New tidyvisor, the package's adviser. tidyvisor() given records and a
question (species distribution, spread, epidemic, range shift, spatial
autocorrelation) checks the data and recommends the analysis, the plots
and the code (tmr_advise()); given a result (growth rate, spread rate,
epidemic or range-shift simulation, species distribution model, Moran's I)
it explains the numbers and the figure in plain sentences
(tmr_interpret()).
plot_sites(), plot_topo() and plot_geo() no longer colour points on
a continuous colour bar: numeric values are shown as ordered classes
with legend keys. In plot_geo() a numeric value is drawn as
proportional symbols (area scaled to the value) by default.
geo_shift_simulate() now normalises the dispersal kernel to sum to 1
over the neighbourhood. Before, propagule pressure grew with the number
of cells within dispersal distance, so on fine grids almost any empty
cell next to the range was colonised, whatever its suitability, and
results depended on the cell size.
geo_shift_simulate(): dispersal_mean is now the true mean
dispersal distance in two dimensions. The exponential kernel used
exp(-d / mean), whose mean distance is twice mean; it now uses
exp(-2 d / mean). Simulated spread for a given dispersal_mean is
therefore slower than before.
geo_shift_simulate() gains a fat-tailed 2Dt kernel (kernel = "2Dt",
kernel_shape), movement resistance (barriers and corridors),
human-assisted long-distance jumps (jump_rate, jump_weights) and
disturbance (clearing, fire or eradication), for invasive species and
land-use change as well as climate.
New geo_dispersal_kernel(): the realised mean distance and
long-distance tail of a kernel on a grid.
New geo_epidemic_simulate(): a stochastic, spatially explicit SIR
model of disease spread among host populations, with the same kernels
and resistance; its final size is checked against theory.
New epi_growth_rate(): epidemic growth rate, doubling time and
reproduction number (Wallinga and Lipsitch 2007) from case counts or a
line list, with a Poisson or negative binomial GLM; intervals and the
fitted incidence band come from the model on the link scale. Checked on
simulated epidemics and the 2014 Ebola epidemic in Sierra Leone.
New geo_spread_rate(): spread rate of an invasion from dated records,
by first records per cell, the invasion front or occupied area, with
confidence intervals and fitted bands from the regression.
geo_shift_ensemble() now measures the unlimited-dispersal range as
all habitat at or above the starting suitability threshold, so it is
comparable with the simulated (limited) range. Before, it summed the
suitability values themselves.
geo_shift_ensemble() gains rescale (default TRUE), which divides
suitability by its present maximum before the simulation, so that
probabilities of presence from a niche model are not taken as
per-generation survival.
predict() for sdm_integrated() gains se_fit and level: the
standard error of the linear predictor (delta method from the fitted
covariance) and a confidence interval formed on the link scale and
back-transformed, so it stays within the bounds of the response.
Checked against glm() with a cloglog link.
sdm_integrated() now fits survey data without presence-only records
(it used to fail), and gives the same fit as a cloglog glm().
New user guide chapter, "Invasions and epidemics": epidemic growth and R (Ebola, Sierra Leone 2014), the spread of the cane toad front, dispersal kernels, resistance and spatial epidemics on Mt Karimui.
New user guide chapter, "Range shifts under climate change", and a case study hindcasting the upslope shifts of New Guinea birds.
New geo_shift_simulate() and geo_shift_summary(): stochastic
colonisation-extinction simulation of range shifts with a negative
exponential dispersal kernel, written in C++ and run in parallel, over a
time series of habitat suitability. It shows dispersal lags, extinction
debt and lost ranges that niche models alone miss. This is the first stage
of a hybrid range-shift model.
New geo_chelsa_area() and geo_chelsa_urls(): CHELSA V2.1 baseline and
CHELSA-CMIP6 projections (five GCMs, three SSPs, three periods) read only
over the area needed, as an ensemble.
New geo_downscale_lapse(): lapse-rate downscaling of coarse temperature
to the resolution of a fine DEM.
New geo_canopy_offset(): forest-canopy buffering of temperature, scaled
by canopy cover.
New sdm_integrated(): an integrated species distribution model that
fits presence-only records (a Poisson point process with bias
covariates) and presence-absence surveys (complementary log-log) with one
shared niche, with optional unimodal (hump-shaped) responses and
physiological limits such as CTmax, for plausible projections to future
climates.
