This package is published on CRAN. Install it with:
I have tried to keep dependencies to a minimum. The package requires
sf
and terra
for loading and merging
geospatial data. If these are not already installed on your machine, the
install_packages
line will ask to install them. We also use
the bcmaps
package in this vignette to define a study
region (but it’s not a requirement of the package).
library(sf)
#> Linking to GEOS 3.9.1, GDAL 3.2.1, PROJ 7.2.1
library(terra)
#> terra version 1.4.22
library(bcmaps)
#> The bcmapsdata package is no longer required to be installed for bcmaps to function.
#> Layers are now cached as needed to 'C:\Users\deank\AppData\Local/R/cache/R/bcmaps' using the bcdata package.
library(rasterbc)
rasterbc
is a data-retrieval tool. Start by setting a
storage directory for the raster layers
datadir_bc
will set the data directory to the path
provided in its first argument. Setting the first argument to
NA
(or failing to call datadir_bc
before
getdata_bc
) will cause the package to select a temporary
directory (base::tempdir
) that is emptied after your R
session ends. We recommended to pick a more permanent location, as in
the example above, so that files only need to be downloaded once.
When quiet=FALSE
(the default), the function will ask
users to confirm that rasterbc
should write files to the
supplied directory, and warn if this directory contains any existing
files/folders. Note that if the storage directory has existing files
with names matching those fetched by the rasterbc
package,
those data will be overwritten by calls of the form
getdata_bc(..., force.dl=TRUE)
(the default is
force.dl=FALSE
).
In future sessions, users can set quiet=TRUE
to skip the
interactive prompt and suppress warnings about existing data.
This path string is stored as an R option. View it using:
Depending on the geographical extent of interest and the number different layers requested, the storage demands can be high. For example, if every layer is downloaded, then around 30 GB of space is needed. Make sure you have selected a drive with enough free space for your project.
To demonstrate this package we’ll need a polygon covering a (relatively) small geographical extent in BC. Start by loading the bcmaps package and grabbing the polygons for the BC provincial boundary and the Central Okanagan Regional District
# define and load the geometry
example.name = 'Regional District of Central Okanagan'
bc.bound.sf = bc_bound()
#> bc_bound_hres was updated on 2021-11-16
districts.sf = regional_districts()
example.sf = districts.sf[districts.sf$ADMIN_AREA_NAME==example.name, ]
# plot against map of BC
blocks = findblocks_bc(type='sfc')
plot(st_geometry(blocks), main=example.name, border='red')
plot(st_geometry(bc.bound.sf), add=TRUE, col=adjustcolor('blue', alpha.f=0.2))
plot(st_geometry(example.sf), add=TRUE, col=adjustcolor('yellow', alpha.f=0.5))
text(st_coordinates(st_centroid(st_geometry(blocks))), labels=blocks$NTS_SNRC, cex=0.5)
The Okanagan polygon is shown in yellow, against a red grid that
partitions the geographic extent of the province into 89 smaller
regions, called mapsheets (here I am calling them “blocks”).
This is the NTS/SNRC
grid used by Natural Resources Canada for their topographic maps, with
each mapsheet identified by a unique
number-letter code. rasterbc
uses this grid to package
data into blocks for distribution. It is lazy-loaded as the
sf
object returned by
findblocks_bc(type='sfc')
, which we copied to variable
blocks
in the chunk above
print(blocks)
#> Simple feature collection with 89 features and 1 field
#> Geometry type: POLYGON
#> Dimension: XY
#> Bounding box: xmin: 199960.5 ymin: 331658 xmax: 1874986 ymax: 1745737
#> Projected CRS: NAD83 / BC Albers
#> First 10 features:
#> NTS_SNRC geometry
#> 1 092B POLYGON ((1299175 340112.5,...
#> 2 092C POLYGON ((1149647 333772, 1...
#> 3 092E POLYGON ((854708.4 444669.3...
#> 4 092F POLYGON ((1001221 442634.1,...
#> 5 092G POLYGON ((1147733 444738.3,...
#> 6 092H POLYGON ((1294127 450980.4,...
#> 7 083C POLYGON ((1548365 807589.3,...
#> 8 083D POLYGON ((1411944 794023.1,...
#> 9 083E POLYGON ((1402477 905139.5,...
