Species distributions

sdm.accessible_area#

The area the species could have reached, where its background is drawn (landschaft's accessible_area, the M of the BAM diagram): the land within a distance of the records (geodesic circles, merged), their convex hull or the regions that hold them (widened by the distance if given), cut at the antimeridian, within a mask.

Input	Type	What it is
occurrencesrequired	any	points, or a table with coordinate columns
regions	vector	polygons (ecoregions, biomes...), for regions
within	vector	polygons the area is cut to

Output	Type	What it is
area	vector	one (multi)polygon in lon/lat

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
method	text	buffer (default), convex_hull or regions
distance	number	the radius in metres (required for buffer)
resolution	integer	the vertices of each circle (default 64)

sdm.area_of_habitat#

The area of habitat of species (Brooks et al. 2019; Strassburg et al. 2020): over the cells of each species' range (within its elevation limits), the shares of the natural classes suitable to it times the cell's area, now and before conversion (the original vegetation); and the habitat it gains per unit of a cell restored. Rows in the order of the ranges' layers.

Input	Type	What it is
rangesrequired	stack	each species' range (cells above 0), by species
land_coverrequired	stack	the share of each class, by name
original	stack	the original share of each natural class
restore	stack	the shares restored land returns to
anthropic	map	the share of anthropic land (where restoring gains)
elevation	map	the elevation, for the species' limits

Output	Type	What it is
table	table	per species (in the ranges' order): its area of habitat now and potential (km²), its range's cells, and whether it counts (1: habitat now and potential, and suitable classes)
gain	stack	the habitat each species gains per unit of a cell restored
loss	stack	with loss: the habitat each species loses per unit of a cell's natural land converted

Setting	Type	What it is
naturalrequired	any	the natural classes (names of the land cover's layers)
suitability	any	each species' suitable classes: {species: [classes]} (the species not listed: every natural class)
include_unlisted	bool	species not in suitability count (default true)
elevation_limits	any	[[species, lowest, highest], ...] (null: none)
cell_area	number	a cell's area in km² (default: the grid's)
loss	bool	also the habitat lost per unit of natural land (conservation, conversion)

sdm.autocorrelation_range#

The range of spatial autocorrelation of the predictors (landschaft's autocorrelation_range, blockCV's cv_spatial_autocor): gstat's variograms (spherical, exponential, Gaussian and Matern) fitted to a sample of every layer, the best one's range; their median and the block size it gives for spatial blocks.

Input	Type	What it is
envrequired	any	the predictors: a stack or a map

Output	Type	What it is
range	number	the median range (metres on lon/lat grids)
block_size	number	the size of spatial blocks (degrees on lon/lat grids)
layers	table	per layer: range, sill, nugget, kappa
models	any	per layer, the variogram model

Setting	Type	What it is
n	integer	the cells sampled per layer (default 5000)
seed	integer	the seed of R's generator (default 1)
deg_to_metre	number	metres per degree on lon/lat (111325)

sdm.background#

Background points for a species distribution model (landschaft's background, dismo's randomPoints): cells with values in every predictor drawn with R's sampling (weighted by the cells' widths on lon/lat grids, or by a bias), within an area, a band of distances from the records or outside their environmental envelope, the records' cells left out.

Input	Type	What it is
envrequired	any	the predictors: a stack or a map
occurrences	any	the records (points or a table)
area	vector	the area to draw within
bias	map	a weight per cell

Output	Type	What it is
points	vector	the points at the cells' centres, with their cell

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
n	integer	the points (default 10000)
distance	any	the largest distance from the records, or [from, to]
envelope	number	outside the records' quantiles [q, 1 - q] of every predictor
exclude_occurrences	bool	leave the records' cells out (default true)
tryf	number	the candidates drawn per point (default 3)
lonlat_correction	bool	weight lon/lat cells by their width (default true)
seed	integer	the seed of R's generator (default 1)

sdm.bias_surface#

The sampling effort as a surface (landschaft's bias_surface): a Gaussian kernel density of the records of a whole group (the same taxa, collectors and decades) on the grid of the predictors, MASS::kde2d's at the cells' centres, with R's bw.nrd in each direction by default. The bias of sdm.background, which draws the background with the records' bias (a target-group background, Phillips et al. 2009; Maxent's bias file). kde2d(h = H) is bandwidth = H / 4.

Input	Type	What it is
occurrencesrequired	any	the records of the target group: points, or a table with coordinate columns
envrequired	any	the predictors: a stack or a map (its grid)

Output	Type	What it is
bias	map	the density (the share of the records per unit of area of the coordinate system: it sums to 1 over the plane) or rescaled
bandwidth	any	the bandwidths used, [x, y]

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
bandwidth	any	the kernel's standard deviation in the grid's units (metres, or degrees on lon/lat): a number, [x, y], or nrd (default: R's bw.nrd in each direction, kde2d's)
isotropic	bool	the larger bandwidth in both directions (default false)
limits	any	[min, max]: bandwidths raised or lowered to them (either null); a floor keeps a clustered group's surface from collapsing on its records
weights	text	a column of the records: each one's weight (specimens, visits); default 1
rescale	any	[lo, hi]: the densities mapped linearly to them, e.g. [0.05, 0.95], so that every cell keeps a chance
mask	bool	no value on the cells without a value in every predictor (default true)

sdm.boyce#

The continuous Boyce index (landschaft's boyce, ecospat's ecospat.boyce): the correlation between the classes of suitability and their ratio of predicted to expected presences, in a moving window (default) or in classes.

Input	Type	What it is
presencesrequired	any	the predictions at the presences
availablerequired	any	the predictions everywhere (or at the background and the presences)

Output	Type	What it is
cor	number	the index (rounded to 3 digits)
table	table	the classes: f_ratio, hs, used

Setting	Type	What it is
column	text	the tables' column of the predictions (default the first)
nclass	any	0 (default, a moving window), a number of classes or their breaks
window	number	the window's width (default a tenth of the range)
res	integer	the window's steps (default 100)
method	text	spearman (default), pearson or kendall
rm_duplicate	bool	drop repeated ratios (default true)

sdm.buffer_folds#

Buffered leave-one-out folds (blockCV's cv_buffer): for every presence (or record) the records farther than the size to train on and the record (with the background within the size) to test.

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates

Output	Type	What it is
folds	any	per fold the records (0-based) to train and to test on

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
sizerequired	number	the buffer's radius (metres on lon/lat records)
presence_background	bool	test on presences (default true)
add_background	bool	and the background near them (default false)

sdm.checkerboard_folds#

A checkerboard of the predictors' cells aggregated by a factor (ENMeval's checkerboard1) or two (checkerboard2, four folds).

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates
envrequired	any	the predictors: a map or a stack (their grid)

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
aggregation	any	the factor (default 2), or two

sdm.clean_occurrences#

Occurrence records with suspicious coordinates flagged or dropped (landschaft's clean_occurrences, CoordinateCleaner's tests): invalid coordinates, equal longitude and latitude, 0/0, capitals, country and province centroids, the sea, towns, outside the country they claim, geographic outliers of their species, GBIF's headquarters, biodiversity institutions, duplicates and artificial hotspots. Each test takes false (skip it), true (its defaults), a number (its main option) or an object of its options. CoordinateCleaner's tables travel with Terraflux; Natural Earth's land, urban areas and countries are downloaded on first use (or connect your own).

Input	Type	What it is
occurrencesrequired	any	points, or a table with coordinate columns
capitals_ref	any	capitals in place of CoordinateCleaner's (points)
centroids_ref	any	centroids in place of CoordinateCleaner's (points)
institutions_ref	any	institutions in place of CoordinateCleaner's (points)
seas_ref	vector	the land, in place of Natural Earth's (polygons)
urban_ref	vector	urban areas in place of Natural Earth's
countries_ref	vector	countries in place of Natural Earth's

Output	Type	What it is
occurrences	any	the records that pass every test, or (flag) all with flag_ and flagged
counts	table	the records each test flagged, and in total

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
species	text	the column of the species (default species, scientificName or label)
action	text	drop (default) or flag the suspicious records
equal	any	equal longitude and latitude (default true)
zeros	any	at 0/0: buffer, degrees around 0/0 (default 0.5)
capitals	any	at a country's capital: buffer (metres, default 10000), geod (true), verify (false)
centroids	any	at a country's or province's centroid: buffer (metres, default 1000), detail (both, country or provinces), geod (true), verify (false)
seas	any	in the sea: scale of Natural Earth's land (110, 50 or 10; default 50)
urban	any	in Natural Earth's urban areas (default false)
countries	any	outside the country it claims: column (the records' ISO codes; a text sets it), ref_col (default iso_a2_eh for two-letter codes, else iso_a3_eh); default false
outliers	any	far from the records of its species: method (quantile, mad or distance), mltpl (5), tdi (km, 1000), size (7), min_occs (7), thinning (false), thinning_res (0.5)
gbif	any	at GBIF's headquarters: buffer (metres, default 1000), geod (true), verify (false)
institutions	any	at a biodiversity institution: buffer (metres, default 100), geod (false), verify (false), verify_mltpl (10)
duplicates	any	duplicated: additions, more columns that must be equal too (default false)
aohi	any	at an artificial hotspot: buffer (metres, default 10000), geod (true), taxa (Aves, Insecta, Mammalia, Plantae); default false

sdm.dataset#

The records of a species distribution model (landschaft's dataset): presences (or a table with a y, presence or pa column of 1 and 0) and background points with the predictors' values at their cells, those without a value in some predictor left out.

Input	Type	What it is
presencesrequired	any	the presences: points or a table of coordinates
background	any	the background points (sdm.background)
envrequired	any	the predictors: a stack (by name) or a map

Output	Type	What it is
records	table	x, y, presence and the predictors

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)

sdm.download_gbif#

Occurrence records of a species from GBIF (landschaft's download_gbif, the occurrence search API, no account needed): the records with coordinates, up to 100 000 per query (narrow it by country, area or years), as points with their Darwin Core terms and GBIF's issues. For larger queries, and the DOI to cite, request a download on gbif.org.

Input	Type	What it is
area	vector	only the records in its bounding box

Output	Type	What it is
occurrences	vector	the records, points in lon/lat
query	any	the query, the taxon matched, the count and whether truncated

Setting	Type	What it is
speciesrequired	any	a scientific name (GBIF's backbone) or a taxon key
country	any	ISO 3166-1 alpha-2 codes (BR), one or a list
year	any	a year, or a range [1990, 2020] or "1990,2020"
basis_of_record	any	PRESERVED_SPECIMEN, HUMAN_OBSERVATION..., one or a list
max_uncertainty	number	drop records less precise than this (metres)
limit	integer	at most this many records (default all, 100 000)
has_geospatial_issue	bool	keep the records GBIF flags with geospatial issues (default false)
workers	integer	the pages asked at the same time (default 4)

sdm.evaluate#

How well predictions separate presences from the background (landschaft's evaluate, dismo's): AUC, the point-biserial correlation and its p-value, the prevalence, the largest TSS and the Boyce index; the confusion table at every threshold (kappa, sensitivity, specificity, TSS...) and the thresholds that optimize them (max_kappa, max_spec_sens, no_omission, equal_prevalence, equal_sens_spec, mtp, p10).

Input	Type	What it is
presencesrequired	any	the predictions at the presences
backgroundrequired	any	at the background points (or absences)

Output	Type	What it is
stats	table	auc, cor, pcor, np, na, prevalence, max_tss, boyce
table	table	the confusion table per threshold
thresholds	table	the optimal thresholds
boyce	table	Boyce's classes: f_ratio, hs, used

Setting	Type	What it is
column	text	the tables' column of the predictions (default the first)
thresholds	any	the thresholds of the table (default dismo's)

sdm.exdet#

Mesgaran's extrapolation detection (landschaft's exdet): NT1, below 0 where a predictor falls outside the reference's range, and inside the ranges NT2, the Mahalanobis distance against the reference's largest (above 1 a novel combination); mic, the most influential covariate, by its 1-based place among the predictors (0: no novelty).

