Validating Water Vapor Retrievals: Lessons Learned and Some Results

1
Validating Water Vapor RetrievalsLessons
Learned and Some Results

• Eric J. Fetzer and Annmarie Eldering
• Jet Propulsion Laboratory
• AIRS Science Team Meeting
• 30 March-1 April 2003
• Greenbelt, Maryland

2
First Order of BusinessJGR Special Issue

• Our proposal for a Special Issue on AIRS
  validation was accepted by the JGR editor
• Have manuscripts ready for peer review in the
  October or November

3
Overview of Todays Talk

• A Motivating Problem
• Understanding water vapor in the subtropical
  eastern Pacific associated with temperature
  inversions
• Analyses (v3.1.9)
• Lots of maps of retrieved quantities
• Intercompare retrievals and sondes along the west
  coast and Hawaii. Increasing complexity
• Intercompare matched profiles
• Compare time series of retrieved and sonde T q
• Create statistical summaries
• Next steps
• First results with v3.5.0 and dedicated sondes

4
Motivating Problem

• What is the water vapor distribution associated
  with retrieved temperature inversions in the
  eastern Pacific?
• Inversion distribution present at October Sci.
  Tm. Meeting, Fall AGU.
• GRL manuscript in preparation
• Should be straightforward
• Over water
• Mostly high pressure
• Sonde stations on West Coast and Hawaii, with
  decent time match.
• Not straightforward.

5
Lessons Learned 0, 1 2

• 0. Temperature is much less variable than water
  vapor.
• Water vapor is highly sporadic in space and time.
• Maxima in water vapor are embedded in cloudy
  regions, where most of our questionable
  retrievals occur.

6
Lesson 1 The highly localized nature of water
vapor (v3.1.9)
1 January 2003 1000-700 mb precipitable water
vapor Full scale 30 mm
700-500 mb precipitable water vapor Full scale
6 mm
7
Lesson 2 Maxima in water vapor are embedded in
cloudy regions with nonzero retrieval_type
1 January 2003 Retrieval type Gray 0 Full
IR Red 10 1st failed Blue 20 MW Strat
IR Green 30 MW Strat IR Black 100 Crash
burn X SST error gt 3 K.
Error with forecast SST Black -5 K Red 5 K.
8
Intercomparing AIRS and Sondes

• Look at December 2002 and January 2003
• Consider operational sondes at 6 locations
• Quillayute, Washington
• Oakland, California
• Vandenberg, California,
• San Diego, California
• Hilo, Hawaii
• Lihue, Hawaii

9
Lesson 3 Highly variable water makes for
nonstationary time series
Precipitable Water Vapor, Lihue, Hawaii 1000-700
mb 700-500 mb layers Blue Sonde, Red AIRS
(diamond nonzero rettyp)
10
Lesson 4 Water vapor at upper levels can be
highly variable
Lihue, 700-500 mb dynamic range 0.1 to 10 mm
precipitable water Dave Tobin UTWV is the
last precipitable 0.1 mm
11
Lesson 5 Locally, conclusions like L.
McMillins in Val ReportCaution Standard
deviation is not robust to outliers
1000-700 mb Lihue 2.7
14.1 Bias Std. Dev. McMillin Globally 3.6
11.6 700-500 mb Lihue
-4.3 44.3 (15th 85th percentiles-19,
12) McMillin Globally 0.0 26
Lihue 700-500 mb 1.95 mm mean
Lihue 1000-700 mb 20.5 mm mean
12
Lesson 6

• Trade wind cumulus regions are challenging
• In Hawaii, a moist boundary layer is overlain by
  very dry, descending air
• This explains the large standard deviation at
  Lihue.
• Need to revisit this problem with dedicated
  sondes
• Andros, Ascencion, others.

13
Lesson 7 Sippican sondes not consistent at
upper levels
The Goldilocks Effect? Vandenberg may have a
combination of high relative humidity and
temperature gt Sippican sensitivity low
bias Oakland too cold? San Diego Hilo too
dry? Vandenberg just right?
UNBIASED few  15-45 BIASED -15-30  15-20
SEE da Silviera, R. B., G. Fisch, L. A. T.
Machado, A. M. Dall Antonia, L. F. Sapucci, D.
Fernandes and J. Nash, (2003) Executive summary
of the WMO intercomparison of GPS radiosondes,
WMO Instruments and Observing Methods Report No.
76, WMO/TD No. 1153.
14
Lesson 7

• Nonzero retrieval types contain potentially
  useful information about water vapor
• THIS IS REFLECTED IN TIME AVERAGED FIELDS

OAKLAND Blue crosses Sonde Water Red Cross
AIRS Water, ret. type 0 Red Diamond AIRS,
ret. type NE 0
15
Lesson 8The 1000-700 mb and 700-500 mb layers
are statistically decoupled
1-16 January 2003 1000-700 mb water (mm)
Also apparent in radiosonde difference
statistics
1-16 January 2003 700-500 mb water (mm)
16
Some Conclusions from this study

• Retrieval type (or RetQAFlag) bias our
  observations toward dry conditions
• The results over the eastern Pacific is similar
  to Larry McMillins conclusions globally.
• Higher variability gt larger standard deviations
  at altitude
• We have sensitivity in 2 km layers.
• We are meeting 15 / 2 km RMS at ARM TWP up to
  300 mb with v3.5.0.

17
Some Questions

• What is our true vertical resolution for water
  vapor?
• Statistically independent 1000-700 and 700-500 mb
  layers imply we can do better
• E. G., can we resolve the abrupt transition from
  moist to dry over trade wind cumulus?
• What is our true RMS uncertainty?
• How strongly does this depend on the vertical
  distribution of q?
• What is the information content of non-zero
  retrieval types?

18
Next Steps

• Extend the analysis of dedicated sondes into
  others regions
• Galapagos (Voemel)
• Caribbean, Ascencion Is. (Schmidlin)
• Chesapeake light platform (McMillan)
• Extend the operational sondes westward over the
  Pacific
• 3. Look for regional variability in statistics
• For example is high variability (45) over
  Hawaii at 700-500 mb seen at Ascencion Is?
• 4. Move over land

19
The Latest TWP Dedicated Sondes and Version
3.5.0 Looking GoodBlack RMS, Red Std Dev,
Blue Bias
18 Nov 2002 to 31 Jan 2003 26 sondes, 45 matches
with rettyp 0 and landfrac 0.
Bias 10, RMS lt35up to 150 mb In 1 km layers
RMS 8-15 in 2 km layers up to 300 mb
100-700 mb 700-500 mb 500-300 mb
Standard layers 100-925 mb, 925-850, etc.