AMSRE Science Team MeetingTelluride, CO

1
Potential Improvements to the AMSR-E L2 Rain over
Land Algorithm

• Ralph Ferraro and Arief Sudradjat
• NOAA/NESDIS, College Park, MD
• Cooperative Institute for Climate Studies,
  College Park, MD

Portions of this talk were presented at the 3rd
TRMM Scientific Conference, February 2008
2
Outline

• GPROF/land algorithm history
• Current status/evaluation
• Areas for improvement possible solutions
• Screening and standardization of flagging
• Diversify the land data base and better
  utilization of Bayesian retrieval
• Convective classification (warm season bias)
• Anomalous surfaces
• Semi-arid land deserts
• Coastlines/small water bodies
• Snow within convective cells
• Summary

3
GPROF-Land Background
Easier to explain versions in terms of TRMM 2A12
product

• 2A12 V4
• GSCAT
• 2A12 V5
• Retrofit/Engineer/Force/Replicate NESDIS
  algorithm within GPROF
• Find profiles that match 22V-85V in NESDIS
  calibration relationship
• 2A12 V6 (current)
• Address deficiencies in unrealistic database and
  NESDIS relationship through match ups with PR and
  TMI
• Convective/Stratiform relationships
• Improved overall performance in terms of bias and
  correlation
• (Improvement) coastline rainfall
• 2A12 V7 (next version)
• Revisit some basic principles
• Will be carried out under auspices of both AMSR-E
  and PMM (focus on HF/cold season)

4
Examples of Good Results

• Produces realistic rainfall fields when compared
  with radar, gauges, climatology, etc.
• GPROF-land is used in virtually all blended
  rainfall products (3B42, CMORPH, CHOMPS, etc.)

Bottom two plots Dave Wolff, GSFC
5
Examples of Problems

• Regional biases
• Compared to PR
• Compared to gauges
• Unrealistic rain rate PDFs in various regions?
• Strange conditional rates
• Underlying surface
• Systematic biases. Why?
• Unrealistic profiles in database
• Bias towards tropical, ocean type systems
• Not enough precip. regimes
• Other dirty laundry
• Surface snow
• Inland water bodies

Liu and Zipser
6
Some old questions
Courtesy of G. Huffman

• Complex, outdated, engineered screening
• How did we get here?
• Developed for SSM/I
• Different FOV size, etc.
• Confusion over zero rain and indeterminate rain.
  Example
• Limitations over cold, dry land. Is this an
  indeterminate rain rate or assumed to be zero?
• Surface e effects
• Evaporation in dryer climate regimes
• Vastly different ice scattering to rain rate
  relationships

! 11/02 Set pixel_status to invalid surface
type and the ! surface rainrate its
negative value for ambiguous ! rain
retrievals ! 03/03 Assigned covect the
probable convective rain. If type of rain
! cannot be determined, convect is left as
missing. ! 09/03 Pixel_status is no longer
set to -21 for ambiguous rain_status !
All retrieved fields are set to their negative
value for ! pixels with an
ambiguous rain_status ! 03/04 All
retrieved fields for a pixel with an ambiguous
rain_status ! are no longer set to
their negative value
7
Improvement Philosophy

• We will not be developing quick fixes
• Experience shows its not real effective
• Look at GPROF improvements across ALL platforms
• GPM era
• Need to think out of the box
• Things get old, fast
• Past screening concept now archaic
• Relied on SSM/I and discriminant functions
• Did not want to rely on others
• Radiometer only
• Data production and delivery reliability
• Lack of experience in the user community
• Synergy with other data centers/research groups

8
Areas for Improvement

• Screening
• Modernize
• Standardization of use of flags
• GPM era, need commonality across platforms!
• Diversify the land data base better utilization
  of Bayesian retrieval
• Land systems, all seasons
• Emissivity models
• All useable measurements
• Convective classification (warm season bias)
• Anomalous surfaces
• Semi-arid land deserts
• Coastlines/small water bodies
• Snow within convective cells

9
Current GPROF Screening Courtesy of G.
Huffman/GSFC
MASKS -21 MASKP -41 MASKI -31 MASKO
-51 MASKC -61
Land (sfc1)
Warm 85H,cold 22V MASKO Other 85H,cold
22V Grody polarization MASKP Ice polarization
MASKP Arid polarization MASKP Not ambiguous 0
Grody cold sfc. MASKI Grody polarization
MASKP Ice polarization MASKP Arid polarization
MASKP missing in 5x5 33 Low
sd Intermediate sd 13 Not ambiguous 0
Compute sdT(85h)
Confused yet?
Really cold sfc. MASKI Ambiguous 14
After this, screening code tries to resolve the
ambiguous regions. Non-scattering zero-rain flags
(-51, -61) arent treated. Regions of positive
(ambiguous) flags are evaluated for surrounding
flag values If perimeter is all negative, the
area is set to the first negative found if
perimeter is zero and negative, the area is set
to 13 if perimeter is zero, the area is set to
0. Unresolved ambiguous values are set to
-15. After this, enough scattering is checked
in the rain code.
10
Standardization of Flags