New geo_climate_scenarios() and geo_shift_ensemble(): ensembles of
climates (GCM x SSP x lapse rate, delta method on a downscaled baseline)
and of range-shift projections with limited and unlimited dispersal;
summary() gives the median change and the probability of losing more
than 30, 50 or 80% of the range.
geo_chelsa() is deprecated in favour of geo_chelsa_area(), which
reads only the area needed. geo_bioclim(source = "chelsa") is unchanged.
geo_bioclim(source = "chelsa", scenario = , year = ) now reads CHELSA-CMIP6 itself (the
30-year period containing year, for the chosen GCM). It used to fall
back to WorldClim CMIP6 at 5 arc-minutes through geodata, mislabelled as
CHELSA. For regions and GCM ensembles, geo_chelsa_area() is recommended.
geo_hypsometry(): land area in each elevation band of a DEM,
inside an optional region, with true cell areas from terra::cellSize()
(area = "ellipsoid").geo_range_shift(): moves species' elevational ranges upslope by
warming / lapse_rate; reports the area now and after warming, the
percentage change, range-shift gaps and summit truncation, with optional
padding of ranges and fixed lower limits for lowland species.geo_patches() and geo_largest_patch(): connected patches of a
mask, with their areas.geo_landcover() and geo_worldcover_tiles(): ESA WorldCover 10 m
land cover read from its cloud-optimised GeoTIFFs, optionally as a
logical layer of chosen classes on the grid of a template raster.geo_elevation() mosaics every tile that a box touches. It downloads
SRTM 1 arc-second tiles first (new geo_srtm_tile_names()), and path
accepts a folder or several .tif, .hgt, .hgt.gz or .zip files,
with a warning when their CRS or resolution differ. Earlier versions kept
only the first tile.geo_raster_crop(), geo_raster_mask() and geo_raster_zonal() accept
a SpatExtent, SpatRaster, SpatVector, sf, sfc, bbox or four
numbers (c(xmin, ymin, xmax, ymax); named in any order), with a clear
error for an invalid box.geo_raster_resample() gains origin and align_to: cells now align to
whole multiples of the resolution, so results match terra::aggregate().geo_raster_zonal() accepts a zone raster and gives area-weighted means
and sums (weight = "area").map_polygons(), map_lines() and
map_points() are the name of data, or the kind of layer when data
is an expression, never a printed object such as <tmr_g_sf[,11]>;
label = NULL leaves a layer out of the legend.SystemRequirements lists GDAL, GEOS and PROJ."centroids" test of geo_clean() was documented but not
implemented: it ran the capitals test instead, so records georeferenced
to the centre of a country passed. It now flags records within
centroids_rad (default 1 km, as CoordinateCleaner) of a country
centroid, and is on by default.geo_clean() on data cleaned before.geo_mde(): a mid-domain null model of species richness over any
bounded domain (island, country, mountain range). Ranges built from
occurrence records keep their size and are placed at random with the
spreading-dye algorithm (Jetz and Rahbek 2001) on an equal-area grid
that follows the real coastline. Returns observed and expected richness,
the simulation interval and a standardised effect size per cell. The
placement kernel is in C++; on a one-cell-wide domain it reproduces the
one-dimensional model of rangemodelR.geo_mde_3d(): the same model in three dimensions for the sea (or
any volume): the water column above a bathymetry raster is divided into
depth bands and ranges grow in six directions, following the coast and
the sea floor. Returns maps per depth band and the vertical richness
profile against the null model.range = "buffer", for patchy habitat such as reefs) or occupied cells.geo_dbscan() gave wrong clusters on Windows: the grid used to find
neighbours built its keys with a 32-bit integer there, so some
neighbours were missed and points were wrongly marked as noise. Keys
are now 64-bit on every platform, and results equal the dbscan package
on Windows too. Re-run geo_dbscan() analyses made on Windows.An audit of every function that takes coordinates or colours found two classes of silent error. The statistics themselves were correct (they match spdep, gstat, dbscan and landscapemetrics); the problem was how input reached them.
Coordinates. Coordinate columns of a plain data frame were always
read as longitude and latitude in degrees. Projected coordinates
(metres, e.g. UTM) given that way were silently treated as degrees,
giving wrong distances and therefore wrong results in geo_moran(),
geo_hotspot(), geo_lisa(), geo_dbscan(), geo_mem(),
geo_distance(), geo_decay(), geo_rarefy_spatial(), geo_buffer(),
geo_centroid() and others. Now:
sf table with
their CRS (exceptions: geo_as_sf() and as_spat_raster() with a
projected crs, and method = "euclidean" in geo_distance(),
geo_mem() and geo_decay(), which are meant for planar
coordinates);geo_moran(), geo_mem(), geo_distance(), geo_decay() and
geo_rarefy_spatial() now accept sf tables, like the other
functions;sf
table and a projected sf table.Results from sf input and from genuine longitude/latitude columns were
always correct and do not change. Analyses that passed projected
coordinates as plain columns should be re-run.
True areas, distances and buffers for longitude/latitude data.
geo_area(), geo_distance_to() and geo_buffer() measured longitude/
latitude data in a global equirectangular projection, which stretches
east-west lengths by 1/cos(latitude): areas and distances were too large
(twice the truth at 60 degrees) and buffers too narrow east-west (a
100 m buffer at 51 degrees was 63 m wide). Areas and distances are now
measured on the sphere by sf, and each buffer is built in an azimuthal
equidistant projection centred on its feature. New tests check the
results against true values. Projected data were never affected.
Results for longitude/latitude data from these three functions should be
recomputed.
The test suite now runs in R CMD check and coverage.
tests/testthat.R was missing, so R CMD check and covr ran none of the
tests (they ran only through devtools::test()); all 663 pass under
R CMD check, coverage is 77 %, and CI now fails below 75 %.
geo_mem() now computes dbMEM as published (Dray, Legendre and
Peres-Neto 2006), reproducing adespatial::dbmem() exactly (same
eigenvectors, eigenvalues and number of MEMs). Before, it computed the
older PCNM eigenvectors (principal coordinates of the truncated distance
matrix): it returned eigenvectors with negative spatial
autocorrelation as well (86 instead of 48 for the Meuse samples), scaled
by the square root of their eigenvalue instead of to variance 1, and
kept numerically zero eigenvectors when distances were in metres.
Models that used geo_mem() axes as predictors should be re-run.
geo_moran() documentation: the variance is the variance under
randomisation (equal to spdep::moran.test(randomisation = TRUE)), not
under normality as stated; the values were always correct.
Reference tests strengthened so they compare full results, and their reference packages (adespatial, dbscan, glmnet, landscapemetrics) added to Suggests so CI runs them: Moran's I and its variance equal spdep's to 1e-10; DBSCAN core points, noise and core partition equal the dbscan package's (border points reachable from two clusters may legitimately join either).
Colours. A named palette for categories (e.g.
c(hot = "red", cold = "blue")) was applied in alphabetical order of
the categories instead of by name in map_points(), map_lines(),
map_polygons(), geo_webmap(), geo_automap(), plot_sites() and
plot_topo(). All now colour each category by its name; a named
palette that misses a category is an error, and factor columns keep
their level order.
geo_hotspot() and geo_lisa() accept sf tables directly.map_raster() gains limits, so rasters in several panels can share one
colour scale and one legend.geo_panels(guides = "keep") no longer overrides each map's own legend
position (including legend = "none").geo_automap() are centred on Greenwich again, features
crossing the date line (Russia, Fiji) are split there instead of being
stretched across the map, and the frame shows the whole globe.inst/extdata/wilhelm-dem.tif), used in the 3D examples.geo_webmap() turns a geo_map() (or one sf or raster layer) into a
self-contained interactive web page built on Leaflet: pan and zoom, four
basemaps, layers that switch on and off, attribute pop-ups, permanent
labels, a scale bar and a legend with the same symbols, classes and presets
as the printed map. Features are embedded as GeoJSON and rasters as images,
so the single HTML file can be emailed or published; Leaflet is loaded with
Subresource Integrity.map_polygons(), map_lines() and map_points() take preset, a
standard cartographic symbol by name (water, forest, protected area,
administrative unit, river, main road, railway, boundary, contour, town,
peak, sample site and more); arguments you set yourself override it.
map_presets() lists them. Polygons gain linetype for dashed outlines.geo_dbscan() clusters points with DBSCAN in C++ (grid index), with
eps in metres on the ground for geographic data and a convex-hull
outline per cluster. It gives the same partition as the dbscan package.geo_landscape() computes FRAGSTATS fragmentation metrics by class (np,
ca, pland, lpi, area_mn, ed, ai, mesh) and for the landscape (with
Shannon's and Simpson's diversity). They match landscapemetrics on
projected rasters to machine precision, and unlike landscapemetrics they
measure geographic (degree) rasters in true hectares and metres.geo_map3d() renders a DEM as a 3D block diagram with sun lighting and
cast shadows, any viewing direction and angle, automatic vertical
exaggeration, hypsometric tints or any raster draped on the surface, and
sites or routes placed on the terrain and hidden behind hills. The
renderer is written in C++ (no OpenGL, rgl or rayshader), so it runs on
servers and in reports and returns an ordinary ggplot. Empty wedges left
by reprojection are trimmed so the block has clean sides.sdm_maxent() fits Maxent as maxnet does (infinitely weighted logistic
regression with the lasso, the same feature classes and default
regularisation), with the lasso path solved in C++ (coordinate descent
with strong rules and active sets, on several cores): no Java and no
glmnet. It agrees with maxnet to a correlation of 0.999 or more on the
Bradypus data (Phillips et al. 2006). It fixes three weaknesses of
maxnet:
standardize = FALSE reproduces maxnet;sdm_ensemble() combines Maxent, GLM, GAM and a C++ random forest in the
spirit of biomod2: spatial block cross-validation with folds balanced for
presences and background, AUC, TSS and the continuous Boyce index,
weights from cross-validated TSS above min_tss, balanced weighting of
presences and background for every algorithm, and predictions of the
weighted mean and the weighted disagreement (SD) between algorithms. The
C++ random forest matches randomForest's accuracy and is about 18 times
faster.
geo_map(crs = "laea") and other named projections now centre on raster
layers too.
geo_stac_search() finds scenes in any STAC catalogue (default: Element
84 Earth Search: Sentinel-2, Landsat, Copernicus DEM) by area, period and
cloud cover, and returns their footprints as an sf table.
geo_stac_read() reads chosen bands over an area only (windowed reads of
cloud-optimised GeoTIFFs), applies the catalogue's scale and offset
(skipping it where Earth Search has already applied it), mosaics tiles and
takes the per-pixel median of several scenes to remove clouds.
geo_stac_collections() lists a catalogue's collections. Needs the
suggested package httr2.
geo_spectral_index() recognises STAC band names (red, nir, ...) and
returns NA where a normalised difference has no signal.
map_curved_labels() writes labels that follow the shape of any
feature: along lines (road and river names), along the long axis or the
outline of polygons, or on an arc above points. Text stays upright, jagged
paths are smoothed, and labels that do not fit are left out.
map_line_labels() is the same for lines. Needs the suggested package
geomtextpath.map_contours() draws contour lines of a raster, with every second
contour thicker and labelled with its height at its straightest stretch.map_zoom() zooms a geo_map() in or out (zoom = 2 is twice as
close), on the centre or on a centre = c(lon, lat), or shows an exact
extent; the scale bar follows the zoom.geo_panels() arranges maps from geo_map(), geo_automap() or any
ggplot in a grid, tagged (a), (b), ..., with matching legends collected
into one and a figure title and caption.geo_facet_map() draws small multiples (by a grouping column, or one
panel per value column) with one classification shared by all panels,
one legend, and the scale bar and north arrow in the first panel only.geo_inset() places a locator map on a larger map. By default
(location = "auto") it takes a corner free of the scale bar, north arrow
and any legend inside the map, and of those the one hiding the least of
the mapped features; a corner name or any c(x, y) position also works.geo_map(legend = "inside") puts the legend in a framed box on the map,
in a corner chosen the same way (or set with legend_corner).
map_legend(position = "inside") does the same for any map.geo_map() and add map_raster() (gradient legend, optional hillshade),
map_polygons(), map_lines(), map_points() and map_labels() (labels
with a white halo). Each layer gets its own legend section with the right
symbol for its geometry, listed top layer first; numbers are classed with
geo_classify(). map_build() returns an ordinary ggplot. Needs the
suggested package ggnewscale.geo_recipe_read(), geo_recipe_write(), geo_recipe_validate() and
geo_recipe_run() read, write, check and run GeoSpaX recipes (version 1 and
2 YAML), so a workflow built in GeoSpaX runs in R and the other way round.
geo_recipe(), recipe_input() and recipe_step() build recipes in R.io.open, interpolate.idw, interpolate.auto (kriging
against IDW by leave-one-out error, with a standard-error map),
aggregate.hexbin, stats.hotspots (Getis-Ord Gi* with FDR-corrected
90/95/99% classes), automap.style, style.apply, layout.auto,
terrain.derive, rs.index, export.map/png/svg/pdf, export.data
and export.report (an HTML report with the maps, decisions and run log).
Steps that only GeoSpaX can run are reported by geo_recipe_validate().inst/recipes: interpolate-samples,
points-to-hotspots and points-to-hexbins.Install the latest version with pak::pak("jmozez/tidymapr").
plot_choropleth() shades polygons by a numeric column, grouped into
classes with geo_classify() (quantile, equal, Fisher-Jenks, pretty,
standard deviation or fixed breaks). It uses colour-blind-safe sequential
palettes from grDevices::hcl.colors() (default "YlGnBu", light for
low values and dark for high). The breaks, counts per class and goodness
of variance fit are kept with the plot.plot_bubble() draws proportional-area circles for counts such as
species richness or abundance, from an sf point table or a lon/lat data
frame, optionally coloured by group and over a background layer. Sites
with a value of zero (sampled, nothing found) are shown as small open
circles, so absences stay on the map. The example maps ant species on the
16 Mt Wilhelm transects, where no ants were found above 2,200 m.geo_basemap(), plot_sites(), plot_geo(), plot_topo() and
plot_sdm() gain new basemaps, all free and needing no API key:
"topo" (OpenTopoMap, with contours and
hillshade), "osm_hot" (OpenStreetMap Humanitarian), "carto_voyager",
"esri_street", "esri_relief" (shaded relief), "esri_terrain" and
"esri_natgeo" (National Geographic), alongside the existing
"esri", "esri_topo", "osm", "carto" and "carto_dark";"hillshade", "contour" (labelled
contour lines) and "terrain" (both). Pass your own DEM with dem
(works offline), or let geo_elevation() download one for the map
extent; set the contour interval with contour_breaks.coord_sf(). You pick the style:
annotation_scale_bar(): "pill", "checker", "line" or "ruler",
in "metric", "imperial" or "both" units. Its length is measured on
the ground when the map is drawn, so it is right on longitude/latitude
and projected maps and in every facet.annotation_north_arrow(): "needle", "compass", "rose" or
"minimal".map_legend(): legend "right", "bottom", "inside" a chosen
corner or "none", with its title, renamed entries (labels), hidden
legends (hide), frame, text, title and key sizes, columns, direction
and order.plot_sites(), plot_geo() and plot_sdm() accept these style names in
scale_bar and north_arrow (TRUE keeps the default, FALSE turns
them off), and plot_sites() and plot_geo() gain scale_units. Their
maps keep the full planned extent around the sites.geo_spectral_diversity() maps diversity from satellite or drone
imagery in Hill numbers (q = 0, 1, 2), on the same scale as species
diversity: bands are reduced by principal components, pixels grouped
into spectral species by k-means (Féret and Asner 2014), and Hill
numbers computed in a moving window in C++ on several cores. Results
equal direct Hill-number calculations to machine precision.geo_spectral_index() computes NDVI, GNDVI, NDRE, NDWI, MNDWI,
NDMI, NBR and NDBI (Sentinel-2 bands recognised automatically), with an
optional Otsu threshold that reports the area above the cut, as in
GeoSpaX's Forestry plug-in.geo_automap() draws a finished map from one call and explains every
choice; geo_automap_explain() returns the decisions as a table for a
methods section. Following GeoSpaX's AutoMap and ArcGIS/QGIS practice:
background such as a DEM
under the sites) gets a stretched ramp with a gradient legend, while
the sites keep their own symbol legend;geo_automap() takes palette and background_palette: "auto", any
grDevices::hcl.pals() name (for example "viridis", "Blues 3",
"Terrain 2") or your own colours. Every map is also a ggplot, so any
scale_fill_*() or scale_colour_*() can replace a ramp with +.geo_diversity_surface() turns a community table into continuous
maps of Hill-number diversity (q = 0, 1, 2) with 95% intervals:
tidydiversity::div_standardise() (Chao and Jost 2012), keeping
its bootstrap standard error;vegan::BCI).plot_diversity_surface() draws the surfaces with a matching row of
uncertainty maps and the sampled sites."Err" variogram model.options(tidymapr.threads = n); by
default tidymapr uses all cores but one. Timings on 4 cores:
