jdherman / r-sensitivity-wrapper

Command-line wrapper for the R sensitivity analysis package

Command-line wrapper for the R Sensitivity Package

(To be used with Sobol, Method of Morris, and FAST Methods)

Requires the R sensitivity package created by Pujol et al. You can get the package here. For more detailed installation instructions, see this post.

CLI wrapper developed for the following papers:

Herman, J.D., J.B. Kollat, P.M. Reed, and T. Wagener. 2013. From maps to movies: high resolution time-varying sensitivity analysis for spatially distributed watershed models. Hydrology and Earth System Sciences Discussions, 10, 10775–10808. Herman, J.D., J.B. Kollat, P.M. Reed, and T. Wagener. 2013. Technical note: Method of Morris effectively reduces the computational demands of global sensitivity analysis for distributed watershed models. Hydrology and Earth System Sciences, 17, 2893–2903.

Parameter files should be created with 3 columns, name, lower bound, and upper bound, e.g.:

P1 0.0 1.0
P2 0.0 5.0
...etc.

Contents:

• sampler.R: generate parameter samples. Example usage:

   Rscript --vanilla --slave sampler.R <method> <N> <parameterFile> > <sampleFile>

• analyzer.R: perform sensitivity analysis on model results. Example:

   Rscript --vanilla --slave analyzer.R <method> <N> <parameterFile> <modelOutputFile> <columnNumber> > <resultsFile>

• Sobol-G Function and other test functions. These are included just for examples of performing decoupled sensitivity analysis. (Sampling -> Evaluations -> Analysis).

Some notes:

• For each of the above files, <method> can be sobol, morris, or fast. Each of these gives a different output format during the analysis step.
• <N> is the number of original samples, which is not the same as the number of parameter sets in sampleFile (these are multiplied approximately by a factor of P, where there are P parameters)
• By default, samples and results will be printed to stdout. In the examples above they are piped into files.
• <columnNumber> starts at 1, not 0. If you do not provide a column number, the default is 1

The full process would look something like this:

module load R

# Perform sampling
Rscript --vanilla --slave sampler.R ${METHOD} ${N} ${PARAM_FILE} > ${NAME}.samples

# Run the model executable with these samples (assuming the model receives parameters over stdin and prints output to stdout, your mileage may vary).
./myModel < ${NAME}.samples > ${NAME}.modelout

# Perform the sensitivity analysis using this model output (column 1)
Rscript --vanilla --slave analyzer.R ${METHOD} ${N} ${PARAM_FILE} ${NAME}.modelout 1 > ${NAME}.results