Argo-O2 data Data Management Real-time and Delayed-mode QC Where we are/ Where we go LPO: V. Thierry, T. Bouinot ALTRAN: J. P. Rannou GEOMAR: A. Kortzinger, H. Bittig MBARI: K. Johnson UW: S. Emerson, S. Bushinsky And SCOR WG 142 on Quality Control Procedures for Oxygen and Other Biogeochemical Sensors On Floats and Gliders
Oxygen data in B files: Data format and data management File format Single profile file (DAC): BRwmo_XXX.nc Multiprofiles file with only Bio data (DAC): wmo_bprof.nc but N_PROF profiles as in the core mlt file Trajectory file with only Bio data (DAC): wmo_brtraj.nc Merged multiprofiles file with core and Bio data (GDAC): wmo_mprof.nc
Data format and data management Parameters Final physical parameter: DOXY Intermediate parameters (associated to different calibration equations /sensor described in the cookbook) TEMP_DOXY TEMP_VOLTAGE_DOXY VOLTAGE_DOXY FREQUENCY_DOXY COUNT_DOXY: who does use this parameter? Need to fill the cookbook part RPHASE_DOXY BPHASE_DOXY DPHASE_DOXY TPHASE_DOXY C1PHASE_DOXY, C2PHASE_DOXY MOLAR_DOXY PHASE_DELAY_DOXY MLPL_DOXY
Data format and data management ARGO-O2 processing manual The initial manual has been reformatted for clarification First part details the scientific basis of the oxygen measurements and computation, describe the different sensors and associated calibration equations in general copy/paste of the Oxygen Sensor manuals Second part is the practical part for the DAC. No need to be an expert to read and use this part. It provides the clear computational method for each case and the way to fill meta data and all required fields First finalized version by end of the year Unit used in the cookbook similar to that used in reference Table 3, might be different from that used in the tech files Tech file: mmoles/l, umole/l, umole/kg Cookbook/argo manual: micromole/kg, micromole/l
Data format and data management Proposition for a dedicated web page Current version of the Cookbook + previous version Oxygen sensor Manuals Aanderaa Aanderaa_TD218_OperatingManual_OxygenOptode_3830_RevSep2002.p df SBE Calibration certificate examples : Aanderaa Calibration_Certificate_Sensor_4330_SNo_259_cal_standard_28_coef.pdf SBE Sample Matlab code : Compute_DOXY_for_Aanderaa_4330_optode.zip Test points : DOXY_from_MOLAR_DOXY.xlsx
Data format and data management Issue to be discussed How to compute final physical parameters that need CTD data and intermediate bio parameters available on different vertical sampling scheme? Interpolation? Take the nearest CTD data point? DOXY = fct (TPHASE_DOXY, TEMP_DOXY, PRES, PSAL) PSAL Interpolated on the DOXY grid or nearest point if interpolation not possible
Data format and data management Issue to be discussed Sometimes raw data (0 or 65535/ffff) are obviously out of reasonable range and will lead for sure to useless oxygen (or bio) data: Do we identify such obvious wrong value at the decoding level and set the raw data to fillvalue as suggested by Claudia -> need to define obvious wrong value Other suggestions: add a QC on raw data use/adapt existing RTQC to detect the derived oxygen data from those wrong value (not sure it works well)
Real-time QC tests What is already existing Agreement of real-time QC tests for O2 data in 2012 Global range test + Spike test + Gradient test + Stuck value test Clear efficiency of the tests
Real-time QC tests Implementation AOML BODC on going Coriolis done CSIO CSIRO INCOIS JMA KMA KORDI MEDS NMDIS
Evaluate and update real-time QC tests if necessary/possible Investigate a climatology-based real-time QC test Real-time QC tests Future needs But Good data in real strong gradients (black sea and maybe elsewhere like OMZ) sometimes identified as bad data Tests cannot identified some obvious bad data (O2 = 0 in North-Atlantic, value greater than 400 mumol/kg, etc..) Investigate regional criterions for the range and gradient tests need feedback from users Update, if necessary/possible, existing real-time QC tests
RT oxygen data cannot be used as is Examples shown come from -78 Argo-O2 floats present in the North- Atlantic -Including 39 floats (red points) deployed in the North-Atlantic by LPO with concomitant reference profiles
Correction of QC flag in delayed mode a) b) Data flaggued as bad data on provor/arvor floats with Aanderaa optode when: Profiles acquire few hours after deployment Hook at the base of the profiles (first 50 to 80 meters from the profile depth)
Correction of QC flag in delayed mode Gradient is large (hook) in the first 50 meter of the ascending profile Time response of the sensor? Oxygen gradient as function of the distance from the bottom of the profile X axis: [O2(z)-O2(z+1)] / [P(z)-P(z+1)] Y axis: P(z)-P(bottom of the profile)
Documented DM method Climatology based corrections (Takeshita et al 2013) Reference profile based correction Correction of DOXY concentration following Takeshita method
Climatology based corrections (Takeshita et al., 2013) First-order errors (offset and drift) in profiling float oxygen data can be corrected by comparing float data to a monthly climatology (WOA09) Calculate the mean oxygen difference O2 between the float and the climatology and evaluate the sensor drift as the average slope of the O2 time-series 1901218 < 1 mumol/kg/an
Climatology based corrections (Takeshita et al., 2013) Determine oxygen dependent correction terms in performing a linear regression on [O2] float vs. [O2] WOA or on [%SAT] float vs. [%PSAT] WOA [O2] WOA = CO [O2] + C1 [O2] x [O2] float [%SAT] WOA = CO [%SAT] + C1 [%PSAT] x [%SAT] float Correction on %PSAT is more appropriate as it takes into account changes in Temperature and Salinity 1901218 Iterative processes that eliminate points (for determining the correction) for which the difference between the WOA atlas and the corrected float data is too large (abs(diff) > 2.8 * std(diff ))
DMQC Climatology based corrections (Takeshita et al., 2013)
Reference profile based corrections, following Takeshita method More accurate oxygen-dependent correction terms can be estimated in comparing float data to a reference profile Especially in areas where the climatology is not well defined or where the ocean is subject to large interannual variability like in the North-Atlantic Ocean Shown by low R2 value R2 for float data corrected from WOA
DMQC Reference profile based corrections, following Takeshita method PSAT_ADJUSTED_WOA PSAT_ADJUSTED_INSITU WOA PSAT_ADJUSTED_INSITU
Reference profile based corrections, following Takeshita method R2 for float data corrected by In situ reference profile DOXYADJ IS DOXYADJ WOA = -2.3 +/- 0.6 mumol/kg
We can now start to work with Argo-O2 data
Where we go? All RT O2 data need to be adjusted in delayed mode In situ calibrated reference profile not always available, especially if floats are deployed from ship of opportunity The 1% precision required for many applications (air-sea gas exchanges, oxygen content change) not reached by the climatology adjustment Optode Aanderaa calibration evolves during storage; despite the multicalibration procedure, an offset is still observed at deployment Optode Aanderaa probably drifts while in water Adapt the existing objective analysis tool (ISAS, Gaillard et al, 2008) to O2 Need to improve sensor (aanderaa, SBE63) New method for correcting data
In Situ Analysis System
Do optodes drift while in water? Drift estimated by comparison to a climatology It has to be considered with caution In general, we considered that there was no drift (drift too small compared to sensor accuracy, time series too short) Exemple float 5902298: drift estimated to -4.4 mumol/kg/year 5902298 5902298
Do optodes drift while in water? 5902305 5902305-11 mumol/kg/an
Do optodes drift while in water?
Do optodes drift while in water? Drift estimates for our 39 floats deployed in the North-Atlantic Optode 4330, multipoint calibration Optode 4330, initial manufacturer calibration Optode 3830, initial manufacturer calibration
Promising new methods Various groups (Henry Bittig and Arne Kortzinger; Ken Johnson; Steve Emerson and Seth Bushinsky) are working on those methods In-situ in-air measurements : as drift check as calibration references New calibration equation as optode is a po2 sensor
Correction from in air measurements (Ken Johnson) Calibration by air oxygen measurements. 22 UW/MBARI floats with optodes now make air oxygen measurements, with some records over 2 years long. Typically, one measurement of air made on each profile. Air calibration applied as a gain correction in concentration (should really be done as a gain in phase) O2 correct = gain x O2 raw No evidence of drift, results agree with Winkler values to within 2 mumol/kg near surface.
UW Float 7601 Station Papa, 50 N. Pacific Air %Sat = 90.9 ± 0.9 % (N>200). Sensor gain correction is 1/0.909 = 1.100 (1.106 when summer values filtered out).
UW Float 7601 Station Papa, 50 N. Pacific Float oxygen corrected with air gain value.
Air Gain Corrected SOCCOM floats 40 South 55 South
Comparison of air gain corrected float 1 to Winkler titration values for pre-soccom float 9254 with air gain from only 4 profiles.
Correction from in air measurements (Steve Emerson and Seth Bushinsky) Percent difference between optode measured po 2 and atmospheric po 2. Each symbol represents a different set of surface air measurements from an optode on a stalk 60 cm above a float deployed at Ocean Station Papa. Error bars are +/- 1 standard deviation from the mean is calculated from total atmospheric pressure (P atm ), mole fraction of oxygen in the atmosphere (X o2 ) and an assumed 100% relative humidity (ph 2 O).
Correction from in air measurements (Steve Emerson and Seth Bushinsky) Optode po 2 has been calibrated to an average of the initial 4 months of atmospheric measurements. 4 month averages are plotted as red lines bounded by gray boxes representing the standard deviation of air measurements during that period. Air measurements shown are nighttime measurements when the pressure and temperature were not changing rapidly (<0.1% atmospheric pressure change, 1 deg C temperature change). Over the 1.5 years shown here, optode measurements have drifted at a rate of - 0.4% relative to the atmospheric calibrations.
New calibration equation (H. Bittig and A. Kortzinger) Optode O 2 calculations should be based on po 2 not O 2 conc. Use po 2 -variant of Uchida et al. 2008 model: (or any other Φ x T po 2 function) Need to be implemented by the DAC and documented in the cookbook Use in situ data to adjust the po 2-7 coefficients (DM correction)
Conclusion The management of Argo-O2 data is on a good track. We progress step by step and simple RT and DM QC procedures already exist Unfortunately, Real Time O2 data still cannot be used as is Our aim now is to improve the data quality and to stabilize the O2 computation and delayed mode adjustment. Many on-going work address those issues (time response, drift, adjustment, sensor evaluation and understanding, etc..) and are discussed as part of a SCOR WG on Quality Control Procedures for Oxygen and Other Biogeochemical Sensors On Floats and Gliders Data management finalize the O2 cookbook create a dedicated webpage RT QC tests Basic RT tests exist and work fairly well Review and adapt the existing tests if necessary, require feedback from users
DMQC method Conclusion Delayed mode adjustment based on climatological data provide correction of first order errors (drift and offset) Correction based on In situ data provide better offset adjustment but in situ reference profile not always available In situ Analysis System will provide a complementary tool to detect bad data points and inconsistent data at basin scale Correction from In Air measurements provide accurate corrections, it is a promising new DMQC method Need In Air measurements Where to store those data, in the tech file? Not possible for SBE 63, is it a problem? New calibration equation (po2) DM adjustment of calibration coefficient New equation used by manufacturer?
Conclusion Optode subject to drift During storage phase Probably at sea also, although the observed drift are small so far. We will need more data (time) Sensor time response Better for pumped sensor (SBE63) Need to know measurements time