a comparison of microwave window channel retrieved and forward-modeled emissivities over the u.s....

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 52, NO. 5, MAY 2014 2395

A Comparison of Microwave Window ChannelRetrieved and Forward-Modeled Emissivities

Over the U.S. Southern Great PlainsSarah Ringerud, Christian Kummerow, Christa Peters-Lidard, Yudong Tian, and Kenneth Harrison

Abstract—An accurate understanding of land surface emissivityin terms of associated surface properties is necessary for improvedpassive microwave remote sensing of the atmosphere, includingwater vapor, clouds, and precipitation, over land. In an effortto advance this understanding, emissivities are calculated for a5◦ latitude by 5◦ longitude region in the U.S. Southern GreatPlains using a combination of land surface model and physicalemissivity model. Results are compared to retrieved values fromthe Advanced Microwave Scanning Radiometer—Earth Observ-ing System passive microwave observations for cloud-free scenesover a six-year period. The resulting emissivities are comparedin the context of surface properties including surface tempera-ture, leaf area index (LAI), soil moisture, and precipitation. Thecomparison confirms that lower frequency channels respond mostdirectly to the surface soil and its dielectric properties. Differencesbetween retrieved and modeled emissivities are generally lowerthan 2%–3% and appear to be a function of soil moisture andLAI at frequencies less than 37 GHz. Agreement is better for thevertical polarization channels. At 89 GHz, a large difference ispresent between retrieved and modeled emissivities in both meanand magnitude of variability, particularly in the summer months.Problems are likely present at higher microwave frequencies inboth the retrieved and modeled products, including the inabilityof the emissivity model to represent liquid water in the form ofdew or precipitation interception on the vegetation canopy.

Index Terms—Emissivity, land surface, passive microwave re-mote sensing, precipitation.

I. INTRODUCTION

THE measurement and mapping of land surface propertiesare a highly valuable product of satellite remote sensing

science. Microwave radiation emitted by the land surface andmeasured using multifrequency dual-polarization satellites con-tains information about the character of the surface itself and itsdynamic properties. In the passive microwave regime over lightto moderately vegetated land surfaces, emission is sensitiveto soil type and soil moisture at lower frequencies, as well

Manuscript received October 29, 2012; revised March 15, 2013; acceptedApril 22, 2013. Date of publication June 27, 2013; date of current versionMarch 3, 2014. This work was supported by NASA’s Precipitation Mea-surement Missions (PMM) Program, NASA Solicitation NNH09ZDA001N,PI: Peters-Lidard.

S. Ringerud and C. Kummerow are with Colorado State University, FortCollins, CO 80523 USA (e-mail: [email protected]; [email protected]).

C. Peters-Lidard, Y. Tian, and K. Harrison are with the NASA GoddardSpace Flight Center, Greenbelt, MD 20771 USA (e-mail: [email protected]; [email protected]; [email protected]).

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.2013.2260759

as properties of the vegetation cover. In addition to real-timemonitoring of flood extent and land use changes, observationsof surface water and vegetation can be applied to weather fore-casting and surface energy budget calculations. A large body ofwork exists in this area (e.g., [1]–[5]). McCabe et al. [6] com-bined passive microwave retrieval techniques with land surfacemodeling by performing soil moisture retrievals using a methodcombining satellite-observed 10.7-GHz horizontally polarizedbrightness temperatures from the Advanced Microwave Scan-ning Radiometer—Earth Observing System (AMSR-E) withsurface data from the North American Land Data AssimilationSystem (NLDAS). This data set was then coupled with aland surface microwave emission model (LSMEM), findingagreement with field campaign measurements of around 3%.A similar investigation by Gao et al. [7] retrieved soil moistureusing brightness temperatures (Tbs) from an airborne L-bandradiometer combined with land surface model (LSM) data andLSMEM, also showing good agreement with field campaignmeasurements. The Gao et al. and McCabe et al. studies bothindicate that the use of LSM and parameter data combinedwith radiative transfer calculations has promising capabilitiesfor passive microwave retrieval of surface soil moisture. Otherworks in this area have used microwave channels and com-parison to in situ data to retrieve information about vegeta-tion, including water content, canopy height, density, structure,and phenology [8], [9]. Liquid water in the form of dew orinterception of precipitation by a vegetation canopy can alsohave observable effects on emission in the passive microwaveregime. Jackson and Moy [10] review previous work in this areaand conclude that water interception and dew have a measurableeffect on microwave Tbs higher than 5 GHz, masking soilemission. Lin and Minnis [11] use ground observations ofdewpoint and skin-air temperature differences combined withSpecial Sensor Microwave Imager (SSM/I) emissivities to sug-gest that dew effects decrease early morning emissivities in OKby roughly 5%, decoupled from the soil moisture variability.The same effect is noted by Moncet et al. [12] in the U.S. cornbelt region during the summer months. Retrieval of any suchexamples of surface information requires knowledge of surfaceemission and emissivity at the frequency and polarization ofinterest.

Accurate global knowledge of land surface characteristicsand associated emissivity offers the potential for future im-provement of physical retrieval of atmospheric quantities, suchas water vapor, clouds, and rainfall, by providing an accurate

0196-2892 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2396 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 52, NO. 5, MAY 2014

background surface over which to calculate radiative transferthrough the atmosphere. A summary of many currently em-ployed techniques utilizing satellite observations and modelingtechniques is presented in Ferraro et al. [13]. Aires et al. [14],for example, use a neural network approach, along with atraining database of simulated data, to retrieve surface tempera-ture, integrated water vapor content, cloud liquid water path,and microwave land surface emissivities in the 19–85-GHzrange from SSM/I Tbs over land. A 2004 paper by Skofronick-Jackson et al. [15] utilized the higher frequency channels of theAMSU-B radiometer to retrieve falling snow over land surfacesusing a physical model and database of previously reportedemissivities for a given snow cover amount. Bauer et al. [16]used climatological emissivity values along with a variationalretrieval scheme to compute rain, snow, and cloud water profilesand to assess retrieval errors over land surfaces in prepara-tion for future focus on high latitude and weak precipitationretrievals, finding that the sounding channels provided a highenough signal-to-noise ratio to be useful for global retrievals.The Microwave Integrated Retrieval System retrieval and dataassimilation system [17] simultaneously retrieves atmosphereand surface states in a 1DVAR approach starting with a firstguess surface emissivity from mean retrieved clear sky values.While this approach yields an estimate of surface emissivity, itdoes so using an emissivity climatology including mean stateand covariance matrices rather than a physical model directlycomputing emissivity as a function of the surface properties.It is desirable then, as a next step in this area, to determinethe feasibility of a coupling between physical models of theatmosphere and a similarly physical model of the surface thatwould supply dynamic surface information.

The retrieval methods cited previously are referred to as“physical” due to their use of geophysical parameter valuescoupled with a radiative transfer model for computation ofTbs. Physical retrievals of precipitation such as the GoddardProfiling Algorithm or GPROF [18] have had great successover the radiometrically cold (emissivities ∼0.5) ocean surfaceand are routinely assimilated by atmospheric models [16]. Withthe upcoming National Aeronautics and Space Administration(NASA) Global Precipitation Measurement Mission (GPM), afocus has been placed on extending the principles of physicalretrieval to land surfaces. This problem is a difficult one as theland surface is highly variable and has emissivity closer to 1in the microwave, making it much more problematic to discernthe signal coming from atmospheric hydrometeors fromthat of the surface. This coupling requires accurate dynamicemissivities, and the GPM passive microwave algorithm teamhas estimated that accuracy of 1% or better is required forthe coupling of emissivity to the physical atmosphere retrievaland separation of dynamic surface emission from emissionby hydrometeors in the atmosphere. Emissivity error istroublesome to define as emissivity is not a directly measurablequantity. If possible, however, such a retrieval could be appliedto the full constellation of radiometers involved in the GPMmission and could lead to greater understanding of globalrainfall and its variability [19].

Spaceborne radiometers measuring passive radiation emittedby the Earth’s surface offer a unique platform for measuring

emissivity. Satellite-derived microwave brightness temperatureobservations include information content about the surfaceemission in cases where the signal has not been completelymasked by absorption in the atmosphere. In the microwaveregion, the Rayleigh–Jeans approximation can be applied, andthe Planck radiance is considered linearly proportional to tem-perature. The observed upwelling Tb at a given polarizationand frequency contains contributions from both the surface andthe atmosphere (downwelling and upwelling) over a specularsurface and can be written as

Tb = εTsfce−τ(0,z∗)/μ + (1− ε)

0∫

z∗

Tatm(z)e−τ(z,0)dτ/μ

+

z∗∫

0

Tatm(z)e−τ(z,z∗)dτ/μ (1)

where ε is the surface emissivity, τ is the optical depth of anatmospheric layer, z∗ is the top of the atmosphere, and μ isthe cosine of the incidence angle. A vegetated land surface isnot a smooth specular reflector nor is it purely Lambertian. Inreality, the surface scattering is some combination of specularand Lambertian components. It has been suggested that this canbe represented by a linear combination of the specular and Lam-bertian contributions for the most accurate representation ofsurface scattering [20]. Over the Southern Great Plains (SGP)area, where the surface type is dominated by agriculture andmight be considered rough and Lambertian during the growingseason and less so during periods of bare soil or snow cover,calculations suggest a difference in incoming Tb on the orderof 0.1 K for AMSR-E-specific calculations. Matzler [20] andPrigent et al. [21] also conclude that, at the 53◦ incidence angle,the specular assumption has little impact, and it is thereforeapplied here to simplify the radiative transfer. Equation (1) canbe solved for the retrieval of emissivity if the atmosphere opticaldepth and surface temperature are known.

Surface emissivity has also been modeled. Physical modelingof surface emissivity using the surface parameters is a complexproblem. Such a model must include radiative transfer fromsubsurface soil (which requires knowing soil type and soilmoisture) to air and then scattering and absorption-emissionthrough the vegetation canopy that varies based upon frequency,vegetation type, and density [leaf area index (LAI)] as well asthe water content of the plants themselves and any interceptedwater or dew. Weng et al. [22] introduce a method for physicalmodeling of emissivity by dividing the surface into three layers(i.e., soil, vegetation, and air) and by performing radiativetransfer through these three media and their boundaries. Thismodel will be described more fully in the following section.The authors perform global comparisons between model resultsand retrieved emissivities, but note that due to the sparsity ofglobally available input data, many parameters must default toassumed average values. While conceptually the emission andscattering of radiation by soil and vegetation are fairly straight-forward, it is neither computationally feasible nor possible interms of availability of required input data to do full radiativetransfer through individual leaves and soil particles over anysignificant area. This makes it difficult to implement surface

RINGERUD et al.: MICROWAVE WINDOW CHANNEL RETRIEVED AND FORWARD-MODELED EMISSIVITIES 2397

emissivity models on a global scale. Comparing retrieved andmodel-derived surface properties is also challenging. This isprimarily due to the problem of defining the surface skintemperature as “seen” by the radiometer. The radiometric tem-perature depends upon the vegetation and associated evapo-ration. Through the latent heat exchange that occurs duringevapotranspiration, the vegetation canopy actively controls itstemperature in contrast to bare soil, as suggested by the smalleramplitude diurnal variability observed over densely vegetatedareas [23]. Skin temperature may also depend upon soil prop-erties, and emission depth varies with frequency [24]. It shouldbe noted as well that contribution of the different componentsof the surface to emissivity is highly frequency dependent. Athigher microwave frequencies (X- and Ka-band), for example,typical satellite viewing angles do not see contribution from soilbeneath a vegetation canopy [25].

LSMs, such as the National Centers for Environmental Pre-diction’s (NCEP) Noah model, contain in their output high-resolution information about surface state, including profiles ofsoil water and temperature, along with vegetation information[26]. The coupling of LSM output to a microwave emissionmodel presents many challenges. The observed microwave sig-nal from the soil depends upon the dielectric profile of the localsoil and is highly frequency dependent [10], [23]. At passive mi-crowave frequencies, the penetration layer is relatively shallowand may not correspond to standard LSM output layering. Inthis paper, we present the results of a multiyear intercomparisonof satellite-retrieved microwave emissivities with emissivitiessimulated from a combined land surface-physical emissivitymodel for a region of the U.S. SGP. It is hoped that a long-term in-depth comparison at multiple frequencies will add toearlier work such as the study by Weng et al. [22] and workin the C-band by the soil moisture community [27]. A centralgoal will be the assessment of how well LSM output can beused as input to a physical emissivity model in the productionof reasonable dynamic land surface emissivities. Model resultswill be compared to retrieved values and will be assessedwith respect to land surface parameters. The following sectionsdescribe data and methods used in the satellite retrieval andLSM + physical emissivity model calculations, followed by aquantitative and qualitative comparison of results from each.

II. CLEAR SKY EMISSIVITY RETRIEVAL

Passive microwave window channel measurements are sen-sitive to soil moisture because of the effects of water on thedielectric properties of the soil. The same measurements aresensitive to vegetation because of the water content of thevegetation (its dielectric properties) as well as scattering andabsorption of radiation by the vegetation itself.

For comparison to the forward model computed emissivities,a clear air emissivity retrieval is developed for the AMSR-Epassive microwave window channels. AMSR-E is a conicallyscanning 12-channel radiometer aboard the NASA EOS Aqua,a polar orbiting sun-synchronous satellite. The retrieval isperformed using intercalibrated level 1 C AMSR-E brightnesstemperatures. The level 1 C standardized format and calibrationwas developed as an initial prototype framework for the GPM

Fig. 1. Comparison of TPW values (in millimeters) over one year (2005, 1418observations) of data from the ERA-Interim reanalysis and ARM SGP soundingobservations.

radiometer constellation and is described at http://mrain.atmos.colostate.edu/LEVEL1C/level1C_overview.html. Pixels are de-termined to be cloud-free using collocated Aqua-Moderate Res-olution Imaging Spectroradiometer (MODIS) 1-km cloud maskinformation [28], for which validation at the SGP site indicates85% agreement with ground-based lidar cloud detection, withmost mischaracterized scenes corresponding to cases of thincloud with low optical depth [29]. Cloud clearing is done in thestrictest sense, designating as clear only those pixels for whichthe MODIS algorithm has determined “Confident Clear” overthe full extent of the largest AMSR-E footprint size. Cloud-freepixels are then combined with coincident ancillary atmosphericinformation, including the surface skin temperature and watervapor, from the European Centre for Medium-Range WeatherForecasts (ECMWF) interim reanalysis [30]. ERA-Interim is areanalysis product, with surface parameters available 3-hourly.Values were interpolated from 1◦ gridded output. Many pos-sible sources of input data exist, particularly for the surfacetemperature. This includes available satellite-retrieved surfacetemperature estimates as well as higher resolution model datasuch as that available from an LSM. ECMWF was chosenhere as an independent data source and one that is routinelyused as ancillary data input for satellite retrievals such asGPROF. ECMWF data have been employed in other emissivitystudies as well, including Holmes et al. [31], where it wasused in conjunction with the Community Microwave EmissionModeling Platform, a forward land surface emissivity modelapplicable to the frequency range 1–20 GHz. It is assumed herethat cloud water content is zero for the clear sky retrievals. Totalprecipitable water (TPW) for the pixel is also taken from theECMWF. As a check of the ECMWF TPW values in the SGPregion, a comparison to TPW calculated using balloon soundingdata from the ARM SGP central facility is shown in Fig. 1 for2005 (1418 observations). There is clearly scatter in the com-parison, but the agreement is generally good, with a mean biasof 1.26 mm, rmse of 8.27, and correlation coefficient of 0.81.

Radiative transfer calculations are performed through theatmospheric column for the pixel, starting with an initial emis-sivity guess and assuming a plane parallel atmosphere with noscattering (i.e., no hydrometeors are present in the column).The resulting simulated brightness temperature is compared


Fig. 2. Map of the continental U.S. The SGP box used for the current study is outlined over KS and OK. Colors indicate International Geosphere-BiosphereProgramme land surface types for 2009.

to the observed Tbs. Emissivity is then adjusted based onthe resulting Tb difference in an iterative process followingthe method of Bytheway and Kummerow [32]. This retrievalscheme is somewhat limited in that it must be assumed thatthe reanalysis surface temperature is equal to the wavelength-dependent surface skin temperature observed by the satellite,which depends on the surface type and may be mitigated byvegetation as discussed in the previous section. Monthly meanvalues have been calculated for the 5◦ SGP box (34:39 N,−100:−95 W) outlined in Fig. 2 over 2004–2011. Resultsare shown in Fig. 3 for the months of January and Juneat 0.5◦ resolution in order to compare with the climatologyavailable from the Tool to Estimate Land-Surface Emissivitiesat Microwave frequencies (TELSEM), a nearly ten-year SSM/I-based climatology derived from 1993–2001 and described inAires et al. [33]. Simple averaging has been applied to theretrieved emissivities within each 0.5◦ box for comparison tothe TELSEM values. Lower emissivity values relative to theTELSEM climatology are present over the full domain in allfrequencies, likely due to differing input surface temperaturedata sets. Spatial variability over the box is similar. Agreementin the monthly mean half-degree emissivity values is within0.01–0.02 in January at 89 GHz over much of the domain. Strik-ing low values occur in the AMSR-E-retrieved data sets on thenorthern edge of the domain that are not observed in the longerTELSEM climatology, likely the result of snow on the groundat the time of AMSR-E overpasses included in the retrievalmean. Emissivity values at 10 and 18.7 GHz show a relativeminimum in south-central KS, an area of mixed cropland andsome pasture/range land (ARM site land use/land cover webpage www.arm.gov). The southeast corner of the grid showsa relative maximum in emissivity in the 10- and 18.7-GHzchannels. The seasonal contrast is most evident in the 89-GHzchannel, where emissivities are significantly lower in June. Thewinters are clearly dryer and less vegetated as evidenced by thehigher emissivity values in January observed in both data setsand at all frequencies throughout the domain.

III. LSM

The land surface modeling portion of this project was im-plemented using NASA Goddard’s Land Information System,a modular platform for running LSMs with various input asdescribed in Kumar et al. [34]. The land model chosen forthis paper is NCEP’s Community Noah Land Surface Model,version 3.2. Noah is a 1-D uncoupled model that simulatessoil moisture and temperature profiles, skin temperature, snowpack depth and water equivalent, and canopy water content bysolving the 1-D surface energy and water balances [26]. For thepurposes of this study, the model was not run over urban areasor bodies of water. Input forcing data come from the NLDASproject phase 2 (NLDAS-2). NLDAS-2 has a grid spacing of0.125◦, nonprecipitation fields come from the analysis fields ofthe NCEP North American Regional Reanalysis (NARR), andthe precipitation fields come from the gauge-only CPC analysisof daily precipitation temporally disaggregated using Stage IIDoppler radar data [35]. MODIS LAI, an eight-day productderived from visible channel spectral reflectance [36], is alsoan input to the LSM. Validation of the MODIS LAI product hasbeen performed by multiple international teams, with resultssuggesting reasonable agreement when compared to field cam-paign measurements with an average overestimation of around10% [37]. The model is run at 1-km resolution. NLDAS-2 andother coarser resolution forcing data were spatially interpolatedto the target resolution with the budget interpolation scheme,based on the spatial interpolation package “ipolates” fromNCEP. A spin-up period of 17 months was included in themodel run.

To get a sense of the model output’s agreement with ob-servations, three years of LSM output soil moisture and soiltemperature, along with the NARR/CPC precipitation field, arecompared to in situ measurements from the USDA AgriculturalResearch Service’s Micronetwork [38]. The observations comefrom a probe measurement at 5 cm depth, while the modeldata points are associated with a 10-cm top model layer, withthe value corresponding to the midpoint at 5 cm. It should


Fig. 3. Monthly mean AMSR-E-retrieved emissivity (left; 2004–2011) compared with TELSEM (right) for the months of January and June over a 5◦ box overSGP for the 10 H, 10 V, 18.7 H, 18.7 V, 89 H, and 89 V channels on AMSR-E. The OK/KS border is visible near 36.5 N.

be noted here that point measurements are being comparedto 1-km resolution model output. Time series comparisons ata single Fort Cobb watershed station (ARS station F102) areshown in Fig. 7 for 2006–2008 in black, with the LSM outputover plotted in cyan. The 5-min observational data reported bythe ARS Micronet station have been averaged into 3-h intervalsconsistent with the LSM output, and missing data have beenremoved. Agreement in location of relative maxima and minima

in the time series is quite good. The model soil temperaturetime series shows a somewhat smaller dynamic range thanthe observations in two of the three years, particularly in thesummer. The soil moisture comparison also shows less dynamicvariability on a day-to-day basis in the model, particularly onthe low end. The agreement is impressive, given the coarsenessof the model information, with similar values and patterns ofvariability. The mean model biases in soil temperature and soil


Fig. 4. 2006–2008 time series of 5-cm soil temperature (top panel) and soilmoisture (bottom panel) for micronetwork station F102 in southwestern OK.Observations are plotted in black, with Noah model results over plotted in cyan.

moisture relative to the ARS Micronet observations for thethree-year period are −1.19 K and 0.04, respectively, with rmseof 6.39 and 0.06. Anomaly correlations are high at 0.83 forthe soil temperature anomalies and 0.79 for the soil moisture.Based upon model sensitivities that will be described in moredetail in the following section (see Fig. 5 and accompanyingtext), this translates to emissivity differences from the soiltemperature bias ranging from almost zero at 6.925 GHz toabout 0.005 at 89 GHz. From the soil moisture, the emissivityvariation associated with the bias is less than 0.01. Some ofthis disagreement is likely connected to the location of the soilmoisture value in each data set. Also, as stated earlier, this isa comparison of a 1-km model grid cell to a point measure-ment. Multiscale comparison of soil moisture is a complex anddifficult problem [6], and the agreement in general trends andpatterns here is encouraging.

IV. PHYSICAL EMISSIVITY MODEL

In order to compare LSM parameters to observed emissivityvalues, a radiative transfer code is required. Weng et al. [22]describe a three-layer model for computing land surface ε for

TABLE IMEAN, MAXIMUM, AND MINIMUM SURFACE PARAMETER VALUES FROM

ONE YEAR OF NOAH MODEL RUN USING NLDAS-2 FORCING FOR A

SINGLE 1-km MODEL PIXEL IN OK, USED FOR SENSITIVITY

CALCULATIONS IN FIG. 4

microwave frequencies in the range 4.9–94 GHz. The model,LandEM, utilizes input surface properties for computation ofthe dielectric properties of the soil and vegetation, computingseparate dielectric constants for each of the three layers, e.g.,soil, vegetation, and the air above. The model then computesemissivity using a two-stream radiative transfer solution thatincludes volumetric scattering and reflection and transmissionat the layer interfaces. Geometric optics calculations are utilizedfor canopy leaves. Inputs to the model are viewing angle,frequency, soil moisture content, vegetation fraction, soil tem-perature, clay fraction, sand fraction, and snow depth. Defaultparameterized values are used for leaf thickness and watercontent per unit of LAI. The specific version used for this paperwas included with the Community Radiative Transfer Model(CRTM) version 2.0.2. Initial validation of the model showedgenerally good agreement with observations for bare soil andgrass but some disagreement over snow-covered surfaces, par-ticularly at higher frequencies [22]. A simple sensitivity anal-ysis was used to demonstrate the relationship between modelemissivity and the input surface properties at SGP. For theanalysis, maximum and minimum values over a one-year periodat an area surface station are used (with the exception of soiltype, which varies sand percentage from 0%–100%), along withthe AMSR-E frequencies and viewing angle. Each parameter isthen varied between maximum and minimum values for com-putation of emissivity while keeping the others constant at theirmean value. Minimum, maximum, and mean values of eachparameter are shown in Table I, and the resulting emissivityspectra are shown in Fig. 5. Polarization difference decreaseswith frequency in all cases for the typical SGP values. Increas-ing the temperature of the first layer of soil under the vegetationproduces a slight decrease in emissivity in all channels, with aslope that increases with frequency. Soil moisture shows theexpected strong connection to emissivity as increased moistureleads to higher dielectric constant and decreased emissivity.This effect decreases as frequency increases. The variation ofsoil type demonstrates that increasing sand content tends toincrease the dielectric constant at the same soil moisture, asmore water is available in the soil, leading to the observeddecrease in emissivity. When clay content (in this case 1—sandcontent) increases, more bound water is in the soil system,leading to a lower dielectric constant [39]. As the leaf area perground area, LAI, is increased in the model, the polarizationdifference decreases to almost zero at the highest values of LAI.The LAI effect is the most nonlinear of the sensitivity results,


Fig. 5. Model emissivity values at the AMSR-E frequencies as a function of the input parameters. Both polarizations are shown for each frequency, with V inthe solid lines and H in the dashed lines.

showing a steepening of the emissivity change at a value of 2–3.Following the sharp increase, there is a leveling off of emissiv-ity at all frequencies, with a decrease apparent at the highestLAI values. The degree of decrease increases with frequency.The representation of microwave emission from snow cover bythe LandEM algorithm is based on dense medium theory andis described in Weng et al. [22]. As very little snow cover isobserved in the SGP area and as passive microwave sensorshistorically have difficulty with quantitative handling of snowon the ground due to the variability of emission based on snowdepth, age, grain size, and other unknowns [21], the modelingand observation of snow cover will not be discussed here.

For the purposes of this study, Noah LSM data as describedin Section III are utilized as input to the Weng et al. emissivitymodel. Emission depth is not considered as a variable in this im-plementation, and values of soil moisture and soil temperaturefrom the topmost model layer (10 cm thickness with reportedvalue at 5 cm) are used as input to the emissivity model.

V. ANALYSIS

Monthly mean emissivity values computed by the Noah +LandEM coupling are shown averaged for 0.5◦ boxes over the

SGP study area in Fig. 6. For comparison with the satellitemethods, footprint averages have been used, and only footprintscorresponding to a clear-sky AMSR-E overpass are included.The identical time period 2004–2011 is used. A bias is evidentwith respect to the satellite retrieval methods plotted in Fig. 3,but there is some consistency in the spatial patterns.

In order to compare modeled and retrieved land surfaceemissivities in the context of associated surface properties, bothretrieved and modeled emissivities are calculated for all monthsand seasons of 2004–2011 over the full 5◦ × 5◦ box shownin Fig. 2. Results of emissivity calculations are first comparedfor the full box in bins defined by MODIS LAI and LSM soilmoisture values nearest the surface. The computed differencebetween retrieved and modeled instantaneous emissivity valuesin each of the soil moisture-LAI bins is plotted in Fig. 7 foreach of the 12 AMSR-E channels over all cloud-free data pointsin the full 2004–2011 data set. Comparisons are computed forthe AMSR-E footprint size at each frequency and include allseasons. Colors indicate the percent difference in instantaneousemissivity values (modeled minus retrieved) as a percent of themean retrieved emissivity in each channel over the full period.White space indicates agreement within 1%. Contours show thenumber of data points falling into each bin, with the plotted


Fig. 6. Monthly mean Noah + LandEM computed emissivities for the monthsof January and June 2004–2011 for a 5◦ box over SGP for the 10 H, 10 V, 18.7 H,18.7 V, 89 H, and 89 V channels at 0.5◦ resolution. Only MODIS cloud-free timesare included for comparison to Fig. 1. The OK/KS border is visible near 36.5 N.

lines indicating bins with 2500, 5000, 10 000, and 20 000 ofthe 3.3 million total matched cloud-free comparison points. Ineach channel, agreement is better in the vertical polarization. At

Fig. 7. (Model-retrieved) emissivity differences as percent of the mean re-trieved emissivity in each channel over the 5◦ SGP region for all seasonsof 2004–2011, as a function of Noah soil moisture and MODIS LAI. Colorsindicate the emissivity differences (in percent), and contours show the totalpoints falling into each bin (3.2 million total).

the lowest frequency, the 6.9-GHz channel often used for soilmoisture remote sensing, the model values are almost uniformlylower than the retrieved emissivities, with differences below 5%.


TABLE IIMEAN SURFACE PARAMETER STATISTICS FOR 2004–2011 TIME SERIES (23 376 DATA POINTS) AT TWO DOE-ARM SURFACE STATIONS

In the 10-, 18.7-, and 23-GHz channels, agreement is quite good,with differences around 1%–2%. The plots for these channelsshow a dipole-type distribution, with the model emissivitiesbeing lower than retrieved values for LAI less than about 2.0and becoming higher than retrieved at larger LAIs. Agreementimproves with increasing LAI in these lower frequencies wheresoil moisture sensitivity is high, particularly in the verticalpolarization, which is slightly less sensitive to soil moisture(Fig. 4). The contours indicate that the higher LAI values areseldom observed, however, and may therefore be less reliablecomparisons. At the higher frequencies of 37 and 89 GHz,the model emissivities are nearly uniformly higher than theretrieved values, although it is notable that, in the LAI-soilmoisture bins most highly populated, differences are generallyless than or equal to 1%. Disagreement increases with LAI andshows little dependence on soil moisture, as would be expectedat these frequencies and their much shallower emission depth.

The analysis of modeled emissivity sensitivity shown inFig. 4, along with the emissivity differences shown in Fig. 7,demonstrates the large influences of LAI and soil moistureon emissivity variability in the SGP area. To explore thesedynamics further, two individual points in the full 5◦ × 5◦

model domain are chosen for closer analysis in the contextof surface parameters as well as temporal variability. Both arepart of the ARM SGP extended facility observation networkcovering most of OK and southern KS. The first is Pawhuska,an area in northeast OK dominated by natural grasses. Thesecond is Halstead, in central KS, and represents an agriculturalregion with rotating crops including winter wheat. Comparisonsof instantaneous emissivities are performed at the resolution ofthe AMSR-E footprint. Time series of the associated surfaceparameters are included, and the methods used to produceemissivities are as described in the previous sections. Meanvalues of MODIS LAI, as well as model output soil moistureand temperature for the two stations, are given in Table II,and the mean emissivities are given in Table III. It is notablethat, while both data points share the same mean temperature,Pawhuska has a much higher LAI, indicating denser vegetationper area, and is also moister, making the sites an illustrativechoice for comparing resulting emissivity dynamics. Biases aregenerally consistent with the mean LAI and soil moisture bindifferences in Fig. 7 and are generally less than 1% for the lowerfrequency channels at Halstead and less than 2% for Pawhuska.The difference in vegetation at the two sites is evident in thethree-year time series of MODIS-derived LAI shown for eachstation in Fig. 8 (top panels). LAI variations clearly appearseasonally and vary from year to year. The seasonal cycle isclearly evident in both the surface skin temperature (secondrow) and the 5-cm soil temperature (third row) at both stations,

TABLE IIIEMISSIVITY STATISTICS COMPARING INSTANTANEOUS

FORWARD-MODELED AND RETRIEVED EMISSIVITY VALUES

FOR 2004–2011 AT TWO DOE-ARM SURFACE STATIONS

with a decreased diurnal variability in the soil as comparedto the surface. The Pawhuska soil temperature shows lessvariability than at Halstead, indicating possible temperaturemitigation under the much higher values of LAI. Soil moistureseries are shown in the fourth row, and a comparison with thelast row plotting rainfall indicates soil moisture’s close connec-tion to precipitation. The time series demonstrate the dynamicvariability of the surface parameters influencing emissivity onmultiple timescales.

As check of mean retrieved emissivity at the two locations,a monthly average emissivity is calculated for a 0.5◦ box over


Fig. 8. January 2005 to December 2007 time series of LAI (top), surface temperature, 5-cm soil temperature, 5-cm volumetric soil moisture, and rainfall (bottom)for LSM data at Halstead (left column) and Pawhuska (right column) stations.

the Halstead and Pawhuska surface stations for 2004–2011 andcompared to the TELSEM data set. Both of these are then com-pared to monthly averages computed from the Noah LSM +LandEM model output for the same 0.5◦ box. Averages arecomputed for each box at the AMSR-E footprint size, and onlymodel values corresponding to a cloud-free retrieval data pointare included. The monthly averages for the 10.65-, 18.7-, and89-GHz (H and V polarization) AMSR-E channels are shownin Fig. 9. Agreement between the two satellite methods is quitegood, with the exception of a fairly constant bias in the 10 Hand 10 V channels. This is not surprising considering that theTELSEM climatology is derived from SSM/I measurementsand must effectively extrapolate to the 10.65-GHz channel.Monthly variability is very similar for the two retrieved serieseven in these channels, while the time series of model values isrelatively flat, particularly for the Halstead footprint. Retrievedemissivities are higher than the modeled values in the monthlymean at Pawhuska station for both polarizations at 10.65 and18.7 GHz. The two agree reasonably well at Halstead inabsolute value, but the model again appears to lack dynamicchanges over the seasonal cycle. The closer agreement at

Halstead may suggest that the water–LAI relationships in theLandEM model are more applicable to the agricultural regionaround that area than to the grazed natural grassland aroundPawhuska. In the 89-GHz channel, modeled emissivities showsubstantially less variability as well as a high bias compared toboth retrieval averages. While it is true that passive microwaveemissivity retrieval may be contaminated at this higher fre-quency due to atmospheric effects including contamination byundetected low-optical-thickness clouds, the fact that the modelvalues show a consistent bias compared to two completelydifferent retrieval schemes using two different cloud-clearingmethods (International Satellite Cloud Climatology Project [40]for TELSEM versus MODIS here) suggests that this is not thesource of disagreement in this case. This channel is sensitive towater vapor amount and vertical distribution, which may be amore likely source of error. Surface temperature discrepanciescan be ruled out here as such differences would be expectedto lead to a consistent high or low bias over all channels,whereas these results show a flip in the bias with increasingfrequency. The highly dynamic nature of the variability in theretrieved values here is suggestive of something not represented


Fig. 9. Monthly mean emissivity values for a 0.5◦ × 0.5◦ box around Halstead (left column) and Pawhuska (right column) stations. Averages computed fromeight years of model data (Noah + LandEM) (solid lines) and eight years AMSR-E retrievals (dashed lines). The TELSEM ten-year climatology is plotted in thedash-dot lines for comparison.


Fig. 10. Emissivity spectra at Halstead (top) and Pawhuska (bottom) for horizontal (right) and vertical (left) polarizations. Spectra shown are computed fromTELSEM (blue), AMSR-E retrievals (red), and LSM + LandEM modeling (black).

in the model, e.g., dew or postprecipitation canopy water, whichbehaves in a radiometric sense completely different from watercontent in the vegetation itself. While it is not possible inthis analysis to state definitively the source of disagreementbetween the methods at 89 GHz, incorrect representations ofthe water vapor profile in the retrieval and radiometric effectsnot captured by the physical model both likely play a role.Fig. 10 shows the emissivity spectral averages for TELSEM,the AMSR-E retrieval, as well as the Noah–LandEM compu-tations. Averages of the retrieved and Noah–LandEM modelemissivities are computed point by point at the location of eachstation at the resolution of the AMSR-E footprint size and onlyfor cloud-free data points over the full 2004–2011 data set.Agreement between the three methods is closer at Halsteadstation in the lower frequencies, and deviation increases forhigher frequencies between the model technique and the twosatellite-derived methods. At Pawhuska station, there is morespread at the lower frequencies and somewhat better agreementwith the model at 89 GHz. The satellite emissivity methodsshow close agreement, with the interesting exception of lowerfrequencies at Pawhuska, where the AMSR-E retrieval seems totrack more with the model emissivities rather than the TELSEMSSM/I climatology.

The associated instantaneous emissivity anomaly compar-isons are shown in Figs. 11 and 12, with mean bias, rmse,and anomaly correlation statistics included in Table III for the2004–2011 period. Anomalies are from the monthly meanscomputed for the entire 2004–2011 period and are calculatedfor instantaneous retrieved and modeled emissivities at theAMSR-E footprint size. For the Halstead location (Fig. 11),anomalies are clearly more dynamic in the retrieval at all fre-quencies, illustrated by the greater spread along the x-direction.Anomaly correlations are around 0.5, with the exception of the

89-GHz channel, which shows less correlation. Large negativeanomalies visible in Fig. 11 are associated with precipitationevents (occurring during a clear-sky overpass following a pre-cipitation event) and are correlated in sign but stronger inmagnitude for the retrieved emissivities versus the modeledvalues. Given that Halstead is a more agricultural surface type,it is possible that irrigation is adding to the soil moisture andincreasing the negative anomalies in the retrieval, while thepositive anomalies are more correlated in the lower frequencychannels. At 89 GHz, emissivities are much more dynamic inthe retrieved data than the modeled values, which are relativelyflat over the eight-year analysis period. Pawhuska (Fig. 12)shows less variability in the lower frequency channels. Morepositive anomalies occur in the retrieval than in the modelemissivities. Anomaly correlations are smaller here as well, par-ticularly in the H polarization, and are close to zero at 89 GHz.This analysis suggests that surface emissivity dynamics, ascaptured by anomalies in the time series, are represented moreaccurately by the forward model at Halstead station. The modeldynamics are flat as compared to variability in the retrievals atall frequencies in both locations. Further improvement in map-ping of dynamic surface and vegetation-type-specific parame-ters may improve agreement and is a part of ongoing research.An adequate representation of the dynamics is particularlylacking at 89 GHz for both locations over the course of theeight-year comparison. In fall 2012, CRTM v2.1 was released,and it includes multiple updates to the land surface emissivitymodel (CRTM v2.1 Users Guide). Future research will includeimplementation of this new version and may improve some ofthese comparisons.

The dew effect common in agricultural regions as discussedin Lin and Minnis [11] and more recently Moncet et al. [12]is difficult to analyze in this context. While canopy water


Fig. 11. 2004–2011 emissivity Noah + LandEM and retrieved emissivity anomalies for the AMSR-E V-pol (left) and H-pol (right) channels at Halstead station.


Fig. 12. 2004–2011 emissivity Noah + LandEM and retrieved emissivity anomalies for the AMSR-E V-pol (left) and H-pol (right) channels at Pawhuska station.


Fig. 13. June 2006 time series of precipitation (top left), canopy water storage (top right), and 89-GHz emissivity from the Noah + LandEM model (bottom left)and AMSR-E clear-sky retrievals (bottom right).

interception and dew are available output via the LSM, thereis as yet no mechanism in the emissivity model to calculatethe radiative effects of the additional liquid atop the vegetationlayer. Fig. 13 shows the canopy interception field from the Noahmodel along with modeled and retrieved 89-GHz emissivitiesfor the single month June 2006. It should be noted that, whilemodeled emissivities are plotted here throughout the rain event,retrieved values are calculated only for cloud-free overpassesand not during the rain itself or cloudy periods, leading togaps in the retrieval plot (lower right). For this summer month,modeled 89-GHz emissivity values are universally higherthan the satellite retrieval. Some cases (e.g., day 7) show acoincident low retrieved 89-GHz emissivity following a time ofprecipitation and increased canopy water storage. The modeldata set shows a slight drop in 89-GHz emissivity following therain event of day 16. This decrease is also present and greaterin magnitude in the retrieved data set. There does not, however,appear to be a distinctive connection between these parameters.The comparison suggests that a dew/intercepted water layermay be a useful addition to the physical emissivity model, butmore investigation is required to attain agreement at this higherfrequency.

VI. SUMMARY AND CONCLUSION

Analysis of land surface emissivities calculated using bothpassive microwave retrieval and physical modeling shows aneed for some improvement in each method. The combinationland surface–physical emissivity model used here does notappear to capture dynamic changes well and is not currentlyready for operational use in physical rainfall retrieval. At thelower frequencies, where emissivity is highly sensitive to soil

moisture, agreement is generally good for areas of mid-rangevegetation density, exemplified by Halstead station, an agricul-tural area. Larger biases are observed at very low or very highvalues of LAI. An improvement over the default leaf thicknessto one that is a function of vegetation type would likely improveagreement, along with a more functional representation ofvegetation content–water relationships. Ideally, a data set of leafthickness measurements could be used; however, calibration ofleaf thickness parameters may be the only viable approach inthe absence of such data. Additionally, the conversion to vege-tation water content would likely be improved by a vegetation-type-specific relationship such as that described in Jacksonet al. [41]. Again, data for the compilation of such an adaptabledynamic parameterization would be necessary and not partic-ularly straightforward and is beyond the scope of this initialinvestigation into model-retrieval comparisons. In the end, wesimply cannot know the thickness, water content, placement,and orientation of every plant and leaf in the domain. Thisplaces a limit on the absolute accuracy of physical emissivitymodeling for real-time satellite retrieval purposes.

The comparison of modeled and retrieved emissivities is acomplicated problem. No validation data set exists, and theinputs required differ, leading to many possible sources oferror. The model contains multiple input data sets, along withuncertainties associated with the land surface modeling schemeitself as well as the radiative transfer within the emissivitymodeling component. This represents a bottom-up calculation,whereas the retrieval scheme requires information about theatmosphere and performs radiative transfer down through theatmospheric column. The requirement of strict cloud clearingin the comparisons here is meant to make the atmosphericcomponent as little an issue as possible.


Given the high correlation of emissivity at the lower fre-quency channels of AMSR-E to changes in soil moisture,precipitation is clearly a major component of dynamic variabil-ity. The magnitude of response to precipitation appears to behigher in the retrieved emissivities, but it should be pointed outthat the retrieval corresponds to an instantaneous observationwhereas the model is a 3-h average. It is possible that emissivitychanges happen on a short timescale as water collects duringand following precipitation and then dries. The timescale ofdrying may be highly variable, and the soil moisture changeat the very surface (affecting emissivity) will be a functionof soil type and previous conditions. Irrigation in this type ofarea will obviously influence the emissivity as well and mustbe considered in assessing model agreement. The use of 3-hmodel data likely affects emissivity changes connected to thesoil temperature as well and influences the models ability tocapture diurnal variability.

Interpretation of the emissivity differences at 89 GHz is lessclear. It is very likely that the retrieval is glossing over known(but not quantified) errors due to unaccounted-for absorptionand scattering in the atmosphere. In fact, Ruston and Vonder-Haar reported in a 2004 study [42] that the impact of a cirruscloud is an order of magnitude greater than differences due tovegetation changes. Clearly, there are errors present as well onthe model side as suggested by the large bias from two differentretrieval methods. In the model, it is possible that unaccounted-for radiative effects of liquid water over the vegetation canopyin the form of intercepted precipitation or dew are having aneffect. There is a lot of work to be done in reconciling the89-GHz channel emissivities.

As discussed previously, emissivity validation and error as-sessment are not straightforward. Given that the LSM is an op-erationally used product, well constrained and tested, it makessense to focus efforts in this area on the physical emissivitymodeling component. This process has already begun in therelease of an updated LandEM coupled with CRTM v2.1, andfuture work will make use of these improvements. It is clearfrom the generally good agreement of 1%–2% at frequenciesless than or equal to 37 GHz with strict clear-sky retrievalsthat LSM output coupled with a physical emissivity model iscapable of producing realistic emissivity values in this region.Improvements in the higher frequency channels and vegetationmodeling as well as a better understanding of land surfacetemperature and penetration depth will lead to a highly valuabletool for understanding microwave land emissivity variabilityand its relationship to surface parameters. Eventual expansionof the techniques explored in this paper to the global scale willoffer further understanding of the surface emissivity dynamics,particularly in areas with dependences on surface parametersthat differ from those at the SGP site, e.g., tropical rainforestand desert areas. The applications of this understanding includeimproved radiative transfer capabilities for physical retrieval ofatmospheric parameters such as cloud water and precipitation.

ACKNOWLEDGMENT

Data were obtained from the ARS Micronet program, op-erated and maintained by the USDA Agricultural Research

Service’s Grazinglands Research Laboratory, and from theAtmospheric Radiation Measurement Program sponsored bythe Office of Science, Office of Biological and EnvironmentalResearch, Climate and Environmental Sciences Division, U.S.Department of Energy.

REFERENCES

[1] N. C. Grody, “Surface identification using satellite microwave radiome-ters,” IEEE Trans. Geosci. Remote Sens., vol. 26, no. 6, pp. 850–859,Nov. 1988.

[2] G. M. Heymsfeld and R. Fulton, “Modulation of SSM/I microwave soilradiances by rainfall,” Remote Sens. Environ., vol. 39, no. 3, pp. 187–202,Mar. 1992.

[3] E. G. Njoku and L. Li, “Retrieval of land surface parameters using passivemicrowave measurements at 6–18 GHz,” IEEE Trans. Geosci. RemoteSens., vol. 37, no. 1, pp. 79–93, Jan. 1999.

[4] F. Weng and N. C. Grody, “Physical retrieval of land surface temperatureusing the Special Sensor Microwave Imager,” J. Geophys. Res., vol. 103,no. D8, pp. 8839–8848, Apr. 1998.

[5] T. J. Jackson, A. Y. Hsu, and P. E. O’Neill, “Surface soil moisture retrievaland mapping using high-frequency microwave satellite observations inthe Southern Great Plains,” J. Hydrometeor., vol. 3, no. 6, pp. 688–699,Dec. 2002.

[6] M. F. McCabe, E. F. Wood, and H. Gao, “Initial soil moisture re-trievals from AMSR-E: Multiscale comparison using in situ data andrainfall patterns over Iowa,” Geophys. Res. Lett., vol. 32, p. L06403,2005, doi: 10.1029/2004GL021222.

[7] H. Gao, E. F. Wood, M. Drusch, W. Crow, and T. J. Jackson, “Usinga microwave emission model to estimate soil moisture from ESTARobservations during SGP99,” J. Hydrometeor., vol. 5, no. 1, pp. 49–63,Feb. 2004.

[8] J. Calvet, J. Wigneron, J. Walker, F. Karbou, A. Chanzy, and C. Albergel,“Sensitivity of passive microwave observations to soil moisture and veg-etation water content: L-band to W-band,” IEEE Trans. Geosci. RemoteSens., vol. 49, no. 4, pp. 1190–1199, Apr. 2011.

[9] M. O. Jones, L. A. Jones, J. S. Kimball, and K. C. McDonald, “Satellitepassive microwave remote sensing for monitoring global land surfacephenology,” Remote Sens. Environ., vol. 115, no. 4, pp. 1102–1114,Apr. 2011.

