Problems with radsstat

[leuliett]

radsstat has two errors: 1) The maximum value is always nan and 2) the option --no-stddev suppresses the output from --minmax.

1. The commit
   57794bc sets z(:,1:) = nan, which causes the max value to always be nan:

swift:~$ radsstat -Sj3 -l -C1 -c
# Statistics of RADS variables (box weight)
# Created: 2021-12-28 16:34:39 UTC: radsstat -Sj3 -l -C1 -c
#
# Satellite : j3/a
# Cycles    :    1 -    1
# Passes    :    1 -  254
#
# Output columns:
#    ( 1) cycle
#    ( 2) date at beginning of cycle [YYMMDD]
#    ( 3) nr of measurements
#    ( 4) mean time [seconds since 1985-01-01 00:00:00 UTC]
# ( 5- 8) mean, stddev, min and max of sea level anomaly [m]
  1 160217   540421  982747942.   0.04301   0.11024  -1.93600       NaN

2. The parameter --no-stddev introduced in commit 95a7f35 also suppresses the output from --minmax.

swift:~$ radsstat -Sj3 -l -C1 -c --no-stddev
# Statistics of RADS variables (box weight)
# Created: 2021-12-28 16:36:33 UTC: radsstat -Sj3 -l -C1 -c --no-stddev
#
# Satellite : j3/a
# Cycles    :    1 -    1
# Passes    :    1 -  254
#
# Output columns:
#    ( 1) cycle
#    ( 2) date at beginning of cycle [YYMMDD]
#    ( 3) nr of measurements
#    ( 4) mean time [seconds since 1985-01-01 00:00:00 UTC]
#    ( 5) mean of sea level anomaly [m]
  1 160217   540421  982747942.   0.04301

[remkos]

I can confirm that both these bugs are present.

[remkos]

I think number 1 is not coming from the change referred to, but rather from the fact that the IEEE fortran standard, nor gfortran, clearly defines what the result of max(x,y) is when either or both x and y are Nan.

My experiments with gfortran shows that

• max(NaN,x) returns NaN
• max(x,NaN) returns x

In most RADS programs, xmin and xmax are first initialised as NaN and then updated using

xmin = min(xmin, x)
xmax = max(xmax, x)

This would mean that both xmin and xmax remain NaN. I do not understand why xmin does appear to be updated; potentially because of the undefined nature of the routine. Using the apparent implementation of min and max in gfortran, this could then be replaced by

xmin = min(x, xmin)
xmax = max(x, xmax)

But maybe better is to avoid the compiler-dependent implementation and go for:

if (.not.(x > xmin)) xmin = x
if (.not.(x < xmax)) xmax = x

[remkos]

Returning assignment to @leuliett to check implementation as per commits above and close this issue.

[leuliett]

Confirm that both issues are fixed by these commits.