geo_lisa() permutation test (500 sites, 999 permutations): 5 s to
0.05 s;geo_interpolate() (300 points to 40,000 cells): 20 s to
0.9 s; new variance = FALSE skips the kriging variance for even
larger grids;geo_hotspot() and geo_lisa(): about
4 times faster;geo_mem() minimum spanning tree (Prim's algorithm): about 10 times
faster and now scales to thousands of sites.coord_projection() draws any tidymapr or ggplot map in a named
projection, added with +: "equal_earth" (true areas, recommended for
world choropleths), "mercator" (World Mercator), "web_mercator"
(matches online tiles), "robinson", "mollweide", "winkel_tripel",
"lonlat", "utm" (zone picked from your data), and "laea",
"albers" and "orthographic" centred on your data. Any EPSG code or
PROJ string also works, and world projections take lon0, for example
lon0 = 150 for a Pacific-centred map.geo_crs() returns the CRS for those names, for reprojecting data.map_graticule() sets the latitude/longitude grid interval and line
style, labels latitudes, longitudes, both or neither (on curved world
projections it labels latitudes only, so longitude labels do not crowd
the pole line), and can draw the edge of the world or, on orthographic
maps, the horizon circle.spData::world (Natural Earth and World Bank, free to use)
for world maps.geo_read() now reads many more formats through GDAL:
.nc, .nc4, .cdf),
HDF4/HDF5 (MODIS, VIIRS, GEDI), GRIB weather files, Sentinel-2 .SAFE
folders and zips, JPEG 2000, ENVI cubes (a .dat/.bin with its
.hdr), ERDAS Imagine, Zarr, VRT, IDRISI, SAGA and ER Mapper;.dxf and .dwg and MicroStation .dgn, with
cad_layer to keep chosen AutoCAD layers (DGN levels) and crs to set
the coordinate system CAD files do not store;.osm/.pbf, S-57 nautical charts and spreadsheets
(.xlsx, .xls, .ods).var argument to pick netCDF, HDF and GRIB variables by name or
number; several variables are combined into one raster, and time steps
stay in terra::time().geo_layers() shows what a file holds before you read it: vector
layers, netCDF/HDF variables and subdatasets, CAD layers or raster bands.geo_hotspot() (Getis-Ord Gi*) and
geo_lisa() (local Moran's I with cluster types, analytical or
permutation p-values) on k-nearest or distance-band neighbours. Results
equal spdep::localG() and spdep::localmoran().geo_interpolate() predicts a numeric column at raster cells or
points by inverse-distance weighting or ordinary kriging (exponential,
spherical or Gaussian variogram, fitted with gstat when not given).
Results equal gstat::idw() and gstat::krige() to numerical precision.geo_flow_direction() (D8),
geo_flow_accumulation(), geo_watershed() (with outlet snapping) and
geo_twi() (topographic wetness index). Accumulation and watersheds equal
terra::flowAccumulation() and terra::watershed() for the same flow
directions.png_ants_sites
holds the published bait GPS of the 16 Mt Wilhelm ant transects (Mottl,
Moses and Klimes 2023, doi:10.5281/zenodo.8191946), and
geo_example_sites() returns their band centroids. The invented Wanang
coordinates and the simulated Wanang stem case study are gone.inst/extdata, WorldClim 2.1), rendered by
data-raw/gallery_maps.R. Basemap tiles are drawn in the right
projection and their zoom follows the map extent.geo_as_sf(), geo_drop_geometry(),
geo_extract(), geo_clean(), geo_gbif(), as_sf(),
as_spat_raster(), geo_as_features().geo_distance(), geo_decay()
(pairs with tidydiversity::beta_decay_model()), geo_moran(),
geo_mem(), geo_rarefy_spatial(), geo_grain(), geo_middomain().plot_sites(), plot_geo(), plot_topo(), geo_basemap(),
geo_save() and theme_tidymapr().geo_read(), geo_write(),
geo_vector_*(), geo_raster_*(), geo_buffer(), geo_crop(),
geo_union() and friends.geo_bioclim(), geo_worldclim(),
geo_chelsa(), geo_elevation(), geo_gadm(), cached under
tools::R_user_dir("tidymapr", "cache").stems_quadrat(), stems_habitat(),
stems_neighbours() and the torus-translation stems_habitat_test();
census summaries stay in tidydiversity.sdm_fit(), sdm_predict(),
sdm_evaluate(), plot_sdm().geo_example_sites(), geo_example_stems(),
geo_example_dem(), geo_example_censuses().