Obtain a formatted output from spatial predictions.

Function to transform a prediction data from various spatial models (e.g., inlabru, PointedSDMs or GLMs tools) into an sf object if it is from points predictions (e.g., fm_vertices) or a data.frame with the corresponding locations if it is from pixel grids (see fm_pixels). Spatial prediction data can also be obtained across a given region using expand.grid(x, y) function, where 'x' and 'y' are geographical coordinates of the grids locations.

Usage

format_predictions(prediction_data, base_map = NULL)

Arguments

prediction_data

A model prediction which may be either an sf or a data.frame object or a raw prediction from the inlabru-like models. The prediction can be on the response or linear predictor scale, depending on whether the output is for a model evaluation or visualization of the prediction.

base_map

An sf polygon having the same crs with the spatial locations used for predictions.

Value

A data.frame for grid-based predictions or sf object for point-based predictions.

See also

Other prediction analyses: generate_maps(), suitability_index()

Examples

if (FALSE) { # \dontrun{
library(dplyr)
library(sf)
set.seed(42)

# Simulate the prediction data
grid_df <- expand.grid(x = 0:50, y = 0:50)
grid_df <- grid_df |>
  mutate(mu = (x + y) / 10 + rnorm(nrow(grid_df))) |>
  mutate(sd = runif(nrow(grid_df), 0.5, 1.5))
head(grid_df)

# Convert the grid into a SpatRaster
grid_r <- terra::rast(grid_df, crs = "epsg:4326")

# a) A standard data.frame returns the same object
field_pp1 <- format_predictions(grid_df)
class(field_pp1)

# b) A grid-based object returns a data.frame
grid_sf <- st_as_sf(grid_df, coords = c("x", "y"), crs = "epsg:4326")
class(grid_sf) <- c("bru_prediction", "sf", "data.frame")

field_pp2 <- format_predictions(grid_sf)
print(class(field_pp2))

# c) A point-based prediction returns the original class
if (require("fmesher", quietly = TRUE)) {
  bnd <- fm_nonconvex_hull(as.matrix(grid_df[, c("x", "y")]), convex = -0.10)
  mesh <- fm_mesh_2d(boundary = bnd, max.edge = c(3, 30), crs = "epsg:4326")
  vt <- fm_vertices(mesh, format = "sf")

  # An inlabru-like prediction at mesh vertices (extended areas are imputed)
  sampled_vals <- terra::extract(grid_r, vt)
  sim_field <- vt |>
    mutate(
      mean = dplyr::coalesce(sampled_vals$mu, mean(grid_df$mu, na.rm = TRUE)),
      sd = dplyr::coalesce(sampled_vals$sd, mean(grid_df$sd, na.rm = TRUE)),
      q0.025 = mean - 1.96 * sd,
      q0.5 = mean,
      q0.975 = mean + 1.96 * sd,
      median = mean
    )
  class(sim_field) <- c("bru_prediction", "sf", "data.frame")

  field_pp3 <- format_predictions(sim_field)
  print(class(field_pp3))
}
} # }