Using a Microwave Emission Model to Estimate Soil Moisture

FEBRUARY 2004 49G A O E T A L .

q 2004 American Meteorological Society

Using a Microwave Emission Model to Estimate Soil Moisture from ESTARObservations during SGP99

HUILIN GAO AND ERIC F. WOOD

Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

MATTHIAS DRUSCH*

Meteorologisches Institut der Universitaet Bonn, Bonn, Germany

WADE CROW AND THOMAS J. JACKSON

USDA ARS Hydrology and Remote Sensing Laboratory, Beltsville, Maryland

(Manuscript received 25 February 2003, in final form 13 July 2003)

ABSTRACT

The 1999 Southern Great Plains Hydrology Experiment (SGP99) provides comprehensive datasets for eval-uating microwave remote sensing of soil moisture algorithms that involve complex physical properties of soilsand vegetation. The Land Surface Microwave Emission Model (LSMEM) is presented and used to retrieve soilmoisture from brightness temperatures collected by the airborne Electronically Scanned Thinned Array Radi-ometer (ESTAR) L-band radiometer. Soil moisture maps for the SGP99 domain are retrieved using LSMEM,surface temperatures computed using the Variable Infiltration Capacity (VIC) land surface model, standard soildatasets, and vegetation parameters estimated through remote sensing. The retrieved soil moisture is validatedusing field-scale and area-averaged soil moisture collected as part of the SGP99 experiment, and had a rmsrange for the area-averaged soil moisture of 1.8%–2.8% volumetric soil moisture.

1. Introduction

Soil moisture is a key factor in understanding land–atmosphere feedbacks. Operational large-scale soilmoisture observational products would likely enhancethe accuracy of numerical weather prediction (NWP)products (e.g., Koster and Suarez 2001), hydrologicalflood forecasting, agricultural drought monitoring, aswell as water cycle research related to climate studies.Observations based on standard in situ instrumentationcan only measure local values and may not adequatelysample land surface heterogeneity. In addition, denseground networks are expensive to install and maintain.Spaceborne microwave radiometry has been recognizedas an effective method for monitoring soil moisture atlarge scales (Owe et al. 1999). In theory, the dielectricconstant of the soil water medium is raised by increasesin soil water content. These variations are detectable by

* Current affiliation: ECMWF, Shinfield Park, Reading, UnitedKingdom.

Corresponding author address: Dr. Eric F. Wood, Department ofCivil and Environmental Engineering, Princeton University, Prince-ton, NJ 08544.E-mail: [email protected]

remote microwave sensors (Njoku 1977). The sensitiv-ity of surface dielectric measurements to soil moistureis higher at lower microwave frequencies. Currently op-erating and scheduled microwave satellite missions thathave been applied to soil moisture retrievals include:the Scanning Multichannel Microwave Radiometer(SMMR) at 6.63 GHz on Nimbus-7, which was launchedin 1978, with data available until 1987 (Owe et al.1992); the Defense Meteorological Satellite Program(DMSP) Special Sensor Microwave Imager (SSM/I) at19.3 GHz, which was launched in 1987 (Jackson 1997);Tropical Rainfall Measuring Mission (TRMM) Micro-wave Imager (TMI) at 10.65 GHz, which was launchedin 1997 (Jackson and Hsu 2001); the Advanced Micro-wave Scanning Radiometer (AMSR) on the Earth Ob-serving System (EOS) Aqua satellite (AMSR-E) at 6.9GHz, which was launched in May 2002 and the Ad-vanced Earth Observing Satellite II (ADEOS-II) AMSR,which was launched in December 2002; and the Eu-ropean Soil Moisture and Ocean Salinity Mission(SMOS) at 1.4 GHz, which has an anticipated launchin 2007 (Kerr et al. 2001).

Algorithm development and validation are essentialbefore global application of soil moisture retrieval al-gorithms. The dielectric properties of wet soil have been

50 VOLUME 5J O U R N A L O F H Y D R O M E T E O R O L O G Y

widely investigated (Wang and Schmugge 1980; Dob-son et al. 1985; Ulaby et al. 1986), as well as the ra-diative characteristics of vegetation (Kirdyashev et al.1979; Wegmuller et al. 1995). Using these results anddata from airborne remote sensing field studies, soilmoisture retrieval algorithms have continued to be re-fined (e.g., Jackson et al. 1995, 1999). However, op-erational application of these algorithms at regionalscales using satellite sensor measurements faces twomajor challenges: the shortage of information regardingthe numerous parameters involved in radiometry phys-ics at large scales and the high within-footprint spatialheterogeneity of land surface variables relative to thelow resolution of spaceborne microwave radiometers(.10 km). This paper contributes to the refinement ofsoil moisture retrieval approaches through applicationof an algorithm based on modeling microwave emis-sions from the surface and information available fromlarge-scale surface vegetation–atmosphere transfer(SVAT) modeling.

There are two major objectives of this paper. The firstobjective is to apply a new soil moisture retrieval ap-proach that utilizes a Land Surface Microwave EmissionModel (LSMEM) for the soil–vegetation–atmosphericsystem with surface temperature data from a SVATmodel that utilizes high-resolution remotely sensed veg-etation and soil data. The second objective is to applythe approach to retrieve soil moisture fields from theairborne Electronically Scanned Thinned Array Antenna(ESTAR) sensor 1.413-GHz brightness temperaturescollected during the 1999 Southern Great Plains Hy-drology Experiment (SGP99) experiment, and makingthem available to the scientific community. Such a dataproduct has not yet been developed. Underlying theseobjectives is the objective to develop a soil moistureretrieval approach suitable for satellite-measured bright-ness temperatures collected at regional to continentalscales. At these scales, the detailed meteorological andvegetation data are unavailable, so operational productsneed to be utilized. Validation of the LSMEM algorithm,using the detailed ground observations available duringSGP99, is an important step. Section 2 presents the soilmoisture retrieval approach based on the LSMEM mod-el, section 3 describes the SGP99 field experiment in-cluding the L-band brightness temperature measure-ments from ESTAR, and section 4 describes theLSMEM model input variables used for the ESTAR-based soil moisture retrievals. The retrieved soil mois-ture fields for the SGP99 domain and validation resultsfor the LSMEM for the SGP99 ground observation sitesare presented in section 5, followed by a discussion andconclusions in section 6.

