make_spatial_info: Build objects related to spatial information
In James-Thorson/VAST: Vector-Autoregressive Spatio-Temporal (VAST) Model

make_spatial_info

R Documentation

Build objects related to spatial information

Description

make_spatial_info builds a tagged list with all the spatial information needed for Data_Fn

Usage

make_spatial_info(
  n_x,
  Lon_i,
  Lat_i,
  Extrapolation_List,
  knot_method = NULL,
  Method = "Mesh",
  anisotropic_mesh = NULL,
  Kmeans = NULL,
  grid_size_km = 50,
  grid_size_LL = 1,
  fine_scale = FALSE,
  Network_sz_LL = NULL,
  iter.max = 1000,
  randomseed = 1,
  nstart = 100,
  DirPath = getwd(),
  Save_Results = FALSE,
  LON_intensity,
  LAT_intensity,
  backwards_compatible_kmeans = FALSE,
  mesh_package = "INLA",
  ...
)

Arguments

`n_x`	the number of vertices in the SPDE mesh (determines the spatial resolution when `Method="Mesh"`)
`Lon_i`	Numeric vector, providing longitude for each sample
`Lat_i`	Numeric vector, providing latitude for each sample
`Extrapolation_List`	the output from `make_extrapolation_info`
`knot_method`	whether to determine location of GMRF vertices based on the location of samples knot_method=`samples` or extrapolation-grid cells within the specified strata `knot_method='grid'`; default `knot_method=NULL` is coerced to knot_method=`samples`
`Method`	a string specifying the form of spatial smoother with the following possible options: `Method="Mesh"` the SPDE method using geometric anisotropy as originally developed by INLA (recommended as default) `Method="Barrier"` the isotropic SPDE method but including barriers (i.e., preventing correlations from traveling over land); note that this feature is still in development and may change without warning `Method="Grid"` a 1st-order autoregressive process in two-dimensional spatial coordinates `Method="Stream_network"` an Ornstein-Uhlenbeck process specified along an acyclic graphc, e.g., representing a stream network `Method="Spherical_mesh"` the isotropic SPDE method with distances defined based on sphierical distances; note that this feature has not been extensively tested and please contract package author if interested in helping to do so
`anisotropic_mesh`	OPTIONAL, anisotropic mesh (if missing, its recalculated from loc_x)
`grid_size_km`	the distance between grid cells for the 2D AR1 grid (determines spatial resolution when `Method="Grid"`)
`grid_size_LL`	the distance between grid cells for the 2D AR1 grid (determines spatial resolution when `Method="Spherical_mesh"`)
`fine_scale`	a Boolean indicating whether to ignore (`fine_scale=FALSE`) or account for (`fine_scale=TRUE`) fine-scale spatial heterogeneity; See details for more informatino
`Network_sz_LL`	data frame with variables `"parent_s", "child_s", "dist_s", "Lat", "Lon"`, with default value `Network_sz_LL=NULL` where non-default values are only needed when `Method="Stream_network"` (determines spatial resolution when `Method="Stream_network"`)
`iter.max`	the number of iterations used per k-means algorithm (default=1000)
`randomseed`	a random number seed
`nstart`	the number of times that the k-means algorithm is run while searching for the best solution (default=100)
`DirPath`	a directory where the function looks for a previously-saved output (default is working directory)
`Save_Results`	a boolean stating whether to save the output (Default=TRUE)
`backwards_compatible_kmeans`	a boolean stating how to deal with changes in the kmeans algorithm implemented in R version 3.6.0, where `backwards_compatible_kmeans==TRUE` modifies the default algorithm to maintain backwards compatibility, and where `backwards_compatible_kmeans==FALSE` breaks backwards compatibility between R versions prior to and after R 3.6.0.
`...`	additional arguments passed to `inla.mesh.create`

Details

fine_scale=TRUE is a new feature starting in V8.0.0 which triggers two major changes:

projecting Gaussian Markov random fields from knots to sampling and extrapolation-grid locations using bilinear interpolation (i.e., piecewise linear smoothing), and
including density covariates individually for extrapolation-grid and sampling locations.

fine_scale=FALSE is designed to be backwards compatible with earlier versions, although V8.0.0 may also require changes to input naming conventions for covariates to specify the same model and attain the same fit.

LON_intensity and LAT_intensity allow users to specify locations that are used by the k-means algorithm to determine the location of knots, where e.g. users can either hard-code the desired knot locations via these inputs (using n_x greater than this number of locations), or use the extrapolation-grid to ensure that knots are located proportional to that grid.