Getting started with fasterRaster
Adam B. Smith
2024-10-24
Source:vignettes/fasterRaster.Rmd
fasterRaster.RmdfasterRaster interfaces with GRASS GIS to process rasters and spatial vector data. It is intended as an add-on to the terra and sf packages, and relies heavily upon them. For most rasters and vectors that are small or medium-sized in memory/disk, those packages will almost always be faster. They may also be faster for very large objects. But when they aren’t, fasterRaster can step in.
Installing fasterRaster
You probably already have fasterRaster installed on your computer, but if not, you can install the latest release version from CRAN using:
install.packages("fasterRaster")and the latest development version using:
remotes::install_github("adamlilith/fasterRaster", dependencies = TRUE)(You may need to install the remotes package first.)
Installing GRASS GIS
fasterRaster uses GRASS to do its operations. You will need to install GRASS using the “stand-alone” installer, available through the GRASS GIS. Be sure to use the “stand-alone” installer, not the “OSGeo4W” installer!
Starting a fasterRaster session
I recommend attaching the data.table, terra, and sf packages before attaching fasterRaster package to avoid function conflicts. The data.table package is not required, but you most surely will use at least one of the other two.
library(terra)
#> terra 1.7.83
library(sf)
#> Linking to GEOS 3.10.2, GDAL 3.4.1, PROJ 8.2.1; sf_use_s2() is TRUE
library(data.table)
#>
#> Attaching package: 'data.table'
#> The following object is masked from 'package:terra':
#>
#> shift
library(fasterRaster)
#> Warning: no DISPLAY variable so Tk is not available
#> fasterRaster 8.4.0.7028
#> To avoid conflicts between functions, please attach the `terra`, `sf`,
#> and `data.table` packages before attaching `fasterRaster` using, for
#> example, `library(terra)`.
#> For guides and table of contents, see `?fasterRaster`.
#>
#> Attaching package: 'fasterRaster'
#> The following object is masked from 'package:data.table':
#>
#> %notin%
#> The following object is masked from 'package:sf':
#>
#> st_coordinates
#> The following object is masked from 'package:stats':
#>
#> kernel
#> The following object is masked from 'package:graphics':
#>
#> grid
#> The following objects are masked from 'package:base':
#>
#> rbind, xorTo begin, you need to tell fasterRaster the full file path of the folder where GRASS is installed on your system. Where this is well depend on your operating system and the version of GRASS installed. Three examples below show you what this might look like, but you may need to change the file path to match your case:
grassDir <- "C:/Program Files/GRASS GIS 8.3" # Windows
grassDir <- "/Applications/GRASS-8.3.app/Contents/Resources" # Mac OS
grassDir <- "/usr/local/grass" # LinuxTo tell fasterRaster where GRASS is
installed, use the faster() function:
faster(grassDir = grassDir)You can also use the faster()
function to set options that affect how fasterRaster
functions run. This includes setting the amount of maximum memory and
number of computer cores allocated to operations.
Importing spatial objects into fasterRaster
GRasters and GVectors
In fasterRaster, rasters are called
GRasters and vectors are called GVectors. The
easiest (but not always fastest) way to start using a
GRaster or GVector is to convert it from one
already in R. In the example below, we use a raster
that comes with the fasterRaster package. The raster
represents elevation of a portion of eastern Madagascar. We first load
the SpatRaster using fastData(),
a helper function for loading example data objects that come with the
fasterRaster package.
madElev <- fastData("madElev") # example SpatRaster
madElev
class : SpatRaster
dimensions : 1024, 626, 1 (nrow, ncol, nlyr)
resolution : 59.85157, 59.85157 (x, y)
extent : 731581.6, 769048.6, 1024437, 1085725 (xmin, xmax, ymin, ymax)
coord. ref. : Tananarive (Paris) / Laborde Grid
source : madElev.tif
name : madElev
min value : 1
max value : 570Now, we do the conversion to a GRaster and a
GVector using fast().
This function can create a GRaster or GVector
from a SpatRaster or a file representing a raster.
elev <- fast(madElev)
elev
class : GRaster
topology : 2D
dimensions : 1024, 626, NA, 1 (nrow, ncol, ndepth, nlyr)
resolution : 59.85157, 59.85157, NA (x, y, z)
extent : 731581.552, 769048.635, 1024437.272, 1085725.279 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
name(s) : madElev
datatype : integer
min. value : 1
max. value : 570Converting rasters and vectors that are already in R
to GRasters usually takes more time than loading them
directly from disk. To load from disk, simply replace the first argument
in fast() with a string representing the folder path and
file name of the raster you want to load into the session. For example,
you can do:
rastFile <- system.file("extdata", "madElev.tif", package = "fasterRaster")
elev2 <- fast(rastFile)Now, let’s create a GVector. The fast()
function can take a SpatVector from the
terra package, an sf object from the
sf package, or a string representing the file path and
file name of a vector file (e.g., a GeoPackage file or a shapefile).
madRivers <- fastData("madRivers") # sf vector
madRivers
Simple feature collection with 11 features and 5 fields
Geometry type: LINESTRING
Dimension: XY
Bounding box: xmin: 731627.1 ymin: 1024541 xmax: 762990.1 ymax: 1085580
Projected CRS: Tananarive (Paris) / Laborde Grid
First 10 features:
F_CODE_DES HYC_DESCRI NAM ISO NAME_0 geometry
1180 River/Stream Perennial/Permanent MANANARA MDG Madagascar LINESTRING (739818.2 108005...
1185 River/Stream Perennial/Permanent MANANARA MDG Madagascar LINESTRING (739818.2 108005...
1197 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (747857.8 108558...
