Available online at www.sciencedirect.com

t 112 (2008) 430–444www.elsevier.com/locate/rse

Remote Sensing of Environmen

Hydrological consistency using multi-sensor remote sensingdata for water and energy cycle studies

M.F. McCabe a,b,⁎, E.F. Wood a, R. Wójcik a, M. Pan a, J. Sheffield a, H. Gao c, H. Su a

a Department of Civil and Environmental Engineering, Princeton University, Princeton New Jersey, 08542, USAb International Space and Response, Los Alamos National Laboratory, Los Alamos, New Mexico, 87545, USAc School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, Georgia, 30332, USA

Received 30 March 2006; received in revised form 28 February 2007; accepted 21 March 2007

Abstract

A multi-sensor/multi-platform approach to water and energy cycle prediction is demonstrated in an effort to understand the variability andfeedback of land surface and atmospheric processes over large space and time scales. Remote sensing-based variables including soil moisture(from AMSR-E), surface heat fluxes (from MODIS) and precipitation rates (from TRMM) are combined with North American RegionalReanalysis derived atmospheric components to examine the degree of hydrological consistency throughout these diverse and independenthydrologic data sets. The study focuses on the influence of the North American Monsoon System (NAMS) over the southwestern United States,and is timed to coincide with the SMEX04 North American Monsoon Experiment (NAME). The study is focused over the Arizona portion of theNAME domain to assist in better characterizing the hydrometeorological processes occurring across Arizona during the summer monsoon period.Results demonstrate that this multi-sensor approach, in combination with available atmospheric observations, can be used to obtain acomprehensive and hydrometeorologically consistent characterization of the land surface water cycle, leading to an improved understanding ofwater and energy cycles within the NAME region and providing a novel framework for future remote observation and analysis of the coupled landsurface–atmosphere system.© 2007 Elsevier Inc. All rights reserved.

Keywords: Remote sensing; Satellite; Hydrology; Hydrometeorology; Climate dynamics; Feedback; Atmospheric processes; Multi-sensor; Data assimilation;Evapotranspiration; Soil moisture; AMSR-E; TRMM; MODIS; Land surface temperature; Hydrological consistency; Hydrological cycle; North American MonsoonSystem; NAMS; SMEX; NAME

1. Introduction

Documenting the global water and energy cycle throughmodeling and observations is a primary goal of the World ClimateResearch Programme's (WRCP) Global Energy and WaterExperiment (GEWEX). Such documentation is needed to enableenhanced understanding of the Earth's climate, including charac-terizing the memories, pathways and feedbacks between key waterand energy cycle (WEC) components. With such enhancedknowledge, there is the potential to develop “improved, observa-tionally based predictions of water and energy cycle consequences

⁎ Corresponding author. International Space and Response, Los AlamosNational Laboratory, Los Alamos, New Mexico, 87545, USA.

E-mail address: mmccabe@lanl.gov (M.F. McCabe).

0034-4257/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2007.03.027

of Earth variability and change” (Belvedere et al., 2005); a centralgoal of theNASAEnergy andWater cycle Study (NEWS) (NASA,2004) and a needed research objective for understanding andquantifying the regional impacts of potential climate shifts.

With NASA's Earth Observing System (EOS), and similarprograms by ESA in Europe and JAXA in Japan, there has beena significant increase in space-based observations that canadvance our knowledge of the surface water and energybudgets. The inherent GEWEX strategy (as well as NEWS) isbuilt around the retrieval of remotely sensed surface observa-tions and their assimilation into process-resolving land surfacemodeling, since characterizing the surface water and energybudgets through in-situ observations alone is infeasible. Thisstrategy recognizes that large-scale applications of energy andwater cycle models are greatly complicated by difficulties in

431M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

representing climate processes at large scales and the scarcity ofland surface observations.

To address the latter and to help the former, the NASA/NEWSstrategy includes the development and deployment of a space-based energy and water cycle global observing system. Ofparticular relevance to current water and energy cycle observa-tions are the following space-borne systems: (1) the TropicalRainfall Measurement Mission (TRMM) satellite that includesboth a precipitation radar and a Microwave Imager (TMI), whichmeasures surface microwave emissions from 10.65 to 85.5 GHzin nine frequency bands, for the retrieval of precipitation (Nesbittet al., 2004) and for land surfacemonitoring (Bindlish et al., 2003;Gao et al., 2006); (2) the EOS Terra and Aqua platforms that pavethe way to the next-generation NPOESS operational satellites,and have sensors across the visible, near-infrared and microwavefrequency bands that allow for the monitoring and retrieval ofincoming solar radiation, albedo, vegetation properties, surfacetemperature, surface emissivity, precipitation, atmospheric watervapor and aerosols. Depending on the frequency, the retrievedsurface emissivity can be used to estimate surface properties,surface soil moisture and snow properties; (3) the GravityRecovery and Climate Experiment (GRACE), which measuresthe change in total column water mass at continental scales(Tapley et al., 2004); (4) QuikSCAT whose radar scatterometermeasurements have been used to detect freeze–thaw transitions athigh latitudes (Kimball et al., 2004); and (5) other relevantoperational satellites and sensors including TOPEX-POSEIDON,which has been used to retrieve surface water stage (Birkett,1998), the Geostationary Orbiting Environmental Satellite(GOES) for solar radiation (Pinker & Laszlo, 1992), Landsatand EO-1 Earth surface imaging, SSM/I for atmospheric watervapor and precipitation, and the Advanced Microwave SoundingUnit (AMSU) and Atmospheric Infrared Sounder (AIRS) foratmospheric temperature and water vapor profiles.

Currently, research satellites and/or sensors are aimed atmeasuring specific components and/or processes of the globalenergy and water cycles. In fact historically, space agencies havecreated separate teams by sensor, focused on the retrieval and/orvalidation of individual water and energy cycle products. Whensatellite retrieved surface water and energy budget variables arecombined with in-situ data like river discharge, budget closure israrely if ever achieved, demonstrating a basic inconsistencyamong the retrieved variables (Pan & Wood, 2006). Identifyingand diagnosing the degree of hydrometeorological closure inland surface and atmospheric models is a critical and oftenoverlooked component of modeling studies, and is likely asymptom of the focus on separate sensing programs.

The North American Monsoon Experiment (NAME) had ascientific focus of better understanding the hydrometeorologyof the southwestern United States during the summer(monsoon) season, with particular attention “to determine thesources and limits of predictability of warm season precipitationover North America” (see description at http://www.etl.noaa.gov/programs/2004/name/). Of key interest is the role of theland surface, and its memory (Koster & Suarez, 2001), indescribing seasonal to inter-annual variations in the NorthAmerican Monsoon System (NAMS) (Adams & Comrie, 1997;

Higgins et al., 1998). To fully understand the hydrometeorologyand surface conditions of the NAME domain, land surface waterand energy cycle variables retrieved from space observationsmust augment in-situ observations. As part of the NAMEexperiment, a soil moisture experiment (SMEX04) wasundertaken to improve the validation of space-based retrievalsof soil moisture that would be used by NAME scientists tobetter understand the role that surface soil moisture plays inland–atmosphere coupling, convection, and the maintenance ofthe NAMS (Small, 2001; Xu et al., 2004; Zhu et al., 2005).

Towards advancing this goal, this paper provides anevaluation of the hydrometeorological consistency of retrievedsurface water and energy cycle variables from space-bornesensors using observations over the Arizona domain of theNAME campaign (a portion of the Tier 1 study area). Someexamples of consistency which are explored in this paperinclude: occurrence/retrieval of precipitation with a correspondingchange in soil moisture; increased soil water availability affectingthe spatial dynamic of surface flux behavior; and surface statesconsistent with atmospheric boundary properties like the convec-tive triggering potential, Humidity Index or condensation liftinglevel. The retrieved variables include precipitation from theTropical Rainfall Measurement Mission (TRMM), soil moisturefrom the Advanced Microwave Scanning Radiometer (AMSR-E),and sensible heat flux derived from MODIS measurements onboard both EOS Terra and Aqua. Atmospheric variables availablefrom the North American Regional Reanalysis (NARR) project areused to help analyze the consistency between surface water andenergy cycle variables and boundary layer indices that indicate thepotential for convection.

