Large raster data processing
Large File Support (LFS)
Affects 32bit systems which are normally limited to 2GB (2^31) per file. As workaround, LFS can be enabled in GRASS GIS at compile time.
- GRASS GIS 6: much of the raster library and modules can be enabled for LFS during compilation
- GRASS GIS 7: raster and vector libraries are LFS enabled.
See also: Large File Support
Workaround: Select a smaller area within a very large dataset before importing in GRASS
Suppose that we have all ASTER GDEM world coverage (>22000 files) and we aim to build a mosaic of Europe and import it in GRASS. A nice reference on how to deal with ASTER GDEM is here. The very first step is to select among the >22000 files those covering our area of interest (Europe). To this aim, we can use use gdaltindex to create an index of all the files:
gdaltindex an_index.shp *.tif
This command will create a polygon shapefile with the footprint of each raster as the polygon shape and the name of the image files represented in the attribute table. After that, we just need to do a spatial select on the shapefile and extract the filenames from the attributes from the selected polygons:
ogr2ogr -f CSV list.csv an_index.shp -spat xmin ymin xmax ymax
will produce a CSV file, from which we can easily copy the list of files in a list.txt.
Here, our area of interest (subset) has the following boundaries:
north: N71: ymax = 72.0001389 south: N34: ymin = 33.9998611 west: W011: xmin = -11.0001389 east: E042: xmax = 43.0001389
At 100m this corresponds to 27.8 billion cells.
Raster map precision types
- CELL DATA TYPE: a raster map from INTEGER type (whole numbers only)
- FCELL DATA TYPE: a raster map from FLOAT type (4 bytes, 7-9 digits precision)
- DCELL DATA TYPE: a raster map from DOUBLE type (8 bytes, 15-17 digits precision)
- NULL: represents "no data" in raster maps, to be distinguished from 0 (zero) data value
- INTEGER MAP: see CELL DATA TYPE
- FLOAT MAP: see FCELL DATA TYPE
- DOUBLE MAP: see DCELL DATA TYPE
See also GRASS raster semantics
Create a mosaic
Now that we have a list of the files interesting our area, and since our files are all in the same datum (WGS84), we can use gdalbuildvrt to create a virtual mosaic. This will take only a few seconds to run.
gdalbuildvrt -input_file_list list.csv mosaic.vrt
Also note that gdal can read Tifs within a zip/gz-file as well.
To import the mosaic into GRASS, we first need to create a WGS84 location, then:
r.external input=/path/to/mosaic.vrt output=mosaic
Split into tiles
Sometimes it is possible to work with tiles, i.e. splitting a big map into chunks. These tiles can be easily generated with. It optionally allows for tiling with overlap.
Raster map precision: reducing file size
Not always double precision is needed for raster map computations. Strategies:
- consider to run with FCELL rather than DCELL precision (see 's float() function).
- you may also multiply with e.g. 100 and then round() to integer (used e.g. for BIOCLIM and WORLDCLIM).
Functions inrelated to precision:
double(x) convert x to double-precision floating point (DCELL) float(x) convert x to single-precision floating point (FCELL) round(x) round x to nearest integer (INT) round(x,y) round x to nearest multiple of y (INT) round(x,y,z) round x to nearest y*i+z for some integer i (INT)
See also GRASS raster semantics
Number of open files limitation
When working with time series or multispectral data, you may experience the following error message (r.series, r.patch etc.):
Too many open files
WARNING: Unable to open raster map <somemap@mapset> ERROR: One or more input raster maps not found
Many operating systems have the limitation of opening 1024 files at the same time. The limits are just to prevent a process from (accidentally or deliberately) consuming too much memory. The descriptor table is in kernel memory, and can't be swapped to disk. However, this can be enlarged. Note that, in GRASS GIS 7.0, each raster map requires two open files: one for the cell/fcell file, one for the null bitmap.
Resource limits are per-process, and inherited. The ulimit command (which is a shell built-in) changes the limits for the current shell process; the new limit will be inherited by any child processes.
Solution for Linux: System-wise change
On most Linux systems, resource limits are set on login by the pam_limits module according to the settings contained in /etc/security/limits.conf /etc/security/limits.d/*.conf. You should be able to edit those files if you have root privilege (also via sudo), but you will need to log in again as a user before any changes take effect. A reboot shouldn't be necessary since changes to limits.conf should take effect for any subsequent logins. A hard limit can't be increased for any existing non-root processes, or those descended from them.
Increasing a hard limit requires root privilege (or at least the CAP_SYS_RESOURCE capability), so the limits have to be set by the program which manages the login (login, xdm, sshd, etc) after the user has identified themself (so it knows which limits to set) but before the process changes its ownership from root to the logged-in user.
sudo su vim /etc/security/limits.conf
Add in /etc/security/limits.conf something like this:
# Limit user nofile - max number of open files * soft nofile 1500 * hard nofile 1800
Restart the user session (i.e., logout, login).
Solution for Windows: System-wise change
From the Microsoft documentation it is not very clear how many files can be opened simulaneously (FIX IF POSSIBLE), here some hints:
Solution for Linux: Session only change
To change the limits for an existing session, you may be able to use something like:
sudo bash ulimit ... su -c bash <username>
However, sudo will probably reset certain environment variables, particularly LD_LIBRARY_PATH, so you may need to reset those in the new shell.
ERROR: G_malloc: unable to allocate xxxx bytes of memory
Be sure to use GRASS GIS 7 which has way better support for large files.
If you happen to use the "memory" option requesting more than 2GB (e.g. memory=3000) check if you are using 64bit executables since 32bit applications can use only up to 2GB memory. In this case it does not matter if the operating system is 64bit or if there is more than 2GB of memory available. You need to use 64bit binaries of GRASS GIS 7.