2. Land surface microwave emission model(LSMEM)

In the reviewed literature, a number of models for thecomputation of microwave emission from land surfaces

exist (Ulaby et al. 1986; Wang and Choudhury 1995;Njoku and Entekhabi 1996). Depending on the specificapplication and frequency range, they represent more orless complex approximations of the vector radiativetransfer equation and distinguish themselves throughdifferent parameterizations for the key processes de-scribing the interaction between radiation and matter.The LSMEM model used in this study is based on asolution for the radiative transfer equation as outlinedin Kerr and Njoku (1990). Following this article, thebrightness temperature of vegetation-covered soils Tbv,p

can be written as

2t 2t 22t*at atT 5 T 1 e (T 1 T e )(1 2 « )ebv,p au ad sky p

2t 2t* 2t*at1 e {« T e 1 T (1 2 v*)(1 2 e )p s y

2t*3 [1 1 (1 2 « )e ]}, (1)p

where Tau and Tad are the upward and downward con-tributions from the atmosphere, Ts is the effective soiltemperature, Ty the vegetation temperature, Tsky the cos-mic radiation, tat the optical depth of the atmosphere,and «p the rough-soil emissivity. For vegetation withcylindrical structure, v* is the single-scattering albedo,t* is the optical depth of the vegetation (Chang et al.1980). For nonisotropic conditions, v* and t* are ef-fective single-scatter albedo and effective optical depthof vegetation (Mo et al. 1982; Jackson et al. 1982).Subscript p indicates polarization dependency in themodel configuration for this study.

Within the LSMEM code multiple options exist tocompute the key parameters «p, and t*, which providea flexible interface to various input data sources andmake the model an appropriate tool for combined hy-drological/data assimilation studies (e.g., Drusch et al.1999; Crow et al. 2001; Drusch et al. 2001; Seuffert etal. 2003). To get rough-soil emissivity «p in LSMEM,the saline dielectric constant is calculated after Kleinand Swift (1977), and the wet soil dielectric constantcan be calculated either after Wang and Schmugge(1980) or after Dobson et al. (1985). Then, using theFresnel equation, the reflectivity of a smooth surface isderived, and the effect of soil roughness is parameter-ized using the equations presented in Wang and Choud-hury (1981). In this investigation, we used the approachby Wang and Schmugge (1980) to compute the soildielectric constant.

Two choices within LSMEM are available to get thevegetation optical depth t*: Effective Medium theoryfor low frequencies by Kirdyashev et al. (1979) and theGeometrical Optics approach as described by Weg-muller et al. (1995). For the Effective Medium theory,t* depends on a vegetation structure parameter, theimaginary part of the dielectric constant of saline water,frequency, the vegetation water content, and the viewingangle. The superiority of the Geometrical Optics ap-proach over the Effective Medium theory is that it isnot restricted by frequency, and it can also estimate v*,


the single-scattering albedo. The drawback is that ad-ditional parameters (dry mass fraction, leaf thickness,and polarization-dependent structure parameters) areneeded. As this study is for the L band, the simplealgorithm by Kirdyashev is used. The single-scatteringalbedo v* is assigned constant (0.1) based on literaturereviews (Ulaby et al. 1983; Pampaloni and Paloscia1986; Wegmuller et al. 1995).

For bare soil the brightness temperature Tbs,p can beobtained from Eq. (1) by setting t* equal to 0. For partlyvegetated surfaces the brightness temperature Tb,p can becalculated introducing the fractional vegetation cover c:

T 5 (1 2 c)T 1 cTb,p bs,p bv,p (2)

In this study, fractional vegetation cover was set to uni-ty. Since it is not realistic to assume homogeneity atESTAR resolution, the parameters involved in the re-trieval are effective parameters for the specific resolu-tion. Aggregation effects due to the nonlinearities inradiative transfer were found to be negligible at L bandin the SGP area (Drusch et al. 1999). Consequently, Tb,p

reduces to Tbv,p. The modeling approach used in pre-vious studies on ESTAR soil moisture retrievals (e.g.,Jackson et al. 1999) requires further approximations.Under the assumption that the contributions of the at-mosphere, the cosmic background radiation, and the sin-gle-scattering albedo are zero, Eqs. (1) and (2) yield

2t* 2t* 2t*T 5 «T e 1 T (1 2 e )(1 1 (1 2 «)e ). (3)b s y

If vegetation temperature is set equal to soil temperature,then

Tb 22t*5 1 1 (« 2 1)e . (4)Ts

Equation (4) forms the basis for the soil moisture re-trieval scheme introduced in Jackson et al. (1982, 1995,1999).

It has been shown in various applications that Eq. (4)leads to very good results when applied to areas withsparse vegetation (e.g., Jackson et al. 1982, 1995, 1999).However, in areas characterized by different vegetationtypes the assumptions outlined earlier may not hold(Ferrazzoli et al. 2002). A second critical assumptionin the earlier modeling approaches is to ignore the tem-perature difference between the soil and vegetation,which can exceed a maximum of 7 K, as reported inJackson et al. (1982). The LSMEM distinguishes be-tween an effective soil temperature, which takes theradiation emission depth into account, and vegetation/surface temperature. In both approximations, soil rough-ness effects are parameterized using the equations pre-sented in Wang and Choudhury (1981). The opticaldepth of vegetation is computed following Kirdyashevet al. (1979). As a result, both rely on the correct cal-ibration with respect to the soil roughness parameterand the vegetation structure coefficient. These quantitiescannot be obtained from large-scale measurements,since at these scales they represent an equivalent effect

rather than geophysical parameters (Choudhury et al.1979).

