landform 0.3.0

A thorough revision of the package. Please read the first two points: with the default settings, results differ from those of earlier versions.

Changes that affect results

The TPI is now standardised, by default, over the whole raster (z-scores of the TPI), as in Weiss (2001) and Jenness (2006), to which the classification thresholds refer (std = "global"). Earlier versions divided the TPI of each cell by the standard deviation of elevation in its own neighbourhood; this is still available with std = "local". In nearly flat areas the local standardisation turns small irregularities of the DTM into large standardised values, so that plains may be classified as a mosaic of valleys and ridges; on perfectly flat areas it is undefined.
By default the focal cell is now excluded from the neighbourhood mean (centre = FALSE), in line with the definition of TPI as the difference between a cell and its surroundings. With global standardisation this has no effect away from the edges.
Results of version 0.2 are reproduced exactly with std = "local", centre = TRUE, apart from the two fixes below.
Fixed: cells with a standardised TPI exactly equal to 0.5 fell into no class (and were coded 0); they are now classified as Upper Slope, so that every cell gets exactly one class.
Fixed: cells on the outer rim of the raster (and next to NA cells), where slope cannot be computed, were left unclassified only when their TPI was close to zero, which biased the class frequencies. They are now consistently left unclassified.
Class areas are now computed with terra::cellSize(), and are therefore correct also for longitude/latitude rasters.

Removed the argument pad from the internal calls to terra::focal(); it belongs to raster::focal() and is rejected by recent versions of terra (reported by Robert J. Hijmans).
Fixed the map colours not matching the legend when some classes were absent.
An invalid class.type now gives an error instead of returning NULL.

Neighbourhoods can be square, circular or annular (shape, inner), and sized in cells or in map units (units; metres for longitude/latitude).
Thresholds can be changed: tpi.thr, lf.thr, slope.thr.
New function tpi(), returning the raw or standardised TPI surface.
New function class_layers(), returning one presence/absence layer per class.
New sample dataset, malta_dtm().
Input checks with informative error messages (window sizes, ln > sn, single-layer input, thresholds, colours).

landform() now returns an object of class "landform" with print(), summary() and plot() methods, holding the classes as a single categorical SpatRaster (classes, with class names and colours), the TPI and slope layers, the descriptive statistics (always computed; now also with mean elevation and slope per class), the input DTM and the settings. The list of per-class rasters returned by earlier versions (result$all, result$Valley, …) is replaced by result$classes and class_layers(result); result[["Descriptive Statistics"]] is replaced by result$stats (or summary(result)).
Class names now follow Jenness (2006) (e.g. “Midslope Ridges, Small Hills in Plains” instead of “Midslopes Ridges-Small Hills in Plains”).

New colour palettes: an ordinal blue-to-brown palette for the Slope Position Classification, and hues grouped by landform family for the Landform Classification (both replace rainbow()). Custom colours via col.
The legend is placed outside the map by default, so that no data are hidden (leg.pos still accepts the usual keywords).
Optional relief shading underneath the classes or the TPI: a slopeshade (shade = "slope", slope in grey levels from white for flat terrain to black for the steepest slopes) or a hillshade (shade = "hill"), with the overlying layer drawn semi-transparently (alpha).
The bar plot of class areas is horizontal and labelled with class names.
plot(x, what = "tpi") maps the standardised TPI surface(s).
plot = FALSE suppresses plotting.