Status of Trajectory Files - Argo · 2014-03-21 · –Checks of available Pres vs Time information...
Transcript of Status of Trajectory Files - Argo · 2014-03-21 · –Checks of available Pres vs Time information...
Status of Trajectory Files
Megan Scanderbeg
Argo Steering Team Meeting
March 2014
Outline
• Progress on V3.0 trajectory files since AST-14
• Update on status of current traj files in North Pacific Ocean from viewpoint of calculating velocities
• Thoughts on real time and delayed mode traj files
Trajectory V3.0 since AST 14
• Trajectory V3.0 format was approved and is in User Manual
• Agreed upon using two-file trajectory system –
WMO_Rtraj.nc and WMO_Dtraj.nc files
• Agreed to split into Core-Argo and B-Argo trajectory files in a
similar fashion to profile files
• DAC cookbook has been published with instructions on how
to create trajectory V3.0 files, including what Measurement
Codes should be included for all current float types
• Webpage created with explanation of new trajectory V3.0 file
format (http://www-argo.ucsd.edu/Traj3files.html)
Timeline for Traj V3.0 to get to GDACs
• Some DACs/PIs have begun creating both real time and delayed mode V3.0 trajectory files in preparation for uploading to DACs/GDACs – BODC, CSIRO, J-P. Rannou, J. Gilson have all created
test files
– Test files for Provor floats done by Rannou can be found at: ftp://ftp.ifremer.fr/ifremer/argo/etc/coriolis-custom/argo-andro-data/
• Most DACs are focusing on switching prof and meta files to V3.0 and then will work on traj files
• Waiting until GDACs can accept V3.0 trajectory files
Creating D V3.0 traj files from ANDRO • What Rannou does to create “D” files from ANDRO
(through 2009): – Checks positions through re-decoding hex messages, additional
qc to remove bad positions – Checks cycle number – Checks Pres/Temp measurements collected during drift phase;
no flagging has been done if determined “bad” – Checks cycle timing variables – Checks of available Pres vs Time information during vertical
phase
• What still needs to be done by PI or Rannou with specific instructions: – Check salinity and adjust if necessary – Flag bad Pres/Temp/Psal during drift – Apply surface pressure adjustments on non auto-correcting
floats ( must be consistent with prof files?? ) – Figure out how to add back in recovered cycles – will DAC create
new prof and traj files to accommodate this? J-P says that 2-3% of cycles are lost in the current netcdf files
Recent Research Papers where Argo traj files used to calculate velocities
• Ollitrault & Colin de Verdiere, 2014 – ANDRO • Ollitrault & Rannou , 2013 – ANDRO • Katsumata, 2010 – Gridding of YoMaHa
• Bostock, 2013 – South Pacific • Kessler, 2013 – Coral Sea • Park, 2013 – Japan/East Sea • Cravatte, 2012 – South Pacific • Czeschel, 2011 – South Pacific • Menna & Poulain, 2010 – Med Sea • Thaillandier, 2010 – assimilating traj files into
Med Sea forecasting • Voet, 2010 – Nordic Seas
Approaching trajectories from user perspective
• Calculating velocities at 1000m in North Pacific Ocean
• Right: A typical year (2010) of Argo coverage
• Doing extrapolation
based on background velocity and inertial current (Park et al, 2004) to estimate location of rise and fall of float • Extrapolation method relies on knowing the times of the rise and fall (AET/TST and TET/DST) and the surface locations
Floats in North Pacific Ocean
– 1393 APEX ( 1217 with Argos, 176 with Iridium )
– 336 SOLO ( 262 SOLO, 74 SOLO-II with Iridium )
– 154 PROVOR
– 103 ARVOR
– 74 NAVIS with Iridium
– 40 Other – NEMO, NINJA, PALACE, ALACE
DAC # of floats
AOML 1135
Coriolis 25
CSIO 50
CSIRO 5
JMA 878
Total 2097
KMA 73
KORDI 38
MEDS 71
NMDIS 2
Total 2285
• JMA float
29047
• Cycle # 2
has cycle
#2 plus all
of cycle 7
and parts
of cycles
8-12
• JULD,
LATITUDE,
LONGITUDE
exactly
the same
• Found
similar
situation
for
several
JMA
floats
Example of erroneous data in a traj file
Cycle 1
Cycle 1
Cycle 2
Cycle 1
Cycle 2
Cycle 3
Cycle 1
Cycle 2
Cycle 3
Cycle 4
Cycle 1
Cycle 2
Cycle 3
Cycle 4
Cycle 5
Cycle 1
Cycle 2
Cycle 3
Cycle 4
Cycle 5
Cycle 6
Cycle 1 Cycle 6
Cycle 2 Cycle 7
Cycle 3
Cycle 4
Cycle 5
Cycle 1 Cycle6
Cycle 2 Cycle 7
Cycle 3 Cycle 8
Cycle 4
Cycle 5
Cycle 1 Cycle 6
Cycle 2 Cycle 7
Cycle 3 Cycle 8
Cycle 4 Cycle 9
Cycle 5
Cycle 1 Cycle 6
Cycle 2 Cycle 7
Cycle 3 Cycle 8
Cycle 4 Cycle 9
Cycle 5 Cycle 10
Cycle 1 Cycle 7
Cycle 2 Cycle 8
Cycle 3 Cycle 9
Cycle 4 Cycle 10
Cycle 5 Cycle 11
Cycle 6
Cycle 1 Cycle 7
Cycle 2 Cycle 8
Cycle 3 Cycle 9
Cycle 4 Cycle 10
Cycle 5 Cycle 11
Cycle 6 Cycle 12
Cycle timing by float type • For PROVOR, ARVOR, SOLO-II, NAVIS, NOVA
and recent APEX with Iridium, most cycle timing information is transmitted and clock drift is minimized
• For SOLO, surface time is constant and well known. Calculate additional time not accounted for in surface fixes and split it, adding half to both sides works fairly well
• For APEX floats with Argos, it becomes more difficult as the surface time is not constant. Must rely on DACs filling the AET and TET or do an estimate of TET independently
• For floats with Iridium, extrapolation is not always possible as sometimes only one location is received. Since GPS fix is fast and accurate, not as critical right now
😊
Applied first algorithm from DAC cookbook to estimate TET from
maximum envelope of Last Message Time for APEX Argos floats
AET is
magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET is cyan
Cycle
nu
mb
er
JULD referenced to cycle #1
5901015
Gives a surface
time that varies
from ~ 9 to 13
hours
Cycle
num
ber
JULD referenced to cycle #1
AET is magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET real time estimate by DAC is magenta
TET by envelope method is cyan
Two different examples of real time
TET estimates currently done by
DACs
One to the right clearly isn’t correct as
it occurs before the last surface fix.
One below is clearly too long after last
surface fix
Examples of anomalous APEX ARGOS floats Odd AET and TST
Cycle
num
ber
JULD referenced to cycle #1
AET is magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET is cyan
5901358
Gives surface
times ~16
hours
Largest
surface time is
~60 hours
• 23 out of 210 cycles are > 16 hours
• Think a decoding error led to anomalous AET and TST
• TET estimation is unaffected, but erroneous surface times for those cycles become clear with visual inspection
• 34 cases like this at
AOML
• 4 cases like this at JMA
• Started removing AET
more than 6 hrs
ahead of first msg
received
• Suggest AOML look at
coding to see if this can
be improved upon since
not as big a problem at
JMA where a large
number of floats
were also analyzed
Bad theoretical cycle time?
• ANDRO work detected an incorrect cycle time in meta.nc
• ANDRO suggests 244 hours, not 245
• If I change it to 244 hours, negative clock drift disappears and the surface times are ~11 hours
Gives a surface
time that varies
from ~10 to 208
hours. Negative
clock drift.
Gives a surface
time ~11 hours.
Still slight clock
drift.
Cycle
num
ber
5901746
JULD referenced to cycle #1
AET is magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET is cyan
• 19 cases like this at
AOML
• 3 cases like this at JMA
• Already check J-P’s
corrected cycle time
•Make sure cycle time is
correct in metadata
Cycle duration anomaly? Clock jump?
Cycle
num
ber
JULD referenced to cycle #1
AET is magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET is cyan
5900634
Gives surface times
that varies from ~ 10 to
14 hours for top section
and ~260 hours for
bottom section
• ANDRO suggests this should be processed by slices to get two separate TET envelopes
• When split into slices, surface times are ~10 to 14 hours
• 3 cases like this at
AOML
• 10 cases like this at JMA
• Not a frequent problem,
but severely bad
estimates of TET
when it occurs
• Stop filling TET in real
time?
• Keep track of these
floats and adjust
break points in real
time?
• Allow DAC to choose?
Anomalous cycle?
Cycle
num
ber
JULD referenced to cycle #1
AET is magenta
TST is cyan
FMT is red
Fixes are blue
LMT is red
TET is cyan
5901685
Surface time for
odd cycle is ~12
hours. Surface
times for all
others is ~24
hours
• TET estimate determined by one odd cycle
• Removed that cycle and reprocessed – gives surface times ~12 hours
• J-P suggests this is a problem with the float and not with decoding
• 6 cases like this at
AOML
• 8 cases like this at JMA
• 3 cases like this at
CSIRO (only had 5
floats in region)
• More investigation
needed as to why this
occurs
Overall APEX floats stats from North Pacific
• 6 floats not used because of bad data in file • 22 floats had major clock drift which wasn’t easily
correctable
• 30 needed extra attention due to clock jump or bad points setting the TET env
• 38 had bad AET times needing attention
• Overall, ~30 floats (2%) excluded, ~70 floats (%6) needing extra help/time
• Still need to look at each float TET estimation plot by eye to ensure it is ok – delayed mode procedure
Zonal velocities at the equator ± 2° in Pacific Ocean 9275 velocity estimates
based on:
213 floats in the region
- 142 APEX ( 78
Argos, 64 Iridium)
- 38 SOLO
- 22 NAVIS with
Iridium
- 11 PROVOR
Excluded 5 out of 78
APEX floats w/ Argos
Removed individual cycles
from 3 other APEX Argos
floats
Mean of 55 estimates per
10° x 1 month bin
Thoughts on Real Time Traj V3.0 files
• Should make files easier to use – clearly labeling each measurement with a code indicates which cycle the measurement belongs in and where in the cycle it occurs
• Encourage DACs to begin creating 3.0 files, even if no additional cycle timing information is added for APEX ARGOS floats
• Consider adding some additional real time tests to prevent ghost messages from becoming part of the file and to prevent data from several cycles being included in one cycle
• Newer float models have the additional cycle times available and they are not included in V2.3 traj files – we are losing valuable information
• Cycle timings will improve as newer float models and high speed communications become a larger part of the Argo data stream
Thoughts on delayed mode for traj files
• Need to start delayed mode quality control on trajectory files
• What needs to be done in dmode and how will it be done? – Some things will vary with float type; others will be the same
– Form a working group to begin documenting dmode process for traj files
• Who will do this dmode on trajectory files? – Float’s owner
– Float expert who understands the float behavior (ie ANDRO work)
– Combination of the two?
• When will it occur? – Yearly when dmode is done on the prof file to apply salinity and pres
adjustments?
– Cycle timing estimates may vary depending on float type
• Look at APEX ARGOS floats yearly to estimate TET?
• Wait until SOLO floats die to estimate cycle times