For a better comparison with previous retrieval stud-ies (e.g., Jackson et al. 1999) the effect of errors in theobservations on the derived soil moisture are neglected.For the application presented in this study, it is notnecessary to retrieve soil moisture through a variationalmethod, which would require the adjoint model of theLSMEM. Since the LSMEM is ‘‘cheap’’ in terms ofcomputational resources, it can be inverted numerically.Starting from a first guess value for volumetric soilmoisture, brightness temperature can be computed usingthe specified vegetation, soil properties, and computedatmospheric contributions, which can be done with anatmospheric radiative transfer model. The optimal soilmoisture value is then retrieved through a simple iter-ative procedure. Figure 1 shows the flowchart ofLSMEM soil moisture retrieval algorithm.

3. 1999 Southern Great Plains Experiment(SGP99)

a. SGP99 data collection

The 1999 Southern Great Plains Hydrology Experi-ment was carried out from 8 to 21 July, 1999 in centralOklahoma. The boundaries for the SGP99 experimentalregion (see http://hydrolab.arsusda.gov/sgp99) were de-fined by the flight path for the airborne measurementsand include three subregions where intensive ground-based sampling was focused. (The experimental plan,including the remote sensing, ground-based, and ancil-lary data collection activities is available at http://hydrolab.arsusda.gov/sgp99 and the reader is referredto this URL for further details.) Table 1 provides a listingof the detailed measurements taken during the experi-ment. The three ground validation subregions are theU.S. Department of Agriculture (USDA) AgricultureResearch Service (ARS) Little Washita watershed (LW)southwest of Chickasha, Oklahoma, the USDA ARSGrazinglands Research Laboratory at El Reno (ER),Oklahoma, and the Department of Energy AtmosphericRadiation Measurement (ARM) Cloud and RadiationTestbed (CART) Central Facility (CF) near Lamont,Oklahoma.

b. Meteorological conditions during the experiment

At the beginning of SGP99, the experimental regionwas generally dry except for the northern portion of thearea. On 10 July, a large, warm season rain event oc-curred over the northern two-thirds of the region. Figure2 shows observed total daily precipitation of the ex-perimental region based on National Environmental Sat-ellite, Data, and Information Service (NESDIS) stage-IV radar–gauge precipitation products. The rainfall to-tals at Oklahoma Mesonet stations at El Reno, LittleWashita, and Central Facility were 107, 49, and 37 mm


FIG. 1. Flowchart of the LSMEM soil moisture retrieval algorithm.

respectively. These two data sources show the rainfallpattern and the high spatial variability of this summerrainfall event. No other rainfall events occurred duringthe experiment, which resulted in a strong drydownfrom 10–20 July and a large dynamic range in observedsoil moisture.

c. ESTAR airborne L-band instrument

ESTAR is an airborne passive microwave L-band ra-diometer centered at 1.413 GHz with a bandwidth of20 MHz. It has been widely used in soil moisture remotesensing studies, including Washita’92 (Jackson et al.1995) and SGP97 (Jackson et al. 1999). ESTAR is ahybrid radiometer; its along-track measurement is ob-tained by real aperture and across track is by syntheticaperture (Le Vine et al. 1990, 2001b). The instrumentwas installed on a P-3B aircraft operated by the NationalAeronautics and Space Administration (NASA) WallopsFlight Facility. Flights were conducted at an altitude of7.5 km. The instrument takes a complete cross-trackscan every 0.25 s. A data record consists of the measuredbrightness temperature from each beam location, thecorresponding time, the (Global Positioning System)

GPS-based aircraft geoposition, and aircraft pitch, roll,and yaw data. The field of view is restricted to 6458to avoid any distortion of the synthesized beam withincidence angle.

As described in the SGP99 campaign documents(http://daac.gsfc.nasa.gov/CAMPAIGNDOCS/SGP99/),postprocessing of the ESTAR data consisted of refiningthe brightness temperature calibration, radio frequencyinterference (RFI) removal, georegistration, and an in-cidence angle correction. Instrument calibrations includ-ed a premission laboratory blackbody measurement anda postmission open-ocean water measurement, the latterflown with salinity and sea surface temperature groundtruth supplied by shipboard measurements. As in the caseof SGP97 (Jackson et al. 1999) the planned four parallellines were modified to compensate for strong RFI in thevicinity of Oklahoma City. This was a critical problembecause of the potential impact on the El Reno studyarea. The flight-line reconfiguration eliminated the strongRFI for measurements over El Reno. (See the experi-mental plan and campaign documents for additional dis-cussion of the RFI.) The data were normalized to nadirusing methods described in Jackson et al. (1995) and LeVine et al. (1994, 2001a), and georeferenced to 0.0058


TABLE 1. SGP99 satellite observing systems, aircraft remote sensing instruments, ground data collection, and regional networks (detailsare available online at http://hydrolab.arsusda.gov/sgp99/sgp99b.htm).

Aircraft observations

Platform* Measurement Frequency during SGP99

PSR Passive radiometer (7.3 GHz) 8, 9, 11, 14, 15, and 19 JulESTAR Passive radiometer (1.4 GHz) 8, 9, 14, 15, 19, and 20 JulPALS Passive radiometer (1.4 and 2.7 GHz); Active radar

(1.2 and 3.1 GHz)8, 9, 11, 12, 13, and 14 Jul

Ground measurements

Platform Measurement Frequency during SGP99

Soil moisture Gravimetric 8–20 JulSoil and surface temperature Infrared thermometers (IRT) 8–20 JulSoil properties Bulk density, roughness Once in fieldsVegetation properties Land cover type

Vegetation water contentOnce in fieldsOnce in fields

Surface heat fluxes Eddy correlation estimates of latent, sensible, andground heat flux

Continuously, in four fields in LittleWashita area

Regional meteorological networks

Platform Measurement Frequency during SGP99

Oklahoma Mesonet Air temperature, soil temperature, soil moisture,wind speed

Every 5 min

ARM CART Radiometric observations, wind, temperature andhumidity sounding systems; Bowen ratio, eddycorrelation, surface meteorology observation, soiltemperature and moisture

Continuously, at the ARM/CARTCentral Facility site

ARS Micronet Rainfall, relative humidity, air temperature, solar ra-diation, soil temperature

Little Washita, every 5 min for cli-mate data, 15 min for soil tem-perature

SSM/I: Special Sensor Microwave Imager; TMI: TRMM Microwave Imager; Landsat TM: Landsat Thematic Mapper; AVHRR: Ad-vanced Very High Resolution Radiometer; GOES: Geostationary Operational Environmental Satellites; PSR: Polarimetric Scanning Radi-ometer; PALS: Passive and Active L and S Band System.