1216 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (739818.2 108005...
1248 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (762990.1 105737...
1256 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (742334.2 106858...
1257 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (731803.7 105391...
1264 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (755911.6 104957...
1300 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (731871 1044531,...
1312 River/Stream Perennial/Permanent UNK MDG Madagascar LINESTRING (750186.1 103441...rivers <- fast(madRivers)
rivers
class : GVector
geometry : 2D lines
dimensions : 11, 11, 5 (geometries, sub-geometries, columns)
extent : 731627.0998, 762990.1321, 1024541.23477, 1085580.45359 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
names : F_CODE_DES HYC_DESCRI NAM ISO NAME_0
type : <chr> <chr> <chr> <chr> <chr>
values : River/Stream Perennial/Perm~ MANANARA MDG Madagascar
River/Stream Perennial/Perm~ MANANARA MDG Madagascar
River/Stream Perennial/Perm~ UNK MDG Madagascar
(and 8 more rows) Operations on GRasters and GVectors
You can do operations on GRasters and
GVectors as if they were SpatRasters,
SpatVectors, and sf objects. For example, you
plot them as if the were any other spatial object:
You can use mathematical operators and functions:
elev_feet <- elev * 3.28084
elev_feet
class : GRaster
topology : 2D
dimensions : 1024, 626, NA, 1 (nrow, ncol, ndepth, nlyr)
resolution : 59.85157, 59.85157, NA (x, y, z)
extent : 731581.552, 769048.635, 1024437.272, 1085725.279 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
name(s) : layer
datatype : double
min. value : 3.2808
max. value : 1870.056log10_elev <- log10(elev)
log10_elev
class : GRaster
topology : 2D
dimensions : 1024, 626, NA, 1 (nrow, ncol, ndepth, nlyr)
resolution : 59.85157, 59.85157, NA (x, y, z)
extent : 731581.552, 769048.635, 1024437.272, 1085725.279 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
name(s) : log
datatype : double
min. value : 0
max. value : 2.75587485567249You can also use the many fasterRaster functions. In general, these functions have the same names as their terra counterparts and often the same arguments. Note that even many terra and fasterRaster functions have the same name, they do not necessarily produce the exact same output. Much care has been taken to ensure they do, but sometimes there are multiple ways to do the same task, so choices made by the authors of terra and GRASS can lead to differences.
The following code 1) creates a raster where cell values reflect the distance between them and the nearest river; b) makes a buffer around the rivers; then c) plots the output:
dist <- distance(elev, rivers)
dist
class : GRaster
topology : 2D
dimensions : 1024, 626, NA, 1 (nrow, ncol, ndepth, nlyr)
resolution : 59.85157, 59.85157, NA (x, y, z)
extent : 731581.552, 769048.635, 1024437.272, 1085725.279 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
name(s) : distance
datatype : double
min. value : 0
max. value : 21310.9411762729
river_buff <- buffer(rivers, 2000)
river_buff
class : GVector
geometry : 2D polygons
dimensions : 1, 5, 0 (geometries, sub-geometries, columns)
extent : 729629.19151, 764989.97343, 1022544.92079, 1087580.24979 (xmin, xmax, ymin, ymax)
coord ref. : Tananarive (Paris) / Laborde Grid
plot(dist)
plot(rivers, col = 'lightblue', add = TRUE)
plot(river_buff, border = 'white', add = TRUE)
And that’s how you get started! Now that you have a raster and a
vector in your fasterRaster “location”, you can start
doing manipulations and analyses using any of the
fasterRaster functions! To see an annotated list of
these functions, use ?fasterRaster.
Converting and saving GRasters and
GVectors
You can convert a GRaster to a SpatRaster
raster using rast():
terra_elev <- rast(elev)To convert a GVector to the terra
package’s SpatVector format or to an sf
vector, use vect()
or st_as_sf():
Finally, you can use writeRaster()
and writeVector()
to save GRasters and GVectors directly to
disk. This will always be faster than using rast(),
vect(), or st_as_sf() then saving the result
from those functions.
elev_temp_file <- tempfile(fileext = ".tif") # save as GeoTIFF
writeRaster(elev, elev_temp_file)
vect_temp_file <- tempfile(fileext = ".shp") # save as shapefile
writeVector(rivers, vect_temp_file)Known issues
Comparability between terra and fasterRaster: As much as possible, fasterRaster functions were written to recreate the output that functions in terra produce. However, owing to implementation choices made by the respective developers of terra and GRASS, outputs are not always the same.
fasterRaster can crash when the temporary folder is cleaned: Some operating systems have automated procedures that clean out the system’s temporary folders when they get too large. This can remove files GRASS is using and fasterRaster is pointing to, rendering them broken. In Windows, this setting can be changed by going to
Settings, thenStorage, thenStorage Sense. Turn off the setting “Keep Windows running smoothly by automatically cleaning up temporary system and app files”.Disk space fills up: As counter to the previous issue, prolonged use of fasterRaster by the same R process can create a lot of temporary files in the GRASS cache that fills your hard drive. fasterRaster does its best to remove these files when they are not needed, so long as the setting
faster('clean')isTRUE(which it is by default). However, temporary files can still accumulate. For example, the operationnew_raster <- 2 * old_raster^3creates a raster file with the^3operation, which is then multipled by 2 to get the desired output. But the raster from the^3operation is still left in the disk cache, even though it does not have a “name” in R. Again, judicious use of the [mow()] function can remove these temporary files.
~ FINIS ~