Global Precipitation Measurement (GPM) Mission
An International Partnership & Precipitation Satellite Constellation for Research on Global Water & Energy Cycle Anticipated Improvements in Precipitation Physics & Understanding of Water Cycle from GPM Mission IGARSS03 -- IEEE International Geoscience & Remote Sensing Symposium [Session on TRMM and GPM] NASA/Goddard Space Flight Center, Greenbelt, MD [tel: ; fax: ; July 21-25, 2003; Centre des Congrs Pierre Baudis, Toulouse, France Improving Precipitation Retrievals
Cloud Macrophysical & Microphysical Fundamentals Determination of: drop size distribution [DSD(r)], mass mixing ratio [q(z)hydro(r)], rain mass flux [Fr(z)], fall velocity [w(z)hydro(r) ], & latent heating [LH(z) ] q(z)hydro(r) = sw (4/3pr3) DSD(r) w(z)hydro(r) = GFO [q(z)hydro(r)] Fr(z) = q(z)hydro(r) w(z)hydro(r) dr LH(z) = C [ Fr(z)/z] RR (z) = Fr(z) / sw RFsur = RR(zsur) Dt Fiorino, S., and E.A. Smith, 2003:Examination of Microphysical Assumptions within TRMM Radiometers Rain Profile Algorithm (2a12) using Satellite, Aircraft, & Surface Datasets from KWAJEX Implementation of Fully Modular OPEN ACCESS Facility Algorithms Accompanied by COMPREHENSIVE TESTING Capability within PPS TRMM & GPM Rainrate Retrieval Simulations Under
Varying Mean Adj Drop Diameter Profiles [ simulations based on Monte Carlo proliferation of Hurricane Bonnie observations ] TRMM Single-Frequency Algorithm (bias due to irretrievable DSD variability) GPM Dual-Frequency Algorithm (near-zero bias & reduced scatter in mid-range) R e t r i v d (mm hr-1) Actual R (mm hr-1) Actual R (mm hr-1) Standard Deviation of R (%) as Function of R in mm hr-1 95% of LH spectra TRMM modal value of LH with R approximately log-normal then sR proportional to R GPM exact variability depends on DSD variability in altitude Percent ~1.5 mm hr-1 ~12.5 mm hr-1 Latent Heating for Hurricane Bonnie from Vertical Derivative of Rain Mass Flux
Droplet Fall Velocities Derived from TRMM Combined Radar-Radiometer Algorithm 2b31 (S. Yang, Z. Haddad, & E. Smith) CAPI View mm hr-1 Rainrate Cross-Section mm hr-1 Latent Heating Cross-Section deg hr-1 Global Water Budget Water Vapor Residence Time
WV Residence Time (RT) = Ave Atmospheric WV Reservoir Path / Ave RR For example, and keeping things simple: RT = 25 mm / 2.5 mm day-1 = 10 days Thus even at 5% retrieval accuracy, and climatic fluctuations of RT on order of less than 10%, many years of global precipitation data will be needed to assess water cycle acceleration -- Dennis Lettenmaier and his colleagues estimate this to be some 60 years. Precipitation Prediction: Key Objective of Water Cycle Research
NOW State of Art Climate Model (CCSM) GOAL Next Generation Climate Model Satellite-based Water Budget of Gulf of Mexico & Caribbean Basins
[P. Santos & E.A. Smith, 2003] Gulf-Caribbean Basins & Upper Air/Buoy Validation Data Sites Design of Algorithm System GOES Combined TRMM-SSM/I & GOES SSM/I GOES Q Uncertainty (%) vs Sample Count (N) Line Integral Validation Study Area, GOES-SSM/I-TRMM Sectors, & ECMWF Grid SSM/I GOES GOES Gulf Basin Caribbean Basin ECMWF Validation Jul Z Fully-Averaged Monthly Framework
Atmosphere Q = 14% (PW + LWP)/t = 0% Vapor Transport to Surroundings Q = (E - P) Vapor-Condensate Storage E = 50% P = 36% Ocean Diurnally-Averaged Monthly Framework
Atmosphere Q = 32% (PW + LWP)/t = 32% P Vapor Transport to Surroundings Vapor-Condensate Storage E = 22% P = 14% Ocean TRMM Precipitation Observations for Improving Hurricane Track Forecast
Case Study 1: Hurricane Bonnie (September 2001) Research from NASA/GSFC Data Assimilation Office (DAO); courtesy of Arthur Hou Assimilation of TRMM precipitation data in global models Improves climate analysis Improves storm track forecast Improves precipitation forecast 5 day forecast of Bonnie storm track from08/20/98 Red:best track(NOAA HRD) Green: forecast from analysis without TRMM data Blue: forecast from analysis with TRMM data Forecast Number TRMM Precipitation Observations for Improving Hurricane Track Forecast
Case Study 2: Cyclone Zoe (December 2002) Research from ECMWF; courtesy of E. Moreau, A. Benedetti, & P. Bauer ECMWF 4D-Var Model Assimilation Reference control run uses wind profilers, ships and buoys, radiosoundings, HIRS data, geostationary satellite data, QuickSCAT wind data, SSM/I radiances in clear-sky areas, and AMSU data. TRMM Radiometer (TMI) and Precipitation Radar (PR) data are included in subsequent runs to assess impact on track prediction. Forecast Time (hour) Research from Fla. State Univ.; courtesy of M. Ou & E.A. Smith
GMS-IR Precipitation Observations for Improving Convective Storm Forecasts Case Study 3: Future Time Data Assimilation over Korean Peninsula using UW-NMS Mesoscale Model (1998, 1999, 2000) Research from Fla. State Univ.; courtesy of M. Ou & E.A. Smith Concerning Data Assimilation
Urban Legend Assimilation of SSM/I & TRMM satellite precipitation observations have improved numerical weather forecasting. Reality NCEP is only operational center producing routine weather forecasts using satellite precipitation data assimilation; impact has been marginal because they use 3d-VAR system which does not exploit time-accumulation effect. ECMWF has not achieved overall positive impact with their 4d-VAR system, because their software is not yet on-line. GSFC-DAO (Hou et al), FSU (Krishnamurti et al., Ou & Smith), ECMWF (Mahfouf and Bauer), and JMA (Nakazawa et al) have shown substantive overall positive impact with satellite precipitation data assimilation schemes that exploit time-accumulation effect -- but none of these assimilation schemes are based on optimal 4d-VAR technique. No operational or research-based numerical weather prediction center has used vertical rainrate profile information; only surface rainrates are being used, and then only to adjust vertical moisture profile, not to generate latent heating. No operational or research-based numerical weather prediction center (with one possible dubious exception) has used either vertically-integrated or vertically-distributed diabatic (latent) heating information from TRMM for data assimilation purposes. No operational or research-based numerical weather prediction center has used independent, objectively- acquired retrieval error characterization information; those few modeling centers using error information in their data assimilation schemes have been limited to constant or conditional error variances fixed in space-time, applied in single column data assimilation mode. Space-time error covariance information has yet to be used anywhere -- this representing the underlying quantitative property under which data can be objectively leveraged and ingested into forecastswithin formal mathematical framework of data assimilation. Operational numerical weather prediction centers have not assimilated precipitation observations wherever model grid predictions are zero, and have resisted making what they view as ad hoc adjustments to allow such calculations to take place -- this is tantamount to defining a models background bias as perfect. NASA Objectives for GPM Missions GV Program
Living -- dynamic research and operations program: generate retrieval error characteristics from independent GV measurements coupled to satellite retrievals. Deploy small network of GV Supersites: sites which operate under data exchange protocol between each sites Supersite Science-Center (SSC) and GPMs Precipitation Processing System (PPS) -- in which PPS-generated satellite retrievals (high bandwidth data packets) relative to given Supersite are passed to SSC and SSC-generated retrieval error characteristics (low bandwidth data packets) are passed to to PPS. What are satellite retrieval error characteristics? Conditional Rainrate Bias Uncertainty: i.e., bias as function of rainrate (accuracy). Error Covariance Structure: pixel-relative, local space-time precision error distribution matrices based on using imperfectly calibrated but high time-space resolution volume-scan radar rainrates as truth data for local space-time auto-correlation structure. Error Characterization (Accuracy)
Bias (B) & Bias Uncertainty (DB) At Supersite B(RRi)tk = j = -NT/2,+NT/2 [1/(NT+1)] [ RRiSR(tj,RRi)RRiPEM(tj,RRi] B(RRi)tk end-to-end uncertainties in PEM {for i = 1 , L rainrate intervals (~5) and time period tk} Based on Physical Error Model based on: physical error model ( passive-active RTE model ) matched satellite radiometer/radar instrument on ground with continuous calibration ( eyeball ) independent measurements of observational inputs needed for error model (DSD profile, T-q profile, surface R) All retrievals from constellation radiometers & other satellite instruments are bias- adjusted according to bias estimate from reference algorithm for core satellite. Error Characterization (Precision)
J(x) = (xb x)T F-1 (xb x) + ( yo H(x))T ( O + P )-1 ( yo H(x)) F, O, & P are error covariance matrices associated with forecast model, observations, & forward model (precip parameterization), where yo , H, & x are observation, forward model, & control variable. Space-Time Observational Error Covariance (O) At Supersite (regional expansion rule based on DPR) O(rrj,j,tj)tk = rj = 0,100 ri = 0,100. j = 0,360 i = 0,360 j = -NT/2,+NT/2 i = -NT/2,+NT/2 [1/NT] [ SR(rrj,j,tj) GV(rMOD(ri+rj,100),MOD(i+j,360),ti+j) ]2 {given polar coordinates (r,) for r out to 100 km and time period tk} Space-Time Autocorrelation Structure Given By volume scanning ground radars ( dual-polarization enables DPR calibration cross-checks ) research-quality, uniformly distributed, dense, & hi-frequency sampled raingage networks GPM Validation Strategy
Tropical Continental Tropical Oceanic I. Basic Rainfall Validation Raingauges/Radars new/existing gauge networks new/existing radar networks Research Quality Data Confidence sanity checks II. GPM Supersites Basic Rainfall Validation hi-lo res gauge/disdrometer networks polarametric Radar system Accurate Physical Validation scientists & technicians staff data acquisition & computer facility meteorological sensor system upfacing multifreq radiometer system Do/DSD variability/vertical structure convective/stratiform partitioning GPM Satellite Data Streams Mid-Lat Continental Continuous Synthesis error variances precip trends Calibration Improvements Algorithm Supersite Products Extratropical Baroclinic III. GPM Field Campaigns GPM Supersites cloud/ precip/radiation/dynamicsprocesses GPM Alg Problem/Bias Regions targeted to specific problems Research cloud macrophysics cloud microphysics cloud-radiation modeling FC Data High Latitude Snow Actual and Potential GPM Ground Validation Sites
Finland U.K. Canada Germany Japan -- CRL-Northern Wakkanai Netherlands Austria U.S. -- NASA-Land DOE/ARM-SGP South Korea France Italy Greece China Spain C Japan -- CRL-Southern Okinawa Israel U.S. -- NASA-Gauge KSC Taiwan India U.S. -- NASA-Ocean Kawajalein/RMI Brazil South Africa Australia Radar GV Site or GV Supersite Regional Raingage Site Both Supersite Template DELIVERY Focused Field Campaigns Legend
Data Acquisition- Analysis Facility Focused Field Campaigns Multiparameter Radar GPM Core Satellite Radar/Radiometer Prototype Instruments Uplk Mtchd Radiom/Radar S-/X-Band Profilers 90 GHz Cloud Radar Meteorological Tower & Sounding System Piloted Site Scientist (3) Technician (3) UAVs Retrieval Error Synthesis DELIVERY Meteorology-Microphysics Aircraft Algorithm Improvement Guidance Validation Analysis 150 km 50-Gage Site Hi-Res Domain Center-Displaced with Uplooking Matched Radiom/Radar [10.7,19,22,37,85,150 GHz/14,35 GHz] Upward S-/X-band Doppler Radar Profilers & 90 GHz Cloud Radar 150 km 5 km Triple Gage Site (3 economy scientific gages) Single Disdrometer/ Triple Gage Site (1 high quality-Large Aperture/ 2 economy scientific gages) 100-Gage Site Lo-Res Domain Centered on Multi-parm Radar Proposed Physical Error Model (PEM)
(E.A. Smith & K-S. Kuo, 2003) Use Ground Eyeball Radar-Radiometer Z-TB measurements in conjunction with matched Core Satellite DPR-GMI measurements to observe both ends of reflectance-radiancetube between satellite and eyeball instruments. Use dual-frequency ground Doppler Radar Profiler measurements (either VHF-UHF or UHF-SBand) to provide initial guess DSD profile to Radiative Transfer Model (RTE model), variationally adjusting DSD profile to within standard error of estimates to optimally match observed Z-TB observations, in which residual mismatch objectively defines bias uncertainty. Take time-average of realization differences vis--vis satellite rainrate algorithm estimates with modeled estimates (diagnosed from resultant model-adjusted DSD profiles) to define conditional bias errors. Based on TRMM analyses, monthly zonally-averagedaccuracies are expected to be approximately 5%. Formulation for Single Scattering Properties
Graupel and aggregate hydrometeors are assumed to be clusters of multi-layered spherical particles. Single-scattering properties of multi-layered sphere is obtained using multi-layered Mie solution (Johnson, 1996) capable of ~7000 layers. One of six canonical configurations is assumed for each hydrometeor, whose single-scattering properties are calculated using consummate solution (interacting dipoles) for ensemble of spheres (Fuller and Kattawar, 1988a-b). Population of hydrometeors is assumed to be composed of such particles with various sizes in specified orientations (e.g., random or oriented).Bulk scattering properties of population are then derived accordingly. 3-Dimensional, Time-Dependent, Deterministic Radiative Transfer Model
used to simulate multiple scattering within absorbing gaseous medium containing 3-dimensional heterogeneous mix of hydrometeors Reverse Monte Carlo Plane-Parallel 3-D Deterministic Model Description: 3-dimensional geometry with heterogeneous composition. Deterministic solution, as opposed to reverse Monte Carlo solution. Picard iterative solution, akin to successive-order-of-scattering solution. Capable of simulating responses to time-dependent sources such as radar pulses. Additional Notes: Picard iteration is arguably most efficient 3-D deterministic RTE solution. Time-dependent solution is obtained by succession of steady-state simulations under momentarily constant medium conditions. Comparison of Radiative Transfer Solution Methods
3D Deterministic Reverse Monte Carlo Plane-Parallel Generality High Low Spatial Heterogeneity 3D 1D Intensity Output In all quadrature directions at all grid points In few directions at few positions layers Computational Demand O(NxNyNzAq2Af2) Moderate S.I.B. O(NxAq2) Application to GV Error Characterization
Matching two-point measurements with radiative transfer simulation by perturbing hydrometeor profile and physical parameters in RTE model. Use hydrometeor profile retrieved from 2-frequency ground-based Doppler profiler (radar) as starting input to RTE model. Perturb model hydrometeor input, to within standard error of measurement, until there is optimal match between simulated reflectances-radiances and those from spaceborne and ground radar-radiometer measurements. Hydrometeor profile determined from best agreement between observed and simulated reflectances-radiances -- result taken as truth for purpose of accuracy assessment of spacecraft retrievals. Note: Effects of small perturbations can be efficiently calculated without running time-consuming model again by first solving adjoint form of RTE (Box et al., 1988, 1989; Polonski and Box, 2002). 1st International GPM GV Research Programme Workshop
November 3-6, Abingdon, England Hosted by Rutherford-Appleton Laboratory & Chilbolten Radar Facility (c/o Dr. John Goddard) Program Committee Eric Smith, Kenji Nakamura, Alberto Mugnai, John Goddard, Carron Wilson, Paul Hwang Main Workshop Objectives 1.Present and share opinions on interests, perspectives, and concerns. 2.Examine conceptual and/or planned GV site templates from NASA, NASDA, ESA, and Other partners. 3.Define scope of international GV research programme. 4.Identify provisional, basic set of GV programme requirements. GPM Mission Design OBJECTIVES Core Satellite Constellation Satellites
Understand horizontal & vertical structure of rainfall, its macro- & micro-physical nature, & its associated latent heating Train & calibrate retrieval algorithms for constellation radiometers Provide sufficient global sampling to significantly reduce uncertainties in short-term rainfall accumulations Extend scientific and societal applications Core Constellation Core Satellite TRMM-like spacecraft (NASA) H2-A rocket launch (JAXA) Non-sun-synchronous orbit ~ 65 inclination ~400 km altitude Dual frequency radar(JAXA/CRL) K-Ka Bands ( GHz) ~ 4 km horizontal resolution ~250 m vertical resolution Multifrequency radiometer (NASA) 10.7,19,22,37,85, (166/1833/7) GHz V&H Constellation Satellites Pre-existing operational-experimental & dedicated satellites with PMW radiometers Revisit time 3-hour goal at ~90% of time Sun-synch & non-sun- synch orbits km altitudes Precipitation Validation Sites for Error Characterization Globally distributed ground validation Sites & Supersites (research quality radars, up-looking GMI/DPR-like radiometer-radar systems, dual-frequency Doppler profiler systems, raingage-disdrometer networks, & T-q soundings) Dense & frequently reporting regional raingage networks Precipitation Processing Center Produces global precipitation products Products defined by GPM partners Optimization and Compromise Potential New Drones/Partners
GPM International Constellation Architecture NPOESS-1 Reference GPM Core (CMIS) NASA-Partner CS (GMI , DPR-Ku/Ka) DMSP-F18/20 b AQUA (GMI) (SSMIS) (AMSR-E) NPOESS-2 TRMM EGPM Co-Op Drone Partners DMSP-F19 (CMIS) DMSP-F16 AQUA Optimization and Compromise (SSMIS) Potential New Drones/Partners (EGPM-PMR , NPR-Ka) DMSP-F17 ADEOS-II NPOESS-Lite FY-3D CORIOLIS FY-3C NPOESS-3 (CMIS) ADEOS-II (PRC-PMR) (CMIS) (PRC-PMR) MEGHA TROPIQUES GCOM-B1 (AMSR) (MADRAS) NPP (back up) (AMSR-FO) TBD (ATMS) Pixel Frequency (log scale)
Assessment of ATMS/WindSat as Rain Instruments TMI (5-frequency) SSM/I (4-frequency) Mean (retr) = 1.33 Bias = 0.04 rms = 0.85 r = 0.92 Mean (retr) = 1.41 Bias = -0.02 rms = 1.13 r = 0.84 ATMS total power radiometer sun-synchronous orbit cross-track scanned 824 km altitude 2300 km swath 0.25 m antenna 23.8 GHz (5.2 deg B/W) 31.4 GHz (5.2 deg B/W) 90.0 GHz (2.2 deg B/W) 33% degradation Retrieved Rainrate (mm hr-1) Retrieved Rainrate (mm hr-1) True Rainrate (mm hr-1) True Rainrate (mm hr-1) ATMS (3-frequency) WindSat (5-frequency) Mean (retr) = 1.55 Bias = 0.02 rms = 1.69 r = 0.66 WINDSAT total power radiometer sun-synchronous orbit conically scanned 830 km altitude 1025 km swath 1.8 m antenna 6.8 GHz (40 x 60 km res) 10.7 GHz (25 x 38 ) 18.7 GHz (16 x 27 ) 23.8 GHz (12 x 20 ) 37.0 GHz (8 x 13 ) Mean (retr) = 1.35 Bias = -0.04 rms = 0.83 r = 0.92 99% degradation 2% improvement Retrieved Rainrate (mm hr-1) Retrieved Rainrate (mm hr-1) Pixel Frequency (log scale) International Unified Physical Validation Program