FIG. 2. Observed total daily precipitation (mm) on 10 Jul 1999 over the SGP99 area based onNESDIS stage IV radar–gauge precipitation products.


FIG. 3. Horizontal component of ESTAR observed brightness temperature (K) on 8, 9, 14, 15,19, and 20 Jul 1999.

latitude by 0.0058 longitude grid (approximately 555 m3 450 m). The grid value is the unweighted average ofall brightness temperatures falling within the grid.

During SGP99, ESTAR measured the horizontal po-larized brightness temperature on 8, 9, 14, 15, 19, and20 July. Figure 3 shows images of these data. A linearsoil moisture regression was reported by Le Vine et al.(2001a) to evaluate the ESTAR observations duringSGP99.

4. Input state variables and parameters for theLSMEM

Besides the horizontal polarized brightness temper-ature, the LSMEM inputs include effective soil tem-

perature, vegetation temperature, soil texture, surfaceroughness, soil bulk density, vegetation water content,and a vegetation structure parameter. For soil moistureretrievals from the ESTAR airborne sensor duringSGP99, all input data were processed onto grids of0.0058 resolution for the area encompassed by 348 to388N in latitude and 2978 to 298.58W in longitude.

a. Soil and vegetation temperatures

The effective soil temperature needed for the soilmoisture retrievals is a function of surface temperature,deep soil temperature, and frequency (Choudhury et al.1982). For SGP97 Jackson (1999) used interpolated


FIG. 4. (a) VIC surface temperature validation compared to allARM CART solar and infrared observing systems (SIROS) sites forJul 1999. (b), (c) The time series of soil temperatures at differentlayers from VIC and Oklahoma Mesonet observations at Apache(34.918N, 98.298W).

Oklahoma Mesonet soil temperatures. However, overcontinental areas lacking intensive ground-based net-works, sparse measurements of soil temperature mayneglect spatial heterogeneity due to soils and land cover,and lead to excessively smooth soil temperature fields.Operational products from land surface models (LSM)offer an alternative to ground-based measurements. Inthis study, the surface and deep soil temperatures wereobtained from the Variable Infiltration Capacity (VIC)land surface model (Liang et al. 1994; Cherkauer et al.2003), running as part of the North American Land DataAssimilation System (NLDAS) (Mitchell et al. 2000;Mitchell et al. 2003). As part of the NLDAS validationactivities, VIC-modeled states were compared to ob-servations. Figure 4a shows VIC surface temperaturevalidation when compared to the ARM/CART solar andinfrared observing system (SIROS) sites for July 1999;Figs. 4b,c show the time series of soil temperatures fromVIC and Oklahoma Mesonet observations at Apache

(34.918N, 98.298W) (see http://climate.envsci.rutgers.edu/luo/research/LDAS/models.vic.php). Based on theVIC NLDAS validation (see also Robock et al. 2003),we believe that the VIC-derived surface temperature andsoil temperatures are suitable for LSMEM soil moistureretrievals. As with the effective soil temperature, theeffective vegetation temperature used in the modelvaries with vegetation structure, vertical canopy tem-perature profile, and frequency. The VIC land surfacemodel has a single surface layer, with the soil temper-ature computed beneath the vegetation (Liang et al.1999.) For the vegetation temperature, it is approxi-mated using the VIC radiometric surface temperature,adjusted for the vegetation emissivity based on its clas-sification.

b. Soil texture

Sand fraction and clay fraction data are used in cal-culating the soil dielectric constant. Soil texture clas-sifications of the SGP99 domain were obtained from thestate soil geographic database (STATSGO), which wasdeveloped by the USDA’s Natural Resources Conser-vation Service (Miller and White 1998), and resampledto a 800-m grid by the SGP99 data team. In this in-vestigation, we further processed the data into grids of0.0058 so to be compatible with ESTAR brightness tem-peratures. This data is shown in Fig. 5.

c. Surface roughness, bulk density, and land coverclassification

The surface roughness and bulk density are from theSGP99 database (ftp://daac.gsfc.nasa.gov/data/sgp99).Following the approach used by Jackson et al. (1999)for the SGP97 data, the SGP99 data team extended theground-measured values using a Land Remote SensingSatellite (Landsat) Thematic Mapper (TM)-derived landcover classification to get surface roughness and bulkdensity across the SGP99 region (Figs. 6a,b). The clas-sification was made with available ‘‘cloud free’’ Land-sat-5 (9 March, 12 May, and 15 July) and Landsat-7scenes (7 and 23 July), from 9 March to 23 July (T.Jackson 2003, personal communication). Using multipleimages typically provides more information to increasethe accuracy of the supervised classification, if therewas no land cover change during the experimental pe-riod. Table 2 shows the land cover statistics for theSGP99 domain.

d. Estimation of the vegetation optical depth

The product of the vegetation water content and veg-etation structure parameter b gives the vegetation opticaldepth for calculating the microwave emission attenuatedby vegetation (Kirdyashev et al. 1979). During SGP99,vegetation water content was measured at the groundsampling sites as indicated in Table 1. Normalized Dif-


FIG. 5. (a) Sand fraction (%) and (b) clay fraction (%) for the SGP99 region.

ferential Vegetation Index (NDVI), derived from TMdata collected on 15 July 1999 (Fig. 6c), provides ameasure of the vegetation greenness for the domain. Forthe ground sampling sites, NDVI values were regressedagainst vegetation water content resulting in the follow-ing relationship:

22VWC (kg m ) 5 1.75 3 NDVI. (5)

Using NDVI from remote sensing and (5), the vegeta-tion water content was estimated over the region.