[10] T. J. Jackson and L. Moy, “Dew effects on passive microwave observationsof land surfaces,” Remote Sens. Environ., vol. 70, no. 2, pp. 129–137,Nov. 1999.

[11] B. Lin and P. Minnis, “Temporal variations of land surface microwaveemissivities over the atmospheric radiation measurement program South-ern Great Plains site,” J. Appl. Meteorol., vol. 39, no. 7, pp. 1103–1115,Jul. 2000.

[12] J. Moncet, P. Liang, J. F. Galantowicz, A. E. Lipton, G. Uymin, C. Prigent,and C. Grassotti, “Land surface microwave emissivities derived fromAMSR-E and MODIS measurements with advanced quality control,” J.Geophys. Res., vol. 116, no. D16, pp. D16104-1–D16104-20, Aug. 2011.

[13] R. Ferraro, C. D. Peters-Lidard, C. Hernandez, F. J. Turk, F. Aires,C. Prigent, X. Lin, S. Boukabara, F. A. Furuzawa, K. Gopalan,K. W. Harrison, F. Karbou, L. Li, C. Liu, H. Masunaga, L. Moy,S. Ringerud, G. M. Skofronick-Jackson, Y. Tian, and N. Wang, “Anevaluation of microwave land surface emissivities over the continentalUnited States to benefit GPM-era precipitation algorithms,” IEEE Trans.Geosci. Remote Sens., vol. 51, no. 1, pp. 378–398, Jan. 2013.

[14] F. Aires, C. Prigent, W. B. Rossow, and M. Rothstein, “A new neuralnetwork approach including first guess for retrieval of atmospheric watervapor, cloud liquid water path, surface temperature, and emissivities overland from satellite microwave observations,” J. Geophys. Res., vol. 106,no. D14, pp. 14 887–14 907, Jul. 2001.

[15] G. M. Skofronick-Jackson, M. Kim, J. A. Weinman, and D. Chang, “Aphysical model to determine snowfall over land by microwave radiome-try,” IEEE Trans. Geosci. Remote Sens., vol. 42, no. 5, pp. 1047–1058,May 2004.

[16] P. Bauer, E. Moreau, and S. Di Michele, “Hydrometeor retrieval accuracyusing microwave window and sounding channel observations,” J. Appl.Meteorol., vol. 44, no. 7, pp. 1016–1032, Jul. 2005.

[17] S.-A. Boukabara, K. Garrett, W. Chen, F. Iturbide-Sanchez,C. Grassotti, C. Kongoli, R. Chen, Q. Liu, B. Yan, FuzhongWeng,R. Ferraro, T. J. Kleespies, and H. Meng, “MiRS: An all-weather 1DVARsatellite data assimilation and retrieval system,” IEEE Trans. Geosci.Remote Sens., vol. 49, no. 9, pp. 3249–3242, Sep. 2011.


[18] C. Kummerow, D. B. Shin, Y. Hong, W. S. Olson, S. Yang, R. F. Adler,J. McCollum, R. Ferraro, G. Petty, and T. T. Wilheit, “The evolution ofthe Goddard Profiling Algorithm (GPROF) for rainfall estimation frompassive microwave sensors,” J. Appl. Meteorol., vol. 40, no. 11, pp. 1801–1820, Nov. 2001.

[19] E. Smith, G. Asrar, Y. Furuhama, A. Ginati, A. Mugnai, K. Nakamura,R. Adler, M. Chou, M. Desbois, J. Durning, J. Entin, F. Einaudi,R. Ferraro, R. Guzzi, P. Houser, P. Hwang, T. Iguchi, P. Joe, R. Kakar,J. Kaye, M. Kojima, C. Kummerow, K. Kuo, D. Lettenmaier, V. Leviz-zani, N. Lu, A. Mehta, C. Morales, P. Morel, T. Nakazawa, S. Neeck,K. Okamoto, R. Oki, G. Raju, J. Shepherd, J. Simpson, B. Sohn,E. Stocker, W. Tao, J. Testud, G. Tripoli, E. Wood, S. Yang, andW. Zhang, “International Global Precipitation Measurement (GPM) pro-gram and mission: An overview,” Adv. Global Change Res., vol. 28,pp. 611–653, 2007.

[20] C. Matzler, “On the determination of surface emissivity from satelliteobservations,” IEEE Geosci. Remote Sens. Lett., vol. 2, no. 2, pp. 160–163, Apr. 2005.

[21] C. Prigent, F. Aires, and W. B. Rossow, “Land surface microwave emis-sivities over the globe for a decade,” Bull. Amer. Meteorol. Soc., vol. 87,no. 11, pp. 1573–1584, Nov. 2006.

[22] F. Weng, Y. B. Yan, and N. C. Grody, “A microwave land emissivity model,”J. Geophys. Res., vol. 106, no. D17, pp. 20 115–20 123, Sep. 2001.

[23] H. Norouzi, W. Rossow, M. Temimi, C. Prigent, M. Azarderakhsh,S. Boukabara, and R. Khanbilvardi, “Using microwave brightness tem-perature diurnal cycle to improve emissivity retrievals over land,” RemoteSens. Environ., vol. 123, pp. 470–482, Aug. 2012.

[24] J. F. Galantowicz, J.-L. Moncet, P. Liang, A. E. Lipton, G. Uymin,C. Prigent, and C. Grassotti, “Subsurface emission effects in AMSR-Emeasurements: Implications for land surface microwave emissivity re-trieval,” J. Geophys. Res., vol. 116, no. D17, pp. D17105-1–D17105-16,Sep. 2011.

[25] P. Pampaloni and S. Paloscia, “Microwave emission and plant water con-tent: A comparison between field measurements and theory,” IEEE Trans.Geosci. Remote Sens., vol. GE-24, no. 6, pp. 900–904, Nov. 1986.

[26] M. B. Ek, K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren,G. Gayno, and J. D. Tarpley, “Implementation of Noah land surfacemodel advances in the National Centers for Environmental Predictionoperational mesoscale Eta model,” J. Geophys. Res., vol. 108, no. D22,pp. 8851–8866, Nov. 2003.

[27] B. Jia and Z. Xie, “Evaluation of the Community Microwave EmissionModel coupled with the community land model over East Asia,” Atmosp.Ocean. Sci. Lett., vol. 4, no. 4, pp. 209–215, 2011.

[28] R. Frey, S. A. Ackerman, Y. Liu, K. I. Strabala, H. Zhang, J. Key, andX. Wang, “Cloud detection with MODIS: Part I. Improvements in theMODIS cloud mask for collection 5,” J. Atmos. Ocean. Technol., vol. 25,no. 7, pp. 1057–1072, Jul. 2008.

[29] S. A. Ackerman, R. E. Holz, R. Frey, E. W. Eloranta, B. C. Maddux, andM. McGill, “Cloud detection with MODIS: Part II: Validation,” J. Atmos.Ocean. Technol., vol. 25, no. 7, pp. 1073–1086, Jul. 2008.

[30] D. P. Dee, S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli,S. Kobayashi, U. Andrae, M. A. Balmaseda, G. Balsamo, P. Bauer,P. Bechtold, A. C. M. Beljaars, L. van de Berg, J. Bidlot, N. Bormann,C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. Haimberger, S. B. Healy,H. Hersbach, E. V. Holm, L. Isaksen, P. Kallberg, M. Kohler,M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J.-J. Morcrette,B.-K. Park, C. Peubey, P. de Rosnay, C. Tavolato, J.-N. Thepaut, andF. Vitart, “The ERA-Interim reanalysis: Configuration and performance ofthe data assimilation system,” Q. J. R. Meteorol. Soc., vol. 137, no. 656,pp. 553–597, Apr. 2011.

[31] R. H. Holmes, M. Drusch, J-P. Wigneron, and R. A. M. de Jeu, “A globalsimulation of microwave emission: Error structures based on output fromECMWF’s operational integrated forecast system,” IEEE Trans. Geosci.Remote Sens., vol. 46, no. 3, pp. 846–856, Mar. 2008.

[32] J. L. Bytheway and C. D. Kummerow, “A physically based screen for pre-cipitation over complex surfaces using passive microwave observations,”IEEE Trans. Geosci. Remote Sens., vol. 48, no. 1, pp. 299–313, Jan. 2010.

[33] F. Aires, C. Prigent, F. Bernardo, C. Jiménez, R. Saunders, and P. Brunel,“A Tool to Estimate Land-Surface Emissivities at Microwave frequencies(TELSEM) for use in numerical weather prediction,” Q. J. R. Meteorol.Soc., vol. 137, no. 656, pp. 690–699, Apr. 2011.

[34] S. Kumar, C. D. Peters-Lidard, Y. Tian, P. R. Houser, J. Geiger, S. Olden,L. Lighty, J. L. Eastman, B. Doty, P. Dirmeyer, J. Adams, K. Mitchell,E. F. Wood, and J. Sheffield, “Land Information System: An interoperableframework for high resolution land surface modeling,” Environ. ModelSoftw., vol. 21, no. 10, pp. 1402–1415, Oct. 2006.

[35] Y. Xia, K. Mitchell, M. Ek, B. Cosgrove, J. Sheffield, L. Luo, C. Alonge,H. Wei, J. Meng, B. Livneh, Q. Duan, and D. Lohmann, “Continental-

scale water and energy flux analysis and validation for the North AmericanLand Data Assimilation System project phase 2 (NLDAS-2): 1. Intercom-parison and application of model products,” J. Geophys. Res., vol. 117,no. D3, p. D03109, Feb. 2012.

[36] C. O. Justice, J. R. G. Townshend, E. F. Vermote, E. Masuoka,R. E. Wolfe, N. Saleous, D. P. Roy, and J. T. Morisette, “An overview ofMODIS land data processing and product status,” Remote Sens. Environ.,vol. 83, no. 1/2, pp. 3–15, Nov. 2002.

[37] W. Yang, B. Tan, D. Huang, M. Rautiainen, N. V. Shabanov, Y. Wang, J. L.Privette, K. F. Huemmrich, R. Fensholt, I. Sandholt, M. Weiss, D. E. Ahl,S. T. Gower, R. R. Nemani, Y. Knyazikhin, and R. B. Myneni, “MODISleaf area index products: From validation to algorithm improvement,”IEEE Trans. Geosci. Remote Sens., vol. 44, no. 7, pp. 1885–1898, Jul. 2006.

[38] R. L. Elliott, F. R. Schiebe, K. C. Crawford, K. D. Peter, and W. E. Puckett,“A unique data capability for natural resources studies,” presented at theInt. Winter Meeting; Amer. Society Agricultural Engineers, Chicago, IL,USA, Dec. 14–17, 1993, Paper No. 932529.

[39] M. Owe and A. A. Van de Griend, “Comparison of soil moisture pen-etration depths for several bare soils at two microwave frequencies andimplications for remote sensing,” Water Resources Res., vol. 34, no. 9,pp. 2319–2327, Sep. 1998.

[40] W. B. Rossow and R. A. Schiffer, “Advances in understanding cloudsfrom ISCCP,” Bull. Amer. Meteorol. Soc., vol. 80, no. 11, pp. 2261–2287,Nov. 1999.

[41] T. J. Jackson, D. Chen, M. Cosh, F. Li, M. Anderson, C. Walthall,P. Doriaswamy, and E. R. Hunt, “Vegetation water content mapping usingLandsat data derived normalized difference water index for corn andsoybeans,” Remote Sens. Environ., vol. 92, no. 4, pp. 475–482, Sep. 2004.

[42] B. C. Ruston and T. H. VonderHaar, “Characterization of summertimemicrowave emissivities from the Special Sensor Microwave Imager overthe conterminous United States,” J. Geophys. Res., vol. 109, no. D19,pp. D19103-1–D19103-5, Oct. 2004.

Sarah Ringerud received the B.S. degree in atmospheric and oceanic sciencesand mathematics from the University of Wisconsin—Madison, Madison, WI,USA, in 2002 and the M.S. degree in atmospheric science from Colorado StateUniversity, Fort Collins, CO, USA, in 2006, where she is currently workingtoward the Ph.D. degree in atmospheric science, focusing on passive microwaveremote sensing over land surfaces.

Christian Kummerow received the Ph.D. degree in atmospheric physics fromthe University of Minnesota, Minneapolis, MN, USA, in 1987.

He is currently a Professor with the Department of Atmospheric Science,Colorado State University (CSU), Fort Collins, CO, USA. His expertise is inremote sensing of clouds and precipitation. His focus is toward understandingof the global hydrologic cycle and how climate change may or may not impactthe availability of water. His current research focuses on determining globalprecipitation and its physical characteristics as shown from space- and ground-based sensors. Before coming to CSU, he served as the NASA Project Scientistfor the Tropical Rainfall Measuring Mission (TRMM). He remains a memberof the Joint TRMM Steering Team, is a member of the Advanced MicrowaveScanning Radiometer team, and plays an active role in planning and definingnew spaceborne missions geared toward a better understanding of the globalwater and energy cycle.

Christa Peters-Lidard received the B.S. degree in geophysics and a minor inmathematics (summa cum laude) from Virginia Polytechnic Institute and StateUniversity, Blacksburg, VA, USA, in 1991 and the M.A. and Ph.D. degreesfrom the Water Resources Program, Department of Civil Engineering andOperations Research, Princeton University, Princeton, NJ, USA, in 1993 and1997, respectively.

She is currently the Chief of the Hydrological Sciences Laboratory, NASA’sGoddard Space Flight Center, Greenbelt, MD, USA, where she has been aPhysical Scientist since 2001. She is currently the Chief Editor of the AmericanMeteorological Society (AMS) Journal of Hydrometeorology. Her researchinterests include land–atmosphere interactions, soil moisture measurement andmodeling, and application of high-performance computing and communicationtechnologies in Earth system modeling.

Dr. Peter-Lidard is an elected Member of the AMS Council, and she waselected as an AMS Fellow in 2012. Her Land Information System team wasawarded the 2005 NASA Software of the Year Award.


Yudong Tian received the Ph.D. degree in atmospheric sciences from theUniversity of California at Los Angeles (UCLA), Los Angeles, CA, USA,in 1999.

He conducted research in climate dynamics and geophysical fluid dynamicsat UCLA. He was also one of the developers of the popular Advanced SpectralAnalysis SSA-MTM Toolkit at UCLA. Between 2000 and 2002, he wasworking in the Internet industry, holding such positions as Systems Managerand Chief Technology Officer. He started to work with the NASA GoddardSpace Flight Center, Greenbelt, MD, USA, in 2002, first on the developmentof the Land Information System, which won NASA’s Software of the YearAward in 2005. More recently, he has been evaluating and validating satellite-based precipitation measurements and also working on land surface emissivitymodeling to support NASA’s Global Precipitation Measurement mission.

Kenneth Harrison received the Ph.D. degree in civil engineering from NorthCarolina State University (NCSU), Raleigh, NC, USA, in 2002.

He conducts research in environmental and water resource system analysisat NCSU. His primary research area is decision making under uncertainty. Ofsecondary interest is the accounting of uncertainty in modeling via Bayesianmethods, the development of computationally based decision support tools,and the optimization of large-scale systems. Applications extend from theadaptive management of water quality, water distribution system modeling,water quantity management, assessment of the external damages of powergeneration, integrated solid waste management, and energy modeling. Mostrecently, he helped to develop extensions in the NASA Land InformationSystem to carry out uncertainty estimation via Markov chain Monte Carlomethods for Bayesian analysis for application to land surface and land surfacecoupled models including microwave emissivity models.

a comparison of microwave window channel retrieved and forward-modeled emissivities over the u.s....

Documents