RADIANCE BIAS CORRECTION HOW GPM CAN HELP AND Sara Zhang 1 , - - PowerPoint PPT Presentation
PRECIPITATION-RELATED RADIANCE BIAS CORRECTION HOW GPM CAN HELP AND Sara Zhang 1 , Philippe Chambon 2 , William Olson 1 , Milija Zupanski 3 and Arthur Hou 1 1 NASA GODDARD SPACE FLIGHT CENTER 2 CNRM-GAME,
Sara Zhang1, Philippe Chambon2, William Olson1, Milija Zupanski3 and Arthur Hou1
1NASA ¡GODDARD ¡SPACE ¡FLIGHT ¡CENTER ¡ 2CNRM-‑GAME, ¡MÉTÉO-‑FRANCE ¡AND ¡CNRS ¡ 3CIRA, ¡COLORADO ¡STATE ¡UNIVERSITY ¡ ¡
AND ¡
THE 6TH WMO SYMPOSIUM ON DATA ASSIMILATION
In satellite data assimilation, bias between observed and model-simulated radiances represents a combination of instrument measurement bias, systematic errors in observation operators, and forecast model errors projected in observation space. Precipitation-sensitive microwave radiances are particularly susceptible to approximations and assumptions on physical properties of precipitation in radiative transfer calculations and model cloud physics schemes.
!hydrometeor phase and amount predicted by model microphysics, storm displacement
hydrometeor shape and size distribution, optical property approximation
FG OBS
Using multi-channel MW radiances, FG-departure statistics are based
both in observations and first guess. Bias model using averaged SIL predictor O-F
FG OBS
Using multi-channel MW radiances, FG-departure statistics are based
both in observations and first guess. Bias model using averaged SIL predictor
O-F
Dual-frequency Precipitation Radar
13.6 GHz (Ku) , 35.5 GHz (Ka)
GPM Microwave Imager
10.7, 18.7, 23.8, 36.5, 89.0, 165.5, 183±8, , 183±3 GHz
GPM-emulated Ku and Ka
from NASA Aircraft-borne APR-2 field campaign
ze = λ 4 π 5 Kw
2
N(D)σ b(D,λ,T )dD
∞
Radar measurements in reflectivity Hydrometeor size distribution N(D) = N0Dµ exp −ΛD
{ }
ze = λ 4 π 5 Kw
2
N(D)σ b(D,λ,T )dD
∞
concentration variability of size proportion of large/small size
Radar measurements in reflectivity Hydrometeor size distribution
frequency back-scattering cross section hydrometeor size distribution
N(D) = N0Dµ exp −ΛD
{ }
ze = λ 4 π 5 Kw
2
N(D)σ b(D,λ,T )dD
∞
DFR = Zku Zka
ρs = αD−β
concentration variability of size proportion of large/small size
Radar measurements in reflectivity Dual-frequency ratio Hydrometeor size distribution Ice-phase hydrometeor density
frequency back-scattering cross section hydrometeor size distribution
N(D) = N0Dµ exp −ΛD
{ }
ze = λ 4 π 5 Kw
2
N(D)σ b(D,λ,T )dD
∞
DFR = Zku Zka
ρs = αD−β
concentration variability of size proportion of large/small size
Radar measurements in reflectivity Dual-frequency ratio Hydrometeor size distribution Ice-phase hydrometeor density DFR is independent of No, and a good proxy for mean mass diameter Dm ρ is inversely proportional to D indicated by observations
frequency back-scattering cross section hydrometeor size distribution
N(D) = N0Dµ exp −ΛD
{ }
ze = λ 4 π 5 Kw
2
N(D)σ b(D,λ,T )dD
∞
DFR = Zku Zka
ρs = αD−β
concentration variability of size proportion of large/small size
Radar measurements in reflectivity Dual-frequency ratio Hydrometeor size distribution Ice-phase hydrometeor density DFR is independent of No, and a good proxy for mean mass diameter Dm ρ is inversely proportional to D indicated by observations
frequency back-scattering cross section hydrometeor size distribution
N(D) = N0Dµ exp −ΛD
{ } Use DFR to infer PSD parameters assumed in the radiative transfer model Use radar-data-adjusted parameters to correct bias in FG MW radiances
Liao, L. and R. Meneghini, 2011: A Study on the Feasibility of Dual-Wavelength Radar for Identification of Hydrometeor Phases.
50, 449–456.
DFR and Ku can identify hydrometeor phases Snow density and diameter are inversely related
89GHz (OBS) DFR (OBS) Radar & MW observations and FG are simulated with different PSD parameters. FG-departures reflect only this bias in radiance observation operator.
89GHz (OBS) DFR (OBS) 89GHz (FG) Radar & MW observations and FG are simulated with different PSD parameters. FG-departures reflect only this bias in radiance observation operator.
Ice-phase particle scattering
bias
BT (K) FG BT (K) OBS
89GHz (OBS) DFR (OBS) 89GHz (FG) Radar & MW observations and FG are simulated with different PSD parameters. FG-departures reflect only this bias in radiance observation operator. DFR of Ka and Ku radar observations are used to infer PSD parameters (Dm and ρ ) FG MW radiances (89GHz) are recalculated using radar-inferred PSD to reduce bias
Ice-phase particle scattering
bias
BT (K) FG BT (K) OBS
DFR 89GHz
OB OBS S FG_B G_BC FG G OB OBS S ES ESTM M FG G
Simulations by Météo-France cloud-scale model (AROME) are used to create synthetic GPM observations. The Goddard cloud-scale ensemble data assimilation system uses WRF with Goddard cloud physics. FG-departures mimic realistic distribution of precipitation-related errors.
Ku Ku DFR DFR FG-Departure 89GHz Surface rain OBS (AROME) FG (WRF)
Averaged DFR with SIL symmetrical sampling FG-departure 89GHz Sample counts
Background Hydrometeors, T, q etc. DPR Zku, Zka estimation of PSD parameters GMI radiance Radar reflectivity Simulation, DFR Ensemble filter analysis to produce increments on model state (mixing ratio of hydrometeors, etc.) Radiance Simulation With DPR-derived bias correction Background Hydrometers, T, q etc. Eventually to adaptive bias correction parameter augmentation and simultaneous estimation in ensemble filter
Biases in precipitation-sensitive radiances are related to approximations and assumptions on hydrometeor PSD in radiative transfer calculations and model cloud physics schemes. GPM dual-frequency precipitation radar data can be used to infer PSD parameters in a physically-derived bias correction scheme for precipitation-affected radiances. Development and implementation are ongoing in OSSE using synthetic GPM observations and in real data assimilation experiments using NASA field campaign observations.
THIS WORK IS A COLLABORATION OF NASA GPM SCIENCE PROGRAM AND MÉTÉO-FRANCE