Check it out: reproj

The reproj package provides a single generic function, reproj(), for transforming coordinates between map projections. It works with matrices, data frames, and anything else that a package author cares to add a method for.

Coordinate transformation is a basic operation on numbers, it shouldn’t require a specific spatial data format.

The generic

library(reproj)

## matrix in, matrix out
reproj(cbind(147, -42), target = "+proj=laea +datum=WGS84", source = "OGC:CRS84")
#>          [,1]     [,2] [,3]
#> [1,] 5969744 -9803200    0

The source and target arguments accept anything that the PROJ library understands: EPSG codes, proj-strings, WKT2, authority:code strings. The output is always a matrix — 2, 3, or 4 columns depending on input and the four argument.

Convenience helpers reproj_xy() and reproj_xyz() force 2- or 3-column output:

reproj_xy(cbind(147, -42), target = "+proj=laea +datum=WGS84", source = "OGC:CRS84")
#>          [,1]     [,2]
#> [1,] 5969744 -9803200

The generic method model

reproj() is an S3 generic. This means any package can import it and add a method for its own spatial type. The method extracts the source CRS from the object and the user only needs to supply the target:

## hypothetical method for a spatial type that carries its own CRS
reproj.my_spatial_thing <- function(x, target, ...) {
  coords <- extract_coords(x)
  source <- get_crs(x)
  result <- reproj(coords, target = target, source = source, ...)
  rebuild_my_thing(x, result, crs = target)
}

This pattern is already used by the silicate and quadmesh packages — their reproj() methods know where the coordinates live and what the source CRS is, so the user just says reproj(my_mesh, target_crs).

The value of this design is that the generic lives in a small, dependency-light package. A downstream package can @importFrom reproj reproj without pulling in GDAL or any heavy spatial stack. The transformation itself is done by the PROJ library via the PROJ R package (preferred) or the proj4 package as fallback.

For example we could write a reproj.sf and a .sfc:

reproj.sfc <- function(x, target, ..., source = NULL) {
  ## handle this format with wk
  handled <- wk::as_wkb(x)
  crs <- wk::wk_crs(handled)$wkt
  ## stop if not a valid crs, we need source specified
  if (is.null(source)) {
    stopifnot(!is.na(crs))
  } else {
    crs <- source
 }
 trans <- PROJ::proj_trans_create(crs, target)
 sf::st_as_sfc(wk::wk_transform(handled, trans))
}

reproj.sf <- function(x, target, ..., source = NULL) {
 sf::st_set_geometry(x, reproj.sfc(sf::st_geometry(x), target, source = source))
}

That’s not very useful since sf::st_transform already exists but demonstrates the utility for any other object, and you can force input of a source if your object does not support that already. We could decompose our object to coordinates and transform those more directly, but anything handled by wk is already supported by this simpler mechanism.

CRS preservation

A key principle: when reproj transforms an object that has a method, the output carries the new CRS. The silicate method rebuilds the mesh with the target CRS attached. The quadmesh method does the same. The idea is that format-specific methods preserve the full object structure while updating both coordinates and CRS metadata.

For plain matrices and data frames, there’s no CRS to carry, and the target and source must be specified.

Under the hood: PROJ package

reproj delegates the actual math to the PROJ R package, which wraps the PROJ C library. The PROJ package provides:

PROJ::proj_trans() — transform coordinates, accepting wk types
PROJ::proj_trans_create() — create a reusable transformation object (a wk_trans)
PROJ::proj_crs_text() — convert between CRS representations (WKT2, proj-string, PROJJSON)

reproj vs sf::st_transform vs terra::project

sf and terra both have projection functions, but they’re tied to their respective data models. sf::st_transform() works on sf objects. terra::project() works on SpatVector and SpatRaster.

reproj works on anything. It’s deliberately format-agnostic: give it a matrix of coordinates, a source CRS, and a target CRS. This makes it useful for packages that don’t depend on sf or terra, and for workflows where spatial data isn’t in a standard spatial format — track data as data frames, coordinates pulled from a netCDF, vertices from an rgl mesh, cell centres computed from a grid specification.

Example: transforming track data

library(reproj)

## some GPS fixes as a plain data frame
tracks <- data.frame(
  lon = c(147.3, 147.4, 147.5),
  lat = c(-42.9, -42.8, -42.7),
  time = as.POSIXct(c("2024-01-01 08:00", "2024-01-01 09:00", "2024-01-01 10:00"))
)

## transform to a local projection
xy <- reproj_xy(
  cbind(tracks$lon, tracks$lat),
  target = "+proj=utm +zone=55 +south +datum=WGS84",
  source = "EPSG:4326"
)

tracks$x <- xy[, 1]
tracks$y <- xy[, 2]

Install from CRAN:

install.packages("reproj")

Links:

Check it out series

This is part of a series of posts for foundational spatial support in R. These posts include: