the status of the noaa/nesdis operational amsu/mhs precipitation algorithm
DESCRIPTION
The Status of the NOAA/NESDIS Operational AMSU/MHS Precipitation Algorithm. Ralph Ferraro NOAA/NESDIS College Park, MD USA Wanchun Chen, Cezar Kongoli, Huan Meng, Paul Pellegrino, Daniel Vila, Nai-Yu Wang, Fuzhong Weng, Limin Zhao. Outline. Review of AMSU and operational algorithm - PowerPoint PPT PresentationTRANSCRIPT
23-27 October 2006 3rd IPWG Melbourne, Australia
The Status of the NOAA/NESDIS Operational AMSU/MHS Precipitation
Algorithm
Ralph FerraroNOAA/NESDIS
College Park, MD USA
Wanchun Chen, Cezar Kongoli, Huan Meng, Paul Pellegrino, Daniel Vila, Nai-Yu Wang, Fuzhong Weng, Limin Zhao
23-27 October 2006 3rd IPWG Melbourne, Australia
Outline
• Review of AMSU and operational algorithm• Recent changes
– NOAA-18– Coastlines
• Upcoming Improvements– Ocean/emission rainfall– LZA bias removal– Snowfall rates
• Future– METOP– NPP– MIRS
23-27 October 2006 3rd IPWG Melbourne, Australia
NOAA AMSU Sensor
•AMSU is a cross-track scanning radiometer (unlike SSM/I, AMSR-E, TMI)
•AMSU-A (45 km nadir FOV)•15 channels•23, 31, 50 – 57 (13), 89 GHz
•AMSU-B (15 km nadir FOV)•89, 150, 183+1,3,7 GHz
•MHS replaces AMSU-B on N-18•89, 157, 183+1,3, 190.3 GHz
Satellite Launch Date LTAN
NOAA-15 13 May 1998 1735
NOAA-16 21 September 2000 1529
NOAA-17 24 June 2002 2218
NOAA-18 20 May 2005 1342
METOP-1 19 October 2006 2130
23-27 October 2006 3rd IPWG Melbourne, Australia
NOAA Produces Operational Products from AMSU
23-27 October 2006 3rd IPWG Melbourne, Australia
Characteristics of the NOAA AMSU Rain Rate Algorithm
• Physical retrieval of IWP and De – 89 & 150 GHz– Use of other window and sounding channels
• Derive needed parameters for retrieval• Filters for possible ambiguous surfaces
– Use of ancillary data & other AMSU derived products
• IWP to RR based on limited CRM data and RTM– RR = A0 + A1*IWP + A2*IWP2
• 183 GHz bands used to identify deep convection– Use another set of ‘A’ coefficients
• 50 GHz bands used to identify snowfall over land
23-27 October 2006 3rd IPWG Melbourne, Australia
Example of Global Real Time Data
23-27 October 2006 3rd IPWG Melbourne, Australia
Example of Regional Retrievals
23-27 October 2006 3rd IPWG Melbourne, Australia
Example of Monthly Data
23-27 October 2006 3rd IPWG Melbourne, Australia
Continuous monitoring of algorithm performance via IPWG validation sites
23-27 October 2006 3rd IPWG Melbourne, Australia
Performance vs. GPCC (8/03 – 12/05)
Mean Mean Mean Bias Bias R R
SSMI AMSU GPCC SSMI AMSU SSMI AMSU
60S-60N 60.9 66.5 75.6 0.80 0.88 0.64 0.69
40S-40N 72.0 76.4 80.5 0.89 0.95 0.70 0.74
20S-20N 102.1 119.3 129.3 0.72 0.94 0.81 0.81
Zonal Mean Land Rainfall (1/04 - 12/05)
0
20
40
60
80
100
120
140
160
-45 -35 -25 -15 -5 5 15 25 35 45 55
Latitude
Rai
nfa
ll (
mm
)
SSMI
AMSU
GPCC
23-27 October 2006 3rd IPWG Melbourne, Australia
AMSU Summary/Limitations• Land
– In general, performs well• Too high in convective situations• Regional biases (of course!), esp. too high in drier regimes
– Better sensitivity to lighter rain rates– Falling snow detection (but not rates)
• Ocean– Restricted to convective precipitation
• Overall, low due to missing precipitation without ice (generally lighter rain intensities)• Rain coverage less than other sensors
– Conditional rain rates too high• LZA bias in IWP
• Coastlines– Not adequately handled
• View angle dependencies– Larger FOV on scan edges results in varying rain rate distributions– Unrealistic PDF’s
23-27 October 2006 3rd IPWG Melbourne, Australia
NOAA-18: MHS replaces AMSU-B
• NOAA, DMSP and METOP will operate POES constellation
• Changes include– Microwave Humidity Sounder (MHS) instead of AMSU-B
• 157 GHz vs. 150; 190.3 GHz vs. 183+7
– MHS will fly on NOAA-N (18), -N’ and METOP (Successful launch 10/19)
• Synthetic NOAA-N 150 and 183+7 GHz based on coincident measurements with NOAA-16
• Operational 29 Sep 2005
23-27 October 2006 3rd IPWG Melbourne, Australia
