a physically-based rainfall rate algorithm for all
TRANSCRIPT
A Physically-based Rainfall Rate Algorithm for All Surfaces: Applicability to All Microwave
Sensors Including TRMM & GPM
Sid-Ahmed Boukabara1
Kevin Garrett2, Leslie Moy2, Flavio Iturbide-Sanchez2, Chris Grassotti2 and Wanchun Chen2
2010 PMM Science Team Meeting Seattle, WA
November 1-4, 2010
1. NOAA/NESDIS/STAR, JCSDA, 2. IMSG Inc.
Agenda
Overview: Microwave Integrated Retrieval System MiRS Approach to Rainfall retrieval over all surfaces MiRS Current Performance for Rainfall Rate MiRS Extension to TMI & GPM/GMI (Proxy Data) Current Limitations and Upcoming Improvements MiRS Contribution to Sfc Emissivity characterization Summary
MiRS Algorithm Description
RR = c0 + c1RWP+c2IWP+c3CLW
Rainfall Rate • Rainfall rate based on relationship to integrated hydrometeor values (not sensor-specific Tbs) • Extension to other sensors does not require regenerating other coefficients • Coefficients derived from MM5 simulations • Same approach used over ocean and land
Vertical Integration and Post-Processing 1D
VAR
O
utpu
ts
Vertical Integration
Post Processing
(Algorithms)
TPW RWP IWP CLW
Core Products
Temp. Profile
Humidity Profile
Emissivity Spectrum
Skin Temperature
Liq. Amount Prof
Ice. Amount Prof
Rain Amount Prof
-Sea Ice Concentration -Snow Water Equivalent -Snow Pack Properties -Land Moisture/Wetness -Rain Rate -Snow Fall Rate -Wind Speed/Vector -Cloud Top -Cloud Thickness -Cloud phase
4
Simultaneous Retrieval?
X is the solution
F(X) Fits Ym within Noise levels
X is a solution
Necessary Condition (but not sufficient)
If X is the set of parameters that impact the radiances Ym, and F the Fwd Operator
If F(X) Does not Fit Ym within Noise
X is not the solution
All parameters are retrieved simultaneously to fit all radiances together
Suggests it is not recommended to use independent algorithms for different parameters, since they don’t guarantee the fit to the radiances
5
Solution-Reaching: Convergence Convergence is reached everywhere: all surfaces, all weather
conditions including precipitating, icy conditions A radiometric solution (whole state vector) is found even when
precip/ice present. With CRTM physical constraints.
Previous version (non convergence when precip/ice present)
Current version
MiRS is applied to a number of microwave sensors, each time gaining robustness and improving validation for Future New Sensors • The exact same executable, forward operator, covariance matrix used for all sensors • Modular design • Cumulative validation and consolidation of MiRS
POES N18/N19
√
DMSP SSMIS
F16/F18
√
AQUA AMSR-E
√
NPP/JPSS ATMS
√ √: Applied Operationally
√: Applied occasionally
√: Tested in Simulation
Metop-A
√
TRMM/GPM/ M-T
TMI, GMI proxy, SAPHIR/MADRAS
√
Current & Planned Capabilities
7
MiRS Approach to Rainfall over All surfaces
Case area after rain event
CPC Figures courtesy http://www.cpc.necp.noaa.gov
CPC real-time 24-hour precipitation from 12Z 2010-10-19, 2010-10-20, 2010-10-22 and 2010-10-23 (from left to right)
MiRS N18 retrieved emissivity at 31 GHz ascending node for 2010-10-19, 2010-10-20, 2010-10-22 and 2010-10-23 (from left to right)
Day in October
Es 19.35V channel 37.0 V channel
Illustration of High Variability of Emissivity (1/2)
Compare 2 days before rain event: 2010-10-19 and 2010-10-20
Compare 1 day before with 1 day after: 2010-10-19 and 2010-10-23
There is a significant difference between emissivities before rain events and during (or shortly after) rain events
Using pre-determined emissivities (monthly, weekly or even daily updates) could potentially lead to issues:
For example: Low emiss values after rain could be interpreted as false rain
Bias:0.00 StdDev:0.01 Bias:0.02 StdDev:0.02
Illustration of High Variability of Emissivity (2/2)
MiRS Approach to handling RR over all surfaces
Emissivity is to be considered highly variable, especially when rain is present, or when it rained a few hours/days ago.
