|
Sections below: Float counts | Meta Files | Tech files | Plots & listings | Surface Pressure | TNPD Argo file format and Surface Pressure correction compliance AuditSpecific comments on this audit, based on GDAC at 1 November 2011The number of floats for each DAC rose as expected, except for Coriolis which actually had 38 fewer floats counted. This could be due to the change in GDAC used, or files removed, or more files with critical faults so they were rejected at the very first stages of the audit process. Coriolis in particular has worked on changing the surface pressure parameter name on a large number of floats. This audit has now ceased accepting "Pres_SurfaceOffsetNotTruncated_dBar" being used with auto-correcting PROVOR floats. Also a large number of floats corrected by reprocessing, particularly at Coriolis. This also affects INCOIS PROVOR floats, which probably accounts for the apparent worsening of statistics there. varying pressure correction in one profilePIs were contacted with details of non-constant PRES-PRES_ADJUSTED - ie the pressure correction appears to varies throughout a profile. These cases should be examined as they indicate something has gone wrong (some PIs already fixed their cases.) multiple SPO namesA jump in floats with multiple SPO names, especially AOML. (due to partial conversion to new names, or deliberate choice??) missing PRES_ADJUSTEDThis appears to be a significant problem. Using our decoding software there are about 140,000 profiles which lack valid PRES_ADJUSTED or PRES_ADJUSTED_QC fields. Refined assessment of "missing PRES_ADJUSTED". Many of these actually arise from null characters in the PRES_ADJUSTED_QC field. Where a char vector contains even one null, the Matlab str2num function returns an empty vector. Correction of this will free up a large amount of extra data, especially from AOML, both for users and to be audited. Some also arise from fillval in PRES_ADJUSTED where QC=3 (happens in CSIRO R files.) Is this wrong?? If not, need to change my test to qc>=3 instead of qc>=4. Recommend DACs trace and correct QC values which do not convert to numerals (ie only numeric values between 48 and 57 are valid.) A listing of all affected files will be provided. The fillvalue (space, numeric 32) should only be found in unfilled variables, and null (numeric 0) should never be present. Lists of these cases for each DAC can be accessed via the table at the bottom of the Plots Index Tables imperfect TNPD commentsCompletely fixed at CSIO and BODC, progress at MEDS and KMA with new processing but a lot of old files still to be changed. Change of GDACThis audit is the first to be based on a copy of the Coriolis GDAC rather than USGODAE. This should make no difference. Jeff Dunn, Nov 2011 CSIRO Argo Technical and Meta file compliance AuditThis page is automatically created by code which scans all files in the Argo GDAC. Previous versions can be accessed via the CSIRO DMQC page The audit purpose and method is described here Compliance summary of this audit This analysis was based on GDAC as at 01-Nov-2011 Float counts in DACs
Meta files
tech files
Plots and listings analysing the Surface Pressure Offset
correction for
all floats for each DAC are accessed through the
Plots Index Tables. Separate tables link to all floats, or only floats
with apparent SP correction anomalies, or only floats with apparent TNPD.
Note: The following analyses were based on assumptions about
behaviour of different types of floats, where "type" was here primarily
defined by the Surface Pressure Offset parameter name. Some of these
assumptions will be wrong, and some types are not analysed at all because I
am unsure of their behaviour. It is probably necessary to also take basic float
model (eg Apex or SOLO) into account to fully determine the pressure correction
treatment required for a float, but it would be better if this could be fully
determined by just the parameter name. My assumptions are recorded in this
table.
Surface Pressure OffsetWe have attempted to test for correct Surface Pressure (SP) correction of profile pressure. The difference between PRES and PRES_ADJUSTED is the DAC-applied pressure correction (assuming PRES is the raw P as reported by the float.) Correct application of this test depends on being able to deduce the correct float type and P adjustment from the meta and tech files. However this is not always possible due to missing or non-standard entries in these files. Our computed pressure correction is p_calc = PRES - (SP - adjustment)where adjustment is the known SP adjustment for a given float type (usually 5dbar). For Apex floats the SP reported with a profile actually relates to the previous profile, because it was recorded just as the float begins to descend. RT profile pressure can only be adjusted with the SP reported with the same profile, but in DM profiles the correct SP is used (ie for profile N, p_calc(N) = PRES(N) - (SP(N+1) - adjustment)So, in the provided plots , for floats which do not auto-correct, the "Calculated Correction" (CC) is For DM profiles: CC = SP(N+1) - adjustment
For RT profiles: CC = SP(N) - adjustment
For this reason, CC will differ from Raw Surface Pressure by the "offset", and
will be displaced by one profile for DM profiles. The "DAC adjustment" should
however agree with the CC.