A natural extension of the research presented here is theincorporation of remotely sensed observations into an assim-ilation framework. While there have been numerous studiesattending to the task of assimilating unique variables intohydrological models (Crow &Wood, 2003; Pan &Wood, 2006;Walker & Houser, 2001), to date there has been relatively littlereported on the assimilation of multi-sensor retrieved variablesinto land surface models, especially at regional scales. Whilenot explicitly examined here, we present some discussion on amethodology to incorporate multiple remote sensing observa-tions into a modeling framework. Section 2 discusses the dataand methodology used for the study, with particular emphasison the remote sensing observations and retrievals. Section 3presents the analysis of hydrological consistency amongstremotely sensed hydrologic variables, and also an extension ofthis idea to examine feedback relationships between select landsurface and atmospheric indices. Section 4 concludes with anoverview of the research, a synopsis of the research questionsaddressed and a discussion of future work.

2. Data and methodology

2.1. NLDAS model forcing

The North American Land Data Assimilation System(NLDAS) (Mitchell et al., 2004) is a multi-institutional effortto provide high quality, spatially continuous meteorological

Table 1Input variables for MODIS-based evapotranspiration prediction

Meteorological variables Data sourceIncoming shortwaveradiation (W m−2)

North American Land Data Assimilation System(NLDAS) interpolated to 0.05°, consistent with thesatellite-based data. Data fields were interpolatedusing a nearest neighbor scheme, minimizingsmoothing and averaging of data.

Downward longwaveradiation (W m−2)

Air temperature (°C)Vapor pressure (kPa)Wind speed (m s−1)Wind direction (°)Atmospheric pressure(kPa)

Aerodynamic parameters Brutsaert (1991, 1999)Satellite-based data MODIS Sensor on board EOS Terra and AquaNominal overpass time 11:00 a.m. (Terra) and 2:30 p.m. (Aqua)Resolution (m) 0.05° Climate Modeling Grid (CMG)Land surface temperatureand emissivity

MOD11C1 and MYD11C1

Land cover type MOD12C1 land coverLeaf area index (LAI) VIC land surface modelAlbedo Based on UMD land classification schemeVegetation height Based on UMD land classification schemeVegetation fraction Relationship of Xavier and Vettorazzi (2004)

432 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

forcing data collated from the best available operationalobservations (Cosgrove et al., 2003). A variety of forcing andoutput data suitable for use in land surface and other modelsimulations is accessible from the system, providing a datasource to understand regional scale processes — particularlywhere extensive ground-based records are scarce. A number ofkey variables were utilized from the NLDAS to aid in derivingremote sensing-based retrievals. Primarily, the data includedmeteorological forcings such as precipitation, wind velocity,humidity, atmospheric pressure, air temperature and downwardshortwave and longwave radiation. Depending on the resolutionrequirements of the derived variables (see details below) thenative NLDAS spatial resolution of 0.125° was eitheraggregated (to 0.25°) or interpolated (to 0.05°) using a simplenearest neighbor scheme to minimize interpolation effects.

2.2. Remote sensing observations

2.2.1. Surface heat fluxes derived from MODIS measurementson EOS Terra and Aqua

Estimates of the latent and sensible heat flux were determinedusing the Surface Energy Balance System (SEBS) model (Su,2002). SEBS was developed to predict heat fluxes usingcombinations of satellite Earth observation data and routinelyavailable meteorological forcing. So far, the SEBS model hasbeen tested over a number of land surface types and scales(McCabe &Wood, 2006; Su et al., 2005; Su et al., 2007). Furtherdetails on the physical basis of themodel can be found in the listedreferences, since only a brief description is offered here.

SEBS requires three broad sets of information to enable theestimation of surface fluxes including; a) data on the landsurface condition consisting of surface albedo, emissivity,surface temperature, fractional vegetation coverage and leafarea index, along with height of the vegetation; b) meteorolo-gical data including atmospheric pressure, air temperature,humidity, and wind speed at a reference height and; c) radiationdata comprising downward solar radiation and downwardlongwave radiation. Given the large-scale application of SEBSin this study (SEBS was run across the conterminous UnitedStates), a focus was placed on utilizing large-scale operationaldata sets for flux prediction. The data included a combination ofNLDAS meteorological forcing; 0.05° Climate Modeling Grid(CMG) data from MODIS comprising land cover type(MOD12C1) and land surface temperature and emissivity(MOD11C1/MYD11C1); the University of Maryland vegeta-tion scheme (Hansen et al., 2000) and associated look up tablesfor vegetation structure and surface parameters; and NLDASdatabase on leaf area index (LAI). A summary of the data setsused to run SEBS is presented in Table 1.

SEBS constrains the surface heat flux estimates byconsidering dry-limit and wet-limit conditions, thus differenti-ating the upper and lower boundaries on the sensible heat fluxestimation. For the dry-limit case (i.e. evident in semi-aridenvironments), latent heat is assumed zero due to the limitationof soil moisture, and the sensible heat flux is at its maximumvalue (limited by the available energy— the difference betweenthe net radiation and ground heat flux). Under the wet-limit

case, the evaporation takes place at a potential rate (λEwet) (i.e.the evaporation is limited only by available energy, under thegiven surface and atmospheric conditions), and the sensible heatflux takes its minimum value (Hwet), which can be estimatedusing a Penman–Monteith parameterization (Monteith, 1981).The dry- and wet-limit constraints allow an expression for theactual evapotranspiration (ET) to be formulated from knowl-edge of these bounds.

Here, surface temperature and emissivity data from theMODISsensor on board EOS Terra (∼11 a.m.) and Aqua (∼2 p.m.) wereused to provide sensible heat flux estimates approximately twotimes per day (dependent on cloud cover, which limits theavailability of the MODIS measurements).

2.2.2. Soil moisture from the Advanced Microwave ScanningRadiometer (AMSR-E)

Near surface soil moistures were derived from the descendingorbit of the AMSR-E satellite between Aug. 1–31, 2004. Earlymorning overpasses of the AMSR-E satellite (around 2 a.m.)were chosen to minimize the influence of daytime surfacetemperature spatial variability and fluctuations (prevalent in thesemi-arid Arizona environment) and also to maximize thepotential of observing the precipitation footprint from theprevious afternoons thunderstorms.

The Land Surface Microwave Emission Model (LSMEM)(Drusch et al., 2004; Gao et al., 2004) was used to estimate soilmoisture from 0.25×0.25° 10.65 GHz brightness temperatures,which were combined with ancillary information required by themodel. LSMEM employs an iterative technique to estimate thesoil moisture. In the first step, a modeled brightness temperatureis calculated, corresponding to an estimated moisture (based onantecedent conditions), early morning air temperature(NLDAS), land cover characteristics derived from MODISdata, and model parameters describing the atmosphere and land

433M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

surface condition. Successive model iterations are performed byvarying the estimated soil moisture, until convergence betweenthe calculated and observed horizontal brightness temperature isachieved. Gao et al. (2004) and McCabe et al. (2005a) providefurther details on model implementation and a description ofLSMEM parameterizations.

Applications of LSMEM over the Southern Great Plains(Gao et al., 2003, 2004) and Iowa (during SMEX02) show thatthe model provides soil moisture retrievals within 4% vol./vol.when compared to coincident ground-based and air-bornemeasurements (McCabe et al., 2005a). The soil moistureretrievals have been observed to be consistent with precipitationpatterns for selected periods (McCabe et al., 2005b). LSMEMmakes assumptions in estimating the soil moisture that havebeen shown to hold true over the sparse vegetation represen-tative of the Arizona region (Jackson et al., 1995, 1999).

2.2.3. Precipitation from the Tropical Rainfall MeasurementMission (TRMM)

Satellite-based estimates of precipitation provide the onlyviable means to determine rainfall distributions over large regionsof the Earth that lack sufficient in-situ gauge measurement. Assuch, they form a particularly useful data set for advancing globalhydrometeorological observations. Data from the TRMM 3B42merged high quality infrared precipitation product (Huffmanet al., 1995) were used to map rainfall patterns and amounts overthe study domain. The TRMM-based product (http://trmm.gsfc.nasa.gov/) provides 3-hourly 0.25×0.25° gridded estimates ofglobal precipitation, over a latitudinal range of ±50°, derived froma variety of satellite and other data sources.

TRMM rainfall estimates are produced in a number ofstages. First, available high quality passive microwave datafrom the TRMM Microwave Imager, as well as SSMI andAMSR-E, are calibrated and converted to precipitationestimates using probability matching of precipitation ratehistograms derived from coincident data. For the infraredcalibrations, merged infrared data provided to the ClimatePrediction Centre (CPC) from a variety of satellites (e.g.NOAA/GOES), are averaged to 0.25° resolution and combinedinto hourly files. Microwave and infrared estimates are thencombined to provide a ‘best’ estimate of rainfall in any 3-hourperiod. Monthly rescaling to observed records of precipitation isthe final step in producing the merged product. All availablemicrowave and infrared estimates are summed over a calendarmonth, producing a monthly multi-satellite product. Availablegauge (in-situ) data are combined with the multi-satelliteproduct to create a post real-time monthly satellite–gaugecombination, which is ultimately used to rescale the 3 hourlyfields, providing the instantaneous precipitation rate at thenominal observation time (see detailed description of the 3B42algorithm at http://trmm.gsfc.nasa.gov/3b42.html).

2.3. Derived atmospheric variables

Data from the North American Regional Reanalysis (NARR)were used to predict two atmospheric indices: the ConvectiveTriggering Potential (CTP) and the Humidity Index (HI). These

variables are used to assess the consistency of the surface flux,soil moisture and precipitation, with respect to diagnosing land–atmospheric coupling. Formal definitions of both indices areprovided below.

2.3.1. Convective Triggering Potential (CTP)The CTP represents a measure of the temperature lapse rate

taken between 100 and 300 mb above the land surface (i.e.approximately 900 mb and 700 mb respectively, assumingsurface pressure at 1000 mb) and provides insight into theboundary layer response to surface flux development. Thetemperature lapse rate offers a means for determining theatmospheres capacity for entrainment, and hence boundary layergrowth. Findell and Eltahir (2003a) describe the CTP as “thearea between the observed temperature sounding and a moistadiabat originating at the observed temperature, 100 mb abovethe surface”. The formulation can be expressed as follows:

CTP ¼Z z100 mb

z300 mb

gðTparcel � TenvÞ

Tenvdz ð1Þ

where z is the pressure level above the surface, g is gravity,Tparcel is the parcel temperature, which is the moist adiabat that isdefined by the observed temperature at 100 mb, and Tenv is thetemperature of the environment. Fundamentally, the formulationfor the CTP follows that of the Convective Available PotentialEnergy (CAPE), a NARR data product, with the primarydifference being the interval over which the integral is taken andthe fact that the parcel temperature profiles can be quite differentsince the CAPE parcel is lifted from the surface, as opposed tothe CTP parcel 100 mb above the surface.

In terms of relative values of CTP and their physicalconsequence, the CTP is large when the lapse rate is close todry adiabatic (∼10 °C/km). When the lapse rate is closer to moistadiabatic (∼5 °C/km), a smaller but still positive CTP is resolved.A negative CTP identifies a temperature inversion in theatmosphere, indicating that the atmosphere would be too stablefor the development of rain. The positive degree of CTP alsodetermines whether sensible heat (higher positive CTP) or latentheat (lower positive CTP) provides a convective advantage.

NARR vertical profile data from 6 a.m. local time (13 UTC)are used in calculating the CTP, and also the Humidity Index(HIlow) which is described below.

2.3.2. Humidity Index (HI)The Humidity Index (HIlow) is defined in Findell and Eltahir

(2003a) as the sum of the dew-point depressions 50 and 100 mbabove the ground surface:

HIlow ¼ ðT950 � Td;950Þ þ ðT850 � Td;850Þ ð2Þ

where Tp is the temperature at pressure level p, and Td,p is thedew-point temperature at pressure level p. HIlow is a generalizedvariation of the Lytinska et al. (1976) original definition as thesum of the dew-point depressions at 850, 700 and 500 mb abovethe surface, and is derived to be appropriate for use in all regions.Findell and Eltahir (2003a) use the index (in combination with the

434 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

CTP) to predict the likelihood of a precipitation event occurringin a particular location, based on early morning soundings of theatmosphere.

In the original version of the Humidity Index (Lytinska et al.,1976), a threshold for rain was established at HI≤30°. In thegeneralized version employed here, the value is reduced toHIlow≤15°. As will be seen in Section 3, rainfall events asdetermined from TRMM data are observed to occur for valuesslightly above this limit. Here, the indices serve as a proxy forassessing whether remote sensing observations of the landsurface (soil moisture and sensible heat flux) impact thedeveloping atmosphere. Further, they may provide a direct linkto illustrate land surface–atmosphere feedback. Indirectly, theyalso serve to validate (through TRMM observations) thecapacity of the NARR-based measures to predict precipitationevents.

NARR fields are based on the National Centers forEnvironmental Prediction (NCEP) operational Eta model.Results have a nominal spatial resolution of 32 km, providedata across 45 vertical layers and return information with atemporal resolution of 3 h. To collocate fields with the AMSR-Ederived soil moisture and TRMM precipitation data, the derivedvariables were resampled to conform to a 0.25° regular grid.Data were interpolated using a nearest neighbor technique,minimizing possible smoothing of the original NARR data.

3. Hydrological consistency and atmospheric feedbackrelationships in observed variables

3.1. Hydrological consistency between remotely sensed data

There exists considerable uncertainty in quantifying the rolethat land surface–atmosphere feedbacks have on NAMSprecipitation. In previous analysis of the NAMS (e.g. Higginset al., 1998; Hu & Feng, 2002; Zhu et al., 2005), studies havetended to focus on output from numerical weather predictionmodels to characterize the nature of the system. To date there hasbeen limited effort to undertake a similar assessment usingcurrently available satellite retrievals— a consequence of a lackof coincident data and lack of focus on integrated retrievals as partof broader energy and water cycle system. The remote sensingdata assembled here provide a unique opportunity to characterizeconsistency and feedback relationships in the NAMS system,using available hydrological observations and derived atmo-spheric variables collected as part of the NAME campaign.

Central to assessing hydrologic consistency, is the inherentdifficulty in validating remote sensing measurements; in partbecause the ground ‘validation’ data represents different spatialscales and, often, different measurements (e.g. gravimetric soilmoisture versus soil emissivity). Generally, there exist fewavenues to robustly evaluate remote observations. Thoseavailable include validation/evaluation through: 1) intensivefield campaigns that are both expensive and spatially/tempo-rally constrained (e.g. HAPEX/MOHIBLY, FIFE/BOREAS,SMEX campaigns, etc.) (Jackson et al., 2004; Kanemasu et al.,1992; Kustas et al., 2005); 2) distributed networks of in-situpoint measurements (e.g. AmeriFlux, CEOP) (Barlow et al.,

1998; Gao et al., 2006; Western et al., 1999), which are limitedby their spatial representativeness; and 3) comparison withmodel output. While valuable insights have resulted from suchcomparisons, each approach has particular restrictions makingtheir use difficult for a thorough evaluation of remote sensingobservations and retrievals.

Our assessment is that the inconsistency in the spatial andtemporal scales between the remote sensing observations andthe measurements available to evaluate them is a majorunresolved problem. The outputs from land surface models,forced by observations, approach the necessary time and spacescales similar to remote sensing measurements, but are limitedby resolution differences (models are necessarily coarse overlarge scales, compared with multi-resolution satellite data) andissues of instantaneous satellite measurements versus timeaveraged model output. Also, model output is possibly the leastdesirable source of evaluation information given concerns aboutthe ability to model land surface processes, and the landscape,comprehensively.

Here, a simple approach towards assessing hydrologicalconsistency is presented, outlining a qualitative examination ofmulti-sensor/multi-platform equivalence. To identify hydrolog-ical consistency within the independent remote sensingretrievals, a number of conditions should be satisfied. Withrespect to the remote sensing variables developed here, thesecriteria include:

a. The surface soil moisture condition (wet or dry) should relateto precipitation (or lack thereof) in the hours preceding theAMSR-E observation;

b. The sensible heat flux (H) estimated from MODIS, shouldcorrespond to the AMSR-E moisture condition e.g. reducedH for an increased soil moisture anomaly; high (ormaintained) H for dryer areas;

c. The wet surface condition identified from AMSR-E (andassociated precipitation) should show strong spatial correla-tion with the early morning HIlow and CTP relationsidentified in (Findell & Eltahir, 2003a): assuming thatthese indices can skillfully predict the atmospheric state;

d. Knowledge of the spatial correlation among HIlow and CTPand AMSR-E should provide information on the nature offeedback relations between the surface and the atmosphereboundary layer.

The following sections explore these criteria by analyzingthe degree of hydrological consistency across three indepen-dently derived remote sensing hydrological data sets: precip-itation, surface soil moisture and surface heat flux. These datasets are also combined with the NARR derived atmosphericvariables (Section 2.3) to provide insight into the nature of theland–atmosphere feedback mechanisms across the Arizonaregion of the NAME domain.

3.1.1. AMSR-E soil moisture and TRMM rainfallThe AMSR-E 10.65 GHz radiometer can be used to retrieve

near daily global soil moisture at resolutions suitable forregional water balance studies. To distinguish between wet and

Fig. 1. Comparison of the AMSR-E soil moisture anomaly for identified rain days during August 2004, over the Arizona domain of the NAME campaign. TRMMrainfall amounts (instantaneous rain-rate) are shown for 3, 6, 9 and 12 h preceding the AMSR-E overpass (∼2 a.m.), so as to capture the afternoon convective processestypical of the NAMS. AMSR-E soil moisture anomaly is the difference between daily values and the monthly average, with dark areas indicating a positive (wet)anomaly.

435M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

Fig. 2. Percentage of pixels satisfying a) AMSR-E soil moisture anomaly greaterthan some value x and having TRMM cumulative rainfall (24 h) greater thanvalue y; and b) TRMM cumulative rainfall (24 h) greater than some value y andhaving an AMSR-E soil moisture anomaly greater than value x.

436 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

dry regions and their potential impact on the NAME regionhydrometeorology, the AMSR-E soil moisture data are recast interms of their anomaly from the monthly average soil moisture.For all available overpasses in August, an average soil moistureestimate was determined for each pixel (32×28 0.25° pixelsacross Arizona). The AMSR-E soil moisture anomaly was thencalculated by subtracting the monthly mean soil moisture fromthe daily satellite observations (i.e. wetter than average moistureproduces a positive anomaly). Analyzing results in this wayallows for the differentiation of consistently wet or dry areasfrom those responding to particular precipitation events. To seekconsistency and validation of the AMSR-E soil moistureanomaly, TRMM derived rainfall amounts were processed for3, 6, 9 and 12 h preceding the 2 a.m. AMSR-E overpass time(descending orbit). It is expected that the time span ofprecipitation should be sufficient to capture the afternoonstorms characteristic of the NAMS.

Fig. 1 details selected positive AMSR-E soil moistureanomaly events (wet soil) distributed throughout the month ofAugust observations. A clear coherence between wet (dark)areas and coincident precipitation is evident throughout thetemporal distributions of TRMM data. In the majority of cases,there is also a strong degree of spatial correlation, particularlyevident in the pattern of Aug. 12, where two localizedanomalies are represented by an equivalent structure in theTRMM precipitation (see also Aug. 2, 5, and 7). In otherinstances, components of the AMSR-E soil moisture anomaliesare well represented, while others are less so. On Aug. 14 and 16for instance, there is a strong agreement between AMSR-E soilmoisture anomaly and TRMM rainfall patterns in the southernportion of the image. However, precipitation identified in otherareas of the TRMM image fail to leave a corresponding imprinton the surface (see central western precipitation in −12 hourTRMM image for Aug. 14). While surface and land coverrelated influences affecting portions of the image cannot beruled out (e.g. complex topography and partial forest cover), itis expected that using a soil moisture anomaly rather thanabsolute values will aid in removing some of these artifacts. Itshould be noted that while AMSR-E has been validated overheavy agricultural vegetation (McCabe et al., 2005a), theinfluence of partial forest cover is less well understood.

When the converse applies (i.e. wet AMSR-E soil moistureanomaly but no corresponding TRMM rainfall, as in the Aug.16 AMSR-E soil moisture anomaly for central easternArizona), these are possibly evidence of rainfall occurringprior to the 12-hour time frame presented here (see Fig. 4). Theprecipitation process is both highly variable in space and time,spanning a broad range of possible values and intensities(light–heavy rain/short–long duration). On the other hand,while soil moisture is also spatially variable, it has a smallertime varying component (temporally stable) and its value isbounded by soil physical properties (soil water potential andporosity), limiting the range expected from remote observation.Given the competing natures of the hydrological processes,there would seem to be a stronger physical basis for observingAMSR-E soil moisture anomalies with no observed rain event(e.g. Aug. 16), than TRMM data showing significant rainfall

but illustrating no corresponding change in soil moisture (e.g.Aug. 14). The contention is strengthened by considering thatinstantaneous rainfall rates are only available at 3-hourlyintervals, increasing the possibility of missing rain eventsduring this time. Indeed, further investigation of the Aug. 16soil moisture anomaly verified a precipitation event in theTRMM rainfall data for Aug. 15 (06 UTC), before the foursuccessive rain events considered here (see Fig. 4).

While strong spatial coherence is evident in the patterns ofAMSR-E soil moisture anomaly and TRMM rainfall shown inFig. 1, quantification of this agreement would facilitate someverification of these patterns. Fig. 2 illustrates two separateexperiments designed to illustrate the level of agreementbetween the AMSR-E and TRMM derived variables. Here wedetermine the percentage of pixels (averaged over the six daysin Fig. 1) that satisfy a) AMSR-E soil moisture anomaliesgreater than some value x, while also having TRMM cumulative

437M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

rainfall (24 h total) greater than some value y; and b) TRMMcumulative rainfall (24 h total) greater than some value y, andalso having an AMSR-E soil moisture anomaly greater thansome value x. For example, in Fig. 2a, 75.6% of pixels have asoil moisture anomaly greater than 4.0% vol./vol. and a TRMM24-h rainfall total greater than 0 mm (i.e. when the soil is wet,there is good agreement that it rained). In Fig. 2b, only 46.9% ofpixels with TRMM 24-h rainfall greater than 0 mm show acorresponding AMSR-E anomaly greater than 4% vol./vol.,indicating that TRMM rainfall is not necessarily a goodindicator of subsequent soil moisture anomaly. On the otherhand, in pixels where the TRMM rain-total is high (greater than10 mm in 24 h), 88.2% show a corresponding soil moistureanomaly of greater than 4% vol./vol., suggesting that heavy rainevents are better represented in the soil moisture record.

These figures are also illustrative of expected trends in the soilmoisture–rainfall relationship. For instance, in Fig. 2a, pixelswithlower soil moisture anomalies are less likely to be coupled withhigh rain-totals (greater than 10mm), than are higher soilmoistureanomalies (i.e. greater than 20% vol./vol.). Indeed, the averagepercentage across the six days for this scenario indicates that it is 3times less likely to occur (10.8% comparedwith 33.3%). Also, thelevel of pixel-to-pixel agreement in Fig. 2a, generally improvesas the soil moisture anomaly increases relative to fixed rain-totals(y-axis), with the exception of 3 runs (91.5, 62.7 and 61.9%)which produce small deviations (∼1–2%) around an otherwiseincreasing trend. Likewise in Fig. 2b, as the rain-total decreaseswith fixed soil moisture anomaly (x-axis), so does the level ofagreement between pixels.

There are a number of difficulties in providing a thoroughstatistical analysis between the different remotely sensedproducts developed here. While there are inevitable scalemismatches between some variables (e.g. 25 km AMSR-E soilmoisture and 5 km MODIS evapotranspiration) which compli-cate the task, even where scales are equivalent, spatial mismatchregularly occurs due to temporal sampling (e.g. 3-hourlyinstantaneous rainfall, 12-hourly soil moisture), resulting inmisrepresentation of storm cells and their subsequent observedfootprint on the ground. Also, as noted previously, the processesexamined here operate on different time scales and span avariety of fast- and slower-components of the hydrologicalcycle. Antecedent conditions also affect the level of agreementi.e. a storm cell imprint is likely to be more clearly observed ona drier rather than a recently wetted surface.

Table 2Pearson correlation coefficient r comparing pixels of increasing AMSR-E soilmoisture anomaly with collocated TRMM precipitation for each of the six daysillustrated in Fig. 1

AMSR-E N4% N8% N12% N16% N20%

Day 2 0.77 (149) 0.76 (62) 0.71 (40) 0.50 (31) 0.40 (24)Day 5 0.28 (302) 0.08 (100) 0.12 (47) 0.02 (17) 0.03 (12)Day 7 0.22 (419) −0.06 (131) −0.28 (46) −0.47 (11) 0.71 (5)Day 12 0.29 (250) 0.15 (62) 0.04 (30) 0.18 (18) 0.17 (12)Day 14 0.49 (411) 0.57 (225) 0.71 (113) 0.69 (61) 0.65 (43)Day 16 0.03 (584) 0.08 (222) 0.02 (134) 0.07 (90) 0.17 (59)

The number of pixels (in brackets) satisfying each criteria represents a fractionof the 32×28 (896 pixels) study domain.

Nevertheless, a comparison between the AMSR-E soilmoisture anomalies and the TRMM rainfall (Fig. 1) is presentedin Table 2. The Pearson correlation coefficient r, is calculatedbetween pixels with a soil moisture anomaly greater than a rangeof values and the collocated cumulative TRMM rainfall for the24 h preceding the AMSR-E overpass. As opposed to the analysispresented in Fig. 2, r values are determined on a daily basis. Ascan be seen, there is a varied level of agreement between the daysconsidered for analysis. Correlation values range from very good,with a high of 0.77 on Day 2, to near zero and even negativecorrelation on Day 16 and Day 7 respectively. The results varygreatly depending on the level at which the AMSR-E soilmoisture anomaly is considered. While no direct comparison canbe made between these varying levels, it would seem that onlyDay 2 and Day 14 display consistent agreement between AMSR-E soil moisture anomaly and TRMM rainfall totals. Day 16presents a consistent near zero correlation, even thoughqualitative comparison from Fig. 1 indicates that some agreementshould exist. Day 7 regularly displays negative correlation,contrary to the clear presence of raining TRMM cells in thevicinity of observed AMSR-E soil moisture anomalies.

Determining a robust means of characterizing the level ofagreement between distinct, yet physically linked processesderived from remotely sensed imagery, remains a considerablechallenge. Additional work is clearly required to bridge the gapbetween qualitative and quantitative analysis before the utilityof these data can be fully exploited.

3.1.2. AMSR-E soil moisture and MODIS-based sensible heatflux

AMSR-E soil moisture anomalies exhibit strong visualcoherence with TRMM precipitation totals upon examination ofrainfall observations over the preceding 12 h. In terms ofhydrological consistency, the soil moisture–precipitation rela-tionship should also be evident in remote measures of thesurface heat fluxes. Using the MODIS sensor on board Terraand Aqua, insight into the changing spatial patterns of surfacefluxes can be observed. Measurements of the sensible heat inthe morning and afternoon following the AMSR-E soil moistureanomaly were combined (Fig. 3) to characterize spatial fluxdistribution.

For each of the AMSR-E soil moisture anomalies, there isconsiderable agreement with the available sensible heatpredictions. Darker areas in Fig. 3 (last column) representregions of low sensible heat (increased evaporative fraction).Regions where no data exist are the result of cloudcontamination of the infrared bands on MODIS. Microwavefrequencies (i.e. AMSR-E) are not subject to the sameatmospheric influences as infrared bands, and incompletecoverage in infrared data is common — particularly whenexamining periods in the vicinity of precipitation events. BothAug. 2 and 5 display increased evaporation (relative to otherareas) in the southeast corner, corresponding well to significantrain events in Fig. 1. Likewise, Aug. 7, 14 and 16 patterns arestrongly related with soil moisture distributions. The spatial fluxfrom Aug. 12 is masked by cloud cover over the rainfall areasfrom Fig. 1. However, rainfall occurring in the 3 h following the

Fig. 3. From left to right: Comparison of the AMSR-E soil moisture anomaly (left column) for identified rain days with; 6 a.m. Humidity Index (HIlow) of the previousday; HIlow for 6 a.m. sounding following the AMSR-E overpass; Convective Triggering Potential (CTP) of the previous day (6 a.m.); and combined map of MODIS-based sensible heat flux for Terra and Aqua.

438 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

Fig. 4. Hydrological consistency between AMSR-E, MODIS sensible heat andTRMM precipitation results for periods not covered in Figs. 1 and 3. Dataillustrate MODIS derived sensible heat for Aug. 12 together with TRMMprecipitation occurring 3 h after the AMSR-E overpass (top). Also shown is anAMSR-E derived soil moisture anomaly for Aug. 16 matched with TRMMprecipitation from 24 h prior to the overpass (bottom).

439M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

AMSR-E overpass (not included in Fig. 1) indicates excellentspatial agreement with the observed flux pattern (see Fig. 4).

While all land surface variables examined here qualify as‘fast’ components of the hydrological cycle, their degree of timestability differs, decreasing from soil moisture (relativelypersistent), to surface fluxes (less persistent), to the highlytime variable precipitation. While it might be expected thatrainfall imparts a footprint onto the near surface soil moisture,extracting a corresponding signal from the surface flux is notnecessarily assured. The AMSR-E soil moisture anomalyrepresents (at best) the top 1–2 cm of the surface layer, andhas been observed (Fig. 1) to illustrate a strong response toprecipitation. However, the evaporative process accounts fortranspiration from vegetation, evaporation from the canopy andalso soil storage. While direct evaporation from the near surfacemoisture storage is likely (see Fig. 3), vegetation andbiophysical influences may hinder a similar consistency inremote observation in areas other than semi-arid environments.

3.2. Consistencies and feedback between the land surface andatmosphere

Considerable spatial coherence and hydrological consistencybetween independent remote sensing data was observed in theprevious analysis. However, these studies do not directlyaddress the occurrence of land–atmospheric feedbacks; anothermeasure of coherence. For this, a proxy for the atmosphericcomponent of the water and energy cycles that can describe thelink to the surface state is required.

Humidity indices (described in Section 2.3) serve as indicatorsof the atmospheric state, characterizing conditions when theatmosphere may be too dry for the development of precipitation.However, if considered separately, the indices' ability to do sounder more humid conditions is not as robust (Mueller et al.,1996). Findell and Eltahir (2003a) suggest that when theConvective TriggeringPotential (CTP) is coupledwith aHumidityIndex (HIlow), an increased ability to identify regions of convectionfrom early morning atmospheric soundings can be achieved.

Unique criteria used to describe the likelihood of precipitationusing HI–CTP relationships have been developed (Findell &Eltahir, 2003a) and include:

• HIlow≥15° or CTPb0 J kg−1: atmosphere very dry and verystable so rainfall cannot occur;

• HIlow≤5° and CTPN0 J kg−1: atmosphere very humid andunstable, so rainfall can occur over wet and dry soils;

• 5°≤HIlow≤15° and CTPN0 J kg−1: the land surface cansignificantly influence the likelihood of rainfall. Here,rainfall over dry soils is more likely for high CTP–highHIlow values, while rainfall over wet soils is more likely inthe low CTP–low HIlow pairs.

Spatially distributed HIlow and CTP data are plotted in Fig. 3together with AMSR-E soil moisture anomaly and MODISsensible heat flux. HIlow values from 6 a.m. soundings for boththe morning preceding the AMSR-E overpass, and alsoimmediately after (approximately 4 h) are included for

comparison. CTP data from the previous morning soundingare also presented.

The HIlow displays considerable spatial agreement with theAMSR-E soil moisture anomaly, with low values of the index(HIlow≤15°) particularly well correlated with subsequentAMSR-E soil moisture anomalies (and hence rainfall distribu-tions). The ability of the HIlow to predict future (+1 day) rainfallevents is considerable. Particularly strong relationships areevident on Aug. 5, 7 and 15, with slightly reduced levels onAug. 2, 12 and 14 (satisfying 15°≤HIlow≤20°). In terms of theCTP data for Arizona, values routinely exceed 1000 J kg−1,which are considerably greater than ranges identified in Findelland Eltahir (2003a). Differences are likely due to disparitybetween the station data used in the analysis of Findell andEltahir (2003a), and the model assimilated NARR output valuesdeveloped here. Overall, for the NARR derived CTP, theprevailing response is rarely less than 400 J kg−1, conditionsassociated with lapse rates approaching dry adiabatic andindicating a convective advantage favoring sensible heat flux.

These results are seemingly in conflict with the criteriaoutlined above and the HIlow–CTP distributions noted in Fig. 13of Findell and Eltahir (2003a). As observed from Fig. 3, theHIlow values for soundings after the AMSR-E soil moisture

Fig. 5. From left to right: AMSR-E soil moisture anomaly for select rain days together with Humidity Index (6 a.m. same day) discriminated for HIlowb15° (whitecircles) and 15°bHIlowb20° (blue circles). TRMM observed precipitation events occurring 9, 12, 15 and 18 h after the 2 a.m. AMSR-E overpass are shown to assesswhether positive or negative feedback can be characterized from knowledge of land surface and atmospheric states. (For interpretation of the references to colour in thisfigure legend, the reader is referred to the web version of this article.)

440 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

anomaly routinely indicate rainfall potential over wet areas —in apparent contradiction to the correspondingly high values ofCTP in the same areas, which should favor negative feedback(i.e. rainfall over dry soils). Only data for August 16 indicatesconvection favoring latent heat, with regions on the southeastborder with New Mexico illustrating CTP values b300 J kg−1.Since the CTP values used here were calculated independentlyof previous works, no direct comparison is possible. Regardless,the data illustrate that HIlow is an effective predictor ofprecipitation in semi-arid environments. In terms of describing,or at least identifying feedback, the HIlow–CTP relationshipwould indicate that the land surface has a significant influenceon triggering potential precipitation events. However, charac-

terizing the dominant mode of this mechanism in Arizona isbeyond the scope of this research, given the limited number ofincidents with which to base conclusion.

One means of establishing periods of either positive ornegative feedback though, is through examining the HIlow dataand its ability to predict subsequent rain events. General analysisof the southwest (see Fig. 2 in Findell & Eltahir, 2003b) suggeststhat the region is dominated by negative feedback — althoughArizona is also overlapped by an atmospherically controlledregion. Atmospherically controlled regions describe a range ofpossible land surface–atmosphere interactions, from very dryand/or very stable atmospheres (no surface influence) toconditions that might provoke rainfall over any land condition

441M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

(wet or dry). In mid-latitude regions of the continental US,D'Odorico and Porporato (2004) demonstrated through atheoretical and experimental analysis, that summer precipitationis significantly influenced by local recycling; with the finding thatsoil moisture anomaly affected the probability of rainfalloccurrence. To identify whether the data here favor one or anotherstate, HIlow values and TRMMdata from 11 a.m. to 11 p.m. of thesame day were analyzed, with results plotted in Fig. 5. Byidentifying soundings immediately following AMSR-E soilmoisture anomaly days (Fig. 1), one can possibly discriminatebetween positive and negative feedback — since both states arerepresented (i.e. as opposed to just using dry days). Ranges ofHumidity Index were plotted, with HIlow values less than 15 °Cshown in white circles and 15°≤HIlow≤20° presented in blue.

Results from Fig. 5 present some interesting trends.Immediately apparent is the strong link between wet areas(positive AMSR-E soil moisture anomaly) and patterns of theHumidity Index. That the index would relate the potential ofrainfall occurring over the previous days rain event, suggestssome degree of positive feedback. However, unless rainfall doesindeed fall upon these regions, the atmospheric component ofthe feedback is not fulfilled. Through examination of thesubsequent TRMM data for the afternoon and eveningfollowing the morning sounding, one can observe the relativelygood prediction of precipitation corresponding to the location ofHIlow. While precipitation does indeed occur over many of thewet areas, it is possibly equally represented over dry areas. Ofprimary interest though, is the fact that the HIlow is convincinglyassociated with AMSR-E soil moisture anomaly values —which were also accurately predicted from the previousmornings sounding (Fig. 3) — indicating that the remotesensing measurements of the land surface do indeed havesignificant consistency with model derived atmosphericvariables.

4. Summary of hydrological consistency and feedback

In order to increase the accuracy in modeling land surfaceprocesses, one can develop more physically descriptiverepresentations, increase the accuracy of measurements, orimprove knowledge through observation of variables notcurrently observed. Here, a month of coincident remotelysensed observations of the hydrological cycle was examined fortheir consistency among component variables. Independentretrievals of soil moisture, sensible heat flux and precipitationwere assessed in terms of their spatial correlation and shown toexhibit significant agreement. In particular, the spatial distribu-tion of TRMM-based precipitation illustrated good agreementwith AMSR-E-based soil moisture anomalies for August 2004,a period coinciding with the NAME–SMEX campaign. Spatialdistributions of sensible heat flux were also well represented inthe rainfall–soil moisture relationship, displaying a significantresponse to moisture forcing. Increased confidence in thecapacity to monitor individual hydrological variables emergedas a consequence of their high level of agreement.

The signature of soil moisture is strongly represented in thesemi-arid Arizona environment, due in part to the intermittent

nature of rainfall and the high evaporative demand of theatmosphere. During the NAMS, these conditions are not alwaysevident, with increased frequency of rainfall and a humidatmosphere not uncommon. Still, a clear link betweenhydrological components is observed in the remote sensingdata. Previous studies demonstrated a significant AMSR-E soilmoisture–rainfall link over more heavily vegetated environ-ments (McCabe et al., 2005b). In such conditions, it might notbe expected that a consistent relationship between the soilmoisture and surface fluxes would exist, since the vegetationlayer acts to diminish and mask this response. Rather, theevaporative response is more likely a function of vegetationtype, particularly when moisture availability is not limited(McCabe & Wood, 2006). However, the ability of remotesensing products to detect hydrological response over a varietyof surface types and conditions indicates their utility for broaderscale application.

Identifying possible feedback relationships between hydro-logical variables was undertaken using NARR-based atmo-spheric indices. Using NARR derived variables to describe thestate of the early morning atmosphere, indices were correlatedwith both past and future precipitation events and subsequentsoil moisture anomalies, offering a pathway towards charac-terizing the degree of feedback within the NAMS system. Astrong correlation was observed between the Humidity Indexand AMSR-E-based soil moisture anomalies for key rainfallevents during August 2004. Further, the Humidity Index wasobserved to persist over wet regions, identifying a possiblepositive feedback mechanism during the summer monsoonperiod over Arizona. An extended analysis of this data overlonger time periods is required to elucidate more clearly thenature of this feedback response, particularly since CTP dataindicated a persistent convective advantage favoring sensibleheat (i.e. negative feedback) — an outcome that was notcompletely supported by coincident remote sensingobservations.

Knowledge of the spatial distributions of individualobservation should allow for improved characterization of allobservations, given the physical link between the processes.However, determining the most appropriate means to utilize therich spatial information present in remote sensing data requiresadditional investigation. Multi-objective model calibrationtechniques (Franks et al., 1998; Gupta et al., 1998; McCabeet al., 2005c) offer one means towards achieving this —although how the spatially distributed information availablefrom remote sensors can be best integrated into such frame-works requires additional research (see McCabe et al., 2005d).Data assimilation schemes, on the other hand, are particularlysuited to incorporating the spatial context of remote sensing datainto land surface models in order to improve their predictionaccuracy. However, these approaches also suffer from limita-tions due to incomplete equivalence in modeled and observedvariables. Recent work using statistically based relationshipsbetween modeled and observed states shows some promise inaddressing this issue (Gao et al., 2007). Furthermore, themajority of assimilation exercises have focused on theincorporation of single variable observations into modeling

442 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

frameworks as opposed to utilizing multiple remotely sensedvariables (see Pan et al., in press for an exception).

However, assimilation represents only one means ofimproving process description. The ability of distributedmodeling approaches to accurately predict, and therefore gainknowledge about land surface processes, relies greatly on thecorrect specification of rainfall amount and location. Furtherdevelopment of relationships such as those observed hereshould lead to improvements of individual hydrologicalcomponents through knowledge of physically related processes.For instance, spatial mismatch between the AMSR-E data andthe TRMM rainfall (excluding uncertainties in the geo-registration of both data) is a likely indicator of spatialmisrepresentation in the rainfall distributions. Likewise, themagnitude of the soil moisture anomaly has been observed tocorrelate well with rainfall amount. As a result, it would seemthat merging these products would allow for improved spatialdescription of the precipitation patterns and increased accuracyof the rainfall amount. Clearly, further work is required tofurther elucidate these observations.

5. Conclusion

A fundamental consideration in establishing an integrativeobservation and modeling approach to water and energy cyclesystems is understanding the capacity to which: 1) independentobservations of the hydrological cycle display consistency amonglinked surface and atmospheric processes; and 2) observedvariables can be merged into a framework designed to enhanceknowledge of Earth-climate dynamics. The North AmericanMonsoon System (NAMS) provides an ideal test-bed underwhich these concepts can be developed and, through the NAMEand SMEX programs, allows for direct insight into the interactionwith land surface and atmospheric processes.

Numerous modeling-based studies have sought to explorethese relationships, focusing on longer temporal periods andspatial extents than are explored here (e.g. Higgins et al., 1998;Hu & Feng, 2002; Zhu et al., 2005). In most of these and similarworks on the NAMS, soil moisture, surface heat fluxes andprecipitation data consist almost exclusively of output fromnumerical weather prediction (NWP) or similar model results.Here, remote sensing observations of surface and atmosphericvariables were analyzed for their hydrological consistency,offering the potential to explore feedback relationships in amore realistic manner than is available through model outputalone. These data represent a unique attempt to characterizeconsistency and feedback relationships within available remotesensing-based hydrological observations and derivable atmo-spheric variables.

Distinguishing hydrological consistency in remote observa-tions is a critical and needed research objective. Currently, acomprehensive or robust framework for integrating multi-sensor, multi-scale remote observations for hydrologicalprediction does not exist. While data assimilation approachesshow promise, the key issue is not one of simply developingmore efficient merging techniques. On one hand, the issue iscomplicated due to an incomplete understanding of both the role

of observational error in modeling and also in accuratelyestimating the magnitude and variability of this. On the otherhand, the predictive skill of hydrologic state ensembles given avariety of error models and remotely sensed information needsto be properly quantified. The initial step in this researchdirection was taken by Crow et al. (2005).

Hydrological consistency is fundamentally an effort to seekimprovement in hydrological prediction, through ensuringconcurrence between unique observations of water and energycycle budget variables. Implementing diverse data and observa-tions into a predictive framework requires the formulation of aholistic modeling philosophy — in contrast to relying solely onsingle process, single variable oriented approaches. While insightinto individual phenomena is no doubt gained through process-based approaches, improved understanding lies in integratingdiverse components, thereby allowing a more complete under-standing of dynamic coupled systems to develop. The analysispresented here offers considerable promise in the ability of remoteobservations to accurately and consistently monitor variations inthe land surface and atmospheric states, and demonstrates ameansthroughwhich diverse information can be effectively integrated toexplore coupled land–atmosphere interactions.

Acknowledgements

Research was funded by NASA project grants 1)NNG04GQ32G: A Terrestrial Evaporation Product UsingMODIS Data; 2) NAG5-11111 Land Surface Modeling Studiesin Support of AQUA AMSR-E Validation; and 3) NAG5-11610:Evaluation of Hydrologic Remote Sensing Observations forImproved Weather Prediction. The NARR derived atmosphericvariables HIlow and CTP, were kindly produced by FrancinaDominguez of the Department of Civil and EnvironmentalEngineering, University of Illinois-Urbana: her effort is greatlyappreciated.

References

Adams, D. K., & Comrie, A. C. (1997). The North American Monsoon. Bulletinof the American Meteorological Society, 78(10), 2197−2213.

Barlow, M., Nigam, S., & Berbery, E. H. (1998). Evolution of the NorthAmerican monsoon system. Journal of Climate, 11(9), 2238−2257.

Belvedere, D. R., Houser, P. R., & Schiffer, R. (2005). Current capabilities andpotential deficiencies of NASA energy and water cycle study (NEWS)product and discovery proposals and other water cycle related activities.GEWEX Conference, Irvine, CA June.

Bindlish, R., Jackson, T. J., Wood, E., Gao, H. L., Starks, P., Bosch, D., et al.(2003). Soil moisture estimates from TRMMMicrowave Imager observationsover the Southern United States. Remote Sensing of Environment, 85(4),507−515.

Birkett, C. (1998). Contribution of the TOPEX NASA radar altimeter to theglobal monitoring of large rivers and wetlands. Water Resources Research,34(5), 1223−1239.

Brutsaert, W. (1991). Evaporation into the Atmosphere — Theory, History andApplications, Netherlands: Kluwer Academic Publishers.

Brutsaert, W. (1999). Aspects of bulk atmospheric boundary layer similarityunder free-convective conditions. Reviews of geophysics, 37, 439−451.

Cosgrove, B. A., Lohmann, D.,Mitchell, K. E., Houser, P. R.,Wood, E. F., Schaake,J. C., et al. (2003). Real-time and retrospective forcing in the North AmericanLand Data Assimilation System (NLDAS) project. Journal of GeophysicalResearch, 108(D22).

443M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

Crow, W. T., Bindlish, R., & Jackson, T. J. (2005). The added value of spacebornepassive microwave retrievals for forecasting rainfall–runoff ratio partitioning.Geophysical Research Letters, 32(18), 1−5.

Crow, W. T., & Wood, E. F. (2003). The assimilation of remotely sensed soilbrightness temperature imagery into a land-surface model using ensembleKalman filtering: A case study based on ESTARmeasurements during SGP97.Advances in Water Resources, 26, 37−149.

D'Odorico, P., & Porporato, A. (2004). Preferential states in soil moistureand climate dynamics. Proceedings of the National Academy of Sciences,101(24), 8848−8851.

Drusch, M., Wood, E. F., Gao, H., & Thiele, A. (2004). Soil moisture retrievalduring the Southern Great Plains Hydrology Experiment 1999: Acomparison between experimental remote sensing data and operationalproducts. Water Resources Research, 40(2).

Findell, K. L., &Eltahir, E. A.B. (2003).Atmospheric controls on the soilmoisture–boundary layer interactions. Part 1: Framework development. Journal ofHydrometeorology, 4(3), 552−569.

Findell, K. L., & Eltahir, E. A. B. (2003). Atmospheric controls on soil moisture–boundary layer interactions. Part II: Feedbacks within the continental UnitedStates. Journal of Hydrometeorology, 4(3), 570−583.

Franks, S. W., Gineste, P., Beven, K. J., & Merot, P. (1998). On constrainingthe predictions of a distributed model: The incorporation of fuzzyestimates of saturated areas into the calibration process. Water ResourcesResearch, 34(4), 787−797.

Gao, H., Wood, E. F., Drusch, M., Crow, W., & Jackson, T. J. (2004). Using amicrowave emission model to estimate soil moisture from ESTARobservations during SGP99. Journal of Hydrometeorology, 5(1), 49−63.

Gao, H., Wood, E. F., Drusch, M., & McCabe, M. F. (2007). Copula derivedobservation operators for assimilating TMI and AMSR-E soil moisture intoland surface models. Journal of Hydrometeorology, 8(3), 413−429.

Gao, H., Wood, E. F., Drusch, M., McCabe, M. F., Jackson, T. J., & Bindlish, R.(2003). Using TRMM/TMI to retrieve soil moisture over Southern UnitedStates from 1998 to 2002. Eos, Transactions of the American GeophysicalUnion Supplement, 84(46).

Gao, H., Wood, E. F., Jackson, T. J., Drusch, M., & Bindlish, R. (2006). UsingTMI to retrieve surface soil moisture over the United States from 1998–2002. Journal of Hydrometeorology, 7(1), 23−38.

Gupta, H. V., Sorooshian, S., &Yapo, P. O. (1998). Toward improved calibration ofhydrological models: Multiple and non-commensurable measures of informa-tion.Water Resources Research, 4, 751−762.

Hansen,M. C., DeFries, R. S., Townshend, J. R. G., & Sohlberg, R. (2000). Globalland cover classification at 1 km spatial resolution using a classification treeapproach. International Journal of Remote Sensing, 21, 1331−1364.

Higgins, R. W., Mo, K. C., & Yao, Y. (1998). Interannual variability of the USsummer precipitation regime with emphasis on the southwestern monsoon.Journal of Climate, 11(10), 2582−2606.

Hu, Q., & Feng, S. (2002). Interannual rainfall variations in the North Americansummer monsoon region: 1900–98. Journal of Climate, 15(10), 1189−1202.

Huffman, G. J., Adler, R. F., Rudolph, B., Schneider, U., & Keehn, P. (1995).Global precipitation estimates based on a technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitationinformation. Journal of Climate, 8, 1284−1295.

Jackson, T. J., Chen, D., Cosh, M. H., Li, F., Anderson, M. C., Walthall, C., et al.(2004). Vegetation water content mapping using Landsat data derivednormalized difference water index for corn and soybeans. Remote Sensing ofEnvironment, 92, 475−482.

Jackson, T. J., LeVine, D.M., Hsu, A. Y., Oldak, A., Starks, P. J., Swift, C. T., et al.(1999). Soil moisture mapping at regional scales using microwave radiometry:The Southern Great Plains Hydrology Experiment. IEEE Transactions onGeoscience and Remote Sensing, 37(5 Part 1), 2136−2151.

Jackson, T. J., Levine,D.M., Swift, C. T., Schmugge, T. J.,& Schiebe, F. R. (1995).Large-area mapping of soil moisture using the ESTAR passive microwaveradiometer in Washita 92. Remote Sensing of Environment, 54(1), 27−37.

Kanemasu, E. T., Verma, S. B., Smith, E. A., Fritschen, L. G., Wesely, M., Field,R. T., et al. (1992). Surface flux measurement in FIFE: An overview. Journalof Geophysical Research, 97(D17), 15,547−15,555.

Kimball, J. S., McDonald, K. C., Running, S. W., & Frolking, S. E. (2004).Satellite radar remote sensing of seasonal growing seasons for boreal and

subalpine evergreen forests. Remote Sensing of Environment, 90(2),243−258.

Koster, R. D., & Suarez, M. J. (2001). Soil moisture memory in climate models.Journal of Hydrometeorology, 2(6), 558−570.

Kustas,W. P., Hatfield, J. L.,&Prueger, J.H. (2005). The SoilMoistureAtmosphereCoupling Experiment (SMACEX): Background, hydrometeorological condi-tions and preliminary findings. Journal of Hydrometeorology, 6(6).

Lytinska, Z., Parfiniewicz, J., & Pinkowski, H. (1976). The prediction of airmass thunderstorms and hails. Proc. WMO Symp. on the Interpretation ofBroad-Scale NWP Products for Local Forecasting Purposes, Warsaw,Poland, WMO.

McCabe, M. F., Franks, S. W., & Kalma, J. D. (2005). Calibration of a landsurface model using multiple data sets. Journal of Hydrology, 302(1–4),209−222.

McCabe,M. F., Gao, H., &Wood, E. F. (2005). An evaluation ofAMSR-E derivedsoil moisture retrievals using ground based, airborne and ancillary data duringSMEX 02. Journal of Hydrometeorology, 6(6), 864−877.

McCabe, M. F., Kalma, J. D., & Franks, S. W. (2005). Spatial and temporalpatterns of land surface fluxes from remotely sensed surface temperatureswithin an uncertainty modelling framework. Hydrology and Earth SystemSciences, 9(5), 467−480.

McCabe, M. F., &Wood, E. F. (2006). Scale influences on the remote estimationof evapotranspiration using multiple satellite sensors. Remote Sensing ofEnvironment, 105(4), 271−285.

McCabe, M. F., Wood, E. F., & Gao, H. (2005). Initial soil moisture retrievalsfrom AMSR-E: Large scale comparisons with SMEX02 field observationsand rainfall patterns over Iowa. Geophysical Research Letters, 32(L06403).doi:10.1029/2004GL021222

Mitchell, K. E., Lohmann,D.,Houser, P. R.,Wood, E. F., Schaake, J. C., Robock,A.,et al. (2004). The multi-institution North American Land Data AssimilationSystem (NLDAS): Utilizing multiple GCIP products and partners in acontinental distributed hydrological modeling system. Journal of GeophysicalResearch, 109(D7). doi:10.1029/2003JD003823

Monteith, J. L. (1981). Evaporation and surface temperature.Quarterly Journal ofthe Royal Meteorological Society, 107, 1−27.

Mueller, C. K., Wilson, J. W., & Crook, N. A. (1996). The utility of sounding andmesonet data to nowcast thunderstorm initiation.Weather and Forecasting, 8,132−146.

NASA. (2004). Predicting energy and water cycle consequences of earth systemvariability and change. A NASA Earth Science Enterprise ImplementationPlan for Energy and Water Cycle Research. Draft Version 1.6. Retrieved[accessed March 26, 2006].

Nesbitt, S.W., Zipser, E. J.,&Kummerow,C.D. (2004).An examination of version-5 rainfall estimates from the TRMMMicrowave Imager, precipitation radar, andrain gauges on global, regional, and storm scales. Journal of AppliedMeteorology, 43(7), 1016−1036.

Pan, M., & Wood, E. F. (2006). Data assimilation for estimating land waterbudget using a constrained ensemble Kalman filter. Journal of Hydrome-teorology, 7(3), 534−547.

Pan, M., Wood, E. F., Wójcik, R., & McCabe, M. F. (in press). Estimation of theregional terrestrial water cycle using multi-sensor remote sensing observationsand data assimilation. Remote Sensing of Environment.

Pinker, R. T., & Laszlo, I. (1992). Modelling surface solar irradiance for satelliteapplications on a global scale. Journal of AppliedMeteorology, 31, 194−211.

Small, E. E. (2001). The influence of soil moisture anomalies on variability of theNorth American monsoon system. Geophysical Research Letters, 28(1),139−142.

Su, B. (2002). The surface energy balance system (SEBS) for the estimation ofturbulent heat fluxes. Hydrology and Earth System Sciences, 6(1), 85−99.

Su, H., McCabe, M. F., Wood, E. F., Su, Z., & Prueger, J. H. (2005). Modelingevapotranspiration during SMACEX02: Comparing two approaches for localand regional scale prediction. Journal of Hydrometeorology, 6(6), 910−922.

Su, H., Wood, E. F., McCabe, M. F., & Su, Z. (2007). Evaluation of remotelysensed evapotranspiration over the CEOP EOP-1 reference sites. Journal ofthe Meteorological Society of Japan, 85A, 439−459.

Tapley, B. D., Bettadpur, S., Ries, J. C., Thompson, P. F., & Watkins, M. M.(2004). GRACE measurements of mass variability in the Earth system.Science, 305(5683), 503−505.

444 M.F. McCabe et al. / Remote Sensing of Environment 112 (2008) 430–444

Walker, J. P., &Houser, P. R. (2001). Amethodology for initializing soil moisturein a global climate model: Assimilation of near-surface soil moistureobservations. Journal of Geophysical Research, 106(D11), 11761−11774.

Western, A.W., Grayson, R. B., &Green, T. R. (1999). The Tarrawarra project: highresolution spatial measurement, modelling and analysis of soil moisture andhydrological response. Hydrological Processes, 13(5), 633−652.

Xavier, A. C., & Vettorazzi, C. A. (2004). Mapping leaf area index throughspectral vegetation indices in a subtropical watershed. International Journalof Remote Sensing, 25(9), 1661−1672.

Xu, J. J., Shuttleworth, W. J., Gao, X., Sorooshian, S., & Small, E. E. (2004). Soilmoisture–precipitation feedback on the North American monsoon system intheMM5-OSUmodel.Quarterly Journal of the Royal Meteorological Society,130(603 PART B), 2873−2890.

Zhu, C., Lettenmaier, D. P., & Cavazos, T. (2005). Role of antecedent land surfaceconditions of North Americanmonsoon rainfall variability. Journal of Climate,18(16), 3104−3121.