Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Progress Towards the Assimilation of Cloud Affected Infrared Radiances at the GMAO

Will McCartyNASA Goddard Space Flight Center

Global Modeling and Assimilation Office

12th JCSDA Technical Review & Science Workshop22 May 2014

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Cloud Height Estimation ErrorsOnly Infrared radiances unaffected by clouds are assimilated

– cloud-free (all channels)– cloudy, but unaffected (only channels sensing above cloud)– Nearly opaque, using a graybody assumption in the forward operator

QC determines the cloud height and screens channels whose weighting functions are sensitive to the retrieved cloud height

In the GSI data assimilation algorithm, a minimum residual method (Eyre and Menzel 1989) is used to determine the cloud height– Assumptions in this method are fundamental sources of error:

• Single layer clouds; Infinitesimally thin, black clouds; treated as fractionally gray

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Forecast Impact of Cloud ContaminationFor clear sky, the misassessment of cloud height can result in

the cloud signals being erroneously projected on the mass (T & q) fields

For cloud-affected observations, the misassessment of cloud height vertically can erroneously project the cloud top temperature signal at the wrong vertical level

The cloud height retrievals were determined to need further refinement. Initially, a verification effort of the retrieval of cloud top pressure (CTP) from AIRS was compared against MODIS v5 retrieved CTP (CO2 Slicing) and CALIPSO lidar-measured CTP

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS CTP vs. MODIS High CTPSource of Contamination

Over-Conservative

Comparing:AIRS Cld Top Pressure (CTP) - ~15 km MODIS MYD06 CTP - 5 km, coincident 3x3- highest CTP reported

1 Jul-30 Sept 2013

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS CTP vs. MODIS High CTPComparing:

AIRS Cld Top Pressure (CTP) - ~15 km MODIS MYD06 CTP - 5 km, coincident 3x3- highest CTP reported

1 Jul-30 Sept 2013

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS CTP vs. MODIS High CTPComparing:

AIRS Cld Top Pressure (CTP) - ~15 km MODIS MYD06 CTP - 5 km, coincident 3x3- highest CTP reported

Considering observations as function of cloud fraction shows biases in confident (opaque) and uncertain (variant) areas of cloudiness

1 Jul-30 Sept 2013

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

MODIS CTP uncertain> 700 hPa and in low fraction

Conservative in areas of low fraction

Apparent bias in areas of high fraction

More low fraction observations by design

0.0 < N < 0.25 #: 279311 0.25 < N < 0.5 #: 112239

0.5 < N < 0.75 #: 85490 0.75 < N < 1.0 #: 138513

AIRS v. MODIS CTP by Fraction

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Compared against L2 CALIPSO-CALIOP determined CTP (1 km, V3-30)

Similar trends, but reduction in lower-troposphere contamination (600-800 hPa)

Includes poles, but separation was considered- No distinct polar signals

Likely sampling issues (spatial and temporal)

1 Jul 2013 – 28 Feb 2014

AIRS CTP vs. CALIPSO Highest CTP

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Forecast Impact of Cloud ContaminationTo quantify these cloud assessment errors as to how they impact

the analysis, two metrics investigated:

• Observation Departures (bias corrected)– Observed minus Forecasted Brightness Temperature (O-F)

• Adjoint-Based Observation Impacts– A 24 hour forecast error projected onto the observations of its initial

analysis using the adjoints of the forecast model and assimilation system

– The measure is a moist energy norm (u, v, T, ps , qv → J/kg)– This method is run routinely in GMAO ops at 0000 UTC – A negative value equates a reduction in error, so NEGATIVE = GOOD

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Impact of AIRS

AIRS is generally larger than any single AMSU-APer radiance, AIRS is significantly less than AMSU-A

– Assumed as redundancy, but can cloud signals get past QC?

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Channel SensitivityT High → Low → Window Water Vapor 4μm T AIRS Jacobians for

temperature and water vapor

The top color bar will serve as reference for the remaining plots

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Impact of AIRS by Channel

Tropospheric observations show consisted normalized impact, but more observations aloft (unaffected by clouds) lead toward greater total impact

T Hi

→ Lo

→ W

inH 2O

v4μ

m

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS CTP vs. MODIS High CTP

Clouds classified into (roughly) ISSCP height classifications as a function of AIRS Cloud Top Pressure

Mid-Level Clouds

700-440 hPa

Low Clouds

1000-700 hPa

High Clouds

440-100 hPa

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – High Clouds

• Negative O-F signal apparent where MODIS/AIRS disagree most towards contamination• Small count relative to total, but show inconsistent impact per observation in this

region

T High→Low→Window Water Vapor

4μm TT High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – High Clouds

• High clouds show that even w/ an AIRS low bias (vertically), impact is generally show observations reduce error

• Channels less effected by these high clouds (< 50) show more neutral impact – effect of the metric

T High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – Mid Clouds

• Cloud signals apparent in lower tropospheric channels • There is an opposite signal in the window (and low-level H2Ov) channels

– Low count – consistent signal. Needs further investigation. Unusual for window channel observations deemed ‘cloudy’ at the mid-level to be assimilated

T High→Low→Window Water Vapor

4μm TT High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – Mid Clouds

• Mid-Level clouds show, again, generally improved impact, though notes of cloud contamination are evident in lower tropospheric channels

• Magnitude of negative impact more pronounced than in high clouds

T High→Low→Window Water Vapor

4μm TT High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – Low Clouds

• Cloud signal again apparent• Positive O-F bias in areas of agreement – potentially a feedback of cloud

contamination having an effect on bias correction• Impact per observation strongly negative in areas of cloud contamination

T High→Low→Window Water Vapor

4μm TT High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Observation Impact – Low Clouds

• Misrepresentation of low clouds relative to MODIS show clear increase in error for channels sensitive to low clouds

• There is a real chance that thin cirrus over small clouds can be a source of error• Magnitude of total degradation due to clouds becomes much more significant w.r.t. overall positive

impact

T High→Low→Window Water Vapor

4μm TT High→Low→Window H2Ov 4μm TT High→Low→Window H2Ov 4μm T

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS ‘Clear’/MODIS Cloudy Observations

• Cloud signal again apparent, and largest magnitude of total degradation• Warm O-F signal near sfc again

T High→ Low → Window H2Ov 4μm TMODIS CTP (hPa)

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

AIRS ‘Clear’/MODIS Cloudy Observations

T High→ Low → Window H2Ov 4μm TMODIS CTP (hPa)

T High→ Low → Window H2Ov 4μm T

• Per observation, the degradation is again apparent

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

• Observations with notable subscale variability generally degrade, and have a negative O-F signal.

• This is using MODIS, which has many of the same limitations as AIRS.

T High → Low → Window H2Ov 4μm TImpact per Ob

O-F

MODIS

CTP (Low – High, hPa)

Heterogeneity in AIRS Observations

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Conclusions/Future Work• Incorporation of CALIPSO datasets was an attempt to overcome issues

with MODIS data– Even though different algorithms, MYD06 (CO2 Slicing/IR Window) and GSI

(MinRes) algorithms build on same assumptions– CALIPSO + GSI Thinning = very limited sample size

• An assimilation study incorporating MODIS CTPs to eliminate potential sources of contamination– In addition to impact assessment, it would help to quantify the effect that cloud

contamination has on variational bias correction (B. Zavodsky)• Incorporation of NCEP updates to CTP retrieval in radiance space• Ultimately, this helps quantify an issue that has been acknowledged

– Various studies have shown improvement by incorporating collocated imagery in QC

– Tom Auligne has done some work with multilevel cloud retrievals in the GSI• This work provides a level of verification necessary to continue efforts to

assimilate cloud-affected IR observations

Global Modeling and Assimilation OfficeGoddard Space Flight CenterNational Aeronautics and Space Administration

Unrelated – CrIS Impacts vs. Apodization