The vegetation structure parameter for vegetation,other than winter wheat, is assigned a value of 0.5, theaverage value for short and long grass according toWang et al. (1980, 1982) and used by Jackson andSchmugge (1991). Since much of the winter wheat hadbeen harvested and consisted of ;10 cm high stubble(similar to the height of short grass), its vegetation pa-rameter is probably larger than that of short grass, whichWang (1980, 1982) estimates to 0.30. Thus a value of0.6 (twice the value for short grass) is assigned forwinter wheat stubble. The land cover classification, de-rived from Landsat TM images, is used to estimate bacross the region (Fig. 6d).

5. Soil moisture retrievals

a. SGP99 regional results

ESTAR brightness temperature data, and other inputvariable and parameters (as described in sections 3 and

4) were used in the LSMEM algorithm to retrieve soilmoisture on a pixel-by-pixel basis for SGP99. Figure 7shows the results.

On 8 and 9 July, the area was mainly dry except inthe northern part, which experienced a total precipitationof 40–80 mm from 29 June to 1 July. Though no ESTARdata were collected from 10 to 13 July, the effect of therainfall on 10 July was still significant on 14 July withthe soil moisture being on average 12% higher as com-pared to 9 July. The soil moisture map is consistent withthe rainfall pattern as shown in Fig. 2. Images for 14,15, 19, and 20 July illustrate the drydown process. Atthe end of the SGP99 experiment, the soil moistureacross the study area was below 10%. From 14 to 20July, soil moisture decreased more than 30% (volu-metric soil moisture) in the northern part of the SGP99domain, decreased about 20% in the middle, and lessthan 10% in the southern part. The soil moisture patternsagree with the following hydrologic characteristics:soils with a high clay fraction and low hydrologic con-ductivity had relatively higher soil moisture, showingless drainage; and areas with high vegetation coverageremained with higher soil moisture, suggesting less soilevaporation, perhaps due to reduced radiation throughthe canopy and higher humidity beneath the vegetation.

b. Comparisons with ground validation sites

The LSMEM-retrieved soil moisture results were val-idated by comparisons with volumetric soil moisture


FIG. 6. (a) Surface roughness, (b) bulk density (g cm23), (c) NDVI, and (d) vegetation parameter b for the SGP99 region.

TABLE 2. Land cover classification statistics over SGP99experiment region.

Land cover classification %

AlfalfaBare soilCornPasture, grazedLegume

1.78.12.3

45.11.2

Pasture, ungrazedTreesUrbanWaterWheat stubble

8.67.43.51.06.9

Bare ground with wheat stubbleBare ground with green vegetationShrubsSand bars and quarriesOutcrops

6.93.03.80.30.3

derived from SGP99 field sampling. During SGP99, themost intensive ground sampling was in the LW area,with less sampling sites at ER and CF. For the validationresults presented here, only sampling sites that coveredan area comparable to the ESTAR footprints (approx-imately 800 m by 800 m) are considered.

The field sampling protocols developed for the ex-periment are as follows: For each site, 14 samples werecollected along two transects separated by 400 m witha sample every 100 m. Each sample was split in halfto provide 0–2.5 cm and 2.5–5.0-cm gravimetric soilmoisture data. From these measurements, site averageand standard deviation were then calculated. The prod-uct of the gravimetric soil moisture and bulk densitydetermines volumetric soil moisture. Since only fourbulk density samples were collected at each field site,bulk density measurements were averaged, when ap-propriate, to provide more representative values. Spe-

cifically, for nearby fields with similar soil properties(LW3–5, LW12–13, LW21–23), an average bulk densitywas computed. For fields not adjacent to other sites, theSGP99-measured bulk density data was averaged withmeasurements from SGP97 for the same field. The 0–5-cm averaged volumetric soil moisture data were usedto compare with the ESTAR-retrieved soil moistures,since the moisture sensing depth at this frequency (1.4GHz) is typically in the 2–5-cm range (Ulaby et al.1986). We feel that this sampling protocol provides areliable dataset for validating the ESTAR-retrieved soilmoisture.

The LSMEM input parameters for the validation siteswere first compared with SGP99 field observations.Vegetation classification and soil texture were, in somecases, inconsistent with site survey data, and were cor-rected. For example, the wheat sites (CF05, LW21, andLW23) were misclassified, which could cause an un-derestimation of vegetation attenuation in these sites.This happened because the winter wheat fields were indifferent stages of harvest during the experimental pe-riod. The SGP regional soil texture data (Miller andWhite 1998) classified the LW winter wheat sites as siltloam (15% clay and 20% sand), while SGP99 field ob-servations suggested more clay in these sites. Thus asilty-clay classification (10% sand and 45% clay) wassubstituted. Table 3 shows the input parameters for allsites, with the sites with adjusted data marked with anasterisk.

For the retrieval validation, soil moisture values forsites without any misclassification were extracted fromthe ESTAR-retrieved soil moisture images, while forthe corrected sites the LSMEM was rerun with modifiedparameters. Figure 8 shows the validation results foreach of the sites, and Fig. 9 shows the validation of theaveraged soil moisture for the sites in each region. The


FIG. 7. LSMEM retrieved soil moisture (%) from ESTAR images during SGP99.

root-mean-square errors are CF 5 2.8%, ER 5 2.3%,LW 5 1.8%, with an average of 2.1% across all sites.As compared to other ESTAR soil moisture retrievalalgorithms, the LSMEM-retrieved soil moisture resultsare excellent for most of the sites. For instance, Jacksonet al. (1999) reported root-mean-square errors of CF 52.7%, ER 5 3.3%, LW 5 2.1%.

Site LW12 had the largest detected inconsistency. For8 and 9 July, the retrieved soil moistures were 8.5% and5.6% lower than field measurements, respectively. Anal-ysis of the field data shows that part of these errorscould be due to the high heterogeneity of this site. Thestandard deviations of observed soil moisture for LW12

on 8 and 9 July is the highest among all the LW samplingsites: 9.95% and 8.77%. Figure 10 shows the field datafor each sampling location for this site during ESTARdata collecting days. Using 8 July as an example, thehighest sample value for this site was 36.9% and thelowest value was 5.0%, sample ID 1 and sample ID 2were only 100 m apart while the reported field valueswere 6.6% and 31.9%. Values for sample ID 2 on otherdays differed significantly from those on the first daysof the experiment, suggesting even higher heterogeneityaround that location. The average value used for vali-dation at LW12 was 19.5%, much larger than that forthe adjacent field LW13 (8.2%). For areas with signif-


TABLE 3. Land cover classification for field sampling sites at CF, ER, and LW.

Site ID Land coverEstimated

NDVIVegetation watercontent (Kg m22) Soil classification

Bulk density(g cm23) Roughness

Estimatedvegetationstructureparameter

CF-04CF-05ER-01ER-05LW-03

Winter wheatWinter wheat*RangelandRangelandRangeland

0.220.250.570.550.49

0.390.441.000.960.86

Silt loamSilt loamSilt loamSilt loamFine sandy loam

1.231.261.291.291.29

0.310.300.300.300.30

0.60.60.50.50.5

LW-04LW-05LW-12LW-13LW-21LW-22LW-23

RangelandRangelandRangelandRangelandWinter wheat*Winter wheatWinter wheat*

0.520.480.500.370.260.150.31

0.910.840.880.650.460.260.54

Fine sandy loamFine sandy loamLoamLoamSilty clay*Silty clay*Silty clay*

1.271.281.291.281.201.241.21

0.300.300.300.300.300.300.34

0.50.50.50.50.60.60.6

* Sites with adjusted data.

icant heterogeneity, due to soils, topography, drainage,and so forth, accurate retrievals of remote sensing soilmoisture data will be challenging. For more discussionon subgrid soil moisture heterogeneity the reader is re-ferred to Charpentier and Groffman (1992), Famigliettiet al. (1999), and Crow and Wood (1999).

As a test, and because the heterogeneity in the SGP99domain is quite small, the second approach was usedto test the retrieval of soil moisture at 0.258 spatial res-olution, and compare these values to soil moisture val-ues retrieved at 0.1258 NLDAS resolution—the reso-lution at which the VIC surface temperature data wereavailable. Figure 11 presents the results, which showsthat using low-resolution brightness temperatures givesalmost the same results as averaging the soil moisturefrom higher-resolution brightness temperatures. This in-dicates that for this region and period, the nonlinearityand heterogeneity has a small impact. Comprehensivetests are outside the scope of this paper, but these initialtests give confidence that a retrieval algorithm based onLSMEM, a land surface model to provide surface tem-peratures, and operational soil and vegetation datasetscan provide a strong basis for satellite soil moistureretrieval. It is currently being tested using TRMM Mi-crowave Imager and AMSR-E data (Wood et al. 2003.)

6. Discussion and conclusions

The use of passive microwave remote sensing for soilmoisture estimation is well established, yet most studiesrely on empirical and semiempirical relationships to re-trieve soil moisture from microwave brightness tem-peratures (e.g., Wang 1983; Owe et al. 1992). A LandSurface Microwave Emission Model (LSMEM) is pre-sented that computes the surface emissions based on theequations of soil emission and vegetation attenuation atmicrowave frequencies. LSMEM is used to retrieve soilmoisture from L-band brightness temperature measure-ments during SGP99, collected from the airborne ES-TAR instrument, surface temperatures from a land sur-

face model, and vegetation parameters estimated fromremote sensing. This approach allows us to exploreLSMEM’s potential in satellite remote sensing retrievalsat coarser resolution using a mix of remote sensing andoperational hydrological modeling products (Wood etal. 2003).

The SGP99 field campaign provided a comprehensivedataset for this investigation. Full-site field observationsof gravimetric soil moisture offered validation data atthe ESTAR footprint resolution. The vegetation and soilparameters were compiled as part of the SGP99 exper-iment and the VIC land surface model provided thesurface temperatures as part of the North American LandData Assimilation System (NLDAS) model output. Atthe same time, the meteorology background during theexperimental period was excellent for testing the re-trieval over a large dynamic range of moisture condi-tions.

LSMEM-retrieved soil moisture, validated by fieldobservations at the three main field sites (CF, ER, andLW), have rms errors in volumetric soil moisture of2.8%, 2.3%, and 1.8%, respectively. Compared to othermicrowave soil moisture retrieval algorithms, theLSMEM performs well (Jackson et al. 1999). This en-courages further application of this physical model inretrieving soil moisture from spaceborne platforms suchas AMSR-E.

To test out the potential of applying hydrologicalmodeling to represent spatial heterogeneity within thefootprints of coarse-resolution spaceborne retrievals, theLSMEM model was run with interpolated OklahomaMesonet surface temperature observations. The vali-dated results by field observations at CF, ER, and LWsites, have rms errors in volumetric soil moisture of3.1%, 2.6%, and 2.0%, respectively. This shows thatusing a LSM-based surface temperature provides ac-curacy comparable to using interpolated temperaturesfrom a dense operational network. It further supportsthe development of off-line land data assimilation sys-


FIG. 8. Validation results for full sampling sites during SGP99.


FIG. 9. Validation of the soil moisture averaged over the samplingsites for the CF, ER, and LW areas. FIG. 11. The 1/48 retrieved soil moisture as compared to soil mois-

ture averaged from 1/88 retrieved soil moisture over the ESTAR ob-served region during SGP99.

FIG. 10. Field measurements at each sampling location in field LW12 during days with ESTARmeasurements.

tems and their data products for applications beyondinitial surface conditions for weather prediction models.Since networks of the density of the Oklahoma Mesonetare unavailable at continental to global scales, the useof LSM-based surface temperatures for satellite soilmoisture retrievals is recommended.

The results presented here offer support that theLSMEM algorithm and approach should work well for

satellite-based measurements of microwave brightnesstemperatures. Because of the low resolution of thesesensors, it remains a challenge to determine the mosteffective way of combining higher-resolution soil, veg-etation, and surface temperature data to retrieve soilmoisture, and for strategies for downscaling either


through active sensors or model-based data assimilation.An additional challenge is to validate satellite products,when significant scale-disparity exists among the point-scale ground measurements, LSM-based estimates, andsatellite-based retrievals.

