OznMon output is inconsistent

[EdwardSafford-NOAA]

The extracted data files (*.ieee_d) are not bitwise identical on successive runs using the same data. The resulting files are the same size but they are not bitwise identical. This should not be so.

[EdwardSafford-NOAA]

The issue is a confluence of a lack of clarity from the NetCDF formatted ozone diagnostic stat files and some bumbling logic on my part.

Here are the results from > ncdump -h netcdf diag_gome_metop-a_anl.2021042106

netcdf diag_gome_metop-a_anl.2021042106 {
dimensions:
        nobs = UNLIMITED ; // (2668 currently)
variables:
        int MPI_Task_Number(nobs) ;
        float Latitude(nobs) ;
        float Longitude(nobs) ;
        float Time(nobs) ;
        float Reference_Pressure(nobs) ;
        int Analysis_Use_Flag(nobs) ;
        float Observation(nobs) ;
        float Inverse_Observation_Error(nobs) ;
        float Obs_Minus_Forecast_adjusted(nobs) ;
        float Obs_Minus_Forecast_unadjusted(nobs) ;
        float Solar_Zenith_Angle(nobs) ;
        float Scan_Position(nobs) ;
        float Row_Anomaly_Index(nobs) ;

// global attributes:
                :date_time = 2021042106 ;
                :Satellite_Sensor = "gome_metop-a" ;
                :Satellite = "metop-a" ;
                :Observation_type = "gome" ;
                :pobs = 0. ;
                :gross = 120. ;
                :tnoise = 2.236 ;
}

The key point is that nobs is the number of observations * number of levels. Currently all ozone instruments have only 1 level except for sbuv2_n19 and ompsnp_npp; both of those have 22 levels.

The lat, lon, and time variables are size nobs but all levels within an obs have the same value. In the OznMon's extraction executables the lat, lon, and time values are stored in a structure that is of size nrecords, which is nobs/nlevel. This is done to maintain internal consistency with the binary formatted diag files and because we don't really need to store 22 redundant values.

Reading the lat, lon, and time vars was the source of the problem. The logic was faulty leading to the assignment of values to the internal lat, lon, and times arrays that were beyond the bounds of the structures that held the data following the read operation. The junk that ended up in the lat, lon, and time arrays didn't match in successive runs using the same test data. (Actually they matched up pretty well given the circumstances.) That logic has been simplified, and the in-code comments expanded to assist in the future. Additionally I found an error in the process of reading the other vars in the diag file (which do have different values for each obs number, level), and fixed that as well.

Successive runs have been made using the same data. The results are bit-wise identical. Additionally samples of the extracted data has been dumped and compared to the output from the ncdump utility. The results match for data sources with both 1 level and 22 levels. Once testing has been completed on other platforms I'll issue a PR.