Code for calculating uncertainties on satellite based land cover, as described in the following papers:
usage: bayeslc [options] control_file.txt > output.txt
The control file must be the last word on the command line.
Options are:
-d %f set the correlation factor to %f
-l %d set the length for the conditional auto regressive model to %d
-n %d set the number of simulations to %d (the higher the better)
-u display this message
The format of the control file is a series of ascii text lines each specifying user defined input and output filenames. Anything after a # character is treated as a comment. Blank lines, or lines consisting only of a comment are ignored. Excluding these the following information is expected, in order:
Line 1: a list of N files containing the input LC proportions
Line 2: a mask file containing 0 where a mask is to be applied and 1 elsewhere
Line 3: a file containing an NxN confusion matrix
Line 4: a file containing the number of "counts"
Line 5: a list of N files to contain the output means
Line 6: a list of N files to contain the output stdvs
Aside from the confusion matrix (which is an NxN ascii matrix) all the input files are plain, headerless ascii text files containing a single matrix, each of which must have the same dimensions as the rest.
The total area of each land cover class for each simulation is written to the stdout.
The example_data directory contains input data for a small portion of southern Africa. This can be run using the following command:
bayeslc -n 100 control.txt > ouput.txt
The file control.txt contains comments explaining the content of each line. Increase the value after the -n to produce increase
the number of samples, the greater the number the better represented the posterior distribution will be, but the longer
the run will take. In Quaife and Cripps (2016) we used a value of 10000.
A quick and dirty python script showImg.py is included to view the input and output maps.
This is open source software (see the license for full details). You can use it without any obligation to the authors. However, if you want help implementing it for a given application we're happy to discuss.
I have only ever tried installing on Linux. The only dependency is the GNU Scientific Library (GSL). As long as you have the GSL and standard development tools installed (specifically gcc and make) it should just be a case of typing make.