Microwave Emissivity Over Ocean in All-Weather Conditions: Validation Using WINDSAT and Airborne GPS Dropsondes

376 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 2, FEBRUARY 2008

Microwave Emissivity Over Ocean in All-WeatherConditions: Validation Using WINDSAT and

Airborne GPS DropsondesSid-Ahmed Boukabara, Senior Member, IEEE, and Fuzhong Weng

Abstract—Emissivity spectra computed by the FASTEM-3model using global positioning system dropsonde wind as inputare compared to emissivity retrieved from stringently coincidentWINDSAT measurements, using a variational approach. This isdone in both clear and rainy conditions to assess the validityof both the emissivity model and the retrieval technique in allconditions. Results of this comparison are presented for verticaland horizontal polarizations, in moderate to high wind conditions.In particular, a slope difference is found, and its potential sourcesare discussed in this paper.

Index Terms—Dropsonde (DS), hurricane, microwave emissiv-ity, retrieval algorithm, variational assimilation.

I. INTRODUCTION

THE MICROWAVE emissivity modeling over sea surfaceshas experienced significant advances in the last decades

and is now considered a well-handled issue [12]. Most opera-tional centers use a microwave emissivity model in the assimi-lation process of surface-sensitive channels, for the purpose ofproducing sea surface wind intensities and/or directions [30]. Alarge number of emissivity models have been published in theliterature—some are based on empirical expressions but fast,and some with explicit treatment of the surface geometry butwith a high computation cost that prevents them from beingoperationally used [2], [11], [15], [27], [31], [33], [34]. Thereare, however, remaining issues in the emissivity modelingrelated to uncertainties in the dielectric constant modeling, thefoam emissivity and coverage parameterizations, and the natureof the relationship between the wind and the surface geometry[1], [6], [25], [26]. There are also wide-open uncertainties onthe accuracy of the emissivity models in rainy conditions [4],[5], [8]. In these conditions, the sea roughness is altered by theamount of rain and the size distribution of the precipitation. Inaddition, the atmospheric signal is impacted by attenuation andscattering of the rain that also freshens the surface.

On the other hand, direct retrievals of emissivity spectrafrom brightness temperatures have been performed, not onlyover land but also over ocean, using variational approaches oranalytical methods [17], [24]. In this paper, we undertook to

Manuscript received February 15, 2007; revised May 16, 2007.The authors are with the National Oceanic and Atmospheric Administration,

National Environmental Satellite, Data, and Information Service/Center forSatellite Applications and Research (STAR), Camp Springs, MD 20746 USA(e-mail: Sid.Boukabara@ noaa.gov; Fuzhong.Weng@ noaa.gov).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TGRS.2007.905366

compare retrievals of emissivity spectra from WINDSAT dataranging from 6 to 37 GHz, to emissivity model simulations us-ing airborne global positioning system (GPS) dropsonde (DS)measurements of wind intensity. These latter were provided tous by the Hurricane Research Division in Miami, FL. They areroutinely taken during Hurricane events, in and around the ac-tive area where airplanes are flown to probe the storm intensityand other parameters. They provide temperature, humidity, andwind vertical profiles, down to a few meters from the surfacelevel. The variational algorithm used to retrieve the emissivityis called the microwave integrated retrieval system (MIRS) [3],where the emissivity spectrum is part of the retrieved statevector. The use of the Community Radiative Transfer Model(CRTM) as the forward operator in MIRS makes it possibleto perform the retrieval in all-weather conditions includingprecipitating ones. Both radiances and Jacobians are providedby CRTM. The problem is ill constrained but is alleviated byperforming the retrieval in a reduced space, selecting only alimited number of degrees of freedom (eigenvalues).

II. MICROWAVE EMISSIVITY MODEL

The microwave emissivity modeling of the sea surface de-pends on a number of parameters: 1) the dielectric constantfor saline water; 2) the surface geometry or roughness; and3) the foam emissivity and coverage. The dielectric constant isitself dependent on the water temperature and the salinity. Theeffect of this latter for high-frequency channels is known to benegligible [14], [23]. The surface roughness is usually modeledby parameterizing the surface large and/or small wave slopedistribution, which is mostly dependent on the surface windvector. The relationship between the wind and the wave dis-tribution usually assumes neutral stability. The geometric optics(GO) models assume that the emissivity is mainly dependent onthe large length waves (larger than the sensor wavelength) andignore the small-scale variability. Two-scale models do accountfor these small perturbations by using the small perturbationmethod but are time prohibitive. The foam is not very wellmodeled and usually depends on regression fits to wind speed.It is neglected in most models for low wind speeds, and largeuncertainties exist for high wind regimes [6], [25], [26]. Theemissivity model used in this paper is the newest version ofthe FASTEM model. This latter is a fast GO model that wasoriginally called FASTEM-1 [11] and was designed to computeocean surface emissivity, given a sea surface temperature, wind

U.S. Government work not protected by U.S. copyright.

BOUKABARA AND WENG: MICROWAVE EMISSIVITY OVER OCEAN IN ALL-WEATHER CONDITIONS 377

speed, and viewing angle for a microwave radiometer channel.Deblonde and English [9] then developed an improved versioncalled FASTEM-2, which takes into account the treatmentof nonspecular reflection. This had the effect of improvingthe simulation of ocean surface emissivity for Special SensorMicrowave/Imager (SSM/I) and Advanced Microwave Sound-ing Unit (AMSU) for larger viewing angles, as described in[9]. FASTEM-2 has been further updated to allow for thedependence of the ocean surface emissivity on the azimuthangle between the wind direction and the line of sight of theinstrument. The resulting FASTEM-3 is also able to predict thebehavior of the third and fourth elements of the Stokes vectoras a function of wind speed and wind direction. The azimuthalvariation of emissivity was based on an empirical model [20].The surface wave distribution in FASTEM-3 is based on a fastregression fit.

III. RETRIEVAL/ASSIMILATION SYSTEM

The inversion system used in this paper is a generic vari-ational retrieval algorithm (1D-VAR) that applies to all mi-crowave sensors with no change to the code. The retrievalof the temperature and moisture profiles is done along withthe precipitating and nonprecipitating cloud parameters. Thesurface boundary is handled by including the surface emissivityand temperature within the retrieval state vector. To alleviatethe limited information content available in the instrumentat hand, the inversion is performed in a reduced eigenvaluespace as mentioned before, which makes the retrieval processstable and mathematically consistent. The mathematical basisof MIRS is a proven and widely used variational approachdescribed in [28], consisting of minimizing the following costfunction:

J(X) =[12(X − X0)T × B−1 × (X − X0)

]

+[12

(Y m − Y (X))T × E−1 × (Y m − Y (X))]

. (1)

The first right term Jb represents the penalty in departing fromthe background value (a priori information), and the secondright term Jr represents the penalty in departing from themeasurements. X0 and B are the mean vector (or background)and covariance matrix of the state vector X , respectively. Eis the measurement and/or modeling error covariance matrix.The measurement vector is represented by Y m, and Y is theforward operator. The following is the solution to the costfunction minimization process, where n is the iteration index(a maximum number of seven iterations is allowed):

∆Xn+1 ={

BKTn

(KnBKT

n + E)−1

}

× [(Y m − Y (Xn)) + Kn∆Xn] (2)

K, in this case, is the Jacobian or derivative of Y with respectto X . At each iteration n, the new optimal departure from thebackground is computed, given the derivatives as well as thecovariance matrices. This is an iterative-based numerical solu-

tion that accommodates moderately nonlinear problems and/orparameters with moderately non-Gaussian distributions. Thewhole geophysical vector is retrieved as one entity, includingthe temperature, moisture, and hydrometeor parameters as wellas skin surface temperature and emissivity vector, ensuringa consistent solution that fits the radiances. In the case ofnonsounder sensors like WINDSAT, the temperature and hu-midity profiles are still part of the retrieved vector but they areusually not varied and remain close to the background values. Inpractical terms, only one or two empirical orthogonal functionsare used for the vertical profiles, and the net effect is that thewater vapor profile is simply scaled so that only the integratedwater vapor gets changed. The forward operator CRTM usedin MIRS was developed at the Joint Center for Satellite DataAssimilation (JCSDA) [32] and produces radiances as wellas Jacobians, for all geophysical parameters. Derivatives arecomputed using K-matrix developed by tangent linear and ad-joint. This is ideal for retrieval and assimilation purposes. Thedifferent components of CRTM, briefly, are the OPTRAN fastatmospheric absorption model [22] and the advanced doublingadding radiative transfer solution for the multiple-scatteringmodeling [21]. The convergence criterion used in MIRS is

ϕ2 =⌊(Y m−Y (X))T × E−1×(Y m − Y (X))

⌋≤ N (3)

where N is the number of channels used for the retrievalprocess. This mathematically means that the convergence isdeclared reached if the residuals between the measurements andthe simulations at any given iteration are less or equal than onestandard deviation of the noise that is assumed in the radiances.

IV. INSTRUMENTAL CONFIGURATION

In this paper, we will use the wind-dedicated missionWINDSAT onboard to Coriolis platform, which was launchedon January 2003, as it offers the opportunity to assess theemissivity modeling in frequencies from 6.8 to 37 GHz in bothhorizontal and vertical polarizations. The WINDSAT sensormeasures primarily the ocean surface wind field at a hori-zontal resolution of 25 km. The secondary measurements ofWINDSAT are the sea surface temperature, the soil moisture,the rain rate, the integrated cloud, the ice/snow characteristics,and the water vapor. The WINDSAT radiometer has five chan-nels at 6.8, 10.7, 18.7, 23.8, and 37.0 GHz. Table I provideskey characteristics of the system. The antenna beams view theEarth at incidence angles ranging from 50◦ to 55◦. The orbit andantenna geometry result in a forward-looking swath of approx-imately 1000 km and an aft-looking swath of about 350 km.The data used in this paper were processed at the Fleet Numeri-cal Meteorology and Oceanography Center and made availablethrough the Jet Propulsion Laboratory.

V. INFORMATION CONTENT ASSESSMENT

Sensitivities of WINDSAT measurements to emissivity inall-weather condition were assessed and are presented here. It isimportant to assess this information content because variationalalgorithms could retrieve products only because of correlations


TABLE IWINDSAT CHARACTERISTICS (SOURCE: NRL/NAVY)

with other derived products and not really because of theexistence of a radiance signal. These correlations are containedwithin the covariance matrix B. To determine if we really dohave signal in our measurements, under precipitating condi-tions, we computed the Jacobians (using CRTM) of brightnesstemperature with respect to emissivity for varying atmosphericwater vapor (total precipitable water, TPW) values, for a rangeof rain water path (RWP) and for a number of cloud amount/iceamount combinations. The TPW was varied between 2 and80 mm, the vertically integrated RWP was varied between0 and 0.6 mm, for cloud amounts of 0 mm (clear sky) and1 mm and for graupel-size ice vertical amounts of 0 and0.6 mm. This is presented in Fig. 1 for the 37-GHz channel.Typically, lower frequency channels present higher signal (notshown). The results show that clear-sky situations (no cloud, norain, and no ice) present the highest sensitivity as one might ex-pect. The ice effect is minimal in this channel, also as expected.This frequency is indeed not very sensitive to scattering by ice.The absorption effect due to cloud and rain is to significantlyreduce the sensitivity to surface. This reduction is modulatedby the amount of humidity in the atmosphere. The lowestsensitivity, approximately 0.4 K for a 0.01 emissivity change, isfound for the most humid and most intense precipitation cases.The strongest sensitivity in clear-sky cases reaches 1.5 K per0.01 emissivity change. The main conclusion of this analysisis that reasonable signal remains in the WINDSAT channelseven when heavy precipitation is present, making it possibleto retrieve emissivity in these conditions.

A similar sensitivity study was performed for all AMSU andMicrowave Humidity Sounder (MHS) channels (not shown)that indicate that saturation occur very rapidly in channelswith frequencies at 89 GHz and higher, making the retrievalof higher frequency emissivities in a variational algorithmtotally dependent on the built-in correlations between the low-frequency and high-frequency emissivities. It is therefore diffi-cult to assess the emissivity model’s validity at high frequenciesin rainy/cloudy conditions.

VI. CASE STUDY

To test with real data and attempt validation, we need asource of ground truth data which simultaneously providesatmospheric temperature, water vapor, and, of course, wind

speed. The vertical profiles are used to side assess other pa-rameters such as TPW which are byproducts of the variationalalgorithm but not the focus of this paper. This constitutes an ad-ditional piece of information in the validation of the emissivity.The standard buoy measurements offer the wind informationbut lack the atmospheric component. The standard radiosondes,on the other hand, offer the measurements of atmosphericprofiles but no information of the wind. High-quality airborneDropsondes (DSs), such as those provided by the HurricaneCenter in Florida, offer a unique opportunity by providingsimultaneous measurements of temperature, moisture, and windprofiles. In fast moving conditions such as in precipitating cells,the time and space collocation errors could be very significantand may overshadow any validation assessment we might at-tempt. Therefore, very stringent collocation criteria must beused. We focused our study on the 2005 Hurricane Denniswhich devastated Cuba and Haiti before making land fall in theU.S. Gulf Coast on July 10. The general view of that hurricaneis presented in Figs. 2 and 3 from the WINDSAT measurementsand from a single channel of the MHS sensor (157 GHz).The WINDSAT signatures are presented for all H polarizationfrom 6.8 to 37 GHz. On top of the brightness temperaturefields are overlaid a number of airborne DS launch spots.The horizontal color bar represents the brightness temperatureintensity, whereas the vertical bar represents the time differencebetween these DSs and the satellite measurements. It is obviousthat the atmospheric effect on the WINDSAT measurementsis to increase the signal due to cloud and rain absorption (noice scattering). This increase is proportional to the frequency.The MHS channel at 157, on the other hand, experiences asignificant scattering effect due to hydrometeors (Fig. 3). Thesignal tends to drop in the middle of the rainy cell for thischannel.

VII. VALIDATION USING GPS DSS

Coriolis/WINDSAT sensor was used to compare retrievals ofemissivity generated using the variational algorithm describedabove with airborne-based GPS dropwindsondes, in both clearcases as well as under precipitating conditions. It is critical thatone gets a clear sense of how accurate the truth measurements(in our case, DSs) are before interpreting any differences be-tween them and the retrievals.

A. GPS Dropwindsondes

These measurements are made in cloudy/rainy conditions(typically during hurricanes and tropical storms) by high-velocity descending GPS DSs. These latter were quality con-trolled using the Hurricane Analysis and Processing System[16]. They operate at altitudes up to 24 km with a descenttime of about 12 min. The measurements are made every halfsecond, which allows a high vertical resolution. Along with thetemperature and moisture, the vertical wind speed profile is alsomeasured by using GPS-based Doppler signal, down to 4–10 mabove the surface. The validation of these DSs was assessedby comparison with standard radiosondes, radars, buoys, aswell as by human visualization of clouds for the saturation


Fig. 1. Sensitivity of the brightness temperature at 37 GHz (horizontal polarization) to emissivity as a function of both the TPW and the RWP for differentcombinations of integrated cloud liquid water (CLW) and graupel-size ice water path (GWP).

check. For a full description of these measurements, see [16].For the wind, an accuracy of 0.5–2 m/s was estimated. Fig. 4shows the locations of the DSs when they were dropped fromthe airplane during Hurricane Dennis. It also highlights theindividual sondes selected for a case-by-case comparison inSection VII-B. It was reported by Hock and Franklin [16] that

almost-coincident measurements of wind profiles by successiveDSs were made to assess the precision of the wind from thesesondes. The main conclusion was that the coincident sondesalways reported differences less than 1 m/s but they mentionthat in cases of poor GPS geometry accuracy, the differencesmay reach 2 m/s.


Fig. 2. WINDSAT brightness temperature responses during Hurricane Dennis in July 2005. Lower frequencies are less sensitive to atmospheric features. Thelocations of the dropwindsondes are overlaid. Note the narrower swath at 6.8-GHz channel.

Fig. 3. Signature of the NOAA-18/MHS 157-GHz channel of the samefeatures depicted in Fig. 2. Note that the measurements presented here occurred3 h before the WINDSAT measurements. This channel is much more sensitiveto precipitation and cloud than WINDSAT channels. Note that MHS swath ismuch wider with 90 scan positions within each scanline.

B. Limitations of the Validation in Extreme Weather Events

Traditional approach to validating retrievals or models bystatistical comparison with ground truth data collected aroundthe measurement’s time/space location is not optimal in thecase of hurricane conditions. The main reason is the fast-moving features involved. Collocation errors are expected tobe dominant in very active areas. Stringent time and spacecriteria must therefore be used, which obviously dramaticallyreduce the total number of coincident collocations. This, in

Fig. 4. Horizontal display of the MIRS convergence metric and the overlaidlocations of the GPS dropwindsondes. A select few sondes were used to doa point-to-point comparison between emissivity spectra as computed frommodels (DS-based wind used as input) and MIRS emissivity retrievals. Theseare highlighted in the plot above. DSs labeled 168 and 171 were used later forthe comparison in clear and rainy conditions, respectively.

turn, renders the empirical assessment statistically meaninglessat best or practically unfeasible at worst. Note that the tighttime and space collocation must be between coincident satellitemeasurements, hurricane events, and ground truth such as DSs.

One of the challenges faced during this comparison is thefact that emissivity models usually require 10-m height windand assume local thermodynamical stability, which is not nec-essarily valid under precipitating conditions. The GPS-basedwind profile sometimes goes down to 10 m but sometimes doesnot for various reasons. The approach adopted in this paper


Fig. 5. Horizontal display of the convergence metric when the multiple scattering is (right) turned ON and when it is (left) turned off. Significant improvementis observed when the multiple-scattering capability is used during the retrieval. Red color means nonconvergence, whereas blue means convergence. This does notmean that the retrieval vector is necessarily correct but at least the solution is radiometrically correct (by fitting the measurements). The 6.8-GHz channel is notpresent (and therefore not used) in the leftmost band of the swath, which explains the subtle difference with the rest of the field.

was to select the reported wind closest to 10-m height. Thedifference in height will contribute to the uncertainty errors.Another type of limitation one should be aware of is whatother studies called representativeness error which relates to thefact that DS measurements are point measurement and do notnecessarily represent what the sensor is measuring within thewhole field of view of the spaceborne sensor. Unfortunately,the number of DSs collocated with satellite measurementsis limited, and therefore, the luxury of averaging within thefootprint to mitigate representativeness errors (or around thetime of the measurement) cannot be afforded.

C. Case-by-Case Validation

Given the limitations discussed in Section VII-B, and forthe purpose of the model versus retrieval comparison, it wascritical to find the as-perfect-as-possible collocation betweenthe satellite measurements and the GPS DSs. Fig. 5 shows afield of the convergence metric (ϕ2) using two retrievals—onewith the multiple-scattering capability turned on and one withthe capability turned off. The retrieval was simultaneously doneusing all WINDSAT channels. This shows how important it isto account for precipitation effects when performing retrievalsin rainy conditions, even when frequencies involved are all low(below 37 GHz). Nonprecipitating cloud is not able, alone, tocompensate for precipitation. Notice that very heavy rain (atrain bands) still exhibits nonconvergence, which suggests thateither 1) the forward model is not accurate enough in theseregions or that 2) the assumed instrument errors are not repre-sentative (too low) or even that 3) the hydrometeor covarianceis not applicable or close enough to the particular case studiedhere. Note also that nonconvergence extends to coastal areas.This is to be expected because the emissivity covariance matrixdoes not represent at all these cases. Figs. 6 and 7 presentthe comparison between several emissivity spectra in clearand precipitating conditions, respectively. These individual DScases were selected in Fig. 4. In both cases, the retrieval reached

Fig. 6. Case of a clear sky/calm sea. Spectra of emissivity 1) computed usingFASTEM-3 emissivity model and GDAS winds as inputs (red, dotted-dashed),2) using the GPS dropwindsondes as inputs (black solid lines), and 3) retrievedusing MIRS variational algorithm and WINDSAT brightness temperaturesusing two options (green solid: retrieval with multiple scattering in the forwardoperator turned on; green dashed: when scattering off). In this plot (calm sea,clear sky), the two options gave same retrievals. 4) The dotted-dashed greenline is the background spectrum used as first guess in the MIRS. Verticaland horizontal polarization channels are shown at the top and the bottom,respectively. Along with the polarizations, the title also holds the DS number(DS#) as well as the altitude (MeasAlt), the DS/satellite time and distancedifferences (DeltaT and DeltaD), the skin temperatures (from GDAS andMIRS), and the wind closest to 10-m height.

convergence, i.e., the resulting emissivities (along with the restof the state vector) fit the measurements within noise levels.

The spectra displayed in these figures are 1) two retrievedspectra from the MIRS system using noncorrected WINDSATbrightness temperatures and in which the scattering model wasturned on or off; 2) the computed emissivity spectrum using theFASTEM-3 model and the DS wind as input (the wind closestto 10-m height); 3) the emissivity spectrum simulated usingFASTEM-3 and Global Data Assimilation System (GDAS)


Fig. 7. Comparison between GPS-based emissivities and MIRS-retrievedspectra for a single case in the middle of the hurricane (precipitating). V andH polarization channels are shown at the top and bottom panels, respectively.In this case, there is a significant impact in using the multiple scattering on(the spectrum is much different from when the scattering is turned off andabsorption only is assumed).

analysis wind as input (included for reference); and 4) thebackground emissivity spectrum is also plotted for referenceto highlight what type of spectral constraint we had during theretrieval. The background emissivity constraints over ocean arebuilt using the same FASTEM-3 model, which is part of theCRTM package we use in the retrieval algorithm.

VIII. RESULTS

In Fig. 6 (clear-sky case), the individual DS measured thewind at a height of 8 m. The MIRS-retrieved surface tem-perature (301.5 K) is roughly similar to the one from theGDAS analysis field (302 K). The distance difference betweenthe DS and the satellite measurement footprint center was4.4 km. The time difference was 1.7 h. The GDAS reportedwind was 6.7 m/s, whereas the DS reported was 8.4 m/s. Thecloseness of surface temperature and wind reporting betweenthe DS and the GDAS is to be expected as GDAS does as-similate those DS measurements. This, in turn, explains whythe corresponding simulated emissivity spectra are close. Thetwo MIRS-based retrieved spectra (using absorption only andaccounting for the full extinction effects) are undistinguishable,confirming that no heavy precipitation was taking place for thisindividual case. In other words, turning on or off the precip-itation handling does not make a difference in the retrievedemissivity. It is worth noting that the retrieved spectrum isdeparting from the background spectrum in both vertical (V)and horizontal (H), consistently, responding therefore to signalfrom the brightness temperature measurements. The retrievedspectrum adopted a slightly larger spectral slope than boththe background and the FASTEM-3-based emissivity (with theGPS-DS wind as input). For V polarization, the low frequency(between 6 and 19 GHz) emissivity values are relatively similarbetween the retrieved spectrum and the DS- and FASTEM-3-based values. The departure between the spectra reaches amaximum at 37 GHz with the retrieved emissivity being higherby as much as 0.04 than the DS and FASTEM value. For

H polarization, the slight slope difference (same direction) iscreating differences in the low and high ends of the frequencyrange, but they do not exceed 0.02. It is important to recallthat the retrieval has reached convergence in this case andis therefore producing final brightness temperatures consistentwith the measurements. The individual channels residuals werearound 2 K for the V channels and 0.8 K for the H channels,with an overall ϕ2 of 1.

Fig. 7 represents another case of comparison, correspondingto precipitating conditions previously identified in Fig. 4. It isworth noting that in this case, the retrieved spectra, with andwithout accounting for the full extinction due to precipitation,are clearly distinct, which confirms that the case is indeedaffected by rain. In this case, the reported skin temperatureby GDAS was 301.8 K and that retrieved by MIRS was303.4 K—a difference of 1.6 K. The wind reported by GDASwas 7.9 m/s, whereas the one reported by the DS was muchhigher at 12.1 m/s. It is expected that assimilation-based windis underestimated in the high end as this is a smoothed valuewithin a grid, whereas the DS measurement is a point measure-ment that has much more spatial variability. The collocationfor this case was almost perfect: less than 15 min in time andless than 6 km in distance, reducing the collocation errors toa minimum. However, the DS wind was reported at 16 m. Itis interesting to note that similar features found in the clearcase are reproduced in this case as well. In particular, theretrieved spectrum (we focus on the scattering-on one) tendsto have a higher spectral slope (than the model) in both V andH channels, but with more dramatic differences than in the caseof clear conditions. It is found that low-frequency emissivityvalues in the V case tend to be similar between the FASTEM-3simulations and the retrieved values, but tend to depart morequickly and reach a maximum difference of 0.05 at 37 GHz.In the case of the H polarization channels, the same remarkfound in the clear conditions could also be drawn, i.e., that theslight difference in the spectral slope introduces differences inthe low and high end of the [6–37 GHz] frequency range butwith differences less than 0.02. It is similarly important to notethat the retrieval had reached convergence in this case as well,with residuals (not shown) being around 3 K for the case ofthe V polarization channels and around 1.5 K in the case of theH polarization channels.

IX. DISCUSSIONS

The comparisons between FASTEM-3-based emissivityspectra and MIRS-retrieved ones, which are presented here, areconsistent with other individual cases not shown. They reveala spectral slope anomaly: the retrieved emissivity spectrumshows a higher slope than that of the model. By focusing onindividual cases, we have reduced the time and space colloca-tion errors to a strict minimum. This was done to isolate onlythe retrieval and emissivity model errors.

The first potential source of discrepancy that comes to mindis the difference between the wind speed measured by the DSand the composite wind speed the sensor views within the fieldof view. However, this difference would have generated a biasacross the spectrum and not such a distinct slope. This could


be confirmed by the emissivity spectra computed using thedifferent winds—the one from DS and the one from GDAS (seeFig. 7). This wind input difference is similar in nature to thedifference due to the reporting at a height different from theexpected 10 m. By the same token, skin temperature difference,besides being small, could not generate such a slope anomaly.An alternative explanation might be that the foam coverageand emissivity are larger in nature than those predicted bythe emissivity model. Foam is known to have much higheremissivity (close to a black body) than saline water. In theirstudy, Rose et al. [29] found that while at 10 GHz, model foamemissivity was close to their measurements, at 36.5 GHz, therewas a significant underestimation of the model. Another sourceof potential source for this difference in slope is the uncertaintyassociated with the seawater dielectric constant model. Thereis indeed a discrepancy in the literature about the dielectricconstant. At high frequencies (85 GHz), there seems to bea consensus that the work of Guillou et al. [14], which wasextended later by Ellison et al. [10] and Lamkaouchi et al. [19],based on laboratory measurements, provides better results thanthat of Klein and Swift [18]. However, at low frequencies, dis-crepancies are reported by Meissner and Wentz [23] based onSSM/I data. In their study, differences between measurementsand simulations at 19-, 22-, and 37-GHz channels showed thatdifferences between the model of Guillou et al. [14] (similarto that employed in FASTEM) and SSM/I radiances generatedifferences that increase with frequency (∼1.5 K at 19/22and ∼3 K around 37 GHz). Small differences were found at85 GHz, which were modulated by the surface temperature.The assumption of atmospheric thermodynamic stability at theair–sea interface is an additional source of uncertainty thatcould explain this slope anomaly. Giampaolo and Ruf [13]showed that air–sea temperature difference had indeed an im-pact on the wind emissivity slope and that impact significantlyvaried with frequency (they used a [18–37 GHz] frequencyrange). Even if their study was performed at nadir, it might bethat these instabilities, not accounted for in the model, are cre-ating frequency-dependent differences and, therefore, a spectralslope anomaly. There is obviously the uncertainty in the seasurface geometry and its unique link to near-surface wind speedthat might also be causing the slope discrepancy. Roughnessof the surface is generally due to gravity and capillary waves,which are sensitive to wind and their dissipation depends onviscosity. However, other parameters play a role that influencesone or both types of waves, such as fetch, swell, nonstationarywind, and sea development stage. The small-scale waves of thesurface are known to impact low frequencies, and therefore,their uncertainty might be a cause for the slope anomaly foundbetween the DS-based and MIRS-based spectra. FASTEM usesa simplified term to account for Bragg scattering followingChoudhury et al. [7].

The spatial resolution could be another source leading indi-rectly to that emissivity slope discrepancy: the lower frequen-cies having larger footprints, the relative impact of the samerainy cell could be different from that on a high-frequency chan-nel with a smaller footprint. This could result in a rain effectwith a frequency-dependent feature. This reasoning applies,however, only to isolated rain cells or isolated cloud patches.

In case of spatially uniform field distribution of rain and cloud,the relative effect of the footprint should be independent offootprint size and therefore of frequency. This is because therelative cloud/rain fraction within the footprint will be the same.

Another type of potential sources for this discrepancy liesin the variational retrieval. Errors in the cloud amount re-trieved could be compensated for by surface emissivity. Theoptical properties of the scatterers retrieved, if not properlymodeled, could also be a potential source for generating thisemissivity spectrum discrepancy. Similarly, if the water vaporcontinuum is not correct, or more exactly, inconsistent withthe measurements, its signature that is bound to be frequencydependent might end up in the retrieved emissivity spectrum.A spectral discrepancy between the measured brightness tem-peratures (cross-channel miscalibration) could also be causingthis emissivity slope anomaly. It is interesting to notice thatboth the clear and rainy cases exhibit similar features, sug-gesting that the general emissivity model may potentially beapplicable under those precipitating events provided perhapsa light tuning. Although it is not possible to conclude basedon a limited number of cases, it seems that the impact ofrain-induced roughness is minimal, judging from the relativebehavior of the emissivity model and the retrieved emissivityspectrum.

In general, within the uncertainties identified, on both theemissivity model side and the variational retrieval side, itis found that the emissivity model agrees with the retrievedspectra within 0.02 for the H polarization channels in the[6–37 GHz]. For the V polarization channels, the differencesare also within 0.02 in the more limited range of [6–19 GHz].Above 19 GHz, the difference increases almost linearly toreach a maximum of 0.04 in clear (0.05 in rainy) case. Thiswas obtained in moderate to relatively high wind conditions.

ACKNOWLEDGMENT

The authors would like to thank S. Feuer and M. L. Blackfrom the Hurricane Research Division of the National Oceanicand Atmospheric Administration (NOAA)’s Atlantic Oceano-graphic and Meteorological Laboratory for kindly providing theDS data, and Q. Liu, Y. Han, and P. Van Delst of the JCSDACRTM team for providing an early version of the radiativetransfer model CRTM. The views expressed in this publicationare those of the authors and do not necessarily represent thoseof NOAA.

REFERENCES

[1] M. A. Aziz, S. C. Reising, W. E. Asher, L. A. Rose, P. W. Gaiser, andK. A. Horgan, “Effects of air–sea interaction parameters on ocean surfacemicrowave emission at 10 and 37 GHz,” IEEE Trans. Geosci. RemoteSens., vol. 43, no. 8, pp. 1763–1774, Aug. 2005.

[2] S. A. Boukabara, L. Eymard, C. Guillou, D. Lemaire, P. Sobieski, andA. Guissard, “Development of a modified two-scale electromagneticmodel simulating both active and passive microwave measurements:Comparison to data remotely sensed over the ocean,” Radio Sci., vol. 37,no. 4, pp. 16-1–16-11, 2002. DOI: 10.1029/1999RS002240.

[3] S.-A. Boukabara, F. Weng, R. Ferraro, L. Zhao, Q. Liu, B. Yan, A. Li,W. Chen, N. Sun, H. Meng, T. Kleespies, C. Kongoli, Y. Han, P. Van Delst,J. Zhao, and C. Dean, “Introducing NOAA’s microwave integrated re-trieval system (MIRS),” in Proc. 15th ITSC, Maratea, Italy, 2006.


[4] F. L. Bliven, P. Sobieski, and C. Craeye, “Rain generated ring-waves:Measurements and modeling for remote sensing,” Int. J. Remote Sens.,vol. 18, no. 1, pp. 221–228, Jan. 1997.

[5] A. Camps, M. Vall-Ilossera, J. Miranda, and N. Duffo, “Emissivity ofthe sea surface roughened by rain: Simulation results,” in Proc. IGARSS,Sydney, Australia, 2001, vol. 5, pp. 2433–2435.

[6] A. Camps, M. Vall-llossera, R. Villarino, N. Reul, B. Chapron,I. Corbella, N. Duffo, F. Torres, J. J. Miranda, R. Sabia, A. Monerris, andR. Rodriguez, “The emissivity of foam-covered water surface at L-band:Theoretical modeling and experimental results from the FROG 2003 fieldexperiment,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 5, pp. 925–937, May 2005.

[7] B. J. Choudhury, T. J. Schmugge, R. W. Newton, and A. T. C. Chang,“Effect of surface roughness on microwave emission of soils,” J. Geophys.Res., vol. 84, pp. 5699–5706, 1979.

[8] C. Craeye, P. Sobieski, and L. F. Bliven, “A relationship between at-mospheric rain reflectivity and elevation variance due to drop impact onthe sea surface,” in Proc. Oceans, 2003, vol. 1, pp. 528–531.

[9] G. Deblonde and S. J. English, “Evaluation of the FASTEM-2 fastmicrowave oceanic surface emissivity model,” in Proc. Tech. ITSC-XI,Budapest, Hungary, Sep. 20–26, 2000, pp. 67–78.

[10] W. J. Ellison, S. J. English, K. Lamkaouchi, A. Balana, E. Obligis,G. Deblonde, T. J. Hewison, P. Bauer, G. Kelly, and L. Eymard, “Acomparison of ocean emissivity models using the Advanced MicrowaveSounding Unit, the Special Sensor Microwave Imager, the TRMM Mi-crowave Imager and airborne radiometer observations,” J. Geophys. Res.,vol. 108, no. D21, p. 4663, 2003.

[11] S. English and T. J. Hewison, “A fast generic millimeter-wave emissivitymodel,” in Proc. SPIE—Microw. Remote Sens. Atmos. Environ., 1998,vol. 3503, pp. 288–300.

[12] S. English, “Issues in ocean emissivity modelling,” in Proc. Talk1st Workshop Remote Sens. Model. Surface Properties, Paris, France,Jun. 2006. [Online]. Available: http://cimss.ssec.wisc.edu/itwg/groups/rtwg/meetings/sfcem/

[13] J. C. Giampaolo and C. S. Ruf, “The effect of atmospheric stability onmicrowave excess emissivity due to wind,” IEEE Trans. Geosci. RemoteSens., vol. 39, no. 10, pp. 2311–2314, Oct. 2001.

[14] C. Guillou, W. Ellison, L. Eymard, K. Lamkaouchi, C. Prigent, G. Delbos,G. Balana, and S. A. Boukabara, “Impact of new permittivity measure-ments on sea surface emissivity modeling in microwaves,” Radio Sci.,vol. 33, no. 3, pp. 649–667, 1998.

[15] A. Guissard, P. Sobieski, and C. Baufays, “A unified approach to bistaticscattering for active and passive remote sensing of rough ocean surfaces,”Trends Geophys. Res., vol. 1, pp. 43–68, 1992.

[16] T. F. Hock and J. L. Franklin, “The NCAR GPS dropwindsonde,” Bull.Amer. Meteorol. Soc., vol. 80, no. 3, pp. 407–420, Mar. 1999.

[17] F. Karbou, C. Prigent, L. Eymard, and J. R. Pardo, “Microwave land emis-sivity calculations using AMSU measurements,” IEEE Trans. Geosci.Remote Sens., vol. 43, no. 5, pp. 948–959, May 2005.

[18] L. A. Klein and C. T. Swift, “An improved model for the dielectricconstant of sea water at microwave frequencies,” IEEE Trans. AntennasPropag., vol. AP-25, no. 1, pp. 104–111, Jan. 1977.

[19] K. Lamkaouchi, A. Balana, G. Delbos, and W. Ellison, “Permittivity mea-surements of lossy liquids in the frequency range 20–110 GHz,” Meas.Sci. Technol., vol. 14, pp. 1–7, 2003.

[20] Q. Liu and F. Weng, “Retrieval of sea surface wind vectors from simu-lated satellite microwave polarimetric measurements,” Radio Sci., vol. 38,no. 4, pp. 8078–8085, 2003.

[21] Q. Liu and F. Weng, “Advanced doubling–adding method for radia-tive transfer in planetary atmospheres,” J. Atmos. Sci., vol. 63, no. 12,pp. 3459–3465, 2006.

[22] L. M. McMillin, L. J. Crone, and T. J. Kleespies, “Atmospheric transmit-tance of an absorbing gas—5: Improvements to the OPTRAN approach,”Appl. Opt., vol. 34, no. 36, pp. 8396–8399, Dec. 1995.

[23] T. Meissner and F. J. Wentz, “The complex dielectric constant of pure andsea water from microwave satellite observations,” IEEE Trans. Geosci.Remote Sens., vol. 42, no. 9, pp. 1836–1849, Sep. 2004.

[24] J. L. Moncet, R. G. Isaacs, and J. D. Hegarty, “Application of theunified retrieval approach to real DMSP sensor data over ocean and land,”in Proc. 2nd Top. Symp. Combined Opt.-Microw. Earth Atmos. Sens.Conf., 1995, pp. 186–188.

[25] S. Padmanabhan, S. C. Reising, W. E. Asher, L. A. Rose, and P. W. Gaiser,“Effects of foam on ocean surface microwave emission inferred fromradiometric observations of reproducible breaking waves,” IEEE Trans.Geosci. Remote Sens., vol. 44, no. 3, pp. 569–583, Mar. 2006.

[26] S. Padmanabhan, S. C. Reising, and W. E. Asher, “Azimuthal depen-dence of the microwave emission from foam generated by breaking

waves at 18.7 and 37 GHz,” in Proc. IEEE MicroRad, Feb. 28, 2006,pp. 131–136.

[27] C. Prigent and P. Abba, “Sea surface equivalent brightness temperature atmillimeter wavelength,” Ann. Geophys., vol. 8, pp. 627–634, 1990.

[28] C. D. Rodgers, “Retrieval of atmospheric temperature and compositionfrom remote measurements of thermal radiation,” Rev. Geophys. SpacePhys., vol. 14, no. 4, pp. 609–624, 1976.

[29] L. A. Rose, W. E. Asher, S. C. Reising, P. W. Gaiser, K. M. St Germain,D. J. Dowgiallo, K. A. Horgan, G. Farquharson, and E. J. Knapp, “Radio-metric measurements of the microwave emissivity of foam,” IEEE Trans.Geosci. Remote Sens., vol. 40, no. 12, pp. 2619–2625, Dec. 2002.

[30] B. Ruston and S. Boukabara, “An overview of current surface assim-ilation in NWP centers,” in Proc. Talk 1st Workshop Remote Sens.Model. Surface Properties, Paris, France, June 2006. [Online]. Available:http://cimss.ssec.wisc.edu/itwg/groups/rtwg/meetings/sfcem/

[31] A. Stogryn, “Electromagnetic scattering from rough, finitely conductingsurface,” Radio Sci. 2 (New Series), vol. 4, pp. 415–428, 1967.

[32] F. Weng, Y. Han, P. Van Delst, Q. Liu, T. Kleespies, B. Yan, andJ. Le Marshall, “JCSDA Community Radiative Transfer Model (CRTM),”in Proc. 14th TOVS Conf., Beijing, China, 2005.

[33] F. J. Wentz, “Model function for ocean microwave brightness tempera-tures,” J. Geophys. Res., vol. 88, no. C3, pp. 1892–1908, 1983.

[34] T. T. Wilheit, “A model for the microwave emissivity of the ocean’ssurface as a function of wind speed,” IEEE Trans. Geosci. Remote Sens.,vol. GRS-17, no. 4, pp. 244–249, Oct. 1979.

Sid-Ahmed Boukabara (SM’06) received the In-génieur degree in electronics and signal processingfrom the National School of Civil Aviation (ENAC),Toulouse, France, in 1994, the M.S. degree fromthe Institut National Polytechnique de Toulouse,Toulouse, France, in 1994, and the Ph.D. degree inremote sensing from Denis Diderot University, Paris,France, in 1997.

He was then involved in the calibration/validationof the European Space Agency’s ERS-2 microwaveradiometer and has worked on the synergistic use

of active and passive microwave measurements. He then joined AER Inc.,Cambridge, MA, as a Staff Scientist, working on the design, implementation,and validation of the NPOESS/CMIS physical retrieval algorithm, on theNASA SeaWinds/QuikSCAT wind vector rain flag, and on the developmentof the atmospheric absorption model MonoRTM, dedicated to microwave andlaser applications. He recently joined the National Oceanic and AtmosphericAdministration, National Environmental Satellite, Data, and InformationService/Center for Satellite Applications and Research (STAR), Camp Springs,MD, where he is leading an effort to develop the capability of assimilatingpassive microwave measurements in all-weather conditions using a combina-tion of variational technique algorithm and the Community Radiative TransferModel. His research interests include radiative transfer modeling includingabsorption and scattering of the surface and the atmosphere, spectroscopy,algorithm development using neural networks, assimilation-type techniques,and statistical approaches.

Dr. Boukabara is a member of the American Meteorological Society and theAmerican Geophysical Union.

Fuzhong Weng received the M.S. degree in radarmeteorology from the Nanjing Institute of Meteo-rology, Nanjing, China, in 1985 and the Ph.D. de-gree from Colorado State University, Fort Collins,in 1992.

He is currently with the National Oceanic andAtmospheric Administration (NOAA), National En-vironmental Satellite, Data, and Information Service/Center for Satellite Applications and Research(STAR), Camp Springs, MD. He is a leading expertin developing various NOAA operational satellite

microwave products and algorithms such as Special Sensor Microwave/Imagerand Advanced Microwave Sounding Unit cloud and precipitation algorithms,and land surface temperature and emissivity algorithms. He is developing newinnovative techniques to advance uses of satellite measurements under cloudyand precipitation areas in NWP models.

Documents

Microwave Emissivity Over Ocean in All-Weather Conditions: Validation Using WINDSAT and Airborne GPS Dropsondes