Coastline Precipitation
• Most passive MW algorithms fail miserably in coastal regions– Emissivity contrast between land and ocean– Different physics package
• Imager approaches– “Extend” land algorithm to coasts
• Bring in scattering rain types
– “Correct” TB’s based on % of FOV filled with land• Computer expensive; used in regional approaches
• Sounders– Utilize channels that are mainly insensitive to surface
23-27 October 2006 3rd IPWG Melbourne, Australia
Improved AMSU Coastline Algorithm
• Utilizes AMSU 53.6, 150, 183+1, 3 and 7 GHz to identify potential rain and a synthetic IWP along coastlines– Compute rain rate in same
manner via IWP
• Also updated land/sea/coast tag
• Implemented into operations 7/31/06
• Substantially better retrievals with minimal false alarms
23-27 October 2006 3rd IPWG Melbourne, Australia
OLD NEW
RADAR
23-27 October 2006 3rd IPWG Melbourne, Australia
Improvement of Oceanic Rain Rates and Removal of Angular Biases
Daniel Vila (details at his poster)
• High oceanic bias attributed to unrealistic PDF’s– Function of LZA, problem traced to IWP retrieval
• Corrected via adjustment with SSM/I PDF’s
• Oceanic rain coverage low– Non-convergence of IWP/De algorithm in mostly
light rain rates• Corrected by adding in emission component
• Low bias at edge of scan due to large FOV
23-27 October 2006 3rd IPWG Melbourne, Australia
Results – April 2005
Current AMSU
Corrected AMSU
SSM/I GPROF V6
Warm/shallow rain
Reduced ocean rainfall
Slight reduction over land
23-27 October 2006 3rd IPWG Melbourne, Australia
Zonal Mean Rain Frequency
•Little change•AMSU>SSMI due to 150 GHz
•Significant improvement due to CLW
23-27 October 2006 3rd IPWG Melbourne, Australia
Zonal Mean Rain Amount
•Reduction in rainfall amounts,mostly in convective zones
•Corrected AMSU much closer to SSM/I
23-27 October 2006 3rd IPWG Melbourne, Australia
Snowfall Rate Algorithm DevelopmentHuan Meng
• Use RTM to retrieve IWP under snow condition – 1 layer & two-streams
• Derive empirical equation connecting IWP with NEXRAD reflectivity
• Adopt an existing reflectivity-snowfall rate equation
• Derive snowfall rate from IWP
23-27 October 2006 3rd IPWG Melbourne, Australia
Matching AMSU-B with Radar Data
• WSR-88D radar reflectivity (Z)• Z is Quality controlled by QCNN
(Quality Control Neural Network) algorithm
• Use radar data from the lowest elevation (0.4º) and within 100 km.
• Assume AMSU-B spatial sensitivity follows Gaussian function within an FOV when matching with radar data
23-27 October 2006 3rd IPWG Melbourne, Australia
Retrieved AMSU IWP
Corresponding NEXRAD Reflectivity
23-27 October 2006 3rd IPWG Melbourne, Australia
Regression between IWP and Z
• Data from 10 snow
cases with 227
matching points
• 3rd order fit between
IWP and Z
= 0.48
23-27 October 2006 3rd IPWG Melbourne, Australia
Choice of Z-R Relationship?IWP-Snowfall Rate Empirical Equations
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0 0.5 1 1.5 2 2.5 3
IWP (kg/m2)
R (
mm
/hr)
|
Carlson&Marshall
Vasiloff
Ohtake&Henmi
Imai
Puhakka
Boucher&Weiler
Sekhon&Srivastava
Fujiyoshi
23-27 October 2006 3rd IPWG Melbourne, Australia
Choice of Z-R Relationship (2)• Sekhon & Srivastava (1970) has the smallest
RMS with the 1-hr-lag observation:
Z = 398 R2.21
C&M VAS OHT IMA PUH B&W S&S FUJ
0.5140 0.4084 0.3722 0.3735 0.3532 0.3522 0.3514 0.3872
23-27 October 2006 3rd IPWG Melbourne, Australia
Example of Snow Rate Retrieval
Lancaster, OH, Feb 15 - 17, 2003
0
0.1
0.2
0.3
0.4
0.5
0.6
0 1000 2000 3000 4000 5000Time (min)
Sn
ow R
ate
(mm
/hr)
| Obs 1-hr late
Retrieval
Newark, OH, Feb 15 - 17, 2003
0
0.2
0.4
0.6
0.8
1
1.2
0 1000 2000 3000 4000 5000Time (min)
Sn
ow R
ate
(mm
/hr)
|
Obs 1-hr lateRetrieval
Pro: Catch basic snowfall patterns
Con: Miss snow or underestimate high snowfall rate
23-27 October 2006 3rd IPWG Melbourne, Australia
Next Steps…• Use more realistic snow characteristics in the RTM
• Explore approaches to classify snowfall type and utilize in snow rate retrieval.
