The CrIMSS EDR Algorithm: Characterization, Optimization, and Validation

25
The CrIMSS EDR Algorithm: Characterization, Optimization, and Validation Murty Divakarla 1 , Christopher Barnet 2 , Xu Liu 3 , Degui Gu 4 , Michael Wilson 1 , Susan Kizer 5 , Xiaozhen Xiong 1 , Eric Maddy 6 , Ralph Ferraro 7 , Robert Knuteson 8 , Denise Hagan 4 , Xia-lin Ma 4 , Changyi Tan 1 , Nicholas Nalli 1 , Anthony Reale 7 , Andrew K Mollner 9 , Wenze Yang 10 , Antonia Gambacorta 1 , Michelle Feltz 11 , Flavio Iturbide-Sanchez 1 , Bomin Sun 1 , and Mitch Goldberg 7 1 I.M. Systems Group, Inc., Rockville, Maryland, USA, 2 Science and Technology Corporation, Columbia, Maryland, USA, 3 NASA Langley Research Center, Hampton, Virginia, USA, 4 Northrop Grumman Aerospace Systems, Redondo Beach, California, USA, 5 Science Systems and Applications, Inc., Hampton, Virginia, USA, 6 Science and Technology Corporation, Hampton, Virginia, USA, 7 NOAA Center for Satellite Applications and Research, College Park, Maryland, USA, 8 Space Science and Engineering Center, University of Wisconsin, Madison, Wisconsin, USA, 9 The Aerospace Corporation, El Segundo, California, USA, 10 ESSIC/CICS, University of Maryland, College Park, Maryland, USA, 11 Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, Wisconsin, USA Abstract The Cross-track Infrared Sounder (CrIS) and the Advanced Technology Microwave Sounder (ATMS) instruments aboard the Suomi National Polar-orbiting Partnership satellite provide high-quality hyperspectral infrared and microwave observations to retrieve atmospheric vertical temperature and moisture proles (AVTP and AVMP) and many other environmental data records (EDRs). The ofcial CrIS and ATMS EDR algorithm, together called the Cross-track Infrared and Microwave Sounding Suite (CrIMSS), produces EDR products on an operational basis through the interface data processing segment. The CrIMSS algorithm group is to assess and ensure that operational EDRs meet beta and provisional maturity requirements and are ready for stages 13 validations. This paper presents a summary of algorithm optimization efforts, as well as characterization and validation of the AVTP and AVMP products using the European Centre for Medium-Range Weather Forecasts (ECMWF) analysis, the Atmospheric Infrared Sounder (AIRS) retrievals, and conventional and dedicated radiosonde observations. The global root-mean-square (RMS) differences between the CrIMSS products and the ECMWF show that the AVTP is meeting the requirements for layers 30300 hPa (1.53 K versus 1.5 K) and 300700 hPa (1.28 K versus 1.5 K). Slightly higher RMS difference for the 700 hPa-surface layer (1.78 K versus 1.6 K) is attributable to land and polar proles. The AVMP product is within the requirements for 300600 hPa (26.8% versus 35%) and is close in meeting the requirements for 600 hPa-surface (25.3% versus 20%). After just one year of maturity, the CrIMSS EDR products are quite comparable to the AIRS heritage algorithm products and show readiness for stages 13 validations. 1. Introduction The Suomi National Polar-orbiting Partnership (SNPP) spacecraft with the Cross-track Infrared Sounder (CrIS) and the Advanced Technology Microwave Sounder (ATMS) instruments provides a continuity of capabilities for operational environmental remote sounding for weather, climate, and other environmental applications. The CrIS instrument is a Fourier transform spectrometer instrument with channels in three bands covering longwave (6551095 cm 1 ), midwave (12101750 cm 1 ), and shortwave (21552550 cm 1 ) bands [Lee et al ., 2010]. Each CrIS eld of view (FOV) has a spatial resolution of 14 km at nadir. The CrIS instantaneously observes nine FOVs at a time in a 3 × 3 array known as a eld of regard (FOR). The instrument provides 30 FORs for each scan of observations. Four scan lines constitute a granule, and a full day of data contains approximately 2700 granules of the CrIS observations. The CrIS instrument is similar to other hyperspectral IR sounding instruments, namely, the European meteorological polar-orbiting satellite Meteorological Operational satellites program (MetOp-A and MetOp-B launched in 2006 and 2012, respectively), Infrared Atmospheric Sounding Interferometer (IASI), and the National Aeronautics and Space Administration (NASA) Atmospheric Infrared Sounder (AIRS) aboard the Aqua satellite (launched in 2002). All of these hyperspectral infrared (IR) sounders are accompanied by microwave (MW) sounding instruments to assist in the generation of high-quality geophysical products in scenes with up to 80% cloud cover. The DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4953 PUBLICATION S Journal of Geophysical Research: Atmospheres RESEARCH ARTICLE 10.1002/2013JD020438 Special Section: Suomi NPP Calibration and Validation Scientic Results Key Points: Optimization and Characterization of CrIMSS EDR Algorithm Validation of CrIMSS AVTP and AVMP retrievals Intercomparison of CrIMSS and AIRS Retrievals Correspondence to: M. Divakarla, [email protected] Citation: Divakarla, M., et al. (2014), The CrIMSS EDR Algorithm: Characterization, Optimization, and Validation, J. Geophys. Res. Atmos., 119, 49534977, doi:10.1002/2013JD020438. Received 27 JUN 2013 Accepted 24 DEC 2013 Accepted article online 3 JAN 2014 Published online 29 APR 2014

Transcript of The CrIMSS EDR Algorithm: Characterization, Optimization, and Validation

The CrIMSS EDR Algorithm: Characterization,Optimization, and ValidationMurty Divakarla1, Christopher Barnet2, Xu Liu3, Degui Gu4, Michael Wilson1, Susan Kizer5,Xiaozhen Xiong1, Eric Maddy6, Ralph Ferraro7, Robert Knuteson8, Denise Hagan4, Xia-lin Ma4,Changyi Tan1, Nicholas Nalli1, Anthony Reale7, Andrew KMollner9, Wenze Yang10, Antonia Gambacorta1,Michelle Feltz11, Flavio Iturbide-Sanchez1, Bomin Sun1, and Mitch Goldberg7

1I.M. Systems Group, Inc., Rockville, Maryland, USA, 2Science and Technology Corporation, Columbia, Maryland, USA, 3NASALangley Research Center, Hampton, Virginia, USA, 4Northrop Grumman Aerospace Systems, Redondo Beach, California,USA, 5Science Systems and Applications, Inc., Hampton, Virginia, USA, 6Science and Technology Corporation, Hampton,Virginia, USA, 7NOAA Center for Satellite Applications and Research, College Park, Maryland, USA, 8Space Science andEngineering Center, University of Wisconsin, Madison, Wisconsin, USA, 9The Aerospace Corporation, El Segundo, California,USA, 10ESSIC/CICS, University of Maryland, College Park, Maryland, USA, 11Department of Atmospheric and OceanicSciences, University of Wisconsin, Madison, Wisconsin, USA

Abstract The Cross-track Infrared Sounder (CrIS) and the Advanced Technology Microwave Sounder(ATMS) instruments aboard the Suomi National Polar-orbiting Partnership satellite provide high-qualityhyperspectral infrared and microwave observations to retrieve atmospheric vertical temperature andmoisture profiles (AVTP and AVMP) and many other environmental data records (EDRs). The official CrIS andATMS EDR algorithm, together called the Cross-track Infrared and Microwave Sounding Suite (CrIMSS),produces EDR products on an operational basis through the interface data processing segment. The CrIMSSalgorithmgroup is to assess and ensure that operational EDRsmeet beta and provisional maturity requirementsand are ready for stages 1–3 validations. This paper presents a summary of algorithm optimization efforts, aswell as characterization and validation of the AVTP and AVMP products using the European Centre forMedium-Range Weather Forecasts (ECMWF) analysis, the Atmospheric Infrared Sounder (AIRS) retrievals, andconventional and dedicated radiosonde observations. The global root-mean-square (RMS) differences betweenthe CrIMSS products and the ECMWF show that the AVTP is meeting the requirements for layers 30–300hPa(1.53K versus 1.5K) and 300–700hPa (1.28K versus 1.5K). Slightly higher RMS difference for the 700hPa-surfacelayer (1.78 K versus 1.6 K) is attributable to land and polar profiles. The AVMP product is within the requirementsfor 300–600hPa (26.8% versus 35%) and is close inmeeting the requirements for 600 hPa-surface (25.3% versus20%). After just one year of maturity, the CrIMSS EDR products are quite comparable to the AIRS heritagealgorithm products and show readiness for stages 1–3 validations.

1. Introduction

The Suomi National Polar-orbiting Partnership (SNPP) spacecraft with the Cross-track Infrared Sounder (CrIS)and the Advanced Technology Microwave Sounder (ATMS) instruments provides a continuity of capabilities foroperational environmental remote sounding for weather, climate, and other environmental applications. TheCrIS instrument is a Fourier transform spectrometer instrument with channels in three bands covering longwave(655–1095 cm�1), midwave (1210–1750 cm�1), and shortwave (2155–2550 cm�1) bands [Lee et al., 2010]. EachCrIS field of view (FOV) has a spatial resolution of 14 km at nadir. The CrIS instantaneously observes nine FOVs ata time in a 3×3 array known as a field of regard (FOR). The instrument provides 30 FORs for each scan ofobservations. Four scan lines constitute a granule, and a full day of data contains approximately 2700 granulesof the CrIS observations. The CrIS instrument is similar to other hyperspectral IR sounding instruments, namely,the European meteorological polar-orbiting satellite Meteorological Operational satellites program (MetOp-Aand MetOp-B launched in 2006 and 2012, respectively), Infrared Atmospheric Sounding Interferometer (IASI),and the National Aeronautics and Space Administration (NASA) Atmospheric Infrared Sounder (AIRS) aboardthe Aqua satellite (launched in 2002).

All of these hyperspectral infrared (IR) sounders are accompanied by microwave (MW) sounding instrumentsto assist in the generation of high-quality geophysical products in scenes with up to 80% cloud cover. The

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4953

PUBLICATIONSJournal of Geophysical Research: Atmospheres

RESEARCH ARTICLE10.1002/2013JD020438

Special Section:Suomi NPP Calibration andValidation Scientific Results

Key Points:• Optimization and Characterization ofCrIMSS EDR Algorithm

• Validation of CrIMSS AVTP and AVMPretrievals

• Intercomparison of CrIMSS and AIRSRetrievals

Correspondence to:M. Divakarla,[email protected]

Citation:Divakarla, M., et al. (2014), The CrIMSSEDR Algorithm: Characterization,Optimization, and Validation,J. Geophys. Res. Atmos., 119, 4953–4977,doi:10.1002/2013JD020438.

Received 27 JUN 2013Accepted 24 DEC 2013Accepted article online 3 JAN 2014Published online 29 APR 2014

IASI instrument is accompanied by the 15-channel advanced microwave sounding unit (AMSU-A) and the5-channel microwave humidity sounder (MHS). Aqua-AIRS is accompanied by the AMSU-A instrument. TheATMS instrument that accompanies CrIS has a combination of channels similar to that of AMSU-A and MHS.Details of these instruments and their channel characteristics are described in many publications [Aumannet al., 2003; Lee et al., 2010; Hilton et al., 2012].

Through the Joint Polar Satellite System (JPSS) Preparatory Project, the CrIS and ATMS instrument suite isexpected to be included in J1 and subsequent satellite launches. Algorithms to process CrIS/ATMS ob-servations into sensor data records (SDRs) and subsequently into Cross-track Infrared and MicrowaveSounding Suite (CrIMSS) environmental data records (EDRs) have been developed. Northrop GrummanAerospace Systems (NGAS) has adapted the atmospheric environmental and research (AER) CrIMSS al-gorithm to retrieve the atmospheric vertical temperature profile (AVTP), atmospheric vertical moistureprofile (AVMP), and atmospheric vertical pressure profile (AVPP) products (http://npp.gsfc.nasa.gov/science/sciencedocuments/2013-01/474-00001-04-02_JPSS-CDFCB-X-Vol-IV-Part-2_0123A.pdf). The CrIMSSalgorithm was implemented by Raytheon to produce EDRs on an operational basis through the InterfaceData Processing Segment (IDPS). The algorithm uses a simultaneous physical retrieval technique andperforms a “MW-only” first stage retrieval, which produces AVTP, AVMP, cloud amount and height,surface skin temperature, and surface emissivity in the ATMS frequency ranges. These products are usedas a first guess for a second stage combined “IR +MW” cloud-clearing algorithm to produce improvedEDR products. If the second stage IR +MW retrieval fails to produce a better EDR product, which canhappen under complex cloudy or overcast scene conditions, the first stage MW-only EDR will be used asthe final EDR product.

One of the goals of the calibration/validation (Cal/Val) team constituted by the JPSS program is to evaluate andensure that the operational EDRsmeet beta and provisional maturity requirements and are ready for stages 1, 2,and 3 validations so that users can start utilizing these products for scientific applications. Beta maturity des-ignates an early release product that is minimally validated andmay have significant errors. The beta product ismade available to users to gain familiarity. Provisional maturity indicates that the product is deemed ready foroperational evaluation. Versioning control commences as the product continues to undergo improvements,and user participation in quality assurance is encouraged. Validated maturity occurs when the product perfor-mance is well defined over a range of representative conditions using a small number of independent mea-surements (stage 1), using a widely distributed set of measurements from many locations and time periods(stage 2), and using statistically robust independent measurements representing global conditions (stage 3). Amore detailed discussion on the product maturity and validation strategies is documented by Barnet (http://www.star.nesdis.noaa.gov/jpss/documents/CalVal/CVPEDRCrISBarnetPublicReleaseSeptember2009.pdf, hereaf-ter referred to as CrIMSS-Cal/Val-DOC-2009; http://www.star.nesdis.noaa.gov/jpss/documents/Status/120808crimss_beta_brief.pptx, 130117crimss_provisional_brief.pptx, hereafter referred as CrIMSS-Pro-Brief-2013). To achieve these goals, the National Oceanic and Atmospheric Administration (NOAA) Center for SatelliteApplications and Research (STAR), in collaboration with the Langley Research Center (LaRC) and NGAS, hasdeveloped the infrastructure to evaluate CrIMSS EDR products by emulating various IDPS versions of the CrIMSSEDR algorithms, namely, the Day-1 CrIMSS EDR algorithm (IDPS-MX5.3, hereafter referred as MX5.3) to the latestversion that is currently in operations (IDPS-MX7.1 as of this manuscript and hereafter referred as MX7.1). Thesepackages were tested by STAR, NGAS, and LaRC through their in-house implementations. Important fixes andimprovements foreseen through research and evaluation were submitted as discrepancy reports to initiate theprocess of code change requests (CCRs) and the development of algorithm change packages (ACPs) for im-plementation into future IDPS builds. One of the key components in our Cal/Val activity during the prelaunchphase was the testing and empirical radiance bias correction of the CrIMSS EDR algorithm using realistic proxydata generated from the current hyperspectral suite of sounders (i.e., IASI/MHS/AMSU-A andAIRS/AMSU-A). Thisenabled us to prepare all necessary tools and methodology needed to efficiently evaluate and tune the CrIMSSoperational EDR algorithm following the SNPP launch. The EDR products from CrIMSS are similar to the currentAqua-AIRS and MetOp-IASI hyperspectral sounder products and provide continuity through SNPP and subse-quent satellite missions. In addition, the CrIMSS EDR provides a new globally evaluated physical approach thatcan be used for many satellite-based hyperspectral atmospheric sounding applications in weather forecasting[Chahine et al., 2006] and other disciplines [Du et al., 2012; Tian et al., 2013]. This paper stands as a forerunner,describing the algorithm optimization leading to stage 1–3 validation and provides users with an insight to the

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4954

uncertainties expected in utilizing the retrieval products. The following sections describe in detail the optimi-zation, characterization, and validation of the CrIMSS EDR as it evolved from prelaunch testing to the currentprovisional product.

2. The CrIMSS EDR Algorithm, IDPS, and Off-line Emulations2.1. The CrIMSS EDR Algorithm

The CrIMSS EDR algorithm consists of seven modules: initialization, preprocessing, ATMS-only retrieval, sceneclassification, joint ATMS/CrIS retrieval, quality control, and postprocessing. Figure 1 is a flow diagramshowing how these modules are logically related.

The initialization module provides “static” data such as a digital elevation map (DEM) and look-up tables(LUTs) that are related to sensor noise, forward model parameters, bias correction, and climatological errorcovariance matrices. The preprocessing module performs functions such as applying the appropriateapodization function for the CrIS spectra, performing precipitation detection, determining land fraction, andcalculating surface pressure based on DEM and numerical weather prediction (NWP) forecast fields. If ATMSdata are available, a two-stage ATMS-only retrieval is performed. The first stage of the ATMS-only retrievalperforms an optimal estimation (OE) retrieval using two global climatology covariance matrices and associ-ated backgrounds (either ocean or land). The second stage retrieval uses more refined covariance matricesand backgrounds based on the retrieved surface skin temperature from the first stage. The output from thismodule includes atmospheric temperature and moisture vertical profiles, surface skin temperature, surfaceemissivities at 22 ATMS channel frequencies, cloud amount, and cloud height. These outputs are then used asthe first guess for the cloud-clearing algorithm that follows. If ATMS data are not available for the current FOR,then NWP forecast fields are used instead. The climatological temperature and moisture constraints are alsotightened by using the covariance derived from NWP forecast models. The scene classification moduleidentifies the cloudiness within the CrIS FOR. Although the CrIMSS EDR algorithm is capable of performingmultiple retrievals based on the number of cloud formations determined, the current baseline is to performonly one retrieval for each CrIS FOR. A cloud clustering algorithm (described in detail in the Cross-trackInfrared Sounder Volume II, Environmental Data Records Algorithm Theoretical Basis Document, http://npp.gsfc.nasa.gov/science/sciencedocuments/ATBD_122011/474-00056_Rev-Baseline.pdf, hereafter referred asCrIMSS-ATBD) is used to classify a scene (nine FOVS within a FOR) as clear, cloudy, or partly cloudy. If the CrISFOR is clear, the radiance spectra are averaged, and an OE retrieval is performed using the noise-reduced CrISradiances and the ATMS radiances. If the scene is overcast within the CrIS FOR or there is no valid CrIS spectralchannel, the ATMS-only retrievals are then used as the final EDR product. The joint ATMS and CrIS retrieval

Figure 1. The CrIMSS EDR algorithm flow showing the seven algorithm modules.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4955

module performs a cloud clearing for those FORs identified as partly cloudy by the scene classificationmodule. The details of this module will be discussed later in this section. The quality control (QC) moduledetermines the overall QC flags calculated at various stages of the CrIMSS algorithm, and these parametersare saved together with the EDR outputs. One of the most important QC flags is the normalized chi-squarevalue (χ2). The chi-square value acts as a gauge of how well the radiances calculated by the CrIMSS forwardmodel agree with the observed ATMS and CrIS radiances relative to the instrument noise and forward modelerrors. The normalized chi-square equation is as follows:

χ2 ¼ ∑nchan

i¼1

Rcalci � Robsi

� �2Ni

= nchan (1)

where nchan is the number of channels,Rcalci is the calculated radiance,Robsi is the observed radiance, and Ni isthe noise variance and forward model error from the ATMS and CrIS sensors [Zavyalov et al., 2013].

The postprocessing module performs the atmospheric vertical pressure profile EDR calculation using thehydrostatic equation and the retrieved temperature and moisture profiles. It also performs vertical averagingconverting 101 level intermediate CrIMSS AVTP and AVMP into 42-layer AVTP and 22-layer AVMP operationalEDR products.

One of the unique characteristics of the CrIMSS EDR algorithm is that it uses an OE method to retrieve atmo-spheric profiles and surface properties simultaneously, thereby avoiding the necessity to estimate the errorsdue to partially fixed parameters when retrieving a particular parameter in a sequential retrieval approach. Sincethe radiance transfer equation is highly nonlinear, the following iterative equation is used:

Xnþ1 � Xa ¼ KTS�1R K þ S�1

x

� ��1KTS�1

R Robs � Rcalc� �þ K Xn � Xað Þ� �

(2)

where the subscripts n and a represent iteration number and a priori, respectively, X represents the state vec-tor that includes atmospheric temperature, moisture, and ozone profiles transformed into the empirical or-thogonal function domain. It also includes surface skin temperature, MW and IR surface emissivity, IRsurface reflectivity, cloud top pressure, cloud thickness, and cloud amount. Xn+1 represents the iterated statevector, K is the Jacobian matrix, Robs are the measured ATMS and the cloud-cleared CrIS radiances, and Rcalcn

are the forward model calculated radiances using the state vector obtained from the nth iteration Xn. Sx is theerror covariance matrix-associated background state vector (Xa), and SR is the error covariance matrix associ-ated with ATMS and CrIS instrument noise and errors due to the forward model. To further handle thenonlinearity of the inverse problem, a D-rad method is used [Lynch et al., 2009]. If the radiance differencebetween the observation and calculation for a particular spectral channel does not agree to the expectedvalue within instrument noise plus forward model error, then an additional error term is added to the corre-sponding diagonal element of the SRmatrix. The magnitude of the error term is proportional to the square oftheRcalcn � Robs term. As the number of iterations increases, the additional term quickly vanishes, and the finalsolution is obtained. The forward model used for the CrIMSS algorithm is the optimal spectral sampling (OSS)technique [Moncet et al., 2008]. Being a monochromatic method, the OSS technique has the advantage overother fast forward model parameterizations in providing the required Jacobians (i.e., the Kmatrix in equation(2)) analytically, with little extra computation time. The details of Jacobian calculations are described in theCrIMSS-ATBD.

The cloud-clearing algorithm used by the CrIMSS EDR algorithm is similar to that adopted by the AIRS science

team (AST) level 2 algorithm [Susskind et al., 2003; Barnet et al., 2005]. The cloud-cleared radiance R̂i;clr forchannel i can be expressed as a linear combination of the measured radiances of the nine CrIS FOVs:

R̂i;clr ¼ Ri;1 þ η1hRi;1 � Ri;Kþ1

iþ : : : þ ηk

hRi;1 � Ri; Kþ2ð Þ� k

iþ : : : þ ηK

hRi;1 � Ri;2

i(3)

where ηk are unknown channel-independent constants, and Ri;k are the measured radiances for the kthFOV. This equation can be solved iteratively by calculating an estimate for R̂i;clr using the ATMS-only re-trieval output as the atmospheric and surface estimate for the CrIS forward model. The values for ηk areobtained by selectively choosing those CrIS channels which have maximum sensitivity toward clouds

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4956

and minimum sensitivity toward surface emissivity. Equation (3) is then used to calculate the cloud-clearedradiances for the whole CrIS spectrum. Once the improved state vectors are obtained using the combinedATMS and cloud-cleared CrIS radiances, the CrIMSS algorithm iteratively improves cloud-clearing parame-ters and state vectors using both equations (2) and (3).

Apart from the CrIMSS IDPS operational algorithm, a suite of retrieval algorithms are in operational use toproduce a variety of EDR products from Aqua-AIRS and AMSU-A instruments [Aumann et al., 2003]. These re-trieval algorithms, hereafter referred to as the AIRS science team algorithms, use the first guess retrieval as aninitial solution by the final IR retrieval module to produce a final retrieval of surface parameters and profiles oftemperature, water vapor, ozone, and other trace gases through an iterated regularized least squares minimi-zation [Susskind et al., 2003]. The AIRS level 2 ATBD (Barnet et al., http://disc.sci.gsfc.nasa.gov/AIRS/documen-tation/20070301_L2_ATBD_signed.pdf) provides a complete description of the standard and support productsavailable from AIRS/AMSU-A instruments. Until recently [i.e., up to AST version V5, Maddy and Barnet, 2008;Susskind et al., 2011], the first guess retrieval was based on a fast eigenvector regression solution using theEuropean Centre for Medium-Range Weather Forecasts analysis fields as the training data set [Goldberg et al.,2003]. The current AST version V6 (AST-V6) is similar to the AST-V5 algorithm [Susskind et al., 2011] with manymajor exceptions, namely, (1) replacement of the first eigenvector cloudy and clear regressions with a two-stepstochastic cloud-clearing and neural network (W. Blackwell, AIRS/AMSU Atmospheric Profile Retrievals UsingStochastic Cloud Clearing and a Neural Network, submitted to IEEE Transactions on Geoscience and RemoteSensing, 2013), (2) surface parameter retrievals using shortwave channels, and (3) use of MODIS climatology toinitiate surface emissivity [Seemann et al., 2008].

NOAA/STAR, with its long-term involvement with the AST, has developed infrastructure for the development,improvement, and validation of the Aqua/AIRS core and research product algorithms [Chahine et al., 2006;Divakarla et al., 2006, 2008; Nalli et al., 2011, 2013;Maddy et al., 2012; Reale et al., 2012]. The AST algorithmwas alsoadapted at STAR to produce MetOp-IASI/AMSU-A/MHS EDRs [Maddy et al., 2011] (A. Gambacorta, “IASI RetrievalATBD,” http://www.star.nesdis.noaa.gov/smcd/spb/iosspdt/qadocs/IASI_Phase2/IASI_ATBD_Phase_II_20111115.pdf) andNOAAUnique CrIS/ATMS EDR products [Gambacorta et al., 2012] (A. Gambacorta, The NOAAUnique CrIS/ATMS Processing System, NUCAPS, manuscript in preparation for submission to Bulletin of the AmericanMeteorological Society, 2014).

In addition to the EDR products from hyperspectral IR sounder instruments, STAR also operates SDR and EDRproduct processing systems specifically tailored for MW-only instruments. The Microwave Surface andPrecipitation Processing System (MSPPS) [Ferraro et al., 2000, 2005] and the Microwave Integrated RetrievalSystem (MiRS) [Boukabara et al., 2011] provide a wide variety of MW-only EDR products. These MW-onlyproducts are useful to compare with the first stage MW-only CrIMSS EDR products for a mutual benefit ofalgorithm improvements.

2.2. IDPS Operations and Off-line Emulations

In 2004, NGAS delivered the AER Fortran 77/90 science code to the SNPP IDPS. Over the period from 2004 to2010, several CrIMSS modifications were carried out to enhance the algorithm, ultimately resulting in theCrIMSS portion of the IDPS operational MX5.3 product. The data are routed to the Government Resource forAlgorithm Verification, Independent Testing, and Evaluation (GRAVITE), which plays a key role in the pathwayto operational implementation of CrIMSS software updates.

Following the launch of SNPP, several algorithm updates were made that can be tied to IDPS versions MX5.3,MX6.3, MX6.6, and MX7.1. All of the LUT updates and software changes were tested and verified in off-lineversions of the science code, in parallel with updates made in the GRAVITE Algorithm Development Area(G-ADA) and the Algorithm Development Library (ADL) prior to delivery to IDPS. The G-ADA is an instance of theIDPS AIX environment, while the ADL provides an algorithm testing and proving environment that emulatesthe programming interfaces available in the IDPS. In addition to the off-line versions of the science code and theADL packages, a Linux version of IDPS operational code was developed by NASA-LaRC [Kizer et al., 2010]. Thisported version of the code removes the reliance on large input databases and extensive code dedicated toSNPP instruments unrelated to CrIMSS. It can therefore quickly process global runs with the flexibility to ingestvarious test data. This ported version has been implemented on various Linux systems at STAR, LaRC, and NASAand was extensively used to test the algorithm’s performance and develop enhancements.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4957

3.0. Data Sets for EDR Product Evaluation

Ahierarchy of data sets as described by Barnet (CrIMSS-Cal/Val-DOC-2009) andNalli et al. [2013] are being used toevaluate CrIMSS EDR products. A global evaluation of EDR products provides an implicit evaluation of cloud-cleared radiances, forward model functioning, and cloud detection capabilities inherent in the CrIMSS EDR al-gorithm. The EDR evaluation thus begins with a global evaluation of CrIMSS EDR products using matched modelforecast/analysis fields that are available globally. The ECMWF analysis fields were used as proxy to truth mea-surements. The National Center for Environmental Predication-Global Forecast System (NCEP-GFS) forecast/analysis fields were used to define surface pressure and other ancillary information needed to run off-lineretrieval algorithms. In addition, matched EDR products from other hyperspectral instruments (Aqua-AIRS andMetOp-IASI) offered a baseline to evaluate global and seasonal patterns generated from CrIMSS EDRs. NOAA-STAR has all of the infrastructure in place to generatematched ECMWF/NCEP-GFS, global radiosonde observation(RAOB) network of measurements, and other correlative data sets through Aqua-AIRS and MetOp-IASI valida-tions [Divakarla et al., 2006, 2008, 2011a; Reale et al., 2012]. The CrIMSS EDR product evaluation was carriedout by adapting the existing infrastructure to the SNPP satellite. A total of four focus days (Table 1) wereselected based on maximum overlap between SNPP and Aqua-AIRS orbits, and matched correlative datasets were generated. Since both EDR products and model forecast/analysis fields are global in nature, atypical day fetches a large number of collocated matches (~324,000/d) and allows assessment of EDRproduct maturity with sufficient sample sizes for different latitude zones (tropics, midlatitudes, and highlatitudes) and for land-sea and day-night categories. The availability of an array of EDR products (CrIMSS,Aqua-AST, and STAR-NUCAPS (NOAA Unique CrIS ATMS Processing System)) with matched truth data setsallowed for an intercomparison and evaluation of EDR products from two different algorithms, namely, theCrIMSS EDR algorithm and the AST-V5/V6.

As a follow up to the global evaluation with model analysis fields, the global RAOB network offers a sizablenumber of collocations when acquired over a long period of time to initiate stages 1–3 validations [Fetzeret al., 2003; Divakarla et al., 2006] (Barnet, CrIMSS-Cal/Val-DOC-2009; Nalli et al. [2013]). Typically, 200–300operational radiosonde observations (RAOBs) are colocated within ±3 h and 100 km radius on a given day.Depending on the satellite orbit characteristics, time and distance collocation criteria (±3 h, 100 km radius),the RAOB-matched data sets may have a skewed geographical distribution and may be representative ofhighly sampled geographic regions and the instrument types used in those RAOBs [Divakarla et al., 2006]. Inthis paper, the utility of RAOB data was demonstrated using a small set of global RAOB matches with theCrIMSS EDR product. The RAOB data set was obtained from the NOAA Products Validation System (NPROVS)operated at NOAA/STAR [Reale et al., 2012].

The third source of truth for the validation of the AVTP product is the Global Positioning System radiooccultation (GPSRO) profiles from the Constellation Observing System for Meteorology, Ionosphere, andClimate (COSMIC) network [Anthes et al., 2008; Anthes, 2011]. The COSMIC dry temperature has an accuracyand structure uncertainty of <0.2 K in the pressure range 200 hPa to 20 hPa. [Ho et al., 2010; Steiner et al.,2013]. The global sampling and long-term stability of the GPSRO observations make it a good candidate for aJPSS EDR validation reference. This data set also provides about 200 matches a day and can offer a largenumber of collocations when collected over a period of time to further verify the inferences made fromglobal data evaluations using the first two data sets. The GPSRO matches for one of the focus days (15 May2012) were utilized to evaluate CrIMSS AVTP EDR products.

Table 1. Focus Day Correlative Data Sets Used to Evaluate the CrIMSS EDR Products

Focus Days and RAOBs IDPS Versions Off-line emulations Correlative Data Sets

15 May 2012 MX5.3 ECMWF/GFS (NOAA-STAR)20 Sep 2012 MX5.3 Aqua-AIRS V6 SDR/EDRs (JPL, NASA)3 Feb 2013 MX6.3 IASI SDR/EDR Products (NOAA-STAR)12 Mar 2013 MX6.6 NUCAPS SDR/EDR Products (NOAA-STAR)

MX5.3, MX6.3, MX7.1 GPSRO Measurements (SSEC)

Global RAOB Matches MX6.6 NPROVS

Dedicated RAOB Matches MX5.3 Aerospace Corporation PMRF RAOBs,Kauai, Hawaii, May, September, 2012

ARM/CART Sites

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4958

The evaluation process is strengthened further through acquisition of measurements with increasing rel-evance to validation. As part of the CrIMSS Cal/Val plan, the JPSS has planned a large number of dedicatedRAOB ascents from the Atmospheric Radiation Measurement Cloud and Radiation Testbed (ARM/CART)sites, intensive field campaigns, and campaigns of opportunity (Barnet, CrIMSS-Cal/Val-DOC-2009;Nalli et al. [2013]). At the time of writing this paper, some of these data sets have been acquired(see Table 3 in Nalli et al. [2013]) and are currently undergoing analysis to generate best estimates as was doneby Tobin et al. [2006]. As part of this effort, the Aerospace Corporation supported the launching of 40 VaisalaRS92 dedicated radiosondes from the Pacific Missile Range Facility (PMRF) in Barking Sands, Hawaii, duringMayand September 2012. The CrIMSS EDR matches with the ECMWF, and the dedicated RAOBs over the PMRF sitewere analyzed to discuss preliminary validation efforts.

Table 1 provides a list of focus days and the data sets collocated to evaluate CrIMSS EDR products. These datasets are available for download at ftp://ftp.star.nesdis.noaa.gov/pub/smcd/spb/ctan/CrIMSS_ALGORITHM.

3.1. EDR Products and Statistical Metrics

The CrIMSS EDR algorithm produces the 42-layer AVTP product, the 22-layer AVMP product, and the 32-layerAVPP product. In addition to these, the product stream produces many intermediate products (IPs) and in-cludes a 101-level ozone profile and 101-level products for AVTP and AVMP derived from both the first stageMW-only step as well as the second stage combined IR +MW retrieval product, cloud-cleared radiances, andinfrared emissivities. The AVPP is a derived product from the AVTP and AVMP products. The NWP modelsurface pressure is used as a boundary term, and the hydrostatic equation is integrated to obtain pressuresat the reporting altitudes (see the CrIMSS-ATBD for details). If the second stage IR+MW retrieval fails toproduce a better EDR product, the first stage MW-only EDR will be reported as the final EDR product in theAVTP and the AVMP. Both the first stage and the second stage AVTP and AVMP products were evaluatedwith matched ECMWF analysis fields for the focus day data sets, with RAOBs for the global RAOB collocateddata sets and with the Aerospace Corporation dedicated RAOB ascents from the PMRF-matched data sets.Aqua-AIRS AST-V6 EDR product statistics were also computed for the same ensemble for the final physicalretrieval [Susskind et al., 2003] that correspond to the second stage CrIMSS IR+MW reported product. TheECMWF/RAOBs were taken as the truth and bias, and the root-mean-square (RMS) differences between thetruth and the AVTP and AVMP products were computed for IR +MW reported retrievals and for theremainder of the MW-only reported retrievals (excluding precipitation cases and overcast cases rejected byboth the first stage and the second stage). Temperature statistics were derived for 1 km layers for 1000hPato 0.01hPa. For water vapor, statistics were computed for column densities converted to integrated columnwater. The water vapor bias was computed as a percentage of the reference (ECMWF or RAOB) watervapor amount in the layer (100X (RET-REF)/REF), where RET stands for the AVMP retrieved product and REFstands for either the ECMWF or the RAOB. The percent error was computed by weighting the RMS dif-ference with the reference (truth) water vapor amount in the layer. Details on the statistical methodologycan be found in Nalli et al. [2013]. The procedure was repeated for different EDR versions (MX5.3, MX6.3,and MX7.1) from off-line emulations and for the IDPS operational product version operated at the time ofthe focus day (MX5.3 for 15 May 2012 and 20 September 2012 and MX6.3 for 3 February 2013). Statisticswere computed for the whole globe for three categories, namely, “all” that includes all the global samples(land/sea/coast), “land”, and “sea.” Statistics were also computed for tropical, midlatitude, and high latitudecases to characterize the retrieval product performance for different regions of the globe.

4. Realization of IDPS MX7.1

The MX7.1 IDPS version that is currently in operations is the product of many improvements andupdates that date back many years. During the period from 2004 to 2010, several CrIMSS codemodifications were carried out to enhance the operational algorithm capability. These modificationsinclude retrieving an ozone profile, making the code robust to potential detector failure in CrIS, usingNWP as a substitute for missing ATMS data, checking for precipitation, refining radiative transfer modelerrors, changing cloud cover logic, adding quality flags, creating more diversified climatologies, andimproving the look-up tables for the OSS forward models. These efforts and subsequent efforts duringthe prelaunch phase with simulated data sets [Gu and Ma, 2010] and with “the CrIS/ATMS focus dayproxy data package” [Barnet et al., ftp://www.star.nesdis.noaa.gov/pub/smcd/spb/murtyd/jpss_reports/

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4959

IPO_Proxy_Rel1p0.pdf; Divakarla et al., 2010a, 2010b] helped to tailor a “Day-1” MX5.3 EDR algorithm buildfor IDPS operations. In the postlaunch phase, the MX5.3 to MX6.3 changes included updates to the infraredand microwave bias LUTs and climatology LUT. The MX6.3 to MX6.6 updates were primarily for softwarechanges with the biggest impact related to a channel indexing error that affected the ozone spectral regionand degraded daytime profile retrievals. Modifications of the infrared noise LUTs as well as several additionalcode changes were made between versions MX6.6 and MX7.1 to optimize parameters used for cloud clearingand the profile retrievals, eventually leading to provisional maturity assessment of MX7.1 [Divakarla et al.,2013] (Barnet, CrIMSS-Pro-Brief-2013). Table 2 shows a complete list of the updates implemented post SNPPlaunch. All of these changes were realized through many prelaunch evaluations with realistic proxy data setsand postlaunch observations, as discussed in the next two subsections.

4.1. Prelaunch to Postlaunch Transition

During the prelaunch phase of the SNPP launch, the CrIMSS EDR algorithm performance was charac-terized by using the prelaunch LUTs in the CrIMSS off-line science code using the CrIS/ATMS proxy datasets generated from IASI/AMSU/MHS observations [Liu et al., 2010, 2012; Gu et al., 2011; Divakarla et al.,2011b, 2011c]. The prelaunch focus day proxy data package has matched MetOp (9:30 A.M./P.M.) IASI/AMSU-A/MHS global SDRs, IASI retrieval products (EDRs) generated at NOAA/STAR, matched NCEP-GFS,and ECMWF analysis fields, and the operational RAOB network. The CrIS and ATMS proxy SDRs werecreated using the algorithms developed by Liu and Kizer [2009a, 2009b] and Jairam [2009], respectively.The ATMS proxy data algorithm uses a regression method to derive ATMS SDRs by taking into accountboth the spectral and spatial differences between the ATMS and AMSU/MHS instruments. The CrIS proxydata algorithm utilizes a double fast Fourier transform transformation of IASI radiance spectra togenerate CrIS radiances. The generated proxy data were used to develop empirical bias LUTs for the CrISand ATMS SDRs to use in the Day-1 algorithm. The proxy SDRs were also ingested into the CrIMSSoperational code to generate CrIMSS EDR products.

The results of the prelaunch evaluation of the CrIS/ATMS proxy SDRs as well as the EDR products werediscussed in a series of presentations at various conferences as well as Sounding Atmosphere Teammeetings [Liu and Kizer, 2009a, 2009b; Kizer et al., 2010; Divakarla et al., 2010a, 2010b; Gu et al., 2011;Divakarla et al., 2011b, 2012]. Figure 2a shows a comparison of the CrIMSS 500 hPa temperature fields withthe corresponding ECMWF temperature fields. Figure 2b shows the RMS differences for proxy CrIMSS AVTPand AVMP products using the ECMWF as the reference. Also shown in Figure 2b is the correspondingIASI-derived AVTP and AVMP product RMS differences with ECMWF. These figures demonstrate apromising CrIMSS prelaunch performance in comparison to other state-of-the-art algorithms. For ex-ample, the RMS difference for the 500 hPa temperature fields shown in Figure 2a is within 1.0 K. Overall,the proxy data were found to be very useful for prelaunch bias characterization, algorithm tuning, andEDR performance evaluation. Empirical bias-tuning procedures developed with proxy SDRs provided“near-to-real” estimates and saved a lot of time in realizing the postlaunch bias-tuning efforts with realCrIS/ATMS observations. The prelaunch EDR product evaluation has helped to identify areas of defi-ciencies in both forward and inverse models, test robustness of the operational code, and check

Table 2. List of Changes for the CrIMSS EDR Algorithm Leading to IDPS MX7.1

Date System Changes

Apr 2012 MX 5.3 • Official beta system. Included bias tuning for CrIS, but not ATMS.

Oct 2012 MX 6.3 • Changed lookup table for ATMS empirical bias corrections.• Updated look-up tables for the CrIS empirical bias corrections.• Updated climatology look-up table.

Feb 2013 MX 6.6 • Fixed CrIS channel indexing that impacts daytime and ozone IP products.

Jul 2013 MX 7.1 • Bug fix for atmospheric noise and sensor noise.• More conservative clear scene detection.• Updated CrIS sensor noise and forward-model LUTs.• Loosened surface constraint for daytime land scenes.• Loosened selection criteria to choose warm ocean climatology instead of cool ocean.• Optimized chi-square threshold parameters to allow higher combined-run yields

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4960

Figure 2. (a) Prelaunch evaluation of the CrIMSS EDR products with focus day proxy data package: retrieved 500 hPa tem-perature map from the CrIMSS AVTP product using (top) the proxy CrIS/ATMS SDRs from 19 October 2007, compared to(bottom) the ECMWF. (b) Prelaunch evaluation of CrIMSS EDR products with focus day proxy data package: (left) the AVTPand (right) AVMP RMS differences with reference to ECMWF for the second stage IR +MW product (solid red) and thecorresponding MW-only product (solid green). Also shown in the figure are the NOAA IASI retrieval system-derived finalphysical retrieval (dashed red) and the MW retrieval (dashed green). The IASI system EDRs were generated with real IASI/AMSU-A/MHS observations, while the CrIMSS EDR algorithm used proxy CrIS/ATMS SDRs.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4961

error-handling capability of the operational code. The optimization of the CrIMSS EDR algorithm with theprelaunch exercises helped for a launch ready performance. With the availability of three hyperspectralIR and MW sounder instruments (CrIS, IASI, and AIRS) and reliable proxy data generation algorithms, thisprocess can be extended to JPSS-J1 and subsequent satellite instrument specifications.

4.2. Postlaunch Updates and Evaluation

One of the major updates in optimizing the algorithm for the postlaunch is the CrIS/ATMS empirical bias-correction (also known as tuning) LUTs. This important update accounts for the differences between theobserved radiances and the forward model calculations used in the retrieval algorithm. Errors in the forwardmodel calculations may be due to incorrect physics, sensitivity to trace gases, surface emissivity uncertainties,and other factors. The observed radiances from the satellite instrument may also have aberrations due toscan-dependent antenna patterns (for the ATMS instrument), changes in instrument optical properties, andother factors. Biases computed between the observed and calculated radiances/brightness temperaturesconstitute the tuning to be applied to the observed radiances/brightness temperatures. Figure 3 shows thepreliminary results of scan position-dependent bias values for each of the 22 ATMS channels. The method-ology used in deriving the bias corrections is described by Gambacorta et al. [2012] (A. Gambacorta et al., “Amethodology for computing systematic biases of the top of atmosphere brightness temperature calculations,Part II: Microwave brightness temperature computations. A case study using the ATMS,” manuscript in sub-mission for IEEE Transactions on Geoscience and Remote Sensing, 2013). Generally, the biases for 50 GHz oxy-gen channels 5–15 are smaller than +/�1 K except for channels 3 and 4, which are more sensitive topolarization and uncertainties in surface properties. The five water vapor channels, 18–22, have biases lessthan 1.5 K with error decreasing as the channel frequency approaches the center of the 183GHz water band.Channels 1, 2, 16, and 17 are very sensitive to the surface emissivity, polarization, and surface skin temper-ature uncertainties and therefore have larger biases.

Figure 4 shows the bias values in the three CrIS spectral bands. The CrIS bias correction is done differently fromthat of the ATMS. It is calculated as the ratio of the observed CrIS radiance to that of the calculated. The empiricalbias tuning is mainly to correct the forward model error, and as such, represents uncertainty in the spectroscopy.We have done some trade studies between bias tuning (using observed minus calculated radiances) and theratio factor. However, the difference in EDR performance is similar between these two methods of bias tuningsince the correction is small (i.e., for small corrections, the finite difference is too small to matter). We chose to usethe tuning in the form of a ratio because it might be more accurate in representing the errors in absorption than

Figure 3. Postlaunch scan-dependent empirical bias correction values for the ATMS channels.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4962

the radiance difference method. Generally, the ratio is close to one for most spectral channels. In addition toforward model errors, some of the deviations from the ideal value of one are caused by incorrect prescription offixed trace gas amounts (CO2, CFCs, CH4, and N2O) used in the CrIS forward model.

MX6.3, MX6.6, and MX7.1 mark the changes to the operational EDR algorithm, with MX7.1 upgrading thealgorithm to provisional status (Barnet, CrIMSS-Pro-Brief-2013) [Divakarla et al., 2013]. These changes areshown below in Table 2. Table 3 shows the impact of these changes on yield for all of the available granulesfor four specific focus days: 15 May 2012, 20 September 2012, 3 February 2013, and 12 March 2013. Theretrieval yields are defined as the percentage of profiles passing the chi-square criteria in equation (1). TheMW-only yields show the percentage of profiles passing the chi-square criteria after the MW-only stage of thealgorithm. Combined yields show the percentage of profiles passing the chi-square criteria after the com-bined IR +MW stage of the algorithm. The IDPS operational system as well as the off-line emulation packagecategorize retrievals as (a) “high quality” for samples meeting the second stage combined IR +MW retrievalconvergence, (b) “low-quality” retrievals for the remainder of the retrievals (excluding high-quality retrievals)that pass the first stage MW-only convergence criteria, and (c) “poor quality,” where the retrieval fails to passthe first stage MW-only convergence criteria. The table also shows the percentage yield for MX7.1 when thegranules are separated into ocean and “land+ coast” categories. Results from the off-line emulations were

Figure 4. Postlaunch empirical bias corrections for the CrIS radiances.

Table 3. Yields for Three Versions of the CrIMSS EDR Algorithm on Four Different Daysa

MX Version Date High-Quality Yield (%) Low-Quality Yield (%) Poor Yield (%) Overall Yield (%)

MX5.3 15 May 2012 4.8 66.3 28.9 71.120 Sep 2012 4.2 70.7 25.1 74.93 Feb 2013 3.4 61.7 34.9 65.112 Mar 2013 3.2 60.7 36.1 63.9

MX6.3 15 May 2012 22.1 66.5 11.4 88.620 Sep 2012 20.2 68.2 11.6 88.43 Feb 2013 19.9 65.2 14.9 85.112 Mar 2013 19.9 65.9 14.2 85.8

MX7.1 Global ALL 15 May 2012 47.4 43.5 9.1 90.920 Sep 2012 47.6 43.6 8.8 91.23 Feb 2013 47.7 40.2 12.1 87.912 Mar 2013 49.7 38.6 11.7 88.3

MX7.1 Global Ocean 15 May 2012 47.9 44.3 7.8 92.220 Sep 2012 47.1 45.6 7.3 92.73 Feb 2013 51.1 41.8 7.1 92.912 Mar 2013 51.6 40.5 7.9 92.1

MX7.1 Global Land and Coast 15 May 2012 46.5 42.1 11.4 88.620 Sep 2012 48.5 40.2 11.3 88.73 Feb 2013 42.0 37.4 20.6 79.412 Mar 2013 45.3 37.6 17.1 82.9

aHigh-quality yields pass the second stage IR +MW criteria. Low-quality yields fail the second stage, but pass the first stageMW-only criteria. Poor yields fail boththe stage criteria. Overall yields are the sum of the high- and low-quality yields.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4963

functionally verified with synchronized IDPS operational code emulations using ADL-4.1 to facilitate the CCRprocess. The AVTP, AVMP, and other intermediate products generated for various emulations and IDPS op-erational products downloaded from the GRAVITE archives were all verified for consistency. The combinedIR +MW run is the desired result, and the percentage of profiles passing this yield have increased substan-tially with each delivery. In situations where the IR +MW run fails, the MW-only run is the next desired result.Finally, a poor yield means that the algorithm failed to provide any good retrieval. This number has decreasedwith each version, showing overall improvements in the algorithm. The changes to the algorithm also posi-tively impact performance, as shown in section 5.

5. Results and Discussion5.1. Global Evaluation With the ECMWF Analysis and the Heritage Algorithm Products

The AVTP and AVMP product evaluations presented in this section are based on the MX7.1 off-line emulations.The CrIMSS EDR algorithm was evaluated with several focus days. The results of focus day evaluationsperformed at STAR and at NASA-LaRC with an off-line emulation for different IDPS versions (MX5.3, MX6.3, andMX7.1) were verified with reprocessed runs generated in the G-ADA (NGAS) environment and in the ADL4.1environment at STAR. Thus, the off-line results presented for MX7.1 should replicate the expected results for anytypical day from the IDPS operations. The majority of evaluations presented here are for the focus day 15 May2012. The results of the evaluation for other focus days show very similar results as tabulated in Table 3.

Figures 5a–5f show 475–535 hPa mean layer temperature retrieval maps generated from the CrIMSS AVTPproducts and other correlative data sets. The second stage IR +MW, the first stage MW-only product, and (toprow) the matched ECMWF analysis fields for SNPP descending orbits and (bottom row) the correspondingretrieval products from AST-V6 Aqua descending orbits are depicted. Figures 6a–6f show the total precipi-table water map derived from the AVMP EDR product and other correlative data sets as like Figure 5.Although both CrIMSS and Aqua orbits overlap, deviations to Aqua and SNPP orbits over a day necessitatedgeneration of ECMWF matches separately for both SNPP CrIMSS and Aqua AIRS retrievals. Except for minordisparities in precipitable water fields (dry bias as shown in the AVMP statistics plots for SNPP MW-only re-trievals discussed in later portions of the text), these figures reveal that the CrIMSS AVTP and AVMP products

Figure 5. The global comparison of the CrIMSS 475–535hPa mean layer temperature retrieval with ECMWF and Aqua-AIRS/AMSU heritage algorithm retrieval: (a)The CrIMSS second stage IR +MW retrieval, (b) the “ATMS-only” retrieval, (c) the corresponding ECMWF analysis, (d) the Aqua-AIRS retrieval, (e) the Aqua-AMSUretrieval, and (f) The corresponding ECMWF analysis.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4964

depict global patterns in a consistent manner with the ECMWF analysis fields and AST-V6 retrievals. Similarresults occur for ascending orbits and other focus days (not shown). Figure 7 shows the global RMS differencesfor the AVTP and AVMP second stage IR+MW products with reference to ECMWF for the MX5.3, MX6.3 (oper-ated until February 2013), and MX7.1 emulations. Minor differences in yield between the tables presentedearlier (Table 3) and shown in the figures are due to the missing ECMWF data matches for some granules.Table 4 shows the AVTP product statistics (RMS differences with respect to truth data sets) for 30–300hPa, 300–700hPa, and 700 sfc and AVMPproduct statistics for 300–600hPa and 600 sfc. The requirements for these layersare also shown in the bottom row. The requirement is set such that you can put retrieval into either category (atrade-off between good statistics and yield). The requirements also suggest at least 50% in the IR+MW categoryto maximize utilization of the CrIS instrument. Experience with the AIRS and IASI systems is that the best trade-off between yield and performance occurs when ~70% of the cases are IR+MW for a mature and robust cloud-clearing algorithm. The MX5.3 (beta maturity) did not meet the requirements except slightly around 400hPa.This is not surprising as the system had no ATMS bias tuning in place whenMX5.3 was implemented in the IDPSoperations. The lack of MW tuning caused a lower first stage MW yield (~66.3%; Table 3) and consequentlyimpacted the second stage retrievals. The lack of ATMS bias tuning and other deficiencies that remained ascode/LUT issues (that were fixed in subsequent versions) resulted in a low second stage yield (IR +MW, ~4%).TheMX6.3 versionwas implemented in the IDPS between October 2012 and February 2013 with changes to thealgorithm and LUTs as discussed in Table 2. An examination of the yield (Table 3) and the improvement in RMSdifference from the MX5.3 to MX7.1 reveals that the optimizations (Table 2) implemented in the MX7.1 haveimproved the yield from 4% to 50% without compromising the EDR product precision. Table 4 also shows thatthe MX7.1 has been meeting the AVTP requirements for the layers 30–300hPa (1.53 K versus 1.5 K) and 300–700hPa (1.28 versus 1.5 K) and is close to meeting the requirements for 700 sfc (1.78 K versus 1.6K). Regardingthe AVMP product, the MX7.1 is yet to meet the requirements for 600hPa sfc (25.3% versus 20%).

Figures 8a and 8b show the RMS and bias statistics for the second stage IR +MW (high quality) and the firststage MW-only (low quality) MX7.1 AVTP and AVMP product, respectively. The global yield of the CrIMSS EDRalgorithm is about 90% (high-quality IR +MW 47%+ low-quality MW-only 43%; Table 3). About 10% of thesamples were rejected by the algorithm, out of which 3% of the samples are precipitation-flagged cases. The

Figure 6. The global comparison of total precipitable water (PCW) computed from the CrIMSS AVMP product and other matched data sets as in Figure 5. (a) The CrIMSSsecond stage IR+MW retrieval, (b) the ATMS-only retrieval, (c) the corresponding ECMWF analysis, (d) the Aqua-AIRS retrieval, (e) the Aqua-AMSU retrieval, and (f) thecorresponding ECMWF analysis.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4965

remaining 7% of the cases were due to issues such as overcast scenes, scenes with aged snow or ice, and badscan data (~0.4% on the 15/5 focus day), where the algorithm failed to provide a good retrieval. The pre-cipitation detection algorithm, as implemented in the MX5.3 and continued through the MX7.1, is functionalbut has been found to have issues (discussed in section 5.5). Consequently, the precipitation flag is found tohave false positives and false negatives and affect the overall statistics presented for cloud-cleared cases. Thecold bias seen at the surface can likely be attributed to minor cloud-clearing deficiencies. Larger RMS dif-ferences (Figure 8a) seen above 200 hPa for the combined IR +MW retrieval are due to biases between theretrieval and the truth measurements as depicted in Figure 8b. The larger bias seen between 200 and 100 hPais due to retrievals accepted by MX7.1 predominantly from polar land/coastal regions (it is the largest in thepolar region but also shows up in the tropics and midlatitudes). Preliminary tests show that by increasing thevalues in the sensor noise LUT and the forward model error LUT in the cloud clearing spectral region, we caneliminate this EDR performance degradation in MX7.1. Thus, the improvements discussed in section 6.0 forthe next IDPS release will alleviate this issue. Despite this aberration, the AVTP requirements are met veryclosely for the layers 30–300hPa (1.53 K versus 1.5 K; Table 4). The combined IR+MW retrieval of MX7.1 alsoshows improvement in the AVMP RMS difference for the layers above 300hPa in comparison to the MX6.3.Table 5 shows the RMS differences for themean layers similar to Table 4 but for the first stageMW-only product.A comparison of the RMS differences for different layers to the requirements (bottom row, Table 5) reveal thatthe AVTPMW-only retrieval is meeting the requirements for 30–300hPa and 700-sfc layers. The AVMPMW-onlyretrieval is meeting requirements for the 300–600hPa and has yet to meet requirements for 600-sfc.

Figure 7. The global evaluation of the CrIMSS IR +MW beta and provisional EDR Products: (left) AVTP and (right) AVMP forIDPS MX5.3 (blue), MX6.3 (green), and MX7.1 (red) emulations. Also shown in dashed magenta line is the IDPS MX5.3 op-erational product as verification to off-line emulation (blue). The RMS differences were computed with reference to ECMWFfor the focus day 15 May 2012. Requirements are shown in dotted red lines.

Table 4. AVTP and AVMP Yield and RMS Difference Statistics for Different Layers of the Atmosphere for MX5.3, MX6.3,and MX7.1 With Requirements for the Second Stage IR +MW Combined Retrieval Product

AVTP(p) AVMP(p)

System IR +MW Yield (%) 30–300 hPa (K) 300–700hPa (K) 700 hPa-sfc (K) 300–600 hPa (%) 600hPa-sfc (%)

MX5.3 4 1.86 1.80 2.31 39.6 25.3MX6.3 22 1.40 1.43 1.93 26.6 25.2MX7.1 47.4 1.53 1.28 1.78 26.8 25.3

Requirements ~50 1.5 1.5 1.6 35 20

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4966

Figure 9 shows the AVTP and AVMP statistics for the second stage IR +MW accepted samples separated intoland and ocean categories. The yield over sea and land is similar (~47%) and indicates that there are no majordifferences in yield for different surface types. The EDR algorithm shows better performance over the sea.Larger RMS differences over land are probably due to the heterogeneity of the land surface and the associ-ated spectral emissivities [Salisbury and D’Aria, 1992, 1994] and contribute to the larger RMS differences seenin the global statistics. In addition, daytime convective buildup in the boundary layer and errors in the

Figure 8. (a) The global evaluation of the CrIMSS MX 7.1 retrievals: (left) AVTP and (right) RMS differences with reference toECMWF for the MX7.1 emulation for the second stage IR +MW (red) and first stage MW-only (green) retrievals.Requirements are shown in dotted red lines. (b) The global evaluation of the CrIMSS MX7.1 retrievals: (left) AVTP and (right)AVMP bias with reference to ECMWF for the MX7.1 emulation for the second stage IR +MW (red) and first stage MW-only(green) retrievals. Note: In this figure, as well as Figure 8b, the second stage retrieval refers to profiles which passed the qualitycriteria for the IR +MW run, while the first stage retrieval refers to the remaining profiles that failed the IR +MW criteria butpassed the MW-only criteria.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4967

interpolated surface pressure due to topography might be other contributing factors for larger RMS differ-ences over the land. Another possibility is that the colocation of ECMWF in time (±3 h) and space may havelarger colocation errors due to spatial and diurnal variability over land. To evaluate the ability of the retrievalalgorithm in retrieving contrasting atmospheric states, the global sample was separated into tropical (23°N–23°S), midlatitude (50°N–23°N, 50°S–23°S), and high-latitude (90°N–50°N, 90°S–50°S) regions, and statistics werecomputed with reference to ECMWF matches (Figure 10). The increase in RMS difference with height (above300 hPa) is evident for all the regions, and the trend resembles global RMS differences. The RMS difference forthe tropical cases (with relatively low variability in truth) is smaller with relatively larger yield (55%) comparedto midlatitude (46%) and high-latitude (41%) cases. For midlatitudes and high latitudes, the RMS difference iscorrespondingly higher above 200 hPa, probably due to larger variability and lower tropopause heights inthose regions. The water vapor RMS differences also show a similar trend with a slightly larger RMS differencefor the midlatitudes because of a larger percentage of land samples.

Figure 11 shows a comparison of the “cloud-free” retrievals as well as the “cloud-cleared” retrievals over thesea for the CrIMSS MX7.1 emulation and the Aqua AST-V6 retrievals. The AST-V6 retrievals generated at theNASA Jet Propulsion Laboratory (JPL) were matched to the CrIMSS EDRs at each FOR to perform this evalu-ation. Although the AST-V6 uses a height-dependent QC in accepting retrieval products for data assimilationpurposes (termed as “pbest” quality), in order to be consistent with the CrIMSS acceptance criteria (where thewhole profile is accepted or rejected based on QC information), the AST-V6 pbest QC flag was used to acceptthe whole profile or reject the whole profile for a comparison with CrIMSS AVTP and AVMP products. Forcloud-free samples, the AIRS “clear” flag in conjunction with the pbest QC flag was used to select a commonsample and then the CrIMSS accepted cases were selected. An examination of the figure reveals that over

Figure 9. The global evaluation of the CrIMSS MX7.1 IR +MW EDR Products: (left) AVTP and (right) AVMP RMS differencesfor land, sea, and all categories.

Table 5. AVTP and AVMP Yield and RMS Difference Statistics for Different Layers of the Atmosphere for MX7.1 WithRequirements for the First Stage MW-only Retrieval Product

AVTP(p) AVMP(p)

System MW only Yield (%) 30–300 hPa (K) 300–700 hPa (K) 700 hPa-sfc (K) 300–600 hPa (%) 600 hPa-sfc (%)

MX7.1 43.5 1.51 1.7 2.2 37.9 26.8

Requirements ~50 1.5 1.5 2.5 40 20

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4968

cloud-free cases, the CrIMSS EDR is performing equally well compared to the AST-V6 heritage algorithmproduct and meeting the 1 K/1 km RMS difference. This suggests that the OE method, being “physical only” isrobust and accurate. Over cloud-cleared cases, the CrIMSS algorithm accepts a larger percentage of samples(51%) compared to the AST-V6-pbest (43%) but also shows a slightly larger RMS difference. The larger RMSdifference with cloud-cleared cases can be attributed to using the MW-only first guess in the CrIMSS EDR

Figure 10. The global evaluation of the CrIMSS MX7.1IR +MW EDR Products: (left) AVTP and (right) AVMP RMS differencesfor different latitude zones: global (90°N–90°S, red), tropical (23°N–23°S, green), midlatitude (50°N–23°N, 50°S–23°S, blue),and high-latitude (90–50°N, 90°S–50°S, cyan) regions.

Figure 11. The global evaluation of the CrIMSS MX7.1 IR+MW EDR Products over sea in comparison to the AST-V6 retrievals:(left) AVTP and (right) AVMP: CrIMSS EDRs (solid lines) relative to Aqua AST-V6 (dashed lines) retrievals for cloud-free (magenta)and cloud-cleared retrievals (blue).

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4969

algorithm in contrast to the neural net (a nonlinear regression) first guess used by the AST-V6 heritage al-gorithm. The fundamental difference between the AST-V6 and the CrIMSS EDR is that the CrIMSS EDR is aphysical-only approach and does not incorporate any knowledge of ECMWF within the retrieval. The largerRMS difference between the CrIMSS EDR and the AST-V6 could be due to either a poorly optimized CrIMSSEDR retrieval or the ability of the AIRS neural network first guess to emulate the ECMWF statistics.

Figure 12 shows the comparison of the CrIMSS second stage IR +MWAVTP (left) RMS and (right) bias statisticswith reference to ECMWF for global ocean (solid blue) and for global ocean within ±60° latitudes (solidbrown). Also shown in the figure are the statistics for global ocean for the AST-V6-pbest final physical retrieval(dashed blue) and the neural net first guess (dashed dotted blue) used by the AST-V6 algorithm. The statisticsfor global ocean within ±60° latitudes are also shown for the AST-V6 physical retrieval as a dashed brown lineand almost entirely overlie the global ocean statistics in dashed blue. With regards to the CrIMSS AVTP per-formance, the percentage of accepted samples is relatively higher for global sea (51%) and has relativelymore accepted cases from polar regions than the AST-V6-pbest (46%). A cold bias at the surface suggestspossible leakage of difficult cloudy cases and suggests improvements to cloud-clearing procedures. The AST-V6 physical retrieval (dashed blue, but mostly covered up by dashed brown) shows relatively smaller bias andRMS differences with ECMWF in comparison to the CrIMSS retrievals (solid blue). Although the AST-V6 showsbetter performance, the AST-V6 physical retrieval does not improve significantly over the AIRS neural net-work solution (dashed dotted blue), indicating that the most of the skill shown for the AIRS V6 is coming fromthe neural network. Given that the neural network is trained from ECMWF, it is likely that the neural networkcontains some statistical knowledge of the ECMWF. The AIRS V6 is a more mature algorithm, and compari-sons such as these will be used to improve and optimize the CrIMSS EDR algorithm. The improvement seen inRMS difference for ±60° latitude ensemble over global ocean cases (solid brown versus solid blue) suggeststhat the CrIMSS EDR algorithm requires a better snow and ice microwave emissivity representation andimprovements to the ATMS bias-tuning procedures using a larger set of confidently clear cases from polarregions. Evaluation of the four different focus days also revealed that the poor yield, where the retrieval failsto pass the first stage MW-only convergence, is more confined to polar regions. The CrIMSS algorithm seemsto have poor MW-only convergence (low-quality yield in Table 3) for the focus days in 2013 (3 February 2013

Figure 12. The evaluation of the CrIMSS MX7.1 IR+MW EDR cloud-cleared AVTP retrievals over sea in comparison to theAST-V6 retrievals. (left) RMS and (right) bias. Solid lines represent the second stage IR +MW CrIMSS AVTP. Solid blue shows allglobal sea profiles, while solid brown includes only sea profiles within ±60° latitude. Similarly, the AST-V6 retrievals are plottedin dashed blue (for all the global sea profiles) and dashed brown (for the sea profiles within ±60° latitude). Dashed dotted blueshows the AST-V6 neural network first guess solution statistics for global sea.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4970

and 12March 2013, austral summer) in comparison to the focus days in 2012 (15 May 2012 and 20 September2012, austral winter and spring) indicating that the microwave emissivities used by the algorithm are deficientfor conditions of melting snow and ice. Improvements currently in progress for the ATMS bias-correction LUTs,snow, and ice microwave emissivity representations will alleviate these issues to realize better retrievals overalland even greater improvement over polar regions.

5.2. Evaluation With Global RAOB Matches

Figure 13 shows the CrIMSS AVTP and AVMP EDR product statistics with reference to global RAOB networkmeasurements generated by NPROVS [Reale et al., 2012]. The data set contains a total of 24,615 matchedcollocations of global RAOBs and CrIMSS MX6.6 IDPS operational AVTP and AVMP retrievals over a period oftwo months (March–April 2013). Collocation criteria of ±3 h in time and a distance match of 100 km are usedin choosing the CrIMSS retrieval that is closest in time and distance to the RAOB. The global yield is about 85%(IR +MW 34%+MW only 51%). The algorithm failed to provide any good retrievals for the remaining 15% ofthe samples. Some of the fixes incorporated into the MX6.6 (Table 2) have helped to improve the yield fromthe earlier version (MX6.3) that operated until February 2013. About 13% of the samples are over the sea(either ship observations or island/coast RAOBs), and the remaining samples are predominantly from 0000 UTto 1200 UT midlatitude land areas, where most of the global conventional RAOBs are concentrated. Thestatistics thus reflect primarily the sounding performance over midlatitude land areas. The results of evaluationwith RAOBs are similar to the results of evaluation with reference to the ECMWF as shown in Figure 8a for thesecond stage IR+MW and the first stage MW-only AVTP and AVMP products. The AVTP and AVMP RMSdifferences for the second stage IR+MW retrieval (Figure 13, red line) are more or less similar to the RMSdifferences observed over land with reference to the ECMWF (Figure 9, green line) and primarily reflectsounding performance over midlatitude land areas, where most of the global RAOB network stations areconcentrated. The larger RMS difference near the surface can also be due to larger collocation error and also theearlier and less optimal versions of the CrIMSS EDR algorithm (MX6.6 versus MX7.1).

Figure 13. The evaluation of the CrIMSS IDPS MX6.6 AVTP and AVMP EDR products with global operational RAOBs: Thedata set contains 60 days (March–April 2013, sample size: 24,615 profiles) of global RAOBs matched to the CrIMSS EDRproducts with a collocation criteria of ±3 h and 100 km in radius. In this figure, the IR +MW lines refer to the 34% of profilesthat passed the IR +MW quality criteria, while the MW-only lines refer to another 51% of profiles that failed the IR +MWquality criteria but passed the MW-Only criteria. This leads to an overall yield of 85%.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4971

5.3. Evaluation With COSMIC Matches

Figure 14 shows a comparison of the CrIMSS EDR MX7.1 AVTP retrieval matches with the COSMIC dry tem-perature profiles coincident in space and within± 1h time. Also shown in the figure are the correspondingglobal RMS differences with ECMWF previously discussed in detail in both section 5.1 and Figure 9. The COSMICdry temperature is valid only at altitudes above the influence of water vapor and a cutoff of 300 hPa was used inthis comparison. The horizontal extent of the GPS occultation raypath is accounted for using the methodologydescribed in M. Feltz et al. (A Methodology for the Validation of Temperature Profiles from HyperspectralInfrared Sounders Using GPS Radio Occultation: Experience with AIRS and COSMIC, submitted to Journal ofGeophysical Research, 2013). The bias trends with altitude observed between the CrIMSS and COSMIC and be-tween the CrIMSS and ECMWF for the focus day (15 May 2012) complement each other, thus making this dataset a viable source of independent verification for long-term evaluations. Although the differences in biases aresmaller, the RMS differences between the CrIMSS and COSMIC and between the CrIMSS and ECMWF deviatefurther from each other above 100hPa. These differences may be attributed to a smaller number of COSMICmatches (N< 200 profiles) in comparison to the global ECMWF matches (47% of 318,600 profiles; Figure 9, redline) and suggests that a single day of matches may not be representative of a result that could be obtainedfrom a longer validation time period. Comparison of the COSMIC and CrIMSS EDR products using a larger set ofmatches (1–10 October, MX5.3 and 22–31 October, MX6.3) show smaller bias and RMS differences as weprogress from the MX5.3 to MX6.3 and subsequent versions (Barnet, CrIMSS-Pro-Brief-2013). Above the tropo-pause, GPSRO should be a more reliable validation source than either the ECMWF or the RAOBs. Therefore, it isimportant to understand why the RMS difference increases for the COSMIC comparisons. We are currentlyanalyzing a larger set of collocations from the three focus days to verify in detail the differences between theCrIMSS EDR products, COSMIC, and ECMWF matches.

5.4. Evaluation With PMRF-Dedicated RAOB Matches

Dedicated RAOB matches from the PMRF site were the first data sets made available to the validation team andare used here for a preliminary validation of CrIMSS MX7.1 EDR products. These RAOB launches were madeduring 12–31 May and 16–30 September 2012 from Kauai (22.05°N, 159.78°W), Hawaii. The data set contains atotal of 30 dedicated RAOB ascents. The ECMWF analysis fields were generated tomatch RAOB locations in timeand space, and the CrIS/ATMS SDR granule matches and the corresponding EDR products from the IDPS version(MX5.3) were downloaded from the GRAVITE. During this period, the operational IDPS version MX5.3 has atypical yield of 4.7% (Table 3) for the second stage IR+MW combined stage. Thus, the most of the CrIMSS EDRsmatched to the dedicated RAOBswere from the first stageMW-only retrievals. To be consistent with the analysisperformed with different data sets, the MX7.1 off-line emulations were performed. The PMRF high-resolution

Figure 14. The comparison of the CrIMSS AVTP bias and the RMS difference relative to coincident ECMWF and COSMIC GPSradio occultation data for the focus day 15 May 2012.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4972

RAOBs were reduced to match the CrIMSS EDR 100-layer product (see Nalli et al. [2013]) for details on themethod used to reduce RAOB data to 100 layers). Figures 15a and Figure 15b show plots of the AVTP and AVMPbias and RMS differences with reference to dedicated RAOB and ECMWF matches for a common ensemble ofsamples accepted by both the first stageMW-only and the second stage IR+MW retrievals. These figures reveal

Figure 15. (a) The CrIMSS AVTP and AVMP biases with matched ECMWF and dedicated RAOBs over the PMRF station atKauai (22.05°N, 159.78°W), Hawaii: IR +MW versus RAOB (solid red), IR +MW versus ECMWF (dashed red), MW-only versusRAOB (solid blue), and MW-only versus ECMWF (dashed blue). This figure (and Figure 15b) differs from the definition of MWOnly in Figures 8 and 13. The IR +MW lines still show statistics for the profiles which passed the IR +MW quality criteria. TheMW-only lines in these figures represent the statistics for the same profiles that passed the IR +MW quality criteria. (b) TheCrIMSS AVTP and AVMP RMS differences with matched ECMWF and dedicated RAOBs over the PMRF station at Kauai(22.05°N, 159.78°W), Hawaii: IR +MW versus RAOB (solid red), IR +MW versus ECMWF (dashed red), MW only versus RAOB(solid blue), and MW only versus ECMWF (dashed blue).

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4973

improved performance (as expected) by the second stage IR+MW in comparison to the first stage MW-onlyretrieval product. The AVTP and AVMP products appear to agree better with the ECMWF than with the RAOBmatches (with the exception of the AVMP near the surface) and the RMS differences with RAOBs, and ECMWFare similar to those observed over global tropics (Figure 10). The first stage MW-only retrieval shows slightlylarger AVTP and AVMP RMS differences (Figure 15b) with both the RAOB and ECMWF matches and the biasesobserved for the MW-only retrieval resemble land/coastal biases due to location of the PMRF site and the landfraction affecting the CrIMSS FOR. Larger AVMP RMS differences observed for MW-only retrieval (Figure 15b) aredue to larger biases with the truth data sets (Figure 15a). It should be noted that the differences between thededicated RAOB and ECMWF statistics for water vapor are due to a small number of radiosondes (i.e., 19) in thisstatistic, and both the bias and RMS are enhanced when there is a dry layer aloft due to the fact that the smallnumber of radiosonde water vapor observations are in the denominator of the calculation (see discussion inNalli et al. [2013]). Although these dedicated RAOBs are not assimilated into NWP models, the agreement be-tween Ethe CMWF and the RAOBs ensures that the RAOB compilations from this site are of good quality.

The percentage of land fraction affecting the CrIMSS FORs is around 5–40% depending on the satellite viewinggeometry and thus may not realistically represent a sounding from a clear or cloud-cleared “sea” location whichoften is the expectation in this type of evaluation. The percent land fraction within the CrIMSS FOR may beimpacting the first stage MW-only solution and consequently the second stage IR+MW solution. Despite thisdrawback, the statistics for the second stage show good agreement with the ECMWF and the RAOBs.

5.5. Known Issues With Identified and Tested Fixes

The CrIMSS EDR precipitation flag is intended to identify precipitation cases in cloudy scenes identified by theCrIMSS algorithm or in scenes where the ATMS instrument would be functioning without CrIS. The flag isused to avoid precipitating cases when evaluating CrIMSS EDR performance. The algorithm that is currently inoperations (implemented in MX5.3 and continuing to MX7.1) was based solely on the AMSU-A [Ferraro et al.,2000] and is not optimized for precipitation detection from ATMS. Additionally, the algorithm was adoptedincorrectly into the initial CrIMSS package and thus has severe limitations and is found to produce falsepositives and false negatives in many regions. Figure 16a shows the rain flag from the SNPP CrIMSS EDR al-gorithm for the 15 May 2012 focus day. Over land areas, the algorithm shows less than 5% of the precipitationoccurrence, and very little rain is detected in the Intertropical Convergence Zone (ITCZ), one of the rainiestregions of the globe. Other areas of inadequate rain detection are also noted in the figure. The current im-plementation of the precipitation flag will thus have implications in evaluating the CrIMSS EDR performance.

To deal with this problem, the CrIMSS team has worked to develop a MSPPS “Day-2 like” algorithm to create arain flag. The MSPPS Day-2 algorithm vastly improved rain detection over both land and ocean by focusingon the AMSU-B high-frequency channels. This was later refined [Vila et al., 2007] and continues to run

Figure 16. Improvements to the CrIMSS EDR algorithm—precipitation flag: (a) Precipitation detection flag as depicted from the current CrIMSS EDR algorithm. (b)Updated precipitation flag using the MSPPS Day-2 like algorithm.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4974

operationally at NOAA/NESDIS (National Environmental Satellite, Data, and Information Service) for bothNOAA 18 and NOAA 19, using MHS in place of the AMSU-B. The ATMS has many similarities to both AMSU-Band MHS. Major exceptions are at 165.5 GHz on the ATMS (with quasi-horizontal polarization) compared to150GHz with quasi-vertical polarization for AMSU-B and 157GHz with quasi-vertical polarization for MHS, aswell as polarization differences at 50.3 GHz (horizontal on ATMS versus vertical for AMSU-B and MHS), whichimpacts the retrieval of ice water path and therefore the precipitation rate. To account for these frequency/polarization differences, a nonlinear regression for 50.3 GHz between the AMSU and ATMS simulations wasdeveloped, and a nonlinear regression that utilizes both ATMS 88.2 GHz and 165GHz channels to simulateAMSU-B 150GHz was developed over both land and ocean. By these means, the ATMS channels areconverted into synthetic AMSU-B channels so that the current MSPPS algorithm can be directly applied togenerate a rain rate and a rain flag for CrIMSS. Figure 16b illustrates the rain flag map using the proposedMSPPS rain algorithm, and the deficiencies noted in Figure 16a have been eliminated. Note the increased rainoccurrences in the ITCZ; much more rain is now detected over land, and the transition from land to ocean isnow continuous. The MSPPS rain frequency statistics were compared with the rain statistics produced by theMiRS [Boukabara et al., 2011; Iturbide-Sanchez et al., 2011] and were found to be in close agreement. Thealgorithm is being integrated into the CrIMSS EDR algorithm, and an ACP is in preparation for a future IDPSbuild. We are currently testing the improvements expected in the CrIMSS performance with the new pre-cipitation flag. In addition to the precipitation flag, a number of minor fixes that may marginally improve theperformance were noted and necessary ACPs have been prepared for the future IDPS build.

6. Summary and Conclusions

The CrIMSS EDR is a baseline operational product utilizing a unique physical-only approach available to theresearch community. The algorithm has been in operations for about 18 months, and after minor changes tothe code and LUT updates to the prelaunch version, the algorithm has shown a remarkably improved per-formance that enabled the declaration of provisional maturity status in January 2013. The EDR algorithmoptimization, characterization, and evaluation with a variety of validation data sets reveal the following:

1. The global yield for the CrIMSS EDR algorithm version MX7.1 is about 90%. The combined IR+MW EDR prod-uct performance (with a yield of 47%, high quality) and the MW-only product performance (with a remainderyield of 43%, low quality) are meeting the AVTP and AVMP global requirements for most of the atmosphere.The algorithm is performing as expected for different categories (land, sea, and coast) and for different re-gimes (tropics, polar, and high latitudes). A slightly larger global RMS difference exceeding the requirementfor the 700hPa sfc is due to larger RMS differences over the land cases and polar regions impacting the globalRMS differences. Temperature RMS differences with ECMWF are very close to meeting the requirements.Water vapor retrievals may require additional algorithm optimization. The AVTP and AVMP EDR productsare ready for stage 1, 2, and 3 validations, and users can start utilizing these products for scientific applications.

2. The evaluation of the CrIMSS AVTP and AVMP EDR products with EDR products from Aqua AST-V6 algo-rithm reveals very similar performance for cloud-free cases. For the cloud-cleared cases, the algorithmis showing a slightly colder bias at the surface for the AVTP product. This requires some optimization.The CrIMSS algorithm also requires a better snow and ice microwave emissivity representation and im-provements to ATMS bias-tuning procedures using a larger set of confidently clear cases. The matchedAIRS/AMSU and CrIS/ATMS SDR and EDR data sets with other correlative measurements over open oceansprovide ample opportunities to develop improved cloud-clearing methodologies and compare the per-formance of two different algorithms over more challenging atmospheric conditions. The algorithms usedby both the AIRS science team and NUCAPS are thus complimentary with the CrIMSS EDR algorithm andcan be combined to define many blended products utilizing strengths of individual algorithms and alle-viating deficiencies for mutual benefit.

3. The precipitation detection routines in the current operational version (MX6.6) have some issues withfalse positives and false negatives. The improved precipitation detection algorithm is expected to beimplemented in a future IDPS build, and the new precipitation flag is expected to improve EDR perfor-mance by properly accounting for false positives and false negatives.

4. Further improvement in the CrIMSS EDR algorithm is planned. This can be done mainly by improving var-ious LUTs such as ATMS and CrIS bias correction LUTs. Better criteria will be used to screen out cloud con-tamination in our bias correction procedure. A better snow and ice MW emissivity representation such as

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4975

those measured by the special sensor microwave/imager and the advanced microwave acanning radiom-eter will be added in the generation of climatological covariance and background LUTs. These new LUTswill improve the accuracy of the MW-only EDR product and the yield over snow and ice surfaces, which inturn will improve the cloud clearing and the second stage IR +MW EDR product. The CrIS land emissivitycovariance matrix and background LUTs will also be refined by using external emissivity databases de-rived from IASI or MODIS [Seemann et al., 2008; Zhou et al., 2011]. By using monthly mean global emissivitymaps from either IASI or MODIS as first guess in the ATMS+CrIS retrieval module, better cloud-clearedCrIS radiances are expected, leading to better CrIMSS EDR performance.

5. The CrIMSS EDR algorithm retrieves ozone profile and is currently an intermediate product. Although theevaluation of ozone profile and total ozone was not discussed in this paper, a preliminary evaluation ofthe CrIMSS retrieved total ozone maps with the Ozone Mapping Spectrometer revealed reasonable com-parisons. However, the number of emissivity hinge points in use in the CrIMSS EDR algorithm is not ade-quate to represent the land surface emissivity spectral variations from 950 to 1095 cm�1. As a result, theradiance fitting ozone spectral region is degraded resulting in lower-quality ozone IP. By adding a new setof emissivity hinge points for the spectral range 950–1095 cm�1, the ozone intermediate product can beimproved in the combined IR +MW retrievals.

6. The current CrIMSS EDR algorithm code is equipped with necessary functions to retrieve other trace gases(e.g., CH4 and N2O), cloud height, and cloud effective emissivity. Although these modules require a littlemore insight and optimization, the infrastructure exists to incorporate these retrievals into the operationalproduct. All of these EDR algorithm improvements summarized here are available as detailed presenta-tions at ftp://www.star.nesdis.noaa.gov/pub/smcd/spb/murtyd/telecons/CrIMSS_ALG.

ReferencesAnthes, R. A. (2011), Exploring Earth’s atmosphere with radio occultation: Contributions to weather, climate and space weather, Atmos. Meas.

Tech., 4, 1077–1103, doi:10.5194/amt-4-1077-2011.Anthes, R. A., et al. (2008), The COSMIC/FORMOSAT-3 mission: Early results, Bull. Am. Meteorol. Soc., 89, 313–333, doi:10.1175/BAMS-89-3-313.Aumann, H. H., et al. (2003), AIRS/AMSU/HSB on the Aqua Mission: Design, science objectives, data products, and processing Systems, IEEE

Trans. Geosci. Remote Sens., 41(2), 253–264, doi:10.1109/TGRS.2002.808356.Barnet, C. D., M. Goldberg, T. King, N. Nalli, W. Wolf, L. Zhou, and J. Wei (2005), Alternative cloud clearing methodologies for the Atmospheric

Infrared Sounder (AIRS), in Atmospheric and Environmental Remote Sensing Data Processing and Utilization: Numerical AtmosphericPrediction and Environmental Monitoring, SPIE Proceedings, vol. 5890, 12 pp., SPIE, San Diego, California, doi:10.1117/12.615238.

Boukabara, S. A., et al. (2011), MiRS: An All-Weather 1DVAR Satellite Data Assimilation & Retrieval System, IEEE Trans. Geosci. Remote Sens.,49(9), 3249–3272, doi:10.1109/TGRS.2011.2158438.

Chahine, M. T., et al. (2006), AIRS: Improving weather forecasting and providing new data on greenhouse gases, Bull. Am. Meteorol. Soc., 87(7),911–926, doi:10.1175/BAMS-87-7-911.

Divakarla, M., C. D. Barnet, M. D. Goldberg, L. M. McMillin, E. Maddy, W. Wolf, L. Zhou, and X. Liu (2006), Validation of Atmospheric InfraredSounder temperature and water vapor retrievals with matched radiosonde measurements and forecasts, J. Geophys. Res., 111, D09S15,doi:10.1029/2005JD006116.

Divakarla, M., et al. (2008), Evaluation of Atmospheric Infrared Sounder ozone profiles and total ozone retrievals with matched ozonesondemeasurements, ECMWF ozone data, and Ozone Monitoring Instrument retrievals, J. Geophys. Res., 113, D15308, doi:10.1029/2007JD009317.

Divakarla, M., et al. (2010a), Preliminary Evaluation of CrIMSS EDR Products with “The CrIS/ATMS Proxy Data Package”, paper presented atSounding Atmosphere Team (SOAT) Meeting, Silver Spring, Md., June 15-17. [Available at ftp://www.star.nesdis.noaa.gov/pub/smcd/spb/nnalli/SOAT/2010-06/.]

Divakarla, M., et al. (2010b), Evaluation of CrIS/ATMS Proxy Radiances/Retrievals with IASI Retrievals, ECMWF Analysis and RAOB Measurements,paper presented at 2010 IEEE IGARSS, Session: Next Generation US Operational Environmental Satellite Systems, Honolulu, HI.

Divakarla, M., et al. (2011a), Validation of IASI Temperature and Water Vapor Retrievals with Global Radiosonde Measurements and ModelForecasts, paper presented at Hyperspectral Imaging and Sounding of the Environment (Optical Society of America), Toronto, Canada,doi:10.1364/AOPT.2011.JWA25.

Divakarla, M., et al. (2011b), Validation of CrIMSS EDR Products with Matched ECMWF Analysis, RAOB Measurements, and IASI Retrievals,paper presented at Seventh Annual Symposium on Future Environmental Satellite Systems, Am. Meteorol. Soc., Seattle, WA. [Available athttps://ams.confex.com/ams/91Annual/webprogram/Paper177630.html.]

Divakarla, M., et al. (2011c), Pre-Launch Evaluation of NPP-CrIMSS EDR Algorithm Products with Matched ECMWF Analysis, RAOBMeasurements, and IASI Retrievals, paper presented at Hyperspectral Imaging and Sounding of the Environment, (Optical Society ofAmerica), Toronto, Canada, doi:10.1364/HISE.2011.HMA2.

Divakarla, M., et al. (2012), Pre-Launch to Post-Launch Transition and Evaluation of CrIMSS EDR Algorithm and Products, paper presented atEighth Annual Symposium on Future Operational Environmental Satellite Systems, Am. Meteorol. Soc., New Orleans, LA. [Available athttps://ams.confex.com/ams/92Annual/webprogram/Paper195956.html.]

Divakarla, M., et al. (2013), Provisional Maturity Assessment of Cross Track Infrared Sounder (CrIS) Temperature andMoisture Profile Products,paper presented at Ninth Annual Symposium on Future Operational Environmental Satellite Systems, Austin, TX. [Available at https://ams.confex.com/ams/93Annual/webprogram/Paper216717.html.]

Du, J., F. Cooper, and S. Fueglistaler (2012), Statistical analysis of global variations of atmospheric relative humidity as observed by AIRS,J. Geophys. Res., 117, D12315, doi:10.1029/2012JD017550.

Ferraro, R. R., F. Weng, N. Grody, and L. Zhao (2000), Precipitation characteristics over land from the NOAA-15 AMSU sensor, Geophys. Res.Lett., 27, 2669–2672.

AcknowledgmentsWe wish to thank the European Centrefor Medium-Range Forecasting groupand the National Center forEnvironmental Prediction for providingthe ECMWF and NCEP data used in thispaper. We also express our sincere ap-preciation to the Jet PropulsionLaboratory and the Goddard EarthSciences Data and Information ServicesCenter for providing the AIRS V6 re-trievals for the focus days. The fundingfor this paper has been provided by theJPSS Program Office and is gratefullyacknowledged. The manuscript con-tents are solely the opinions of the au-thors and do not constitute a statementof policy, decision, or position on behalfof NOAA, NASA, or the U.S. Government.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4976

Ferraro, R. R., F. Weng, N. Grody, L. Zhao, H. Meng, C. Kongoli, P. Pellegrino, S. Qiu, and C. Dean (2005), NOAA operational hydrologicalproducts derived from the AMSU, IEEE Trans. Geosci. Remote Sens., 43, 1036–1049, doi:10.1109/TGRS.2004.843249.

Fetzer, E., et al. (2003), AIRS/AMSU/HSB validation, IEEE Trans. Geosci. Remote Sens., 41(2), 418–431, doi:10.1109/TGRS.2002.808293.Gambacorta, A., et al. (2012), The NOAA Unique CrIS/ATMS processing system (NUCAPS): First light results, paper presented at Proceedings

of the International TOVS Working Group Meeting, ITSC-XVIII, Toulouse, France.Goldberg, M. D., Y. Qu, L. M. McMillin, W. Wolf, L. Zhou, and M. Divakarla (2003), AIRS Near-Real-Time Products and Algorithms in Support of

Operational Numerical Weather Prediction, IEEE Trans. Geosci. Remote Sens., 41(2), 379–389, doi:10.1109/TGRS.2002.808307.Gu, D., and X. Ma (2010), CrIMSS Retrieval Algorithm Testing with Proxy Data, paper presented at 6th Annual Symposium on Future National

Operational Environmental Satellite Systems-NPOESS and GOES-R, Am. Meteorol. Soc., Atlanta, GA. [Available at https://ams.confex.com/ams/17Air17Sat9Coas/techprogram/paper_174911.htm.]

Gu, D., et al. (2011), Testing and Tuning CrIMSS EDR Algorithm with Proxy Data in Preparation for NPP Post-launch EDR Product Validation,paper presented at Seventh Annual Symposium on Future Environmental Satellite Systems, Am. Meteorol. Soc., Seattle, WA. [Available athttps://ams.confex.com/ams/91Annual/webprogram/Paper187505.html.]

Hilton, F., et al. (2012), Hyperspectral Earth observation from IASI: Five years of accomplishments, Bull. Am. Meteorol. Soc., 93(3), 347–370,doi:10.1175/BAMS-D-11-00027.1.

Ho, S.-P., Y.-H. Kuo, W. Schreiner, and X. Zhou (2010), Using SI-traceable Global Positioning System radio occultation measurements forclimate monitoring, Bull. Am. Meteorol. Soc., 91(7), S36–S37.

Iturbide-Sanchez, F., S.-A. Boukabara, R. Chen, K. Garrett, C. Grassotti, W. Chen and F. Weng (2011), Assessment of a variational inversion systemfor rainfall rate over land and water surfaces, IEEE Trans. Geosci. Remote Sens., 49(9), 3311–3333, doi:10.1109/TGRS.2011.2119375.

Jairam, L. (2009), Algorithm Theoretical Basis Document - ATMS Proxy Data Generator, Release 1.0, MIT Lincoln Laboratory Report, 15 pp., MITLincoln Laboratory, Lexington, Mass.

Kizer, S., et al. (2010), Porting and Testing NPOESS CrIMSS EDR Algorithms, paper presented at 2010 IEEE IGARSS, Session: IR AtmosphericSounding and Calibration, Honolulu, HI.

Lee, F., et al. (2010), NPOESS: Next-Generation Operational Global Earth Observations, Bull. Am. Meteorol. Soc., 91, 727–740, doi:10.1175/2009BAMS2953.1.

Liu, X., and S. Kizer (2009a), CrIMSS OPS Code Porting and Proxy Data Testing, paper presented at Sounding Atmosphere Team (SOAT)Meeting, Silver Spring, Md. [Available at ftp://www.star.nesdis.noaa.gov/pub/smcd/spb/nnalli/SOAT/2009-05/.]

Liu, X., and S. Kizer (2009b), Testing CrIMSS EDR Algorithm Using Synthetic and Proxy Data, paper presented at Sounding Atmosphere Team(SOAT) Meeting, Logan, UT. [Available at ftp://www.star.nesdis.noaa.gov/pub/smcd/spb/nnalli/SOAT/2009-09/.]

Liu, X., S. Kizer, A. Larar, W. Smith, D. Zhou, C. Barnet, M. Divakarla, G. Guo, and W. Blackwell (2010), Porting And Testing NPOESS CrIMSS EDRAlgorithms, in Proc. SPIE, 7857, Multispectral, Hyperspectral, and Ultraspectral Remote Sensing Technology, Techniques, and ApplicationsIII, 4 pp., SPIE, Incheon, Republic of Korea, doi:10.1117/12.869428.

Liu, X., et al. (2012), Retrieving Atmospheric Temperature And Moisture Profiles From Suomi NPP CrIS/ATMS Sensors Using CrIMSS EDRAlgorithm”, Proceeding of International Geoscience and Remote Sensing Symposium (IGARSS), Munich, Germany, 1956–1959,doi:10.1109/IGARSS.2012.6350816.

Lynch, R., J. Moncet, and X. Liu (2009), Efficient nonlinear inversion for atmospheric sounding and other applications, Appl. Opt., 48(10),1790–1796, doi:10.1364/AO.48.001790.

Maddy, E. S., and C. D. Barnet (2008), Vertical Resolution Estimates in Version 5 of AIRS Operational Retrievals, IEEE Trans. Geosci. Remote Sens.,46, 2375–2384, doi:10.1109/TGRS.2008.917498.

Maddy, E. S., et al. (2011), Using MetOP-A AVHRR Clear-Sky Measurements to Cloud-Clear MetOP-A IASI Column Radiances, J. Atmos. OceanicTechnol., 28, 1104–1116, doi:10.1175/JTECH-D-10-05045.1.

Maddy, E. S., et al. (2012), On the Effect of Dust Aerosols on AIRS and IASI Operational Level 2 Products, Geophys. Res. Lett., 39, L10809,doi:10.1029/2012GL052070.

Moncet, J.-L., G. Uymin, A. E. Lipton, and H. E. Snell (2008), Infrared Radiance Modeling by Optimal Spectral Sampling, J. Atmos. Sci., 65,3917–3934, doi:10.1175/2008JAS2711.1.

Nalli, N. R., et al. (2011), Multi-year observations of the tropical Atlantic atmosphere: Multidisciplinary applications of the NOAA Aerosols andOcean Science Expeditions (AEROSE), Bull. Am. Meteorol. Soc., 92, 765–789, doi:10.1175/2011BAMS2997.1.

Nalli, N. R., et al. (2013), Validation of satellite sounder environmental data records: Application to the Cross-track Infrared MicrowaveSounder Suite, J. Geophys. Res. Atmos., 118, 13,628–13,643, doi:10.1002/2013JD020436.

Reale, T., B. Sun, F. Tilley, and M. Pettey (2012), The NOAA Products Validation System (NPROVS), J. Atmos. Oceanic Technol., 29, 629–645,doi:10.1175/JTECH-D-11-00072.1.

Salisbury, J. W., and D. M. D’Aria (1992), Emissivity of terrestrial materials in the 8-14μm atmospheric window, Remote Sens. Environ., 42, 83–106.Salisbury, J. W., and D. M. D’Aria (1994), Emissivity of terrestrial materials in the 3-5μm atmospheric window, Remote Sens. Environ., 47, 345–361.Seemann, S. W., E. E. Borbas, R. O. Knuteson, G. R. Stephenson, and H.-L. Huang (2008), Development of a Global Infrared Land Surface

Emissivity Database for Application to Clear Sky Sounding Retrievals from Multispectral Satellite Radiance Measurements, J. Appl.Meteorol. Climatol., 47, 108–123, doi:10.1175/2007JAMC1590.1.

Steiner, A. K., et al. (2013), Quantification of structural uncertainty in climate data records from GPS radio occultation, Atmos. Chem. Phys., 13,1469–1484, doi:10.5194/acp-13-1469-2013.

Susskind, J., C. D. Barnet, and J. Blaisdell (2003), Retrieval of atmospheric and surface parameters from AIRS/AMSU/HSB data under cloudyconditions, IEEE Trans. Geosci. Remote Sens., 41(2), 390–409, doi:10.1109/TGRS.2002.808236.

Susskind, J., J. M. Blaisdell, L. Iredell, and F. Keita (2011), Improved Temperature Sounding and Quality Control Methodology Using AIRS/AMSUData: The AIRS Science Team Version 5 Retrieval Algorithm, IEEE Trans. Geosci. Remote Sens., 49(3), 883–907, doi:10.1109/TGRS.2010.2070508.

Tian, B., E. J. Fetzer, B. H. Kahn, J. Teixeira, E. Manning, and T. Hearty (2013), Evaluating CMIP5 Models using AIRS Tropospheric AirTemperature and Specific Humidity Climatology, J. Geophys. Res. Atmos., 118, 114–134, doi:10.1029/2012JD018607.

Tobin, D. C., H. E. Revercomb, R. O. Knuteson, B. M. Lesht, L. L. Strow, S. E. Hannon, W. F. Feltz, L. A. Moy, E. J. Fetzer, and T. S. Cress (2006), ARMsite atmospheric state best estimates for AIRS temperature and water vapor retrieval validation, J. Geophys. Res., 111, D09S14, doi:10.1029/2005JD006103.

Vila, D., R. Ferraro, and R. Joyce (2007), Evaluation and Improvement of AMSU Precipitation Retrievals, J. Geophys. Res., 112, D20119,doi:10.1029/2007JD008617.

Zavyalov, V., et al. (2013), Noise performance of the CrlS instrument, J. Geophys. Res. Atmos., 118, 13,108–13,120, doi:10.1002/2013JD020457.Zhou, D. K., A. M. Larar, X. Liu, W. L. Smith, L. L. Strow, P. Yang, P. Schlüssel, and X. Calbet (2011), Global land surface emissivity retrieved from

satellite ultraspectral IR measurements, IEEE Trans. Geosci. Remote Sens., 49, 1277–1290, doi:10.1109/TGRS.2010.2051036.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD020438

DIVAKARLA ET AL. ©2014. American Geophysical Union. All Rights Reserved. 4977