The Aquarius L2 dataset is a complex dataset requiring quite some expertise in working with remote sensing data, in part because of the complexity of the flagging regimen that is used.
The latter reflect the fact that Aquarius is a proof of concept mission for remotely sensed salinity observation, and there remain cal/val investigations that are ongoing by the Aquarius science team and associated research community.
Working correctly with these L2 data requires some expertise and familiarity with the data. In addition to the user guide, you may want to consult the algorithm description document (ATBD) and v2.0 validation analysis document in particular. All available Technical documentation can be accessed from https://podaac.jpl.nasa.gov/aquarius.
The third dimension (=4) in the RadiometerFlag 3d-array variable (data block 4083 x beams 3 x 4) is "MaxRadiometer Flags".
This is defined in Table 10 of the user guide (p48) … see below and attached figure Table10extract.png
radiometer_flags: Unsigned Integer (32-bit) 4083 x 3 x 4 (Blocks x Beams x Max. Radiometer Flags)
Radiometer data quality flags:
Each bit represents a data quality condition that was detected for that beam and block. For each flag condition, *** up to Max Radiometer Flags (last array dimension) individual flags are set per beam ***. ***Table 12 presents the condition associated with each flag, including the thresholds used; the meaning of the last dimension***; the use as a flag or mask at Level-2, and the associated radiometer science data fields. Moderate and severe flags are mutually exclusive (e.g., the moderate flag is 0 if the severe flag is set). Any unused array elements are set to 0. This array has attributes that provide the names of the algorithms used in determining the setting of the flags. The algorithms associated with these names are described in “Flag Documentation”, D. M. Le Vine (17 July 2012 draft).
Further qualifications of what MaxFlags dimension is for each radiometer flag condition (0-31) given in table 12 of the user guide on p50.. (see attached figure Table12extract.png)
Typically the third dimension =4 reflects the number of polarizations per beam (3) and block (4093).
Regarding the interpretation of the typically large values in each cell (different from the assigned flag condition bit values spanning 0-31 in table 12 as you point out):
the values in cells of the 3D flag array represent combinations of those bitflag conditions that apply for a given data block (with 1.44s sampling interval), beam and polarization.
The flag values are bitpacked, and thus will need to be unpack to get the combination of individual flags applied to that data array cell for a given block,beam, polarization to then filter against.
So the procedure is to:
1) first, unpack the bits encoded in that single longinteger array cell value to figure out what combination of flag conditions are indicated for that data cell.
2) test cells for unacceptable flag conditions: depending on your specific analysis purpose and region of interest etc, you may need to filter out data elements from the complete data array that you do not want to include in your analysis because the likelihood of the associated retrieved salinity value for that cell is not accurate or has high error associated with it. For example, you may want to exclude salinity values for data cells where windspeed is flagged as high or RFI is flagged as high because the these conditions are not conducive to valid salinity retrieval. If you do not filter such values out, then when you compute say regional salinity statistics you may get spurious results.
One would likely apply filters (based on desired flag conditions compared to unpacked bitflag values in this array) uniformly across all block x beam x polarization dimensions. However, the flexibility is there given the structure of the radiometer flag variable to test specific conditions and have different condition filter sets for say different beams.
Again figuring out the appropriate combinations of flag conditions to filter out data useful for your needs will take quite some experimentation and real expertise/understanding of the Aquarius L2 dataset on your part. In developing the Aquarius L3 products, the Aquarius science team have applied the combination of flags and mask settings based on their experience that will yield mapped data best serving the needs of the typical user. Unless you absolutely need to work with the swath data and have a good understanding of the Aquarius data and its limitations, the we recommend that you consider working with the L3 mapped data.