• Add more snow cases to improve the accuracy of IWP-Z regression.
• Implement experimental retrievals for CONUS winter 2006-07
23-27 October 2006 3rd IPWG Melbourne, Australia
Future for AMSU/MHS• Implement near term algorithm improvements
– FOV biases – CLW component for ocean rain– Experimental snowfall rates over CONUS
• Longer term– Improved AMSU physics package
• “MIRS” – Microwave Integrated Retrieval System– Temperature and moisture sounding– Bayesian precipitation retrieval using O2 and H2O channels
– Reprocessing of entire AMSU time series– Snowfall rates
• Transition to operations– METOP-1 (Oct07)
• Jan/Feb 2007?• Pipeline processing
• Prepare for NPP/ATMS
23-27 October 2006 3rd IPWG Melbourne, Australia
Backup Slides
23-27 October 2006 3rd IPWG Melbourne, Australia
http://www.orbit.nesdis.noaa.gov/corp/scsb/mspps/
23-27 October 2006 3rd IPWG Melbourne, Australia
AMSU Climate Products shown in NatureMichael Evans, April 2006
23-27 October 2006 3rd IPWG Melbourne, Australia
Falling Snow over Land from AMSU• Use of AMSU-B 183 GHz bands along with AMSU-
A 53.6 GHz allows for expansion of current algorithm to over cold and snow covered surfaces:– AMSU-B channels allow for detection of scattering
associated with precipitation, but surface blind when “sufficient moisture” exists
– AMSU-A channel 5 allows for discrimination between “rain” and “snow”
• Feature added in 11/03, snowfall detection only (assigned arbitrary rate of 0.1 mm/hr)
• Validation over CONUS winter 2003-04
23-27 October 2006 3rd IPWG Melbourne, Australia
Summary/Limitations
• Algorithm Performance– Can detect snowfall associated with synoptic scale systems– Low false alarms– Can increase region of application by lowering TB54L threshold
up to 5 K• Some increase in false alarms
• Limitations– Relative moist atmospheres - -5 to 0 C– Southern extent of snow pack/temperate latitudes– Precip layer needs to extend to ~4-5 km or higher– No signal in extreme cold climate regimes and shallow snow
23-27 October 2006 3rd IPWG Melbourne, Australia
Retrieving Ice Water Path Using Radiative Transfer Model
• One-layer two-stream RTM (Yan & Weng, 2006, to be submitted to JGR)
7180
150
89
31
23
1)(
B
B
B
B
B
TT
S
e
T
T
T
T
T
AEAA
T
D
V
II: ice water path
V: total precipitable water
De: cloud particle effective diameter
Ts: surface temperature
A: derivatives of TBi over I, V, De, & Ts
E: error matrix
TBi: brightness temperatures at 23.8, 31.4, 89, 150, and 180±7 GHz
• Use iteration scheme. Iteration stops if ΔTBi ≤ δi (i = 1, 5).
• Adopt fast one-layer RTM to meet near real-time operational requirement.
23-27 October 2006 3rd IPWG Melbourne, Australia
Choice of Z-R Empirical Eq – Cont’d• Observation Data: hourly snowfall observations from 12 weather stations• Highest correlation coefficient between retrieved R and snowfall
observations with 1-hr lag:
0-hr late 0.5-hr late 1-hr late 2-hr late 3-hr late 4-hr late
0.1281 0.2370 0.3239 0.2898 0.2122 0.1921
Retrieved R represents snow in the atmosphere because channel 183±7 GHz is sensitive to 600 – 700 mb. The time lag represents the time it takes the snow to fall to the ground.
23-27 October 2006 3rd IPWG Melbourne, Australia
Application to AMSU Measurements
Radar ReflectivityAMSU Precipitation Rates 2004 Nov 24
23-27 October 2006 3rd IPWG Melbourne, Australia
0-10 mm 10-30mm
Unrealistic peak location Inability to retrieve high rain rates for large LZA
The position of the peak in the histograms (systematic bias), is corrected performing a Gaussian PDF with µ (peak histogram position) and б (standard deviation of observed distribution) depending on LZA
and latitude. For high rain rates, a linear correction scheme with the slope depending on SSM/I and AMSU-B footprint ratio is proposed to normalize AMSU-B derived rain rates.
23-27 October 2006 3rd IPWG Melbourne, Australia
Adding Emission Component to Oceanic Rain
Mean AMSU-B derived rain rate for different CLW values. CI is the convective index computed by using the three 183 GHz channels. Satellite: NOAA 15 - 60N-60S: Year: 2005.
RR vs. CLW: Proposed correction scheme
23-27 October 2006 3rd IPWG Melbourne, Australia
Results – April 2005
Mean rain rate of AMSU-B NOAA-15 operational derived rain rates (mm/day) (upper panel), corrected values (middle panel) and mean derived rain rates (mm/day) for SSM/I F-13 GPROF v6.0 for April 2005.
Monthly mean absolute bias (upper panel) and RMSE (bottom panel) for UNCORR and CORR algorithm compared with SSM/I GPROF V6.0 estimates for 2005.