Interpretting the plots: For complete agreement in pressure correction, the "DAC Adjustment" (blue squares), should agree with the CC (red stars). Missing "DAC Adjustment" symbols can occur where there is no PRES_ADJUSTED profile (eg if QC=4). Missing CC symbols can occur where the DAC has provided an SP estimate (possibly by interpolation) but the automated process has not [due to different processing or because the tech files are not quite up-to-date.] Disagreement is likewise visible where "MPP minus CC" (cyan diamond) does not overlay "Min Corrected Pressure" (magenta square). The latter quantity is just the near-surface PRES_ADJUSTED value. Note that the raw SP symbols (blue diamond) are shown exactly as found in the tech files. That is, their x-coordinate is the cycle number corresponding to that value in the tech files. A light gray background is used where disagreement is between 0.1 and 1.0 dbar. This occurs frequently due to DAC smoothing of the SP series, or for other reasons. A dark gray background indicates disagreement greater than 1.0 dbar. Some of these cases are due to deficiencies in our automatic procedures, but some others appear to be related to processing faults at the DACs. An olive green background behind the CC value indicates an alternative correction method (such as allowed for SOLO floats) gives better agreement with the DAC value than the primary method. The CC value shown is then as derived by the alternative method. For example, SOLO floats are auto-correcting, so the primary method is "no correction". However, a secondary method - the normal SP(n+1) correction - may be applied in DM if there is evidence of strong drift. A very pale green background indicates that the DAC has explained how their adjustment is correct, or pointed out that the CSIRO audit software has made a mistake. Any disagreement in these profile is ignored when collating statistics.
See the table which summarizes the results of this pressure correction audit for all DACs. Lists: for each DAC a text file (accessable through the Plots Index Tables) lists the result of checking the surface pressure correction for every float. The floats are grouped by PI. The lists give the number of DM profiles, total number of profiles, assumed surface pressure offset, and counts of profiles which agree exactly or disagree with my automated correction. Please note that in many cases the disagreement is because your DAC correction is a better choice than my automated one. Format example: 5900033 DM 264/295 SPoffs=5 Diffs(dbar) =0[271] <1 [0] >1 [ 1 *] Of 295 profiles for this float, 264 have been DM processed. 271 profiles exactly
agree with my automated correction, 0 disagree by a small amount, but 1
disagrees by more than 1 dbar.
TNPDTruncated Negative Drifting Pressure (TNPD) is a term used to describe Apex floats which truncate negative surface pressure values (all firmware prior to APF9) and the pressure sensor develops a strong negative drift. The usual cause of strong negative drift in these sensors is the "microleak" fault associated with Druck sensors. The pressure in profiles affected by TNPD cannot be corrected for surface pressure offset. The first step in detecting TNPD is to determine if a float is of the type where such an fault might occur. We have attempted to do this using the tech and meta files alone, but this is not always possible. The next step is to find where the reported SP (corrected for 5dbar adjustment if necessary) has "flat spots" on zero, indicating that negative values are being truncated. Previously a float with 80% SP=0 was declared to be TNPD. We now say that any susceptible float is in a TNPD state after the last believable positive SP value. This is because the strong negative drift condition is typically monotonic, so if such a condition has commenced then typical variability will not give rise to small positive SP values. The main difficulty of this approach [assuming the basic assumptions are correct] is determining when isolated positive values are natural perturbations about a small negative value, or when they are just sensor error spikes. In the present analysis I have used (for the first time) the crude test that a positive value > 1dbar which has 2 zero values immediately either side of it, is a spike that can be ignored. [This differs to the standard despiking process used to prepare the SP series for pressure correction because here we just want a simple way of assessing where an SP series is devoid of true +ve values.] Obviously this is very imperfect - there is no substitute for skilled DM analysis, and in very many cases the DAC will be right and our algorithm will be wrong. Another deficiency of our algorithm is that is [apparently] extremely unlikely that microleaking will start after years of correct operation. For example, if SP is 0 for the last 5 of 200 profiles then the algorithm says TNPD commenced at profile 196. The time of onset of TNPD is examined in this set of TNPD plots. That page also includes plots which examining whether, for different sensor types, a large negative drift can be followed by recovery to plausible positive SP values. "Start of TNPD" in the surface pressure plots provided indicates where TNPD appears to commences, according to the above algorithm, if that is at least 6 months before the last DM profile. If the condition is met with less than 6 months of subsequent DM profiles, then 'Potential TNPD' is displayed. Where some of the float series has been DM processed, extra information is provided for the DM profiles in an extra frame at the top of the figure. See the table which summarizes the results of this TNPD treatment audit for all DACs. Lists: for each DAC two text files (accessable through the
Plots Index Tables) list the result
of checking for TNPD. The floats are grouped by PI.
The "short" list gives the cycle number ("pn") where TNPD appears to start,
and counts of profiles which appear to require the TNPD comment string and do not
have it ("missed"),
or do have the comment string ("good"), or have a comment string which is not
precisely the standard string ("imperfect"). The list also notes where we have
NOT detected TNPD but have found the TNPD comment string.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Legal Notice and Disclaimer | Copyright | Privacy | Page author: Jeff Dunn | Contact: Jeff.Dunn@csiro.au