The SGP99-retrieved soil moisture fields developedin this study will be made available to the communitythrough the SGP99 data center (ftp://daac.gsfc.nasa.gov). The retrieved ESTAR soil moisture products canbe compared with soil moisture products derived fromC-band and X-band radiometers, whose penetrationdepth is less, and sensitivity to vegetation is greater,than at L band. During SGP99, these higher-frequencyproducts include soil moisture from the airborne polar-imetric scanning radiometer (PSR) C-band passive ra-diometer and the satellite X-band TMI radiometer. In-tercomparison studies among L, C, and X bands areimportant due to the launch of the Advanced MicrowaveScanning Radiometer on board NASA Aqua platform(AMSR-E) and National Space Development Agency ofJapan (NASDA) ADEOS-II, which includes both C andX bands as well as the X-band on TMI, which has flownsince 1997 (Wood et al. 2003), and the anticipated L-band sensor on the Soil Moisture Ocean Salinity(SMOS) satellite.

Acknowledgments. This research is based on supportthrough NASA Grants NAG5-9635 and NAG5-11111.The authors acknowledge the availability of the SGP99data, which is available through the SGP99 web site. Inaddition, we acknowledge Dr. L. Luo for providing thevalidation data used in Fig. 4. The authors are gratefulto the NLDAS project, of which the second author is acoinvestigator, for providing the NLDAS model output.We also acknowledge the availability of the OklahomaMesonet data, made available to the SGP99 experimentthrough fundings provided by NASA Land Surface Hy-drology Program and NOAA Office of Global Program.M. Drusch was funded through the German ClimateResearch Project DEKLIM (01LD0006).

REFERENCES

Chang, S. L., J. A. Kong, and L. Tsang, 1980: Radiative transfertheory for passive microwave remote sensing of a two-layerrandom medium with cylindrical structures. J. Appl. Phys., 51,5588–5593.

Charpentier, M. A., and P. M. Groffman, 1992: Soil moisture variabilitywithin remote sensing pixels. J. Geophys. Res., 97, 18 987–18 995.

Cherkauer, K. A., L. C. Bowling, and D. P. Lettenmaier, 2003: Var-iable Infiltration Capacity (VIC) Cold Land Process Model up-dates. Global Planet. Change, 38 (1–2), 151–159.

Choudhury, B. J., T. J. Schmugge, A. Chang, and R. W. Newton,1979: Effect of surface roughness on the microwave emissionfrom soils. J. Geophys. Res., 84, 5699–5706.

——, ——, and T. Mo, 1982: A parameterization of effective tem-perature for microwave emission. J. Geophys. Res., 87, 1301–1304.

Crow, W. T., and E. F. Wood, 1999: Multi-scale dynamics of soilmoisture variability observed during SGP’97. Geophy. Res. Lett.,26, 3485–3488.

——, M. Drusch, and E. F. Wood, 2001: Effects of land-surfaceheterogeneity on AMSR-E soil moisture retrieval. IEEE Trans.Geosci. Remote Sens., 39, 1622–1631.

Dobson, M. C., F. T. Ulaby, M. T. Hallikainen, and M. A. El-Rayes,1985: Microwave dielectric behavior of wet soil—Part II: Di-electric mixing models. IEEE Trans. Geosci. Remote Sens., 23,35–46.

Drusch, M., E. F. Wood, and C. Simmer, 1999: Up-scaling effects inpassive microwave remote sensing: ESTAR 1.4 GHz measure-ments during SGP97. Geophys. Res. Lett., 26, 879–882.

——, ——, and T. J. Jackson, 2001: Vegetative and atmosphericcorrections for soil moisture retrieval from passive microwaveremote sensing data: Results from the Southern Great PlainsHydrology Experiment 1997. J. Hydrometeor., 2, 181–192.

Famiglietti, J. S., and Coauthors, 1999: Ground-based investigationof soil moisture variability within remote sensing footprints dur-ing the Southern Great Plains 1997 (SGP97) Hydrology Exper-iment. Water Resour. Res., 35, 1839–1851.

Ferrazzoli, P., L. Guerriero, and J. P. Wigneron, 2002: Simulating L-band emission of forests in view of future satellite applications.IEEE Trans. Geosci. Remote Sens., 40, 2700–2708.

Jackson, T. J., 1997: Soil moisture estimation using special satellitemicrowave/imager satellite data over a grassland region. WaterResour. Res., 33, 1475–1484.

——, and T. J. Schmugge, 1991: Vegetation effects on the microwaveemission from soils. Remote Sens. Environ., 36, 203–219.

——, and A. Y. Hsu, 2001: Soil moisture and TRMM microwaveimager relationships in the Southern Great Plains 1999 (SGP99)Experiment. IEEE Trans. Geosci. Remote Sens., 39, 1632–1642.

——, ——, and J. R. Wang, 1982: Passive microwave remote sensingof soil moisture under vegetation canopies. Water Resour. Res.,18, 1137–1142.

——, D. M. Le Vine, C. T. Swift, T. J. Schmugge, and F. R. Schiebe,1995: Large area mapping of soil moisture using the ESTARpassive microwave radiometer in Washita’92. Remote Sens. En-viron., 53, 27–37.

——, ——, A. Y. Hsu, A. Oldak, P. J. Starks, C. T. Swift, J. D. Isham,and M. Haken, 1999: Soil moisture mapping at regional scalesusing microwave radiometry: The Southern Great Plains Hy-drology Experiment. IEEE Trans. Geosci. Remote Sens., 37,2136–2151.

Kerr, Y. H., and E. G. Njoku, 1990: A semiempirical model for in-terpreting microwave emission from semiarid land surfaces asseen from space. IEEE Trans. Geosci. Remote Sens., 28, 384–393.

——, P. Waldteufel, J. P. Wigneron, J. M. Martinuzzi, J. Font, andM. Berger, 2001: Soil moisture retrieval from space: The SoilMoisture and Ocean Salinity (SMOS) mission. IEEE Trans.Geosci. Remote Sens., 39, 1729–1735.

