Skip to content

Latest commit

 

History

History
114 lines (81 loc) · 4.52 KB

README.md

File metadata and controls

114 lines (81 loc) · 4.52 KB
Raw

stormwavecluster

This is a collection of R scripts and data used for our statistical modelling of coastal storm waves, which is described in this_paper.

Structure

The sub-folders are:

./Analysis - 'driver' scripts used to perform the high level analysis, and an associated vignette.

./Data - location for data used in the analysis.

./R - self-contained R scripts used for the analysis. These include:

  • nhpoisp -- Code for fitting non-homogeneous Poisson processes

  • wave_dispersion -- Code to solve the airy wave dispersion relation

  • nearest_index_sorted -- Fast Rcpp code for doing a nearest-neighbour search on a sorted numeric vector. This is a simple task, but we were unable to find a similarly fast and memory efficient way of doing this in native R.

  • evmix_fit -- Convenience wrappers for fitting extreme value mixture models

  • tpxo7.2 -- Convenience interface for the tpxo7.2 tidal prediction software. The latter must be installed separately.

Information on using and testing the above codes is provided within their directories.

Dependencies

The tpxo7.2 package should be installed to use the tpxo7.2 interface. However, the remainder of the code and analysis will run fine even if this is not done, so if you just want to look at the statistical analysis, this can be skipped.

The code relies on many different R packages, which can be obtained within R using the command:

install.packages(c( 
    'knitr', 'devtools', 'Rcpp',  'forecast',
    'CDVine', 'evmix', 'MCMCpack', 'numDeriv', 
    'optimx', 'logspline', 'ismev', 'VineCopula',
    'Matching', 'TwoCop'))

At the time of writing there is a bug (of significance for our analysis) in the version of VineCopula available on CRAN. This version was downloaded with the above command. The bug is fixed in a version of VineCopula on github, which can be obtained within R using the command:

devtools::install_github("tnagler/VineCopula")

The above requires that you are in an environment where you can build R packages. This is usually automatically supported on linux and mac, but may require pre-installation of Rtools on windows, see: https://cran.r-project.org/bin/windows/Rtools/ .

Testing

To run all the tests for the 'R' folder, start R and execute the test_all.R script in the current directory:

    source('test_all.R')

This may take awhile to run (currently the extreme value mixture model tests take tens of minutes on my machine, as do the tests of non-homogeneous-Poisson process fitting).

If you are only interested in a single piece of code in the R folder, then look there for a corresponding test script.

Advice on adapting the analysis to another site

The Analysis code is not expected to be applied to another site without modification. This is because it includes choices which may not generalise well to other locations (e.g. which probability distributions best fit a site), and hard-coded site specific details (e.g. time-zones, assumptions about structure of source data, etc.). Also, it relies heavily on non-linear optimization techniques, which may need tweaking to converge with other datasets.

Therefore, we strongly advise against its use as a black-box code. Nonetheless, it is provided here to be useful as:

  • a source of examples using routines in ./R, which are more generic
  • an example analysis which can be adapted to another site by an experienced user
  • to make our work more reproducible and transparent

Bugs, maintainence and contributions

Bugs relating can be raised on the 'issues' page of the github site. Consider making a github pull request with any fixes. However, if you would like to make a major contribution, you should discuss this with the package maintainer first ([email protected]) to ensure it will be accepted.

Acknowledgements

Development of this code was supported through the Bushfire and Natural Hazards Cooperative Research Centre, as part of the project Resilience to clustered disaster events on the coast: Storm Surge.

This project would be impractical without the efforts of the community who develop and maintain the R software, and its large ecosystem of packages.