMB to extract these values as a data frame.