A Flexible Algorithm for Network Design Based on Information Theory

This code is for the design of an optimal observing network and is written in R. It has been written with networks for atmospheric sampling in mind, however, the algorithm is applicable to any network design problem where the relationship between the variables to be determined and the observations is linear, and the uncertainties are Gaussian distributed. The algorithm uses the metric of maximizing the Information Content of the observations and an incremental optimization method. The code can also be used to determine the locations of new sites when there is an existing network, and for observations of a different type (e.g., isotopes versus mixing ratios).

To use the code, the user needs to supply the following files:

• A list of potential observation sites, i.e., those from which to select, in ascii format (csv). This list needs to include the location of the sites as well as the measurement uncertainties (see example).
• A list of existing sites (if any) in ascii format (csv) (see example).
• The relationship between the unknown variables and the potential observations (as well as for existing observations, if any). These are the Jacobian matrices, or in atmospheric sciences, source-receptor relationships, SRRs. The default is one file (in NetCDF format) per observation corresponding to the potential and, if any, the existing sites.
• A prior estimate of the unknown variables (for atmospheric sciences, the fluxes)  in NetCDF format.
• A land-sea mask (if only land variables are to be considered) in NetCDF format.
• A region mask (if only a region is to be considered) in NetCDF format.

The paths, files and settings are specified in the file “network_config.r”, which is called by the main script “optimize_network.r”.