Raster data processing in GRASS GIS

Raster maps in general

A "raster map" is a data layer consisting of a gridded array of cells. It has a certain number of rows and columns, with a data point (or null value indicator) in each cell. These may exist as a 2D grid or as a 3D cube made up of many smaller cubes, i.e. a stack of 2D grids.

The geographic boundaries of the raster map are described by the north, south, east, and west fields. These values describe the lines which bound the map at its edges. These lines do NOT pass through the center of the grid cells at the edge of the map, but along the edge of the map itself. i.e. the geographic extent of the map is described by the outer bounds of all cells within the map.

As a general rule in GRASS GIS:

1: Raster output maps have their bounds and resolution equal to those of the current computational region.
2: Raster input maps are automatically cropped/padded and rescaled (using nearest-neighbour resampling) to match the current region.
3: Raster input maps are automatically masked if a raster map named MASK exists. The MASK is only applied when reading maps from the disk.

There are a few exceptions to this: r.in.* programs read the data cell-for-cell, with no resampling. When reading non-georeferenced data, the imported map will usually have its lower-left corner at (0,0) in the location’s coordinate system; the user needs to use r.region to "place" the imported map.

Some programs which need to perform specific types of resampling (e.g. r.resamp.rst) read the input maps at their original resolution then do the resampling themselves.

r.proj has to deal with two regions (source and destination) simultaneously; both will have an impact upon the final result.

Raster import and export

The module r.in.gdal offers a common interface for many different raster formats. Additionally, it also offers options such as on-the-fly location creation or extension of the default region to match the extent of the imported raster map. For special cases, other import modules are available. The full map is always imported.

For importing scanned maps, the user will need to create a x,y-location, scan the map in the desired resolution and save it into an appropriate raster format (e.g. tiff, jpeg, png, pbm) and then use r.in.gdal to import it. Based on reference points the scanned map can be recified to obtain geocoded data.

Raster maps are exported with r.out.gdal into common formats. Also r.out.bin, r.out.vtk, r.out.ascii and other export modules are available. They export the data according to the current region settings. If those differ from the original map, the map is resampled on the fly (nearest neighbor algorithm). In other words, the output will have as many rows and columns as the current region. To export maps with various grid spacings (e.g, 500x500 or 200x500), you can just change the region resolution with g.region and then export the map. The resampling is done with nearest neighbor algorithm in this case. If you want some other form of resampling, first change the region, then explicitly resample the map with e.g. r.resamp.interp or r.resamp.stats, then export the resampled map.

GRASS GIS raster map exchange between different locations (same projection) can be done in a lossless way using the r.pack and r.unpack modules.

Metadata

The r.info module displays general information about a map such as region extent, data range, data type, creation history, and other metadata. Metadata such as map title, units, vertical datum etc. can be updated with r.support. Timestamps are managed with r.timestamp. Region extent and resolution are mangaged with r.region.

Raster map operations

Resampling methods and interpolation methods

GRASS raster map processing is always performed in the current region settings (see g.region), i.e. the current region extent and current raster resolution is used. If the resolution differs from that of the input raster map(s), on-the-fly resampling is performed (nearest neighbor resampling). If this is not desired, the input map(s) has/have to be resampled beforehand with one of the dedicated modules.

The built-in nearest-neighbour resampling of raster data calculates the centre of each region cell, and takes the value of the raster cell in which that point falls.

If the point falls exactly upon a grid line, the exact result will be determined by the direction of any rounding error. One consequence of this is that downsampling by a factor which is an even integer will always sample exactly on the boundary between cells, meaning that the result is ill-defined.

The following modules are available for reinterpolation of "filled" raster maps (continuous data) to a different resolution:

r.resample uses the built-in resampling, so it should produce identical results as the on-the-fly resampling done via the raster import modules.
r.resamp.interp Resampling with nearest neighbor, bilinear, and bicubic method: method=nearest uses the same algorithm as r.resample, but not the same code, so it may not produce identical results in cases which are decided by the rounding of floating-point numbers.
For r.resamp.interp method=bilinear and method=bicubic, the raster values are treated as samples at each raster cell’s centre, defining a piecewise-continuous surface. The resulting raster values are obtained by sampling the surface at each region cell’s centre. As the algorithm only interpolates, and doesn’t extrapolate, a margin of 0.5 (for bilinear) or 1.5 (for bicubic) cells is lost from the extent of the original raster. Any samples taken within this margin will be null.
r.resamp.rst Regularized Spline with Tension (RST) interpolation 2D: Behaves similarly, i.e. it computes a surface assuming that the values are samples at each raster cell’s centre, and samples the surface at each region cell’s centre.
r.resamp.bspline Bicubic or bilinear spline interpolation with Tykhonov regularization.
For r.resamp.stats without -w, the value of each region cell is the chosen aggregate of the values from all of the raster cells whose centres fall within the bounds of the region cell.
With -w, the samples are weighted according to the proportion of the raster cell which falls within the bounds of the region cell, so the result is normally unaffected by rounding error (a minuscule difference in the position of the boundary results in the addition or subtraction of a sample weighted by a minuscule factor; also, The min and max aggregates can’t use weights, so -w has no effect for those).
r.fillnulls for Regularized Spline with Tension (RST) interpolation 2D for hole filling (e.g., SRTM DEM)

Furthermore, there are modules available for reinterpolation of "sparse" (scattered points or lines) maps:

Inverse distance weighted average (IDW) interpolation (r.surf.idw)
Interpolating from contour lines (r.contour)
Various vector modules for interpolation

For Lidar and similar data, r.in.lidar and r.in.xyz support loading and binning of ungridded x,y,z ASCII data into a new raster map. The user may choose from a variety of statistical methods in creating the new raster map.

Otherwise, for interpolation of scattered data, use the v.surf.* set of modules.

Raster MASKs

If a raster map named "MASK" exists, most GRASS raster modules will operate only on data falling inside the masked area, and treat any data falling outside of the mask as if its value were NULL. The mask is only applied when reading an existing GRASS raster map, for example when used in a module as an input map.

The mask is read as an integer map. If MASK is actually a floating-point map, the values will be converted to integers using the map’s quantisation rules (this defaults to round-to-nearest, but can be changed with r.quant).

(see r.mask)

Raster map statistics

A couple of commands are available to calculate local statistics (r.neighbors), and global statistics (r.statistics, r.surf.area). Profiles and transects can be generated (d.profile, r.profile, r.transect) as well as histograms (d.histogram) and polar diagrams (d.polar). Univariate statistics (r.univar) and reports are also available (r.report,r.stats, r.volume). Since r.univar may be slow for extended statistics these can be calculated using r.stats.quantile. Without a zones input raster, the r.quantile module will be significantly more efficient for calculating percentiles with large maps. For calculating univariate statistics from a raster map based on vector polygon map and upload statistics to new attribute columns, see v.rast.stats. Category or object oriented statistics can be computed with r.statistics. For floating-point cover map support for this, see the alternative r.stats.zonal. For quantile calculations with support for floating-point cover maps, see the alternative r.stats.quantile.

Raster map algebra and aggregation

The r.mapcalc command provides raster map algebra methods. The r.resamp.stats command resamples raster map layers using various aggregation methods, the r.statistics command aggregates one map based on a second map. r.resamp.interp resamples raster map layers using interpolation.

Regression analysis

Both linear (r.regression.line) and multiple regression (r.regression.multi) are supported.

Hydrologic modeling toolbox

Watershed modeling related modules are r.basins.fill, r.water.outlet, r.watershed, and r.terraflow. Water flow related modules are r.carve, r.drain, r.fill.dir, r.fillnulls, r.flow, and r.topidx. Flooding can be simulated with r.lake. Hydrologic simulation model are available as r.sim.sediment, r.sim.water, and r.topmodel.

Raster format

In GRASS GIS, raster data can be stored as 2D or 3D grids.

2D raster maps

2D rasters support three data types (for technical details, please refer to the Wiki article GRASS raster semantics):

32bit signed integer (CELL),
single-precision floating-point (FCELL), and
double-precision floating-point (DCELL).

In most GRASS GIS resources, 2D raster maps are usually called "raster" maps.

3D raster maps

The 3D raster map type is usually called "3D raster" but other names like "RASTER3D", "voxel", "volume", "GRID3D" or "3d cell" are yet common. 3D rasters support only single- and double-precision floating-point. 3D raster’s single-precision data type is most often called "float", and the double-precision one "double".

No-data management and data portability

GRASS GIS distinguishes NULL and zero. When working with NULL data, it is important to know that operations on NULL cells lead to NULL cells.

The GRASS GIS raster format is architecture independent and portable between 32bit and 64bit machines.

Raster compression

All GRASS GIS raster map types are by default ZSTD compressed if available, otherwise ZLIB compressed. Through the environment variable GRASS_COMPRESSOR the compression method can be set to RLE, ZLIB, LZ4, BZIP2, or ZSTD.

Important: the NULL file compression can be turned off with export GRASS_COMPRESS_NULLS=0. Raster maps with NULL file compression can only be opened with GRASS GIS 7.2.0 or later. NULL file compression for a particular raster map can be managed with r.null -z.

Integer (CELL type) raster maps can be compressed with RLE if the environment variable GRASS_COMPRESSOR exists and is set to RLE. However, this is not recommended.

Floating point (FCELL, DCELL) raster maps never use RLE compression; they are either compressed with ZLIB, LZ4, BZIP2, ZSTD or are uncompressed.

RLE: DEPRECATED Run-Length Encoding, poor compression ratio but fast. It is kept for backwards compatibility to read raster maps created with GRASS 6. It is only used for raster maps of type CELL. FCELL and DCELL maps are never and have never been compressed with RLE.
ZLIB: ZLIB’s deflate is the default compression method for all raster maps, if ZSTD is not available. GRASS GIS 7 uses by default 1 as ZLIB compression level which is the best compromise between speed and compression ratio, also when compared to other available compression methods. Valid levels are in the range [1, 9] and can be set with the environment variable GRASS_ZLIB_LEVEL.
LZ4: LZ4 is a very fast compression method, about as fast as no compression. Decompression is also very fast. The compression ratio is generally higher than for RLE but worse than for ZLIB. LZ4 is recommended if disk space is not a limiting factor.
BZIP2: BZIP2 can provide compression ratios much higher than the other methods, but only for large raster maps (> 10000 columns). For large raster maps, disk space consumption can be reduced by 30 - 50% when using BZIP2 instead of ZLIB’s deflate. BZIP2 is the slowest compression and decompression method. However, if reading from / writing to a storage device is the limiting factor, BZIP2 compression can speed up raster map processing. Be aware that for smaller raster maps, BZIP2 compression ratio can be worse than other compression methods.
ZSTD: ZSTD (Zstandard) provides compression ratios higher than ZLIB but lower than BZIP2 (for large data). ZSTD compresses up to 4x faster than ZLIB, and usually decompresses 6x faster than ZLIB. ZSTD is the default compression method if available.

In the internal cellhd file, the value for "compressed" is 1 for RLE, 2 for ZLIB, 3 for LZ4,4 for BZIP2, and 5 for ZSTD.

Obviously, decompression is controlled by the raster map’s compression, not the environment variable.

SOURCE CODE

Available at: Raster data processing in GRASS GIS source code (history)