ENSEMBLES EN3 v1c - changes from v1b
====================================

All the NetCDF files have been replaced: 

a) The in situ temperature (NetCDF variable TEMP) was 
incorrect in previous versions - now corrected.  
The potential temperature (POTM) was unaffected.  
If TEMP is used the POTM_QC flags should be checked to 
screen out rejected values.  

b) I have also rejected NINJA, TSK floats (3 floats in 
2002-2003).  

c) In 1950 and 1951 there are minor changes near the Black Sea. 

d) I have added 2006!  The data sources were GTSPP and 
Argo data from the Coriolis Argo GDAC, both copied in March 2007.

Bruce Ingleby   4 May 2007


ENSEMBLES EN3 v1b - changes from v1a
====================================

The NetCDF files for 1994 to 2005 have been replaced by 
v1b versions with the following changes: 

a) From late 1994 PALACE (profiling float) reports in the 
tropical Indian Ocean and selected floats from other areas 
were rejected (230 floats in total, most of these reported 
only every 25 or 30 days and gave profiles over the top 900 m 
or 1000 m).  A further 58 floats are rejected below 950 m as 
their lowest levels seem unreliable.  Most of the rejected 
floats stopped reporting in July 2001 if not earlier.  
The affected profiles are from the WOD05 (PFL) database, they 
were not in WOD01 or GTSPP and hence not in EN2.  

I first noticed a possible problem when I was processing 1995 
and 1996, but gave the data the benefit of the doubt at the time.  
However it became clear that in our objective analyses the 
tropical Indian ocean was exceptionally cold in the top 300 m 
from 1995 to mid-2001 compared to other years, and observation 
statistics suggested that the PALACE floats were particularly 
cold near the thermocline - consistent with a depth error.  
A few of the worst affected values will have been rejected by 
the automatic QC in v1a, however most were passed by the QC - 
tolerances are large near the thermocline because there can be 
large real fluctuations there - some problems can only be 
reliably detected by looking at composite statistics.  

b) A problem with SOLO, FSI Argo floats was reported on 16 Feb 2007, 
see  http://www-argo.ucsd.edu/Acpres_offset.html 
I have rejected these data (type 852; 287 floats, approx. 6% of Argo 
reports in 2005, started being used circa 2000) and also NINJA, SBE 
(type 856; 10 floats, only about 0.1% of Argo reports) which also 
showed severe problems in our statistics.  

c) To help in detecting the affected floats minor changes were 
made to the processing/storage of metadata: 
- For GTSPP more profiles now have the WMO code table 1770 type 
  assigned (stored in INST_REFERENCE in the NetCDF files; 
  in the GTSPP decode I am now using PFR$ as well as PEQ$).  
- Also for GTSPP WMO code table 4770 is now stored in PROJECT_NAME.  
- For WOD05 PFL data the "WMO number" is now stored at the end 
  of PROJECT_NAME instead of the NODC platform number.  
  This "WMO number" (not available for a few early floats) is the 
  same as the Id used in the Argo GDAC, or with a Q in front for 
  the GTSPP callsign.  

Bruce Ingleby  19 March 2007