NEWS.md

In stars: Spatiotemporal Arrays, Raster and Vector Data Cubes

version 0.7-0

• use values column from RAT table, and merge duplicate labels; #761

• st_extract() no longer stops when GDAL's InterpolateAtPoint returns error value(s); #760

• add st_as_stars() methods for CFVariable and CFDataset, provided by package ncdfCF; #756, #757 by @pvanlaake

• try harder to convert CFTime times to Date when they are dates.

• if set, options("stars.regular") controls the threshold below which dimension coordinates are considered regular, with default 1.0e-4

• removed dependency on PCICt, added dependency on CFtime; #754 by @pvanlaake

• handling of dimensions with a single value is now more consistent; #754

• slice.stars_proxy() works (again) for single-file proxy objects; #751, #527

• [.stars() with a character selector selects on names of a dimension if it has names; #747

• handle "intervals" dimension in aggregate.stars() and st_as_sf.stars(); #745

• fix c.stars() for the case where some of the objects have a single slice on the along= dimension; #743

• st_redimension() (and by that, merge.stars()) use abind() for numeric variables rather than c(), reducing memory overhead.

• fix problem with reading HDF4 files; #741 by Alexys Rodriguez

• use, by default, normalizePath() on filename in read_mdim(); #735

• st_extract() accepts empty points as target; #734

• st_sfc2xy() passes ... on to st_as_stars(); #733

version 0.6-8

• address /.difftime issue new in R-devel rev 87670

• c.stars() is more strict when combining time sequences; #703

• fix plotting when breaks contain duplicates; #728

• fix st_as_stars.im(); #727 and #648, thanks to Barry Rowlingson

version 0.6-7

• st_extract() fix if points coincide with boundary grid cell centers and bilinear interpolation is used; #720

• st_extract() if used with GDAL 3.10.0 uses InterpolateAtPoints, allowing for cubic and cubicspline interpolators (requiring sf >= 1.0-19).

• Ops.stars() (math ops) now also recycle arrays in the first argument; #718

• c.stars() verifies semantic equivalence of objects' CRS; #703

• initial support for read_mdim() to work with proxy = TRUE; #659

version 0.6-6

• skip cubble tests for cubble version 0.3.1; https://github.com/huizezhang-sherry/cubble/issues/30

• st_transform.stars transforms geometries in array elements

• mutate.stars (and others) handle attribute names with spaces in them; #689

• st_crop() gains an argument normalize; when set to TRUE st_normalize() is called on the returned value; #685, #686

• constrain reading full GEOLOCATION arrays to the case where they are 2-D; #678

version 0.6-5

• fix st_as_stars.Spatial() for Spatial gridded objects with non-square grid cells, see https://github.com/r-spatial/gstat/issues/123

• add prcomp() methods for stars and stars_proxy objects, working on attributes or last dimension

• st_rasterize() with align=TRUE returns NA values where there are no data; #668

• read_mdim() reads tables with composity type, returning a data.frame in that case; #659

• st_rotate() transforms a rotated grid back to a curvilinear grid in unrotated coordinates.

• aggregate.stars() deals with functions that return more than one number, putting these in a new dimension like st_apply() does

• st_as_stars.data.frame() and st_as_stars.sf() better handle non-raster data cubes

• plot.stars() only resets layout when needed (more than one sub-plot, or key present)

• fixed st_as_stars.im(); #648

• st_crs<-.stars() is less critical on existing CRS being of class crs

• c.stars() with a single (valid) argument and along specified adds a dimension; #646

• st_join.stars() keeps attributes from x complete; #643

• st_as_stars.list() requires a named list, and will set names of array dimensions if not present

version 0.6-4

• plot.stars() has a fill argument that shifts unused plotting space between sub-maps to the bottom or right side of the plotting area

• in plot.stars(), the key.width default is sensitive to par("ps"), the pointsize graphics parameter

• plot.stars() and image.stars() are sensitive to cex.axis, for axes and key (requires sf >= 1.0-14); #642

• move lwgeom dependency to Suggests; using st_transform_proj() on vector data cubes requires loading lwgeom first

version 0.6-3

• st_downsample() has argument offset to pixel-shift downsampled images

• st_downsample() has argument FUN to compute sub-tile aggregates; #604

• st_as_stars.bbox() retains factor values; #640

• fix CRAN error in test script

• st_crop() works (and warns) for the case when the crop area does not overlap with the area of the object; #638

version 0.6-2

• split.stars() accepts stars objects with multiple attributes; #635

• [.stars() supports NA values in dimension ranges for vector geometry (sfc) dimensions, resulting in empty geometries

• st_extract() supports extracting points values from curvilinear grids (when not proxy); #632

• read_mdim() reads curvilinear rasters (geolocation arrays), so far only at full extent/resolution

• st_as_stars.stars() accepts curvilinear argument with lon/lat array names present in .x

• consistently use OGC:CRS84 instead of EPSG:4326

• setting values = NULL in st_set_dimensions() removes dimension values

• more gracefully handle plotting of global coverage curvilinear grids; #632

• image.stars() plots images (e.g. of cross sections) when x and/or y are singular or absent; #628

• st_as_stars.cubble_df() adds conversion from cubble; cubble::as_cubble() methods converts back

• [<-.stars() accepts for i an indicator (numeric length character) to the array to be replaced; #626

• plot.stars() gains an argument key.lab to set the legend key label (requires sf >= 1.0-13)

version 0.6-1

• remove rgdal dependency

• read_stars() fixes combining bands with different block sizes; #623

• st_warp() gets (proper) default value for threshold; #618

• read_mdim() reads "raster" with single pixel (as a point)

• [.stars(), as in r[x] allows x to be a logical stars object

• [<-.stars_proxy() clones environment, so that after r[r > 100] = NA we don't get infinite recursion when realizing r

• read_stars() avoids reading a raster twice to determine how to choose proxy; proxy can now be set as (and defaults to) the number of cells (bands * rows * columns) above which data will not be read in memory but returned as stars_proxy object; #609

• fix using RasterIO in read_stars() when proxy=TRUE; #608

• plot.stars() hook function can handle arguments row, col, nrow, ncol, nr, value and bbox; #600

• fix handling of categorical rasters with colors but without category labels; #595, fallout of #565

• fix subsetting of proxy objects over a time range; #596

version 0.6-0

• write_stars() writes scaled and/or shifted values when using argument scale_offset; #589

• aggregate.stars_proxy() implements aggregation with non-spatial by objects (lazily)

• fix [.stars_proxy() when selecting dimension 3 and higher; #561

• in plot.stars(), col can also be a palette function

• st_res() returns spatial resolutions, and optionally all dimension resolutions; #557 thanks to Krzysztof Dyba

• read_stars() shortens band or array names that contain a common start or ending, unless names would become empty or shorten=FALSE was set; e.g. shorten="B" puts a B before shortened array names

• printing stars dimension tables omits fields with only NULL or NA values, unless print(..., all = TRUE) is given

• improve reading categorical rasters, which now avoids calling factor(); #565 thanks to Krzysztof Dyba

• read_mdim() will read bounds arrays for coordinates using the bounds attribute, and accepts a bounds argument to specify them when that attribute is missing

• time() returns time stamps of a time dimension

• st_cells() returns the cell index for a set of point coordinates, provided as sf or sfc object; #558

• reading & writing vector data cubes: read_mdim() reads CF compliant vector geometries, and reconstructs them into an sfc dimension; write_mdim() writes them.

• write_mdim() uses GDAL multidimensional array API;

• read_mdim() uses arguments offset, count and step to read sub-arrays or strided arrays (requires sf >= 1.0-9)

version 0.5-6

• export read_mdim(), a reader using GDAL's multidimensional array API (for sf <= 1.0-8)

• remove tos_O1_2001-2002.nc from packaged datasets to keep source package size below 5 Mb

• as.POSIXct.stars() converts PCICt dimensions to POSIXct values.

• improve handling of PCICt 360 or 365 day calendars; read them in read_mdim (requires sf >= 1.0-9)

• read_stars() reads factor levels better from attribute table; #484 thanks to @ailich

• read_stars() puts band names from band_meta DESCRIPTION= tags into values;

• improve handling of categorical rasters, and their exchange with terra; #484

• plot() handles auto colors better for factor arrays

• read_ncdf() handles units more formally in setting crs; #533

• print message that dimensions of proxy objects do not reflect unevaluated operations; #530

• passing na.action = na.omit to geom_stars() removes NA values; #532

• read_stars() detects curvilinear grids automatically; #513

• st_warp() warps curvilinear grids (using Euclidean distances only on coordinates); #513

• Ops.stars() errors when (common) dimension are not identical; #506

• guess_raster() accepts empty rows/columns and sparse grids; #509

• speed up rgb plotting; #503

• Added a new helper function st_tile() to specify the block parameters (nXOff, nYOff, nXsize, nYSize) required by RasterIO argument in read_stars(); #492 thanks to Krzysztof Dyba

version 0.5-5

• st_as_stars.bbox() creates an empy raster file if proxy = TRUE; #489

• st_rasterize() has option align = TRUE to use a template for aligning the new raster to; #489

• adrop.stars() with missing dimensions no longer drops x/y raster dimensions; #485

• aggregate.stars() propagates units of arrays; #477

version 0.5-4

• c.stars() fails if it tries to merge arrays with different units; #475

• For NetCDF files, read_stars() uses the long_name as array name; #475

• add rename() method; #470

• refresh CRS of packaged L7_ETMs.tif; #466

• as.data.frame.stars() works for mixed regular and rectilinear dimension; #458

• plot.stars() plots curvilinear rasters with color table, or without table but col argument passed; #456

• st_extract() accepts a matrix with points as at argument, for when performance is important; see e.g. https://github.com/rspatial/terra/issues/341

• fix bug in st_crop() when cropping area is larger than grid; #455

• export st_downsample(), e.g. to be used by tmap; https://github.com/r-tmap/tmap/issues/597

• argument downsample in plot.stars() and st_as_stars.stars_proxy() and st_downsample() has the same effect (removed a one-offset between them).

• st_redimension() works for curvilinear grids; #441

• downsample is propagated to subexpressions like r[r < 50] = NA

• predict.stars() obtains an argument drop_dimensions that, if TRUE, drops dimensions from the prediction data.frame; #362

• extend options in st_rgb(), #432, by Gabo Gaona

• allow subsetting with [ by using labels, e.g. of band names.

version 0.5-3

• read_stars() accepts a function (or list with functions) as first argument, allowing for saving stars objects that read from package directories resolving platform-dependent paths at run-time

• handle categorical rasters starting at value 0 (by adding 1, and warning); #428

• add %in% method; #424

• read_stars() gains an argument tolerance to control tolerance in dimension value comparisons; #414

• binary Ops (like +, -, * etc.) work for stars_proxy objects; #390

• st_rasterize() rasterizes multiple attributes, and handles factors (when sf >= 0.9-9)

• write_stars() deals better with stars_proxy objects; #404

• fix regression in reading some stars_proxy objects; #379

• add [<- (partially) and is.na methods for stars_proxy objects; #402

• add replace_na() methods; #402

version 0.5-2

• read and write factor levels as GDAL category names; write color table; #392

• handle normalize_path for choosing to proxy; #391

• ignore units when there are different units across bands of a subdataset

• speed up st_rgb() using faster st_apply() approach; #315, #390

• improve handling of crs in Spatial objects (avoid loss of wkt comments)

• correctly write band subsets for smaller proxy objects; #291

• write arbitrarily cropped proxy objects; #291

• speed up st_apply() when a function is provided that works on chunks at a time; #390

• warn when breaks = "quantile" results in a single class; #388

• fix [ bug selecting bands in proxy objects; #388

• for stars_proxy objects, write_stars() writes all objects into a multi-layer file; #385

• multi-file proxy objects can be st_warp()ed with use_gdal = TRUE; #385

version 0.5-1

• fix weird GDAL-related bug in stars2 vignette

• read_ncdf() does not take time as mid-points of regular intervals, but as starting points; #378

version 0.5-0

• fix handling of rasters with color tables; #375

• st_apply() and other methods for stars_proxy objects handle ... ; #374

• add st_bbox(), st_crs() methods for terra's SpatVector objects; https://github.com/r-tmap/tmap/issues/536

• add st_bbox(), st_crs() and st_as_stars() methods for terra's SpatRaster objects; https://github.com/r-tmap/tmap/issues/536

• allow for multi-resolution attributes in stars_proxy objects (e.g., all gray scale sentinel-2 bands); see vignettes 2 and 7 for examples.

• plot() defaults to a categorical color scale when plotting a factor variable; https://github.com/r-tmap/tmap/issues/526

• st_extract() extracts space-time points if time_column is specified, and handles time intervals; #352

• add [[<-.stars method, which is now called by $<-.stars, so that array names can be set programmatically

• add transmute() methods

• plot.stars() calls droplevels if a factor array has any NA levels; #339

• read_stars() reads NaNs as NA; #333

• improve st_extract() method for both stars and stars_proxy objects; interpolation options are reduced to bilinear; #322, #279, #290

• better handle categorical rasters that do not start at value 1; #329

• plot layout can be controlled with mfrow = c(nr, nc) argument

• stars_proxy objects have a normalized path; #331

• cropping or selecting with bbox treats cells always as small polygons; #330

• add faster st_extract() method for stars objects; #322

• added vignette: "How raster functions map to stars functions", by Sebastien Rochette; #122, #325

• fix bug in dimension values field when downsampling; #324

• write_stars() also writes out band names; #323

• add rgdal to Suggests:

• each call_list entry of a stars_proxy object carries its proper calling environment; #309

• st_as_sf.stars() copes with zero attribute (empty) stars objects

• add st_set_bbox() generic, to set raster extent, motivated by #315

• set up tic, with great help from @pat-s, #313

• get rid of more proj4strings for representing coordinate reference systems; #312

• as(x, "Spatial") correctly handles from dimension values different from one

• read_stars() now sets the BANDNAME GDAL metadata item, or else the band's GetDescription() as the band's dimension values

• st_as_stars.data.frame() reads simple tables (non-raster data) if dims has length less than 2

• band descriptions are in the band dimension values

• dimension tables are simpler, and are shown properly in Rstudio

• st_rgb() gains a probs argument, to cut off and stretch based on quantiles

• as(x, "Raster") merges multiple attributes before converting to raster brick

version 0.4-3

• fix bug in st_as_stars.Raster; set crs to the one assigned by raster; https://github.com/r-tmap/tmap/issues/471

• add s2 to Suggests:

• new function st_rgb() collapses (reduces) a dimension to rgb hex value; #302

version 0.4-2

• aggregate.stars() handles arrays with NA values now correctly; brought up in #299 by Thorsten Simon

• aggregate.stars() gains an argument exact which, if TRUE, calls exactextractr for polygonal aggregation; #289

• read_stars() reads all subdatasets with dimensions equal to first, and warns when ignoring others; #296

• make copying over of dimensions somewhat easier; #295

• st_as_stars.Raster() tries to read from file if the raster object is not an in-memory object.

• write_stars() normalizes path, as read_stars already did; #293

• merge() for proxy objects acts, and is no longer lazy; #290

• st_as_stars.Raster() returns a proxy object if the raster layer is on disk

• add st_extract() to extract e.g. time series from grids at point locations; #279; #290

• read_stars() chooses a value for proxy that depends on the data dimensions; #281

• x/y range subsetting of stars_proxy objects now only reads that range, similar to how crop already did this.

• st_warp() preserves levels and colors; https://github.com/r-tmap/tmap/issues/429

• st_crop() works with bounding boxes larger than the downsampled bounding box; #276

• st_crop() has a non-zero default for epsilon (bounding box shrinkage) to exclude cells touching the crop bounding box; #275

• image.stars() (and hence plot.stars) gains an extent argument for setting the extent of a plot; https://github.com/r-spatial/sf/issues/1193

version 0.4-1

• st_warp() (stars native) flips longitudes a full cycle; #256, #264, #269

• handle axis order in st_transform (requires sf >= 0.9-1)

• depend on sf 0.9-0

• adapt to cubelyr split-off from dplyr; add cubelyr to Suggests:; https://github.com/hadley/cubelyr/issues/2

• add droplevels() method

• handle color tables, category tables and raster attribute tables read through GDAL; #128, #245; https://github.com/r-spatial/mapview/issues/208

• handle dimension crs specially, for proxy objects now.

• handle new-style crs objects from upcoming sf, moving away from proj4strings

• handle full crs objects as refsys element in a spatial dimensions, rather than proj4string only

• st_raster_type(x) reveals the raster type of x; #248, https://github.com/r-tmap/tmap/issues/368

• add st_as_stars.OpenStreetMap() method; #241 by @mtennekes

• add st_flip() to flip arrays along one or more dimensions without changing dimension properties

• add as.owin() method to convert (2D raster) stars objects to spatstat owin; https://github.com/r-spatial/sf/issues/1233

• for temporal aggregation, aggregate.stars now also takes by arguments like "week", "month", or "5 days"

• add st_as_stars() method for xts objects; improve as.xts for stars objects

• skip some tests on solaris

version 0.4-0

• plot() now uses all data to figure out breaks, in order to also find extremes; #216

• st_mosaic() creates mosaics from spatially disjoint rasters; #210

• 205 large refactoring of read_ncdf, by David Blodgett and Mike Sumner, affecting #199, #89, #30, #86, #175

• allow for funny units like m s**-1; #201

• add contour() method for stars objects; #196

• plot uses rasterImage by default if available; #194

• the x and y raster dimensions can be set and unset with xy argument in st_set_dimensions; #190

• retain factor levels with dimension values when set in st_set_dimensions; #188

• add conversion from stars_proxy to Raster: #193

• make plotting of multiple curvilinear grids work

• plot by default no cell borders in case of curvilinear, rotated or sheared grids

• robustify handling of units

• allow read_ncdf() to ignore bounds

• scale was applied wrongly on multi-band images; #189, this requires sf >= 0.7-5

• .nc is now recognized correctly by write_stars and will write a NetCDF file; #186

• [ subset now works correctly with negative or logical indices; #184, #185

• NA values for float32 grids are now correctly detected; #182, this requires sf >= 0.7-5

• cropping of a stars_proxy object now works; #179

• st_apply() can now loop over Raster layers; examples in #176

version 0.3-1

• st_as_stars.bbox now has an ncells and pretty argument, to better choose default raster dimensions

• geom_stars now works with stars_proxy objects, but needs downsample to be set; #21

• NA values in Float32 rasters are now read correctly with read_stars; #182

• handle bounds, when given, in read_ncdf

• provide time parsing (POSIXct, PCICt) for read_ncdf; #115

version 0.3-0

• add st_area method to return raster grid cell sizes; #99

• fix st_warp with use_gdal=TRUE, allowing for multi-band warps

• add st_get_dimension_values to get the values of a particular dimension (if meaningful); #100

• allow for setting intervals as dimension values; see examples of st_dimensions

• add st_contour, and clean up st_as_sf; #99

• experimental color table support; https://github.com/r-spatial/mapview/issues/208

• rewrote vignettes, added vignettes; #99

• deprecate st_write.stars for write_stars; #96

• use "native" R array-factor support

• support for PCICt 360- and 365-day calendars; #29

• remove import of ncdf4 in favor of RNetCDF, now in line with practice in ncmeta package. Thanks to David Blodgett for motivation and testing (see #87, #94).

• st_as_sf uses date/time column names when appropriate

• allow missing trailing comma's when subsetting: a[1,,] and a[1,] now do the same.

• move rlang to Imports: ; rewrite [ subset using rlang.

• add conversion to and from Spatial* classes, including the gridded ones, taking care of factor variables

• depend on sf 0.7-2

• add logz support for log-scale keys to plot and image

version 0.2-0

• vignettes now use an external package, starsdata, for larger dataset examples

• support [<-.stars e.g. to mask out values; support is.na.stars to replace NA masks

• support cut methods and factor arrays (plot, subset); #56

• add st_rasterize, which uses GDALRasterize to rasterize an sf object; #13

• st_as_sf.stars now uses GDAL(F)Polygonize when give a regular or sheared grid grid cells are not points, and returns contour bands using GDALContourGenerateEx (requiring GDAL 2.4.0) in case cells are points; #13

• support curvilinear grids; see #54 and the data_model vignette

• add vignette about how stars_proxy objects work

• stars_proxy objects defer processing of st_apply jobs until after subsampling; #50

• allow reading sections of a raster, raster at a lower resolution, selected bands; #48

• allow reading vectors (arrays) with more than 2^31 elements; #48

• fold all higher dimensions into the third dimension before coercing to Raster; #40

version 0.1-1

• add meta data reader to read_stars

version 0.1-0

• add merge (merge attributes into array dimension) and split (split dimension over attributes)

• interface to sf, raster and spacetime

• improve plotting

• handle logical arrays in plot

• add st_apply, analogous to apply

• add cropping/masking when used as x[buf] with buf an sf, sfc or bbox object; masking when x[buf, crop = FALSE]

• add Ops (+,-,/ etc) and Math (sqrt, pow, sin etc) methods

• add dimnames and dimnames<- methods for stars objects

• downsample large grids to only plot pixels actually shown

• can plot rectilinear grids (but will plot rgb images as regular grids)

• rgb argument to image works

• [ array-like subsetting works; first index is attribute selector

version 0.0

• interface the 9 C++ gdal utils through gdal_utils() (now part of sf)

stars documentation built on Dec. 14, 2025, 9:06 a.m.