#> 10 082E POLYGON ((1440286 461355.4,...
Let’s download Canada’s 1:250,000 digital elevation model (CDEM) layers corresponding to the yellow polygon. For the full BC extent, these rasters would occupy around 1.2GB of space. But we only want the smaller extent corresponding to the polygon. There are three blocks (totalling about 20 MB) which overlap with our region of interest
fetch them using the command:
You should see progress bars for a series of three downloads, and
once finished, the paths of the downloaded files are printed to the
console. Note that if a block has been downloaded already (eg.
by a getdata_bc
call with a different geo
argument), the existing copy will be detected, and the download skipped.
eg. repeat the call…
getdata_bc(geo=example.sf, collection='dem', varname='dem')
#> all 3 block(s) found in local data storage. Nothing to download
#> [1] "C:/rasterbc_data/dem/blocks/dem_092H.tif"
#> [2] "C:/rasterbc_data/dem/blocks/dem_082E.tif"
#> [3] "C:/rasterbc_data/dem/blocks/dem_082L.tif"
… and nothing is downloaded, because the data are there already.
Verify by loading one of the files as SpatRaster
:
tif.path = file.path(datadir_bc(), 'dem/blocks/dem_092H.tif')
#> current data storage path: C:/rasterbc_data
example.raster = terra::rast(tif.path)
print(example.raster)
#> class : SpatRaster
#> dimensions : 1212, 1525, 1 (nrow, ncol, nlyr)
#> resolution : 100, 100 (x, y)
#> extent : 1286588, 1439088, 450888, 572088 (xmin, xmax, ymin, ymax)
#> coord. ref. : +proj=aea +lat_0=45 +lon_0=-126 +lat_1=50 +lat_2=58.5 +x_0=1000000 +y_0=0 +datum=NAD83 +units=m +no_defs
#> source : dem_092H.tif
#> name : dem_092H
#> min value : 7.653875
#> max value : 2608.961
plot(example.raster, main='elevation (metres)')
To display the elevation data for the entire district, we need to
combine the three blocks downloaded earlier. This can be done using
opendata_bc
, which loads all required blocks, merges them
into a single layer, crops and masks as needed, and then loads into
memory the returned SpatRaster
object:
example.tif = opendata_bc(example.sf, collection='dem', varname='dem')
#> creating mosaic of 3 block(s)
#> clipping layer...masking layer...done
print(example.tif)
#> class : SpatRaster
#> dimensions : 622, 893, 1 (nrow, ncol, nlyr)
#> resolution : 100, 100 (x, y)
#> extent : 1426588, 1515888, 539188, 601388 (xmin, xmax, ymin, ymax)
#> coord. ref. : +proj=aea +lat_0=45 +lon_0=-126 +lat_1=50 +lat_2=58.5 +x_0=1000000 +y_0=0 +datum=NAD83 +units=m +no_defs
#> source : memory
#> name : dem
#> min value : 340.43
#> max value : 2153.29
plot(example.tif, main=paste(example.name, ': elevation'))
plot(st_geometry(example.sf), add=TRUE)
plot(st_geometry(blocks), add=TRUE, border='red')
If you’re opening the downloaded data right away, you can skip the
getdata_bc
call. The argument dl
(by default
TRUE
) in opendata_bc
will cause it to call
getdata_bc
automatically to download any missing
blocks.