• Current flags were designed to
• Diagnostics for algorithm developers
• Geared for limited set of early users
• GPCP - Key concern was zero rain vs.
  indeterminate rain
• New, desired attributes include
• Proper survey of user community of their needs
• A nomenclature that everyone can understand
• A set of flags that is useful for everyone
• Commonality across all sensors in GPM era
• Will be handled through PMM

11
New Approaches for Sfc. Identification
Through utilization of both operational and
research data sets that are routinely available,
we should be able to conquer the land
surface screening problem with relative ease.
85 GHz Emissivity
12
Convective/Stratiform

• Current method (a combination of 3 published
  methods) was improvement in V6 from V5, but still
  has problems. This is the culprit of the warm
  season bias.
• Why?
• Resolution TMI is coarser than PR? Spreads out
  rain?
• Beam filling? Underestimate intense rain

13
GPROF databases and Rain Retrieval

• Examine and improve land databases
• GPROF land profiles are limited in diversity
• Predominantly from tropical oceanic systems
• Not diverse nor realistic enough for true
  physical retrieval
• Look at excellent data sets from a variety of
  sources
• C3VP, CloudSat, NOAAs HMT (West Coast
  U.S./Winter)
• Expand previous TMI/PR studies with full 10-year
  data
• Working with PMM science team members
• Retrieval needs to expand to more than just
  22V-85V
• Published studies show other combinations are
  better in different precipitation regimes

14
Special (Seasonal) Situations

• False snow (ambiguous) inside of thunderstorm
• Usually the heaviest RRs in storm system
• V6 attempts to filter these out, but it doesnt
  always work
• Why? Ice scattering at 22V (rare) similar
  characteristic to snow cover
• Solution ? use of snowcover climatology and land
  surface temperature
• Arid regions (e.g., Sahel)
• Screening removes potential rain areas BUT
• Also isnt 100 effective, so false alarms result
• Also removes true rain in other regions
• False alarms create unrealistic conditional rain
  rates
• Solution ?better utilization of emissivity
  calculations and surface type information
• Whats going on with small water bodies?
• Is it called coast or land?
• Applying the wrong algorithm to the wrong surface
  type? OR
• Is it a mixed pixel effect?
• Does it have seasonal effects due to vegetation
  cover?
• Impact is at the low end (onset) of rainfall
• Solution ?Examine current land/sea/coast mask as
  a starting point

15
AMSR-E L, W, C
Are inland water bodies important? Coastline
improvement not intended here!
Impact of thick coast on retrievals?
16
Preliminary investigation Rain/No Rain
Global
Africa
Rain
Rain
No Rain
No Rain
Snow Cover
Snow Cover
Grody Rule Rain if TB22Vgt264 K and
TB22Vgt175.49TB85V GPROF uses TB21V-TB89Vgt5 K
Grody rule GSCAT rule spatial filter
17
Preliminary investigation Desert/Semi-Arid
Global
Africa
DESERT
ARID
DESERT
ARID
ARID
DESERT
No rain if PD19gt20 OR TB85gt253 K and PD19gt7
18
Summary

• GPROF-land is utilized in a number of
  applications and has proven useful
• Despite this, feedback of current version by the
  user community suggest several critical issues
  that need to be resolved for next upgrade
• We will try to prioritize which problems we can
  solve
• Looking towards more generic fixes for GPM era
  rather than band aids for TRMM and AMSR-E
• PMM Team Meeting (August 2008) plans to address
  some of these through WGs
• Land surface/emissivity WG (Gail S-J./Christa
  P-L./Ralph F.)
• Rain detection WG (Guosheng L./Grant P.)

19
Backup Slides
20
NOAA HMT 2005
T and q profiles from HMT RAOB and current land
database
HMT S-band radar reflectivity Freezing level is
around 2.5 km Database lacks such profiles
21
TRMM
Courtesy of M. Diederich
AMSR-E
SSM/I
22
High Resolution Land/Water Mask
23
Filtering out smaller water bodies
24
B02 2002-06-18 (2340) to 2004-09-24(1404)
B03 2004-09-24(1404) to 2004-11-04(0046)
B04 2004-11-04(0046) to 2005-04-01(428)
B05 2005-04-01(428) to 2005-07-06 (428)
B06 2005-07-06 (428) to 2005-08-23 (427)
B07 2005-08-23 (427) to 2006-03-10 (931)
B08 2006-03-10 (931) to 2006-11-13 (0258)
v9 2002-06-18 (2340) to 2003-12-31(2359)
25
(No Transcript)