Exploring The Potential for Multipatch Soil-Moisture Retrievals Using ...

1114

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 40, NO. 5, MAY 2002

Exploring The Potential for Multipatch Soil-Moisture Retrievals Using Multiparameter Optimization Techniques Eleanor J. Burke, Luis A. Bastidas, and W. James Shuttleworth

Abstract—This paper explores the potential to retrieve surface soil moisture and optical depth simultaneously for several different patches of land cover in a single pixel from dual polarization, multiangle microwave brightness temperature observations such as will be provided by, for instance, the Soil Moisture and Ocean Salinity (SMOS) mission. MICRO-SWEAT, a coupled land-surface and microwave emission model, was used in a year-long simulation to define the patch-specific soil moisture, optical depth, and synthetic, pixel-average microwave brightness temperatures similar to those that will be provided by SMOS. The microwave emission component of MICRO-SWEAT also forms the basis of an exploratory retrieval algorithm in which the difference between (synthetic) observations of microwave brightness temperatures and modeled, pixel-average microwave brightness temperatures for different input values of soil moisture and optical depth is minimized using the shuffled complex evolution (SCE) optimization procedure. Results are presented for two synthetic pixels, one with eight patches, where only the soil moisture is retrieved, and one with five patches, where both the soil moisture and the optical depth are retrieved. Index Terms—Passive microwave, SMOS, soil moisture, retrieval, vegetation.

I. INTRODUCTION

P

ASSIVE microwave radiometers, preferably operating at L-band (21 cm, 1.4 GHz) but also at C-band (6 cm, 5.6 GHz) frequencies have promise as prospective tools for measuring the moisture of the top 5 cm of the soil [1]–[3], this being a key variable in the hydrologic cycle. Data from many truck- and aircraft-based experiments have been used to evaluate the relationship between microwave brightness temperatures and near-surface soil moisture and to investigate the impact of the soil type, surface roughness, and vegetation canopy (if present) on this relationship [4], [5]. Such research suggests potential for the retrieval of soil moisture over large areas. The prospects for a satellite-based radiometer flown with the specific objective of providing global estimates of remotely sensed soil moisture have recently improved. Operating at C-band, the Advanced Microwave Scanning Radiometer

Manuscript received June 13, 2001; revised February 8, 2002. E. J. Burke was supported by NOAA Project NA96GP0412. L. A. Bastidas was supported by NOAA Project NA86GP0324 and NSF-EAR-9876800. W. J. Shuttleworth was supported by NASA Project NAG5-7554. The authors are with the Department of Hydrology and Water Resources, University of Arizona, Tucson, AZ 85721 USA (e-mail: [email protected]). Publisher Item Identifier S 0196-2892(02)04816-7.

(AMSR) [6] instrument will provide surface soil-moisture estimates under sparse canopies. In addition, the Soil Moisture and Ocean Salinity (SMOS) [7] L-band mission has been selected by the European Space Agency (ESA) for an extended Phase A study in preparation for launch around 2005. As a byproduct of their retrieval, both of these new satellites will also provide estimates of the optical depth of the vegetation, a variable that is indicative of the amount of water in the vegetation canopy. To give context and relevance, in this study synthetic SMOS data for a range of look angles were the basis of the investigation. However, investigation of the feasibility for multipatch retrieval presented could just as well have been framed around any set of microwave brightness temperatures for the same heterogeneous ground area observed at several different look angles. The SMOS mission will consist of a dual-polarized, two-dimensional (2-D) microwave interferometric radiometer operating at 1.4 GHz, which will be launched in a sun-synchronous orbit with a nominal resolution of 50 km and a repeat time of 2–3 days [8]. The instantaneous field of view of the radiometer is two-dimensional, extending both along and across the satellite path. As the satellite orbits, it collects a series of 2-D images that can be analyzed to provide estimates of microwave brightness temperatures for the same pixel at a set of distinct look angles. Many ( 33) look angles are sampled along the path of the satellite, ranging from nadir to 55 . However, the number of angles sampled decreases significantly further from the satellite path, and the range of look angles decreases, with the sampled angles being, in general, further from nadir [8]. By applying a simple model optimization technique with the dual-polarization, multiangle data, the proposed SMOS retrieval algorithm [8] is able to obtain simultaneous estimates of surface soil moisture and vegetation optical depth for the (assumed uniform) land cover in each pixel. This paper investigates whether there is potential for retrieving these variables for several different land covers in each pixel; the extent of land-cover complexity for which multipatch retrieval may be possible is also explored. Previous attempts have been made [9], [10] to obtain high-resolution soil-moisture information from low-resolution microwave brightness temperatures, but these attempts required additional detailed patch-specific information. In this paper, the MICRO-SWEAT model, a coupled land-surface and microwave emission model, is used to predict synthetic microwave brightness temperatures at a range of SMOS-related emission angles for hypothetical pixels with mixed land cover. The microwave emission component of MICRO-SWEAT along

0196-2892/02$17.00 © 2002 IEEE

BURKE et al.: MULTIPATCH SOIL-MOISTURE RETRIEVALS USING MULTIPARAMETER OPTIMIZATION TECHNIQUES

with advanced multiparameter optimization techniques are used to investigate whether there is enough information in these synthetic data for multipatch retrieval of surface soil moisture and vegetation optical depth.

II. MATERIALS AND METHODS A. MICRO-SWEAT The MICRO-SWEAT model developed by Burke et al. [11], [12] merges a model of surface energy and water exchanges with a description of the microwave emission of the whole soil–vegetation–atmosphere interface. The land-surface model, SWEAT, simulates the evolution of soil–water content and temperature by coupling equations that describe their transfer through the soil profile [13]. The thermal properties of the soil are defined as a function of soil type following [14]. Subsurface processes are linked to the atmosphere through an evaporation model which incorporates the surface energy balance and which calculates the latent and sensible heat. In the presence of vegetation, the effect of plants is included in the energy and water balance equations and surface boundary conditions [13]. The profiles of soil–water content and temperature calculated by SWEAT form the basis of the microwave emission calculation. In MICRO-SWEAT, the Wang and Schmugge [15] model is used to estimate the dielectric properties of the soil profile using the soil particle size distribution and the soil profile water content and temperature profiles from SWEAT. The soil dielectric and soil-temperature profiles are used to predict the brightness temperatures at the soil surface using the Wilheit [16] model for coherent propagation of electromagnetic radiation through a stratified medium. Any overlying vegetation will absorb the microwave emission from the soil surface at L-band and, assuming that scattering is negligible, will also contribute its own emission. The brightness temperatures detected by a radiometer are therefore the sum of the emission from the soil, the upward emission from the canopy, and the downward emission from the canopy that is reflected by the soil surface. In addition, the canopy will absorb a proportion of all the energy passing through it. In MICRO-SWEAT, a simple two-parameter model expressed in terms of optical depth and single-scattering albedo is used to account for the effects of vegetation on microwave emission from the soil [17]. The optical depth defines the amount of absorption and emission by the canopy, while the single-scattering albedo determines the importance of scattering relative to absorption within the medium. At L-band, scattering is usually assumed to be small, and the single-scattering albedo is set to zero. Examples of the verification of the MICRO-SWEAT model include comparison of time series of simulated microwave brightness temperatures against both truck- and aircraft-based radiometry measurements for a range of bare and vegetated soils and different look angles [10], [11], [18]–[20]. In these studies, the predicted time series given by MICRO-SWEAT captured both the diurnal variation in brightness temperatures as well as day-to-day changes as the soil surface wets and dries. Typically, the root mean square of the error in predicted brightness temperatures has been found to be less than 5 K.

