Usage

If you run max_diff_nc.sh with argument -h, you will get a short usage message:

$ max_diff_nc.sh -h
Usage: max_diff_nc.sh [OPTION]... file file-or-directory

Compute and locate difference between NetCDF files.

Options:
   -h: this help message
   -s: report identical variables
   -m: compute average order of magnitude
   -i: compare variables with same varid, regardless of variable name
   -q: only report names of variables which differ, without maximum difference
   -v <var>: only compare variable with name <var>

max_diff_nc.sh requires two arguments: the files to compare. If the second argument is a directory then max_diff_nc.sh will look in this directory for a file with the same basename as the first argument. The order of the files is important. The first file is taken as the reference file in order to compute relative differences. If you permute the files, the absolute differences will not change but, in general, the relative differences will change.

Consider a particular variable name (or variable ID, for option -i). The program reads one array of values from each input file, corresponding to that variable name (or variable ID, for option -i). Let us call data_old the array from the first input file and data_new the array from the second input file. We assume the arrays have the same shape (the program will stop if they do not). The program computes the “absolute difference”:

$$|\mathrm{data\_old} - \mathrm{data\_new}|$$

and the “relative difference”:

$$\left|\frac{\mathrm{data\_old} - \mathrm{data\_new}}{\mathrm{data\_old}} \right|$$

The program takes special care of zero values. We construct a mask that tells us, for each position (set of indices in each dimension), whether there is a zero value at that position in either array. That is, in the Fortran program, we define the logical array zero:

zero = data_old == 0. .or. data_new == 0

The program only computes the relative difference where zero is false. For the absolute difference, the program gives two results: a global result, for the whole arrays, and a result for the portions of the arrays where zero is true.