It is also highly variable on a footprint-by-footprint level, especially in heterogeneous areas (rivers, coasts, mountains, etc)
MiRS approach is to include emissivity as part of the state vector, along with hydrometeor
Rely on physical constraints and forward operator/Jacobian to distinguish emisssivity and hydrometeor signals
10
MiRS TPW Weekly composite
MiRS Emiss Weekly composite
Ex. of MiRS emissivity capturing heterogeneous signal (allowing a smooth retrieval of other products along the coast line)
11
Added Value of Emissivity Handling: Same RR algorithm Over Both Ocean and Land
Image taken from IPWG web site: credit to Daniel Villa
No discontinuity at coasts (MiRS applies to both land and ocean)
12
MiRS Current Performance for Rainfall Rate
Independent Validation (IPWG) 1/2
Provide MiRS PE over CONUS, S.A. and Australia
Independent assessment versus radars/gauges
Comparison to other precipitation algorithms
Images courtesy of John Janowiak, UMD; Daniel Vila,
CPTEC; and Elizabeth Ebert,
Australian Bureau of Meteorology
Independent Validation (IPWG) 2/2
Monitor a running time series of statistics relative to rain gauges
Intercomparison with other PE algorithms and radar
Caution: algorithms perfs depend on how many sensors are used
MiRS Testbed: Climatology & Vertical Structure
MiRS Hydrometeor Profiles
TRMM 2A12 Hydrometeor Profiles
Hydrometeor Profiles/Vertical Cross Sections
Rainfall Climatologies
Monthly Averaged MSPPS NOAA-18 Rainfall Rate for 11-2009
Monthly Averaged MiRS NOAA-18 Rainfall Rate for 2009
MiRS
Heritage
16
MiRS Extension to TRMM/TMI and GPM/GMI
Work initiated a month ago: in progress. These are draft results
MiRS for TRMM/TMI
Example of retrieved rainfall rate from MiRS on TMI data at ~5 km resolution (left) compared to TRMM 2A12 (right) for 2010-09-19
MiRS has been extended to TRMM/TMI (work still in progress)
Current issues being addressed: - Non-convergence
- Coastal false alarm signal
Extension of MiRS to GPM/GMI (1/2)
18
GPM/GMI proxy data (simulated brightness temperatures) were generated to test MiRS algorithm.
Simulations performed using CRTM forward model and ECMWF geophysical inputs
Simulations over all surfaces TRMM/TMI metadata used (for
scanning geometry, angle, swath, time, etc) and also for emissivity
Simulations performed daily at NOAA.
Goal: Make sure the algorithm is ready on day-1 for GPM/GMI data (switch between proxy data flow and real data stream)
Example: GMI simulated 36.5 GHz H-pol TB
Current issue being addressed: Apparent pixel shift
GMI is similar to TMI with additional high frequency channels (166 and 183 GHz)
We look forward to using L1B data from GPM simulator (Matsui et al)
Extension of MiRS to GPM/GMI (2/2)
19
MiRS has been applied on the GPM/GMI Proxy data. All products are being assessed, including RR, Emissivity, TPW, etc
Draft Results: Work is still in progress to optimize the emissivity covariance for GMI and TMI
GMI Emiss @ 36.5 GHz H-pol
GMI TPW
Current Limitations & Planned Improvements
Sensor Applicability
Current Limitations Planned Improvements
Importance/ Difficulty
All sensors Current atmospheric covariance is a single covariance used globally
Current effort aims at developing stratified covariances, by latitude and season
Important (to improve warm season perfs)
All sensors Rain Rate relationship (w 1DVAR hydrometeors) is also a single relationship, used globally
Investigate the stratification of rainrate relationship by season/latitude
Important (to improve warm season perfs)
All sensors Very low false alarm rate but Low detection Rate, especially for light rain, due to compensation of light rain signal by other parameters (such as WV)
Make sure high frequency channels have a stronger weight in the Chi-Square computation
Moderate
TMI/GMI Important coastal False Alarm
Improve emissivity covariance (not mature yet for these sensors)
Low
21
MiRS Contribution to Surface Emissivity Characterization
Global Variationally-based Inversion of Emissivity: Routine Assessment
22
MiRS inverts emissivities for all channels, including high-frequency (Inversion performed in EOF space)
Emissivity is assessed by comparing it to analytically-inverted emissivity
Surface Emissivity Inter-Comparison 12/01/2007 – 02/28/2009
Frequency (GHz)
Es
Es
MiRS N18
GDAS
MiRS N18 minus GDAS
Emis
sivi
ty d
iffer
ence
(MiR
S-A
naly
t)
Frequency (GHz)
Ocean __ Sea Ice (Antartic) ___ Sea Ice (Arctic) ___ Sea Ice (First Year) ___
Desert __ Amazon __ Wet Land __ Snow __
Intercomparison between MiRS variational emissivities and analytical ones
Differences within 2%. Larger diffs noticed for snow (~8%) & Arctic sea-ice (3%). Questions: Tskin used in analytical emiss from GDAS accurate enough?