Any simple features object of class sf
or
sfc
can be used for the argument geo
, provided
its geometry intersects with the provincial boundary of BC. The intended
usage is for the user to delineate their region of interest as a
(MULTI)POLYGON
object (here, example.sf
is a
MULTIPOLYGON
). Geometries of other classes (such
SpatialPolygons
, as defined by sp
; or data
frames containing coordinates of vertices) can often be coerced to
sf
using a command like
sf::st_as_sf(other_geometry_class_object)
. Alternatively,
users can directly download individual blocks by specifying their
NTS/SNRC codes, eg. the next chunk plots the DEM again, using
the codes:
example.codes = findblocks_bc(example.sf)
example.tif = opendata_bc(example.codes, collection='dem', varname='dem')
#> creating mosaic of 3 block(s)
#> done
plot(example.tif, main=paste('NTS/SNRC mapsheets ', paste(example.codes, collapse=', '), ': elevation'))
plot(st_geometry(blocks), add=TRUE, border='red')
plot(st_geometry(example.sf), add=TRUE)
text(st_coordinates(st_centroid(st_geometry(blocks))), labels=blocks$NTS_SNRC, cex=0.5)
If you forget which files have been downloaded, you can either check
the directory data.dir
using your file browser (subfolder
‘/dem/blocks’, in this case), or use listfiles_bc
to get a
logical vector indicating which files are currently found in your local
storage directory:
is.downloaded = listdata_bc(collection='dem', varname='dem', simple=TRUE)
paste('downloaded: ', sum(is.downloaded), '/', length(is.downloaded)) |> print()
#> [1] "downloaded: 3 / 89"
This shows that of the 89 blocks for the variable name ‘dem’ (in the
collection ‘dem’), we have downloaded three so far. Notice the return
value of listdata_bc
is a named vector, with names
indicating the destination filenames and paths. This shows where they
will be written by getdata_bc
. All filenames are either of
the form ‘varname_mapsheet.tif’ (as in this example) or else
varname_year_mapsheet.tif (for time-series data).
By default, the listdata_bc
function prints a list of
all available layers. eg. in the ‘dem’ collection we also have
‘aspect’ and ‘slope’:
listdata_bc(collection='dem', verbose=2)
#> year description unit tiles
#> dem NA digital elevation map (metres above sea level) 3/89
#> slope NA derived from digital elevation map (degrees above horizontal) 3/89
#> aspect NA derived from digital elevation map (degrees counterclockwise from north) 0/89
Notice the ‘dem’ blocks that were downloaded earlier. We merged these
blocks in the opendata_bc
function call that created
example.tif
. Currently, this layer resides in memory and
can be accessed via the R object example.tif
. To save a
copy, one can use the terra::writeRaster
function:
dem.path = file.path(getOption('rasterbc.data.dir'), 'dem', 'example_dem.tif')
terra::writeRaster(example.tif, dem.path, overwrite=TRUE)
getdata_bc
writes all of its data inside a ‘blocks’
subdirectory (in this case ‘/dem/blocks’), and the subfolder of the data
directory corresponding to the collection (in this case ‘/dem’) is, by
default, left empty. So it is a good place to store and organize such
derivative files, where they can be loaded more quickly (in future),
eg.
terra::rast(dem.path)
#> class : SpatRaster
#> dimensions : 2459, 2983, 1 (nrow, ncol, nlyr)
#> resolution : 100, 100 (x, y)
#> extent : 1286588, 1584888, 450888, 696788 (xmin, xmax, ymin, ymax)
#> coord. ref. : +proj=aea +lat_0=45 +lon_0=-126 +lat_1=50 +lat_2=58.5 +x_0=1000000 +y_0=0 +datum=NAD83 +units=m +no_defs
#> source : example_dem.tif
#> name : dem
#> min value : 7.653875
#> max value : 2918.951
If you’re finished with rasterbc
and want to remove all
of the stored data, or if you simply want to free up space, the entire
data directory or any of its contents can be deleted using your file
browser. This will not break the rasterbc
installation.
However, all downloaded data will be erased and you will need to run
datadir_bc
again before using the other package
functions.
Note that the bgcz
collection data are factors, which
are then encoded in the geotiff files as integer codes.
opendata_bc
returns a plot-ready categorical raster in this
case, with coded levels replaced by factor names. The complete lookup
tables are also stored in the lazy loaded list object
metadata_bc
.
lookup.list = rasterbc::metadata_bc$bgcz$metadata$coding
print(lookup.list$zone)
#> [1] "BAFA"
#> [2] "BG"
#> [3] "BWBS"
#> [4] "CDF"
#> [5] "CMA"
#> [6] "CWH"
#> [7] "ESSF"
#> [8] "ICH"
#> [9] "IDF"
#> [10] "IMA"
#> [11] "MH"
#> [12] "MS"
#> [13] "PP"
#> [14] "SBPS"
#> [15] "SBS"
#> [16] "SWB"
#> [17] NA
For example we have zone, 1 = Boreal Altai Fescue Alpine (BAFA), 2 = Bunchgrass, etc. See the documentation for the bgcz source script for links to a complete description of all codes.
The code below plots this data for the example region: