Awesome

TSAR

Time-Series computation Automation for Raster data

These scripts offer functionality to:

• compute multi-temporal statistics
• apply transforming computations (e.g. time series smoothing)

over the third (i.e. time) dimension of 3D raster stacks.

Input is a raster stack which might be pointing to a single file or virtually to a list of files with same spatial extent. A master process divides the number of lines by the number of parallel processes and passes each instructions on which subset to read from the stack. Each process then reads its subset and computes a set of functions (workers) over the time dimension.

Once finished the computation results are gathered by the master process and combined along the y-axis.

The defined workers take a vector as input (i.e. a pixel time series) and can output a single value (e.g. a multi-temporal statistic) or several values (e.g. time series smoothing). The number of resulting bands is checked for a single pixel before the parallel processing is started to determine whether any of the resulting files might already exist. If all files computed by one worker already exist and no file is to be overwritten then this worker is deleted from the list and not passed to the individual processes.

The names of the computed bands can either be explicitely defined by the user or are otherwise generated from the names of the workers. In this case, if individual workers return several values, each band will carry a numeric suffix (e.g. worker1_1, worker1_2, worker2_1, etc.).

Depending on whether the output is to be a single file or separate files, an ENVI stack or several GeoTiff files are written. In case an ENVI stack is written, the file itself is named as defined by out.name and the bandnames are written to the HDR file. If individual GeoTiffs are written, out.name is going to be a directory with the bandnames being the names of the GeoTiff files.

Examples

# load a raster stack
ras = raster::stack(c("timestep1.tif", "timestep2.tif", "timestep3.tif"))

# define the NA value; this might not be necessary if the value is stored in the files
# the input files need to all have the same NA value
raster::NAvalue(ras)=-99

# define a list of functions with names
workers = list(minimum=function(x)return(min(x, na.rm=T)),
               average=function(x)return(mean(x, na.rm=T)),
               maximum=function(x)return(max(x, na.rm=T)),
               quantiles=function(x)return(quantile(x,probs=c(.25, .5, .75),na.rm=T,names=F))))

# define the cluster setup
# In this case a total of 30 processes will be started: three nodes with 10 processes each
# the node list is expected to contain a number of node names accessible via SSH without password
processes = 10
nodes = c("node1", "node2", "node3")

# the scheme for writing the results
# since separate is true, outname will be a directory containing several GeoTiff files
outname = "/path/to/write"
separate = TRUE

tsar::tsar(raster.name=ras, workers=workers, cores=processes, out.name=outname, 
           separate=separate, na.out=-99, overwrite=FALSE, verbose=TRUE, nodelist=nodes)

# the following files would be written:
# /path/to/write/minimum.tif
# /path/to/write/average.tif
# /path/to/write/maximum.tif
# /path/to/write/quantiles_1.tif
# /path/to/write/quantiles_2.tif
# /path/to/write/quantiles_3.tif

# alternatively, if separate=F, a single ENVI stack would be written:
# /path/to/write
# /path/to/write.hdr

# in this scenario the band names of the ENVI stack would be the same as the names of 
# the single GeoTiffs (i.e. minimum, average, ...)

# the automatically generated output names can be overridden by parameter out.bandnames in tsar::tsar

Installation

This package is still under development and not yet officially registered on CRAN. You can install it like this:

library(devtools)
install_github("johntruckenbrodt/tsar")