Is assumption of specularity valid for snow and sea-ice?
MiRS TRMM Emissivities (in clear and rainy conditions)
Frequency (GHz)
Non Precip., vertical polar.
horizontal polarization
Rainy, vertical polar.
horizontal polarization
Non Precip & Rainy
Aver
aged
Em
issi
vitie
s us
ing
over
land
Rain changes dramatically the emissivities: - Water-type spectral shape - Lower values - Higher V-H contrast
Summary MiRS is a variational algorithm (1DVAR) and can be applied to virtually any
microwave sensor MiRS uses CRTM as forward and jacobian operators Retrieves sounding & surface parameters simultaneously, including
hydrometeor profiles, rain rate & surface emissivity Applicable over all surfaces (emissivity is part of the state vector), allowing
a spot-by-spot variability of the surface emissivity. Extensively assessed both internally and independently. Applicability in all-weather conditions (including rainy) Run operationally at NOAA for N18, N19, SSMIS F16, F18 and Metop-A,
and being integrated for NPP/JPSS ATMS MiRS is also currently being extended to support GPM (GMI) and Megha-
Tropiques (MADRAS and SAPHIR) Current enhancements to the algorithm expected to improve performances
of hydrometeor retrievals for all sensors We look forward to using GV data when they become available (plan to
extend CRTM, and therefore MiRS to airborne setups) and GPM simulator. Variational Emissivities from MiRS are available (all surfaces, for all
frequencies) as well as corresponding covariances. MiRS is a community algorithm (available publicly), benefiting from
community-driven improvements, suggestions, scrutiny and assessment.
26
BACKUP SECTION
27
1D-Variational Retrieval/Assimilation
MiRS Algorithm
Measured Radiances
Initi
al S
tate
Vec
tor
Solution Reached
Forward Operator (CRTM)
Simulated Radiances Comparison: Fit
Within Noise Level ?
Update State Vector
New State Vector
Yes
No Jacobians
Geophysical Covariance
Matrix B
Measurement & RTM
Uncertainty Matrix E
Geophysical Mean
Background
Climatology (Retrieval Mode) Forecast Field (1D-Assimilation Mode)
MiRS is a physical algorithm:
Solution constrained by
(1) geophysical covariance, (2) fitting measurements, (3) physical Jacobians, (4) physical radiative transfer and (5) simultaneous retrieval of all parameters
28
MiRS General Overview
Radiances
Rapid Algorithms
(Regression)
Advanced Retrieval (1DVAR)
Vertical Integration &
Post-processing
selection
1st Guess
MIRS Products
Vertical Integration and Post-Processing 1D
VAR
O
utpu
ts
Vertical Integration
Post Processing
(Algorithms)
TPW RWP IWP CLW
Core Products
Temp. Profile
Humidity Profile
Emissivity Spectrum
Skin Temperature
Liq. Amount Prof
Ice. Amount Prof
Rain Amount Prof
-Sea Ice Concentration -Snow Water Equivalent -Snow Pack Properties -Land Moisture/Wetness -Rain Rate -Snow Fall Rate -Wind Speed/Vector -Cloud Top -Cloud Thickness -Cloud phase
29
All-Weather and All-Surfaces
Scattering Effect
Scattering Effect
Absorption
Surface
sensor Major Parameters for RT: • Sensing Frequency • Absorption and scattering properties of material • Geometry of material/wavelength interaction • Vertical Distribution • Temperature of absorbing layers • Pressure at which wavelength/absorber interaction occurs • Amount of absorbent(s) • Shape, diameter, phase, mixture of scatterers.
Sounding Retrieval: • Temperature • Moisture
Instead of guessing and then removing the impact of cloud and rain and ice on TBs (very hard), MiRS approach is to account for cloud, rain and ice within its state vector.
It is highly non-linear way of using cloud/rain/ice-impacted radiances.
To account for cloud, rain, ice, we add the following in the state vector: • Cloud (non-precipitating) • Liquid Precipitation • Frozen precipitation
To handle surface-sensitive channels, we add the following in the state vector: • Skin temperature • Surface emissivity (proxy parameter for all surface parameters)
MiRS TMI Surface Emissivity- channel libraries can be built
85.5 V 19.35 H
37.0 H
10.65H
19.35V
37.0V 85.5H
21.3H 10.65V
MiRS F18 SSMI/S Oct. 18, RR mm/hr
MiRS F18 SSMI/S Oct. 23, RR mm/hr
MiRS F18 SSMI/S Oct. 27, RR mm/hr
MiRS F18 SSMI/S Oct. 18, 37V Emis
MiRS F18 SSMI/S RR Oct. 23, 37V Emis
MiRS F18 SSMI/S RR Oct. 27, 37V Emis
October
Es 19.35V channel 37.0 V channel
18th 23rd 27th
Surface Emissivity – Rain Event
MiRS F18 SSMI/S (RPW > 0.05mm)
------ VPol ------ HPol ------ RCP
Rain Rate on Oct 18, mm/hr
Rain Rate on Oct 23rd, mm/hr
Rain Rate on Oct 27, mm/hr
Emissivity in box on 18th Emissivity in box on 23rd Emissivity in box on 27th
Frequency in GHz Frequency in GHz Frequency in GHz
Surface Emissivity – Rain Event
MiRS F18 SSMI/S (RPW > 0.05mm)
Day in October
Es
Rain Rate on Oct. 18, mm/hr
Rain Rate on Oct. 23, mm/hr
Rain Rate on Oct 27, mm/hr
19.35V channel 37.0 V channel
October
Es
MiRS F18 SSMI/S Oct. 18, RR
19.35V channel 37.0 V channel
MiRS F18 SSMI/S Oct. 27, RR
MiRS F18 SSMI/S Oct.18, 37V E MiRS SSMI/S Oct22, 37V E
18th 23rd 27th
MiRS F18 SSMI/S Oct. 22, RR
MiRS F18 SSMI/S Oct. 23, RR
MiRS SSMI/S Oct. 23, 37V E MiRS SSMI/S Oct. 27, 37V E
Channel Freq. (MHz): 1 = 50.3 H 2 = 52.8 H 3 = 53.6 H 4 = 54.4 H 5 = 55.5 H
6 = 57.29 RCP 7 = 59.4 RCP
8 = 150 H 9,10,11 = 183.31 H 12,13 = 19.35 H/V
14 = 22.235 V 15,16 = 37 H/V
17,18 = 91.655 V/H 19 = 63.28 RCP
20-24 = 60.79 RCP
Note: difference in color bar range
F18 SSMI/S Surface Emissivity Correlation Matrix In
Non & Rainy Conditions over Land
Rainy (RPW>0.05mm) Land Surf. Emissivity Correlation Matrix from 5,000 scenes Oct. 2010
Non-Precipitating Land Surface Emissivity Correlation Matrix from 53,000 scenes Oct. 2010
Rainy (RPW>0.05mm) Land Surf. Emissivity Correlation Matrix from 5,000 scenes Oct. 2010
Non-Precipitating Land Surface Emissivity Correlation Matrix from 53,000 scenes Oct. 2010
Channel Freq. (MHz): 1 = 50.3 H 2 = 52.8 H 3 = 53.6 H 4 = 54.4 H 5 = 55.5 H
6 = 57.29 RCP 7 = 59.4 RCP
8 = 150 H 9,10,11 = 183.31 H 12,13 = 19.35 H/V
14 = 22.235 V 15,16 = 37 H/V
17,18 = 91.655 V/H 19 = 63.28 RCP
20-24 = 60.79 RCP
Note: difference in color bar range
SSMI/S Surface Emissivity Correlation Matrix Clear & Rainy Conditions over Land
MiRS Testbed: Stats
Comparison to 24 hr. PE to CPC Rain Gauge Analysis 2010-06-26 (top) and time series of PoD and Correlation from May 2009 – Aug. 2010
38
Parameters are Retrieved Simultaneously: How is Emissivity constrained?
X is the solution
F(X) Fits Ym within Noise levels
X is a solution
Necessary Condition (but not sufficient)
If X is the set of parameters that impact the radiances Ym, and F the Fwd Operator
If F(X) Does not Fit Ym within Noise
X is not the solution
All parameters are retrieved simultaneously to fit all radiances together
Suggests it is not recommended to use independent algorithms for different parameters, since they don’t guarantee the fit to the radiances
Emissivity is simply added to the state vector to be retrieved in the Bayesian Algorithm
It is constrained by: Background covariance (spectral constraints) Physical constraints (CRTM) Fitting the radiances
Atmospheric Covariances Generation
Water Vapor (ECMWF 60) Temperature (ECMWF 60) Cloud Liquid Water (WRF)
Graupel (WRF) Rain Water (WRF) . Representative sets of profiles of temp., moisture, hydrometeors are gathered from different sources (ECMWF, MM5, WRF, GDAS, etc)
. For multiple seasons and a number of latitude ranges (tropical, midlatitude, high latitude).
. These datasets are then used to generate stratified covariances to be used by MiRS.
Future improvement: Stratification of Covariances
Mid-latitude Profiles
Tropical Profiles
Rain
Rain
Ice
Ice
WRF Model Simulation
WRF Model Simulation
Differences in vertical structures of ice, cloud, rain
Differences in how temperature and moisture correlate to hydrometeors
Differences in how rainfall rate relate to integrated values of rain and ice (IWP, RWP)
Atmospheric Covariance Matrix
New Atmospheric Background Covariance Matrix based on ECMWF 60, and WRF simulations over tropic oceans
performed during SON season
Cloud liquid, Rain and Ice water from WRF
MiRS Current Atmospheric Background Covariance Matrix based on Global ECMWF 60, and tropic-ocean
MM5 simulations
Temperature, Water Vapor and CLW from ECMWF 60
Rain and Ice water from MM5
Temperature and Water Vapor from ECMWF 60
Noticeable Differences noticed in covariances, especially in hydrometeors. Impact assessment on RR performances in progress
More Information Publications
S.A. Boukabara, F. Weng and Q. Liu, Passive Microwave Remote Sensing of Extreme Weather Events Using NOAA-18 AMSUA and MHS. IEEE Trans. on Geoscience and Remote Sensing, July 2007. Vol 45, (7), 2228-2246
S.A. Boukabara, F. Weng, Microwave Emissivity Retrieval over Ocean in All-Weather Conditions. Validation Using Airborne GPS-Dropsondes. IEEE Trans Geos Remote Sens, 46, 376-384, 2007
S.-A. Boukabara, K. Garrett, and W. Chen, “Global Coverage of Total Precipitable Water using a Microwave Variational Algorithm,” IEEE TGARS, vol. 48, Sept. 2010
F. Iturbide-Sanchez, S.-A. Boukabara, R. Chen, K. Garrett, C. Grassotti, W. Chen, and F. Weng, “Assessment of a Variational Inversion System for Rainfall Rate over Land and Water Surfaces,” Submitted IEEE TGARS, July 2010.
S.-A. Boukabara et al. “MiRS: An All-Weather 1DVAR Satellite Data Assimilation and Retrieval System,” Submitted IEEE TGARS, May 2010.
Website http://mirs.nedsis.noaa.gov
For More Information: