Satellitebased Estimation of Precipitation Using Passive Opaque Microwave Radiometry

1
Satellite-based Estimation of PrecipitationUsing
Passive Opaque Microwave Radiometry

• Frederick W. Chen, Laura J. Bickmeier, William J.
  Blackwell, R. Vincent Leslie
• MIT Lincoln Laboratory (Lexington, MA, USA)
• David H. Staelin, Chinnawat Pop Surussavadee
• MIT Research Laboratory of Electronics
  (Cambridge, MA, USA)
• 3rd Workshop of the International Precipitation
  Working Group
• Melbourne, VIC, Australia
• 24 October 2006

This work was sponsored by the National
Aeronautics and Space Administration under
Contract NNG 04HZ53C, Grant NNG 04HZ51C, and
Grant NAG5-13652, and the National Oceanic and
Atmospheric Administration under Air Force
Contract FA8721-05-C-0002. Opinions,
interpretations, conclusions, and recommendations
are those of the author and are not necessarily
endorsed by the United States Government.
2
Outline

• Physical basis
• Algorithm development
• AMSU (Advanced Microwave Sounding Unit)
• ATMS (Advanced Technology Microwave Sounder)
• Future work
• Summary

3
Physical Basis
OPAQUE BANDS
TRANSPARENT BANDS

• Transparent channels (or window channels)
• Warm water vapor signatures over cold ocean
• Scattering signatures due to ice particles over
  land
• Opaque channels
• Varying atmospheric opacity
• Sensitive primarily to specific layers of
  atmosphere

4
54-GHz and 183-GHz Weighting Functions
54-GHz
183-GHz
5
Estimation of Precipitation Rate with Opaque ?W
Channels(54-GHz and 183-GHz)

• Precipitation rate ? humidity vertical wind
  velocity
• Absolute humidity
• 54-GHz band reveal temperature profile
• 54-GHz and 183-GHz bands reveal water vapor
  profile
• Vertical wind velocity
• Stronger vertical wind ?
• Stronger vertical winds results in increased
  backscattering of cold space radiation
• Perturbations (cold spots) in 54-GHz data reveal
  cloud-top altitude
• Absolute albedos reveal hydrometeor abundance
• Relative albedos (54 vs. 183-GHz) reveal
  hydrometeor size

Greater hydrometeors size Greater hydrometeor
abundance Higher cloud-top altitude
6
Particle Sizes Revealed in NAST-M Data
54 GHz
118 GHz
?TB
183 GHz
425 GHz
Visible
Leslie Staelin, IEEE TGRS, 10/2004
7
AMSU Radiometry

• Passive ?W sounder
• AMSU-A
• 12 channels in opaque 54-GHz O2 band
• Window channels near 23.8, 31.4, and 89.0 GHz
• AMSU-B
• 3 channels in opaque 183.31-GHz H2O band
• Window channels near 89.0 and 150.0 GHz

8
General Structure of AMSU Algorithm(Chen and
Staelin, IEEE TGRS, 2/2003)

• Signal processing
• Regional Laplacian interpolation
• Image sharpening
• Principal component analysis
• Neural net
• 2-layer feedforward neural net
• 1st layer tanh transfer function
• 2nd layer linear transfer function

9
Signal Processing Components

• Neural-net correction of angle-dependent
  variations in TBs
• Cloud-clearing via regional Laplacian
  interpolation
• Temperature profile characterization
• Cloud-top altitude characterization
• Principal component analysis for dimensionality
  reduction
• Temperature profile PCs
• Window channel / H2O profile PCs
• Image sharpening
• AMSU-A data sharpened to AMSU-B resolution

10
The Algorithm Precipitation Masks
Precipitation-Induced Perturbations
IMAGE SHARPENING
PRECIPITATION DETECTION
CORRUPT DATA DETECTION
LIMB--SURFACE CORRECTION
REGIONAL LAPLACIAN INTERPOLATION
11
The Algorithm Neural Net
Trained to NEXRAD
12
Final Output
13
Example of Global Retrieval
14
ATMS

• Similar to AMSU
• To be launched on NPP (2009) NPOESS satellites
• NPP NPOESS Preparatory Project
• Improvements over AMSU
• Additional channels in 54-GHz and 183-GHz bands
• Better resolution in 54-GHz band
• Better sampling
• Nyquist sampling of 54-GHz data
• Identical sampling of all channels

15
Simulating ATMS TBs

• MM5 Atmospheric Circulation Model
• Provides temperature profile, water vapor
  profile, hydrometeor profile,
• Used Goddard hydrometeor model (Tao Simpson,
  1993)
• Radiative Transfer
• TBSCAT due to Rosenkranz (IEEE TGRS, 8/2002)
• Multi-stream, initial-value
• Improved hydrometeor modeling due to Surussavadee
  Staelin (IEEE TGRS, 10/2006)
• Filtering
• Accurate matching of TBs on MM5 grid to ATMS
  resolution and geolocation using satellite
  geometry toolbox for MATLAB
• Computing angular offset of surface locations
  from boresight
• Computing satellite zenith angles from scan angle
• Computing geolocation from scan angle

16
MM5 Rain Rate Typhoon Pongsona, 2002/12/8
17
AMSU vs. ATMS, 1837 GHz
Observed AMSU
Simulated ATMS

• Simulated ATMS 1837 GHz data shows reasonable
  agreement with AMSU-B
• Morphology difference between AMSU observations
  and MM5 predicted radiances is due to the
  inaccuracy of the NCEP analyses used to
  initialize the MM5 model

18
AMSU vs. ATMS, 50.3 GHz
Observed AMSU
Simulated ATMS

• Simulated ATMS 50.3-GHz data with finer
  resolution and sampling shows finer features than
  AMSU-A
• Intense eyewall signature in simulated ATMS
  50.3-GHz data due to NCEP initialization
  limited 5-hr MM5 spinup time producing excess of
  large ice particles

19
Future Developments

• Adapting Chen-Staelin algorithm (IEEE TGRS,
  2/2003) for ATMS
• Exploiting Nyquist sampling in the 54-GHz band
• Using methods from window-channel-based
  algorithms
• Improving the signal processing of Chen-Staelin
  algorithm
• Improving neural net training
• Representations of circular data

20
Recently Launched Future Instruments

• Similar to AMSU-A/B
• AMSU/MHS on NOAA-18 (2005)
• AMSU/MHS on NOAA-N, METOP-1, METOP-2, METOP-3
• ATMS
• NPP (2009)
• NPOESS
• ?W instruments on geostationary satellites?
• lt 1 hr revisit times

21
Summary

• Physical basis of precipitation estimation using
  opaque ?W channels
• Atmospheric sounding capabilities of opaque ?W
  channels
• Cloud shape and particle size distribution from
  NAST-M 54-, 118-, 183-, and 425-GHz data
• AMSU precipitation algorithm
• Relies primarily on 54-GHz and 183-GHz opaque
  bands
• Signal processing regional Laplacian
  interpolation, principal component analysis,
  image sharpening
• ATMS precipitation algorithm development
• Simulation system using MM5/TBSCAT

22
Backup Slides
23
NAST-M

• NAST NPOESS Aircraft Sounder Testbed
• Risk-reduction effort by NPOESS Integrated
  Program Office
• Cooperative effort of NASA, NOAA, DoD
• Equipped with 54-, 118-, 183-, and 425-GHz
  radiometers
• Flown on high-altitude aircraft
• ER-2 (NASA)
• Proteus (Scaled Composites)
• 2.5-km resolution near nadir

24
Scattering in the 54-GHz and 183-GHz Bands
0.7 mm
2.4 mm
25
AMSU Geographical Coverage

• Aboard NOAA-15, NOAA-16, NOAA-17
• Nearly identical AMSU/HSB on Aqua

26
AMSU-A/B Sampling Resolution
AMSU-A
AMSU-B

• AMSU-A
• 3 1/3 sampling (50 km near nadir)
• 3.3 resolution (50 km near nadir)
• AMSU-B
• 1.1 resolution (15 km near nadir)
• 1.1 sampling (15-km near nadir)

27
15-km AMSU vs. NEXRAD Comparison
28
RMS Discrepancies (mm/h)between AMSU and NEXRAD
29
Features of ATMS vs. AMSU

• Channel set
• Similar to AMSU
• Additional 51.76-GHz channel
• Additional 183.314.5-GHz 183.311.8-GHz
• 165.5-GHz replaces 150-GHz on AMSU-B
• No 89.0-GHz 15-km channel (available on AMSU-B)
• Resolution
• 54-GHz and 89-GHz 2.2 vs. 3.33 on AMSU
• 23.8- and 31.4-GHz 5.2 vs. 3.33 on AMSU
• Sampling
• 165.5-GHz, 183-GHz Similar to AMSU-B
• Other channels 3 finer than AMSU-A cross-track
  along-track
• All channels sampled at the same locations
• Nyquist sampling of 54-GHz and 89-GHz
• Similar sensitivity

30
ATMS AMSU Footprints
31
ATMS AMSU Footprints (Near Nadir)
32
ATMS Rain Rate Retrieval Algorithm

• Completely new algorithm
• Neural net
• Inputs
• All 22 channels
• sec(satellite zenith angle)
• Training, validation, and testing sets
• MM5 data over Typhoon Pongsona
• 1 time step (1521 data points) each for training,
  validation, and testing

33
ATMS vs. MM5, 1.1
34
ATMS vs. MM5, 5.2
35
Representations of Geolocation

• Rectangular (2-D)
• Discontinuity across 180 E/W (Intl Date Line)
• Topological distortion around 90 N 90 S (Geo.
  N S Poles)

• Cylindrical (3-D)
• Continuity across 180 E/W
• Topological distortion around 90 N 90 S

• Spherical (3-D)
• Continuity across 180 E/W
• No topological distortion around 90 N and 90 S

36
GeolocationComparing the Representations

• Spherical representation produces the lowest RMS
  errors
• RMS error with 10 weights biases
• Linear 0.16
• Cylindrical 0.16
• Spherical 0.01
• Weights biases needed for RMS error lt 1.5
  10-2
• Rectangular 23
• Cylindrical 18
• Spherical 6

RECTANGULAR
CYLINDRICAL
SPHERICAL