Kirdyashev, K. P., A. A. Chukhlantsev, and A. M. Shutko, 1979:Microwave radiation of the Earth’s surface in the presence of avegetation cover. Radio Eng. Electron. Phys., 24, 37–44.

Klein, L. A., and C. T. Swift, 1977: An improved model for thedielectric constant of sea water at microwave frequencies. IEEETrans. Antennas Propag., 25, 104–111.

Koster, R. D., and M. J. Suarez, 2001: Soil moisture memory inclimate models. J. Hydrometeor., 2, 558–570.

Le Vine, D. M., M. Kao, C. T. Swift, A. Griffis, and A. B. Tanner,1990: Initial results in the development of a synthetic aperturemicrowave radiometer. IEEE Trans. Geosci. Remote Sens., 28,614–619.

——, A. J. Griffis, C. T. Swift, and T. J. Jackson, 1994: ESTAR: Asynthetic aperture microwave radiometer for remote sensing ap-plications. Proc. IEEE, 82, 1787–1801.

——, T. J. Jackson, C. T. Swift, M. Haken, and S. W. Bidwell, 2001a:ESTAR measurements during the Southern Great Plains exper-iment (SGP99). IEEE Trans. Geosci. Remote Sens., 39, 1680–1685.

——, C. T. Swift, and M. Haken, 2001b: Development of the synthetic


aperture microwave radiometer, ESTAR. IEEE Trans. Geosci.Remote Sens., 39, 199–202.

Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: Asimple hydrologically based model of land surface water andenergy fluxes for general circulation models. J. Geophys. Res.,99, 14 415–14 428.

——, E. F. Wood, and D. P. Lettenmaier, 1999: Modeling ground heatflux in land surface parameterization schemes. J. Geophys. Res.,104 (D8), 9581–9600.

Miller, D. A., and R. A. White, cited 1998: A conterminous UnitedStates multilayer soil characteristics dataset for regional climateand hydrology modeling. Earth Interactions, 2. [Available onlineat http://EarthInteractions.org.]

Mitchell, K., and Coauthors, 2000: The collaborative GCIP Land DataAssimilation System (LDAS) project and supportive NCEP un-coupled land-surface model initiatives. Preprints, 15th Conf. Hy-drology, Long Beach, CA, Amer. Meteor. Soc., 1–4.

——, and Coauthors, 2003: The multi-institution North AmericanLand Data Assimilation System (NLDAS) Project: Utilizingmultiple GCIP products and partners in a continental distributedhydrological modeling system. J. Geophys. Res., in press.

Mo, T., B. J. Choudhury, T. J. Schmugge, and J. R. Wang, 1982: Amodel for microwave emission from vegetation covered fields.J. Geophys. Res., 87, 11 229–11 237.

Njoku, E. G., 1977: Theory for passive microwave remote sensingof near-surface soil moisture. J. Geophys. Res., 82, 3108–3118.

——, and D. Entekhabi, 1996: Passive microwave remote sensing ofsoil moisture. J. Hydrol., 184, 101–129.

Owe, M., A. A. Van de Griend, and A. T. C. Chang, 1992: Surfacemoisture and satellite microwave observations in semiarid south-ern Africa. Water Resour. Res., 28, 829–839.

——, ——, R. de Jeu, J. J. de Vries, E. Seyhan, and E. T. Engman,1999: Estimating soil moisture from satellite microwave obser-vations: Past and ongoing projects, and relevance to GCIP. J.Geophys. Res., 104, 19 735–19 742.

Pampaloni, P., and S. Paloscia, 1986: Microwave emission and plantwater content—A comparison between field measurements andtheory. IEEE Trans. Geosci. Remote Sens., 24, 900–905.

Robock, A., and Coauthors, 2003: Evaluation of the North AmericanLand Data Assimilation System over the Southern Great Plainsduring the warm season. J. Geophys. Res., 108, 8846, doi:10.1029/2002JD003245.

Seuffert, G., P. Viterbo, J. F. Mahfouf, H. Wilker, M. Drusch, and J.C. Calvet, 2003: Soil moisture analysis combining screen-levelparameters and microwave brightness temperature: A test withfield data. Geophys. Res. Lett., 30, 1498, doi:10.1029/2003GL017128.

Ulaby, F. T., R. K. Moore, and A. K. Fung, 1986: Microwave RemoteSensing: Active and Passive. Vol. 3, From Theory to Applica-tions. Artech House, 1120 pp.

Wang, J. R., 1983: Passive microwave sensing of soil moisture con-tent: The effect of soil bulk density and surface roughness. Re-mote Sens. Environ., 13, 329–344.

——, and T. J. Schmugge, 1980: An empirical model for the complexdielectric permittivity of soils as a function of water content.IEEE Trans. Geosci. Remote Sens., 18, 288–295.

——, and B. J. Choudhury, 1981: Remote sensing of soil moisturecontent over bare field at 1.4 GHz frequency. J. Geophys. Res.,86, 5277–5282.

——, and ——, 1995: Passive microwave radiation from soil: Ex-amples of emission models and observations. Passive MicrowaveRemote Sensing of Land–Atmosphere Interactions, B. J. Choud-hury et al., Eds., VSP, 423–460.

——, and Coauthors, 1980: Remote measurements of soil moistureby microwave radiometers at BARC test site. NASA Tech.Memo. 80720.

——, and Coauthors, 1982: Remote measurements of soil moistureby microwave radiometers at BARC test site II. NASA Tech.Memo. 83954.

Wegmuller, U., C. Matzler, and E. G. Njoku, 1995: Canopy opacitymodels. Passive Microwave Remote Sensing of Land–Atmo-sphere Interactions, B. J. Choudhury et al., Eds., VSP, 375–387.

Wood, E. F., H. Gao, M. Drusch, T. Jackson, and R. Bindlish, 2003:Soil moisture retrievals over the Southern Great Plains: Com-parisons between experimental remote sensing data and opera-tional products. Proc. IGARSS, Toulouse, France, IEEE, CD-ROM, 03.0801.

Documents

Using a Microwave Emission Model to Estimate Soil Moisture