Clouds and Precipitation

Christian Kummerow Colorado State University JCSDA

December 1, 2015College park, MD

CloudsActive/Passive

A CloudSat view of Hurricane Ileana(23 Aug. 2006)

CloudsGeo/Leo

Satellite radiometers measure (>1980)

emitted, reflected, scattered radiation

cloud detectioninverse radiative transfer

cloud properties

Passive systems better cloud top or bulk cloud properties

Active systems have better layer information and cloud base, but sparse sampling

Cloud Properties from Space

1 Oct 2012 6

IR-NIR-VIS Radiometers good spatial resolution (1-5km), 1 to 5 radiometric channels: depending on day-night1) COD,CT (assumption on microphysics) 2) spectral difference (VIS-NIR) -> CRE, CWP

IR Sounders 15km res, sounding CO2 abs band (5-8 channels): sensitive to thin Ci (COD>0.1), day&night1) CP,CEM (no assumption on microphysics) 2) spectral difference (8-12mm) -> CRE, CWP (only Ci)

multi-angle VIS Radiometers 1/20km res, only day, only sensitive to clouds with COD>2: Ci over low cld -> low cldmulti-angle scattering -> cloud top polarization -> CT

independent phase

CALIPSO, HIRS,TOVS,AIRS, MODIS

ISCCP, PATMOSx, ATSR-GRAPE

MISR, POLDER

Ci over low clouds : Interpretation of Cloud height

20% of all cloudy scenes (from CALIPSO)

7

Interprétation des propriétés nuageuses

CA 0.68 ± 0.03 (+ 0.05 subvisible Ci); global monthly variability: 0.27; interannual variability: 2-4%

CAHR (hgh clds out of all clds) depends on sensitivity to thin Ci (30% spread) (misidentified as midlevel clouds by ISCCP, ATSR, POLDER)

CAHR 50% (incl. subvis Ci), 42% (COD>0.1), 20% (COD>2); CAMR 16% (±5%); CALR 42% (±5%)

CAEH (CAH weighted by CEMH) agrees better : 0.17

Global Averages: total & height-stratified

CALIPSO only considers uppermost layers to better compare with the other datasets

Effective Cloud AmountCloud Amount

CALIPSO: including subvis Ci, T(cld top) Passive Remote Sensing: T(rad. cld height)

=> CTH(CALIPSO-ST) should be highest & nearest to tropopause,

Better agreement for low-level clouds because these are less diffusive

Cloud Temperature

9

Interprétation des propriétés nuageuses

CAIR+CAWR=100%(except AIRS/TOVS: ice < 230K, liquid > 260K;missing 35% correspond to mixed phase clouds)

CREI, CREW agree quite well: 25 mm (± 2 mm) / 14 mm (± 1 mm) CWP / COD depend on retrieval filtering:• ATSR OE valid only for 40% of all clouds• MODIS-ST only for COD > 1• AIRS / TOVS ice < 230 K, semi-transp.

cirrus

Global Averages: ice - liquidRetrieval of bulk microphysical properties needs thermodynamical phase distinction:• polarization (POLDER, CALIPSO)• multi-spectral (PATMOS-x, MODIS, ATSR)• temperature (ISCCP, AIRS, TOVS)

Cloud Water Content from Microwave

F16 F17

Minimal dependence on microphysics but • Liquid only• Ocean only

Clouds Different observing systems have different sensitivity to cloud

detection, but use additional channels, polarizations & vu angles plus “a-priori / experience” to constrain.

Cloud simulators can account for different detectors but generally need more information (Reff, shape) than model can supply.

Argue that retrievals and data assimilation are the same with “a-priori / experience” replaced by “model forecast”. Can the retrieval “algorithm” be included in model covariance? If so, then radiance and product assimilation become consistent.

New high space/time resolution cloud images (e.g. GOES-R or Himawari-8) contain information about cloud growth rates and potentially latent heating. Rate of cloud changes also speak to cloud processes.

When do clouds begin to precipitate?

Fractional occurrence of rainfall as a function of percentiles of cloud depth, stratified by percentiles of PIA (colored curves) for warm, single-layered clouds (a) observed by CloudSat and (b) simulated with RAMS. Nodes along each PIA curve represent the median cloud depth values within each cloud depth percentile bin.

Models begin to precipitate at fixed LWCReal clouds do not.

Precipitation

The GPM Core Observatory carries two advanced instruments that allow us to view precipitation (rain, snow, ice) in new ways and serve as a connector between the GPM Core and measurements taken on other partner satellitesGPM Microwave Imager (GMI): 10-183 GHz13 channels provide an integrated picture of the energy emitted by precipitation, including light rain to heavy rain to falling snow. Like an X-Ray.Dual-frequency Precipitation Radar (DPR): Ku-Ka bandsTwo different radar frequencies that measure precipitation in 3-D throughout the atmospheric column. Like a CT Scan. Built by JAXA Non-Sun-Synchronous orbit at 65o inclination

(Arctic to the Antarctic Circle) at 407 km

corrected Ze

Zm

GPM Cross section of reflectivities (Typhoon Ita)

Can reflectivities be directly assimilated?

- very sensitive to size and shape of drops

Radar typically uses additional constraint to determine DSD. Ground based radars use gauges to adjust DSD . Spaceborne radars use total attenuation (does not work as well over land).

Two frequencies can also constrain DSD as ΔZ contains some DSD information.

Model needs variable DSD and melting particles for this to work

Imaging Radiometers

Various estimates for spillover correction (eta) for each GMI channel. Final values are indicated by solid yellow line (Courtesy Tom Wilheit).

Rev G (Final)

10v 10h 18v 18h 23v 36v 36h 89v 89h 166v 166h 183±3 183±7

GMI Calibration Summary

Cold calibration No evidence of emission from either the main reflector or cold

calibration subreflector. See figure on the left. (based on analysis by Spencer Farrar at UCF)

Magnetic anomalies Along-track magnetic anomalies due to spacecraft flying through

Earth’s magnetic field Cross-track magnetic anomalies due to magnetic latches for

GMI cover (10H worst effect) Correction developed and applied. Residual anomalies are very

small. Polarization check based on nadir view

Analysis of nadir view looks by Spencer Farrar found differences between V & H Ta within ~0.3K over ocean and ~0.2K over land (consistent with expectation for no polarization difference at nadir)

Land nadir polarization difference provides strong constraint on any calibration adjustment as it would have to leave cosmic background temperature unchanged. Remaining options include adjustment to warm load and/or spillover correction

10v 10h 18v 18h 23v 36v 36h 89v 89h 166v 166h 183±3 183±7

5.0

4.5

4.0

3.5

3.0

2.5

Ta (K

)

Observed vs. Expected GMI Cold Space Ta

Observed vs. expected antenna temperatures by channel based on analysis of data from deep space calibration maneuver (Courtesy Spencer Farrar, Univ. Central Florida).

Change in spillover correction (eta) based on inertial hold analysis (Courtesy David Draper, Ball Aerospace).

• Spillover Corrections– Majority of the antenna pattern missing the Earth is in the spillover region (i.e. the

radiation from behind antenna that comes in around the edges of the main reflector into the feed horns.

– Forward part of antenna pattern measured by Ball at near field range pre-launch, but spillover region could not be measured so they used two different models, which gave different answers.

– Initial spillover corrections (Eta) for 166 and 183 channels were 1.0 (unphysical)– Data from 2 inertial hold maneuvers were analyzed by David Draper at Ball

Aerospace, who derived new Eta values for all of the channels.– The resulting Eta values (see table/figure on the left) are based on physical

observations rather than models (as used initially). These values are also not tuned to match any radiative transfer model.

•Summary– The GMI calibration appears to be at least as good and likely better than any

other window-channel radiometer.– A conservative estimate for the absolute calibration errors of the GMI window

channels are < 1K– Comparisons of the GMI 166 and 183 GHz channels with the MHS cross-track

sounders show differences of < 0.5K

Channel Previous

(ηF)Final(ηG)

ΔTb (ocean)

TotalError

10v 0.94435 0.95404 1.7 0.4510h 0.94369 0.95404 1.0 0.2418v 0.93968 0.95603 3.3 0.6118h 0.94082 0.95603 2.0 0.4223v 0.96601 0.97075 1.1 0.4236v 0.99590 0.99535 -0.1 0.1436h 0.99590 0.99535 -0.1 0.1089v 0.99810 0.99734 -0.2 0.1289h 0.99810 0.99734 -0.2 0.11

166v 1.00000 0.98814 -3.2 0.26166h 1.00000 0.98814 -3.2 0.26

183±3v 1.00000 0.99212 -2.1 0.24183±7v 1.00000 0.99212 -2.1 0.24

Passive Microwave Signatures (Ocean vs Land)

TRMM/GPM Retrievals Bayesian Inversion

~10 km

TB observed

TB model #3

TB model #2

TB model #1

Database

Bayes, T. and R. Prices, 1763: An Essay towards solving a problem in the Doctrine of Chance. By the late Rev. Mr. Bayes, communicated by Mr. Price, in a letter to John Canton, M.A. and F.R.S. Philos. Trans. R. Soc. London, 53, 370-418.

Extra-tropical Low seen by GMI

GMI Precipitation vs Surface Radar

Precipitation

• GPM provides outstanding rain estimates from its combination of radars and radiometer

• Radiometer rain retrievals using GPM databases will be available by end of January

• Precipitation retrieval is already quasi- Optimal Estimation. Can the a-priori information be used in a covariance framework?

Under-constraint retrievals given the observations.

Bayesian formulation:

Goddard PROFiling algorithm (GPROF): ECMWF one-dimensional + four-dimensional variational analysis (1D+4D-Var):

- Constrained by observationally generateda-priori database consisting of PR/TMI observations and CRM simulations

- x: microphysical profiles

- Constrained by ECMWF model’s First Guess(FG) and the 1D cloud model

- x: thermodynamic profiles - Only channel 19h, 19v, and 22v used.

weightingmicrophysics profiles thermodynamic profiles

GPROF algorithm and ECMWF 1D+4D-Var

a-priori

Case studies

Data:6619 1D-Var retrievals[60°S, 60°N]

SSM/I Tb vectorthermodynamic profilesmicrophysical profiles

10° by 10° area

Pixel 3

Tb departures (K) 19V 19H 22V

GPROF maximumlikelihood departures -0.191 -1.105 -0.222

ECMWF 1D-Var FG departures -13.212 -23.246 -4.788

ECMWF 1D-Varanalysis departures -0.684 0.603 -1.809

• FG Tb negative too much emission

• Analysis reduces both CWP and RWP

• Tbs match while microphysics has discrepancy

Tb ✔CWP/RWP ✖

Final thoughts

• Data assimilation uses radiances to constrain model forecast to obtain best “state”, while retrievals use ancillary data, including “experience” to invert the radiances for the best “state”

• The two use very different language but are really doing very similar things. They can be merged with some work for the benefit of both.

• Rapid updates from GOES-R (Himawari-8) speak to process more than state. Planning underway for “Cloud and Precipitation Processes Mission”. Need to bridge that gap as well.