Input	Type	What it is
referencerequired	any	the conditions the model knows: a table of the predictors (e.g. its presences and background) or a stack
envrequired	any	the new conditions: a stack of the predictors (by name), or a table

Output	Type	What it is
result	any	a stack on the new conditions' grid, or a table

sdm.fit#

A species distribution model with cross-validation (landschaft's fit): MaxNet (the default), a GLM (scikit-learn's StandardScaler and LogisticRegression) or a random forest (scikit-learn's RandomForestClassifier, the same trees); one fit per fold of the records (AUC of the training and test records and their difference, the largest TSS, Boyce's index, the omission rates at the minimum and 10th percentile training presence), their means and standard deviations, the records' out-of-fold predictions (for models stacked on them), and the model on every record with its threshold of maximum sensitivity plus specificity.

Input	Type	What it is
recordsrequired	any	the records: predictors, presence (1 or 0) and fold columns

Output	Type	What it is
model	sdm_model	the model on every record
metrics	table	per fold: fold, auc_train, auc_test, auc_diff, tss, boyce, or_mtp, or_10p, n_test
summary	table	the metrics' means and standard deviations
threshold	number	the threshold of maximum sensitivity plus specificity
predictions	any	with cross-validation, the records with a column prediction: the suitability from the model fitted without the record's fold (none in fold 0), to score the model on what it did not see or to feed a model stacked on it (an ensemble, a calibration, the distance to an expert range: Merow et al. 2017)

Setting	Type	What it is
predictors	any	the predictor columns (default every other column)
presence	text	the presence column (default presence)
fold	text	the fold column (default fold)
categorical	any	the predictors that are categories (classes)
classes	text	the feature classes: default (maxnet's, by each fit's presences: l, lq, lqh or lqph; with cross-validation a fold's model may get fewer than the final one: give the letters to fix them), or letters of l, q, p, h, t
regmult	number	the regularisation multiplier (default 1)
add_samples_to_background	bool	add the presences to the background (true)
type	text	the output: cloglog (default), logistic
clamp	bool	clamp to the training ranges (default true)
cv	bool	cross-validate over the folds (default true)
model	text	maxnet (default), glm or random_forest
C	number	glm: the inverse of the L2 penalty (default 1)
standardize	bool	glm: standardize the predictors first (default true)
n_estimators	integer	random_forest: the trees (default 100)
min_samples_leaf	integer	random_forest: default 1
min_samples_split	integer	random_forest: default 2
max_depth	integer	random_forest: default none
max_features	any	random_forest: the features tried per split: sqrt (default), log2, all, a number or a share
bootstrap	bool	random_forest: default true
random_state	integer	random_forest: scikit-learn's seed (default 1)
weights	text	the records' weights for the GLM and the forest: balanced (default: presences 1, the background their number over its), equal; a weight column of the records is taken as it is

sdm.importance#

The permutation importance of a model's predictors (landschaft's importance): the mean drop of AUC over the records when each predictor is shuffled (numpy's generator), and its standard deviation; the most important first.

Input	Type	What it is
modelrequired	sdm_model	a fitted model
recordsrequired	any	the records: predictors, presence (1 or 0) and fold columns

Output	Type	What it is
importance	table	importance and sd per predictor, the most important first
predictors	any	the predictors, most important first

Setting	Type	What it is
predictors	any	the predictor columns (default every other column)
presence	text	the presence column (default presence)
fold	text	the fold column (default fold)
n_repeats	integer	the shuffles per predictor (default 5)
seed	integer	numpy's seed (default 1)

sdm.jackknife_folds#

Jackknife folds (ENMeval's): every presence its own fold; background 0.

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column

sdm.maxnet#

A MaxNet species distribution model (landschaft's MaxNet, the maxnet package): Maxent as a lasso logistic regression of presences against background points over linear, quadratic, product, hinge, threshold and categorical features of the predictors (glmnet's path, maxnet's regularisation).

Input	Type	What it is
data	table	the predictors and a presence column (1 presence, 0 background)
presences	table	or the predictors at the presences
background	table	and at the background points

Output	Type	What it is
model	sdm_model	the fitted model
coefficients	any	every feature with a nonzero coefficient and its coefficient

Setting	Type	What it is
presence	text	the presence column of data (default presence)
categorical	any	the predictors that are categories (classes)
classes	text	the feature classes: default, or letters of l, q, p, h, t
regmult	number	the regularisation multiplier (default 1)
add_samples_to_background	bool	add the presences to the background (default true)
type	text	the predictions' output: cloglog (default), logistic, exponential or link
clamp	bool	clamp the predictors to the training ranges (default true)

sdm.maxnet_predict#

A model's predictions (maxnet's predict, the others' predict_proba): the suitability of every cell of a stack of the predictors (by name), or of every row of a table; for a MaxNet cloglog, logistic, exponential or the link, for a GLM or a forest the probability of presence.

Input	Type	What it is
modelrequired	sdm_model	a fitted model (sdm.maxnet, sdm.fit)
envrequired	any	a stack of the predictors (by name), or a table

Output	Type	What it is
prediction	any	a map on the stack's grid, or a table

Setting	Type	What it is
type	text	the output (default the model's)
clamp	bool	clamp to the training ranges (default the model's)

sdm.mess#

Elith's multivariate environmental similarity surface (landschaft's mess, dismo's): per predictor the similarity of each value to the reference's (negative outside their range), and the least of them; negative where the model extrapolates.

Input	Type	What it is
referencerequired	any	the conditions the model knows: a table of the predictors (e.g. its presences and background) or a stack
envrequired	any	the new conditions: a stack of the predictors (by name), or a table

Output	Type	What it is
result	any	a stack on the new conditions' grid, or a table

Setting	Type	What it is
full	bool	every predictor's MESS too (default false)

sdm.migrate#

A species' dispersal through the habitat suitability of successive periods (MigClim 1.6's cellular automaton, the same run for a seed): every step, a suitable empty cell is colonised from an occupied cell within the kernel's distance with the kernel's probability (times the source's propagule production while it matures, times the sink's suitability), unless barriers lie between; long-distance dispersal to random cells; cells that turn unsuitable stay until their period ends. The final map is MigClim's: 1 the initial cells, a step's number (period × 100 + step) where it colonised, 30000 suitable but not reached, below 0 lost, 0 absent.

Input	Type	What it is
initialrequired	map	the initial distribution: 1 present, 0 absent
suitabilityrequired	raster	the habitat suitability of each period, 0 to 1000 (a stack, a loop's series, rasters)
barriers	map	the barriers (1)

Output	Type	What it is
map	map	the final state (MigClim's codes)
occupied	map	1 where the species is at the end, 0 elsewhere
steps	stack	the state after every dispersal step
table	table	per step: period, step, step_id, unlimited, no_dispersal, occupied, absent, colonized, decolonized, ldd

Setting	Type	What it is
steps	integer	dispersal steps per period (default 1)
kernelrequired	any	the probability of colonisation at 1, 2... cells: [1, 0.4, 0.1]
ini_maturity	integer	the age a cell first disperses (default 1)
full_maturity	integer	the age it fully disperses (default 1)
propagules	any	the propagule production of the ages from ini_maturity to full_maturity (default 1 each)
threshold	integer	suitability below it 0, else 1000 (default 0: none)
ldd_frequency	number	long-distance dispersal's frequency (default 0)
ldd_min	integer	its smallest distance in cells (default 10)
ldd_max	integer	its largest distance in cells (default 15)
barrier_type	text	strong (default: two of five lines) or weak (all five)
seed	integer	the seed of the C runtime's rand() (default 1)

sdm.mop#

The mobility-oriented parity of new conditions (landschaft's mop, the mop package): the mean distance of each to its nearest reference conditions (Euclidean or Mahalanobis), and where the predictors fall outside the reference's ranges (basic; simple: how many; detailed: which, below and above, as sums of 2^k).

Input	Type	What it is
referencerequired	any	the conditions the model knows: a table of the predictors (e.g. its presences and background) or a stack
envrequired	any	the new conditions: a stack of the predictors (by name), or a table

Output	Type	What it is
result	any	a stack on the new conditions' grid, or a table

Setting	Type	What it is
type	text	basic (default), simple or detailed
distance	text	euclidean (default), mahalanobis or none
where	text	the conditions whose distance to take: in_range (default), out_range or all
percentage	number	the share of the reference nearest, (0, 100] (default 1)
scale	bool	scale the predictors (Euclidean; default false)
center	bool	center them (Euclidean; default false)
rescale	bool	the distances rescaled to 0-1 (default false)

sdm.omission_rate#

The share of test presences predicted below the training presences' minimum (mtp) or 10th percentile (p10, default) (landschaft's omission_rate, ENMeval's).

Input	Type	What it is
trainrequired	any	the predictions at the training presences
testrequired	any	at the test presences

Output	Type	What it is
rate	number	the omission rate

Setting	Type	What it is
column	text	the tables' column of the predictions (default the first)
threshold	text	p10 (default) or mtp

sdm.one_per_cell#

A record per cell of a raster (landschaft's one_per_cell): the first in each cell, those outside the raster left out (and with drop_missing those on cells without a value in any layer).

Input	Type	What it is
occurrencesrequired	any	points, or a table with coordinate columns
rasterrequired	any	a map or a stack: its grid

Output	Type	What it is
occurrences	any	the records kept

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
drop_missing	bool	drop records on cells without values (default false)

sdm.prepare#

The preparation of a species distribution model in one node (landschaft's prepare): the records cleaned (CoordinateCleaner's tests), one per cell, thinned; the accessible area; the predictors without collinearity in it; background points in it; the predictors' values at the records; the folds. Each step takes false (skip it), true, a number (its main setting) or an object of the settings of its node; one seed for every random step. The report counts what each step removed.

Input	Type	What it is
occurrencesrequired	any	the records: points, or a table with coordinate columns
envrequired	any	the predictors: a stack (by name) or a map
area	vector	the accessible area, in place of the area step
background	any	background points, in place of the background step
regions	vector	regions for the area step (method regions)
within	vector	polygons the area is cut to

Output	Type	What it is
records	table	x, y, presence, the predictors and the fold
env	any	the predictors kept
area	vector	the accessible area (else the predictors' extent)
presences	any	the records left
background_points	any	the background points
report	any	per step: the presences, those removed and why
settings	any	every setting used

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
species	text	keep only this species (the species column)
clean	any	sdm.clean_occurrences' settings (default true)
one_per_cell	bool	a record per cell of the predictors (default true)
thin	any	sdm.thin: the distance in metres or its settings (off)
area	any	sdm.accessible_area: the buffer's distance in metres or its settings (off: the whole extent)
predictors	any	sdm.select_predictors: the VIF threshold (10) or settings
background	any	sdm.background: the points (10000) or settings
folds	any	k spatial blocks (5), or {method: blocks, quadrants, checkerboard, random, jackknife or clusters, and its settings}; blocks take size "auto" (the predictors' range of autocorrelation)
seed	integer	the seed of every random step (default 1)

sdm.quadrant_folds#

The four quadrants of the presences (ENMeval's block partition): near-equal counts split along the orientation's axes, the background by the same lines.

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
orientation	text	lat_lon (default), lon_lat, lon_lon or lat_lat

sdm.random_folds#

Random folds of the presences (landschaft's random_folds, ENMeval's randomkfold): k groups of near-equal size drawn with R's generator; background 0.

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
k	integer	the folds (default 5)
seed	integer	the seed of R's generator (default 1)

sdm.reference_data#

The geographic reference data of the cleaning of records (landschaft's reference_data): CoordinateCleaner's centroids and capitals of countries and provinces (countryref, at the centroids), biodiversity institutions and artificial hotspots (aohi), and Natural Earth's land (land110, land50, land10), urban areas (urban50) and countries (countries50), downloaded on first use.

Output	Type	What it is
features	vector	the points or polygons, with their attributes

Setting	Type	What it is
namerequired	text	the data set

sdm.response_curves#

A model's response curves (landschaft's response_curves): each predictor over its range in the records, the others at their mean or median; a column of values and one of suitability per predictor.

Input	Type	What it is
modelrequired	sdm_model	a fitted model
recordsrequired	any	the records: predictors, presence (1 or 0) and fold columns

Output	Type	What it is
curves	table	and _suitability columns

Setting	Type	What it is
predictors	any	the predictor columns (default every other column)
presence	text	the presence column (default presence)
fold	text	the fold column (default fold)
of	any	the predictors to draw (default all)
n	integer	the steps (default 100)
fixed	text	the others at their mean (default) or median

sdm.select_predictors#

Predictors without collinearity (landschaft's select_predictors, usdm's vifcor and vifstep): the most correlated pairs above a correlation dropped one by one (the one of larger VIF), then the predictors of the largest VIF above a threshold; those to keep stay.

Input	Type	What it is
envrequired	any	the predictors: a stack (by name) or a table
area	vector	the area of the stack to look at

Output	Type	What it is
env	any	the predictors kept: a stack or a table
selection	any	kept, excluded (in order) and the kept ones' VIF

Setting	Type	What it is
size	integer	the cells sampled on large grids (default 5000; null: all)
seed	integer	the seed of R's generator (default 1)
vif	number	the VIF threshold (default 10; null: none)
cor	number	the correlation threshold (default none)
keep	any	the predictors to keep
method	text	of the correlation: pearson (default), spearman, kendall

sdm.spatial_blocks#

Spatial blocks (blockCV's cv_spatial): the records in hexagons or squares of a size (or rows and columns), the blocks assigned to k folds at random (the most balanced of the iterations), systematically or as a checkerboard.

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
size	number	the blocks' size (metres on lon/lat records)
rows_cols	any	else their rows and columns (default [10, 10])
k	integer	the folds (default 5)
hexagon	bool	hexagons (default) or squares
flat_top	bool	flat-topped hexagons (default false)
selection	text	random (default), systematic or checkerboard
iteration	integer	the random assignments tried (default 100)
offset	any	the blocks' offset, [x, y] shares of a block
extend	number	squares: the extent grown by this percentage
deg_to_metre	number	metres per degree on lon/lat (111325)
seed	integer	the seed of R's generator (default 1)

sdm.spatial_clusters#

Spatial or environmental clusters (blockCV's cv_cluster): R's kmeans of the records' coordinates, or of the predictors at the records (scaled as terra).

Input	Type	What it is
recordsrequired	any	the records: points, or a table of coordinates
env	any	the predictors: environmental clusters

Output	Type	What it is
records	any	the records with their fold

Setting	Type	What it is
presence	text	their presence column, 1 or 0 (default presence)
x	text	the table's x column
y	text	the table's y column
k	integer	the clusters (default 5)
scale	bool	scale the predictors (default true)
algorithm	text	Hartigan-Wong (default), Lloyd, Forgy or MacQueen
seed	integer	the seed of R's generator (default 1)

sdm.thin#

Occurrence records thinned (landschaft's thin, spThin's thin.algorithm): records dropped at random until none is closer than the distance to another (great-circle, on lon/lat), the repetition that keeps the most.

Input	Type	What it is
occurrencesrequired	any	points, or a table with coordinate columns

Output	Type	What it is
occurrences	any	the records kept
kept	any	how many each repetition kept, most first

Setting	Type	What it is
x	text	the table's x column (default x, lon, longitude...)
y	text	the table's y column (default y, lat, latitude...)
distancerequired	number	the least distance between records, in metres
reps	integer	the repetitions (default 10)
seed	integer	the seed of R's generator (default 1)

sdm.threshold#

A threshold of suitability from presences and background (landschaft's threshold, dismo's): max_spec_sens (default), max_kappa, no_omission, equal_prevalence, equal_sens_spec, mtp, p10, or the highest that keeps a sensitivity.

Input	Type	What it is
presencesrequired	any	the predictions at the presences
backgroundrequired	any	at the background points

Output	Type	What it is
threshold	number	the threshold

Setting	Type	What it is
column	text	the tables' column of the predictions (default the first)
method	text	the criterion (default max_spec_sens)
sensitivity	number	of the method sensitivity (default 0.9)

sdm.vif#

The variance inflation factor of every predictor (landschaft's vif, usdm's vif): 1 / (1 - R^2) of its regression on the others (R's lm), over the cells of the stack (a sample of them on large grids) or the rows of a table.

Input	Type	What it is
envrequired	any	the predictors: a stack (by name) or a table
area	vector	the area of the stack to look at

Output	Type	What it is
vif	table	a column per predictor, its VIF

Setting	Type	What it is
size	integer	the cells sampled on large grids (default 5000; null: all)
seed	integer	the seed of R's generator (default 1)