estimation of clouds in atmospheric models tomislava vukicevic cira/csu and paos/cu
TRANSCRIPT
Estimation of clouds in atmospheric models
Tomislava VukicevicCIRA/CSU and PAOS/CU
Motivation for accurate information on cloud
properties • GCM, NWP and CRM
– Development and validation of cloud parameterizations
– Initialization in NWP
• Assessment of current climate – Hydrologic trends– Interaction with other climate system
components
Observation sources
• Special site measurements (ARM)• Field experiments• Satellite remote sensing • Ground based remote sensing
Mostly indirect Retrieval content limited relative to desired
information Spatial distribution Model quantities not observable
4D cloud data assimilation
Satellite radiance
CRM with bulk cloud
microphysics
+
Sensitive to atmospheric hydrology
High spatial and possibly temporal resolution
GOES Wavelength Central Detector
Channel (µm) Wavelength Resolution (µm)
(km)___________________________________________
1 0.52-0.72 0.7 12 3.78-4.03 3.9 4 3 6.47-7.02 6.7 8
3 G12 5.77-7.33 6.5 4 4 10.2-11.2 10.7 4 5 11.5-12.5 12.0 4
6 G12 12.9-13.7 13.3 8
GOES imager
15 minute data
VISNear IR
Diff between ice and water
clouds
IR water vapor
IR clouds and surface
IR clouds, surface and low level
vapor
CRM RAMS• Bulk, 2 moment cloud microphysics for ice:
pristine ice, aggregates, snow, graupel and hail• 1 moment for liquid: cloud droplets and and rain• Prognostic mixing ratio and number
concentration in 3D• Assumed Gamma distribution with prescribed
width• Nonhydrostatic dynamics • Regional simulations with initial and boundary
conditions from weather analysis
Technique
• Nonlinear 4DVAR • Full physics nonlinear forward
model • No approximations in adjoint of
RAMS with cloud microphysics• Quasi-Newton minimization of cost
function with preconditioning
Mapping from CRM to GOES VIS and IR operator
ytXHy )( Greenwald et al. 2003
Gas absorption: OPTRAN (McMillin et al., 1995) Cloud properties: Anomalous Diffraction Theory
Solar: SHDOM (Evans, 1998)
IR: Eddington two-stream (Deeter and Evans 1998)
4D assimilation of GOES imager IRerror statistics
(model – observation)
mean = 0.3 K
sd = 5.9 K
mean = 33 K
sd = 8.2 K
prior posterior
Brightness Temperature
Brightness Temperature
Vukicevic et al, 2004, 2005
Verification of the estimate in 4D cloud study against independent obs
ARM Cloud Radar reflectivity
Before assimilation
After assimilation
observations
Time
Thick ice cloud
Liquid cloud
Heig
ht
km
1 hour
Thick ice cloud
Thin ice cloud
More observations better result
Single channel assimilations, 30 min frequency
2-channel assimilation, 30 min frequency
2-channel assimilation, 15 min frequency
GuessWorst
Best
m7.10 m0.12
Tb errors
Conclusions• Successful estimation:
– Information content in the model enhanced consistent with the the observation information content
– Stronger observational constraint narrower error distribution
• But, model was applied as strong constraint
Linear model error addedas in other NDVAR studies
Did not work : no convergence
• Conclusion Linear generic model error not appropriate
in cloud estimation
• Suggested approach Physically based model error model
parameter estimation
Comparison of state and parameter estimation
Lorenz 3 component system
Estimation technique: Markov Chain Monte Carlo (MCMC)
Estimation of parameters
State solutions within estimation period
PDFs of parameters after estimation
Estimation period
Forecast period
Estimation of initial condition (state)
PDFs of initial condition components after estimation
State solutions in forecast using mean of distribution as best estimate
State solutions in forecast using maxima of distribution as best estimate
X Y
X forecast forecast
Estimation of state
Observations without errors
XY
Erroneous observations
Derivation of suitable form of parameterization for estimation
Possible solution Extend information from the
measurements into 3D+time
CRM information is not accurate but has skill
CRM simulation in 600 by 17
domain started from crude 4D weather analysis
Mixed phase
Pristine ice
Liquid cloud
rain
Horizontal circulation
Vertical circulation
Ground based
Satellite
Sensitivity to reducing frequency of observations
mean = -0.6 K
sd = 9.7 K
observations
Posterior all obsPosterior less obs
mean = 0.3 K
sd = 5.9 K
Less observations
flat distribution
less accuracy
Sensitivity to channels
Sensitivity to clouds in 10.7nm and 12.0 nm is very similar.
Are both channels needed?
Ch 4 alone
Ch 5 alone
4 and 5 together
4 and 5 together
Ch 4 prior
Ch 5 prior
Model – Observations brightness temperature
Yes
Complementary information
more accuracy
Study conclusions• Ice cloud well specified by GOES imager IR
channels 4 and 5 and CRM when all observations were used
• Weaker observational constraint wider error distribution, less accuracy
• Modeled liquid cloud not improved below ice cloud– No observational constraint: need other
measurements, different channels
• Modeled cloud environment only slightly improved– Weak observational constraint: need other
observations
What next?• Add more satellite observations
– Visible channel– GOES sounder – Microwave for precipitation– Other IR
• Add ground based measurements – ARM
• Goal is to test how much constraint is there in the observations for variety of cloud cases
Back to original motivation Problems
1. Retrievals from any of the measurements cannot fully verify parameterizations
Solution: Assimilation of satellite and other observations into CRM(s) to represent 4D cloudy atmosphere
2. Current cloud climate trends and role of clouds inconclusive because the current retrievals are not accurate enough or the observation information content is insufficient
Solution: Systematic assimilation of satellite and other observations into future NWP with CRM resolution OR cloud properties 1D retrievals with multi
channel measurements in high spatial resolution
modelsObservational
operators
States and
parameters
Adjoint models
VIS and IR information content analysis Example for case with mixed phase clouds
VisibleVisible
Near IRNear IR
IR IR
•Vertical and horizontal variability
•Sensitivity to multiple cloud layers
Greenwald et al, 2004
Sensitivity by optical properties and hydrometeor type
prior Observations
posterior
+ =
Model 3D
cloud
2D
Tb
Sequence every 15 min End time shown
4D assimilation of GOES imager IRmulti-layered non-convective case