1115

B. Multiparameter Optimization Techniques This paper discusses use of the single-criterion, multiparameter optimization technique developed at the University of Arizona, namely the shuffled complex evolution (SCE-UA) [21], [22] algorithm, to retrieve both the near-surface soil moisture and vegetation optical depth. In general, a numerical model might have parameters that observations. The distance beneed to be calibrated from model-simulated responses and the observatween the tions is defined by an objective function ( ) such as the root mean square error (RMSE). The goal of model calibration is to find the preferred values for the parameters that minimize within the feasible set of parameters. In a single-criterion optimization, the observations used are only one type. The SCE-UA algorithm is a general-purpose global optimization method designed to handle many of the response surface problems encountered in the calibration of nonlinear simulation models. It randomly samples the feasible parameter space to select a population of points. The population is then partitioned into several “complexes,” each of which evolves independently in a manner based on the downhill simplex algorithm [23]. The population is periodically “shuffled,” and new complexes are formed so that the information gained by previous complexes is shared. As the search progresses, the entire population tends to converge toward the neighborhood of the global optimum value for the objective function. These steps are repeated until prescribed termination rules are satisfied [21], [22]. The SCE-UA has been used mainly to calibrate the values of the parameters in hydrological models against field observations. However, the algorithm is not application-specific, and it can readily be applied to deduce estimates of surface soil moisture and vegetation optical depth (the parameters to be calpatches of different land-cover type that lie ibrated) for the within an individual pixel. In this study, the objective function to be minimized is the RMSE between the synthetic and modeled pixel-average brightness temperatures for all available look angles and both polarizations. C. Multipatch Retrieval MICRO-SWEAT provides the basis for the multipatch retrieval. When calculating the synthetic data, it was assumed that the microwave brightness temperatures for vertically and horizontally polarized microwave emission from patches of vegetation, bare soil, and free water can be realistically calculated using MICRO-SWEAT for the (SMOS-related) range of look angles. These individual microwave brightness temperatures are then added in proportion to the prescribed fractional area of each contributing patch to calculate the synthetic area-average SMOS measurements for the mixed land-cover pixel. The retrieval process also uses the microwave emission component of MICRO-SWEAT. Because the same model is used to calculate both the synthetic data and to retrieve the parameter values, in some experiments a small, random noise with a standard deviation of 0.1 K was added to the synthetic microwave brightness temperatures to simulate measurement errors. It was assumed that the other parameters required during retrieval, namely the soil-particle size distribution, soil and vegetation temperatures,

1116


soil-surface roughness (assumed to be zero in this study), detection frequency, and the proportion of the pixel covered by each land cover, are adequately known from ancillary (modeled or measured) data. The retrieved soil moisture is that for the top 5 cm of soil. However, the microwave emission is partly dependent on the soil moisture at depths below this, and in order to retrieve the correct soil moisture, the soil moisture deeper in the profile must be modified along with the surface soil moisture. Burke et al. [10] demonstrated that this can be done by assuming that the relationship between the surface soil moisture and the deep soil moisture is correctly predicted by SWEAT, and that the deep soil moisture can be adequately parameterized as a simple empirical function of the surface soil moisture from the history of SWEAT simulations. With this assumption, the synthetic pixel-average , at look angle and polarization brightness temperature, , can be taken to be a function of just 2 unknown parameters, namely , the soil moisture in the surface layer of each patch, and , the vegetation optical depth of each patch, the latter being assumed to be independent of both polarization and look angle. Thus

(1) is the brightness temperature of any open water surwhere face present in the pixel calculated using the Wilheit [16] model; is the area of free water present in the pixel; is the total is the area covered by each number of patches in the pixel; is the deep-soil moisture; and are the tempatch; and peratures of the soil and vegetation, respectively; and are the mass fractions of sand and clay, respectively. The unknown parameters in this function, and , can be deduced by minimizing an objective function, , which is the RMSE bedifferent angles and tween the (synthetic) measurements at two polarizations and the value calculated by (1), i.e., by minimizing

(2) In this paper, two example retrievals were made in which a) only soil moisture was estimated, or b) both soil moisture and vegetation optical depth were evaluated for all the patches in the pixel. Because the parameter space in which optimization is made is so complex, the SCE-UA algorithm generally identified several local minima presumably in addition to the true global minimum. Consequently, the set of parameters with the minimum value of the objective function was assumed to be the required set corresponding to the global minimum. In the experiments undertaken in this study, ten optimizations were made, each with a different random start. The likelihood of finding the true global minimum increases with an increased number of replications. Hence, limiting the optimizations to ten contributes to the error in the retrieved values of parameters.

D. Synthetic SMOS Measurements A one-year MICRO-SWEAT run was made for each simulated patch of vegetation using data from the Southern Great Plains 1997 experiment (SGP97 homepage) [24] to force the model. The weighted area-average microwave brightness temperatures were then calculated for the range of polarizations and look angles appropriate to a typical, on-axis SMOS pixel [8], i.e., for 33 angles (some duplicated) in the range 0–55 and two polarizations. These became the synthetic SMOS measurements from which parameters (i.e., patch-specific surface soil moisture and, in some experiments, patch-specific vegetation optical depth) for the patches were retrieved using the SCE-UA algorithm every three days at 6:00 a.m., i.e., for the proposed repeat and overpass time of the SMOS satellite. Results are given for two synthetic pixels. III. RESULTS A. Eight-Patch Pixel The first example pixel had eight patches of equal area, namely, bare sandy soil (92% sand, 3% clay), open water, short grass growing in both sand and clay (15% sand, 55% clay), an agricultural crop growing in both sand and clay, and shrubs growing in both the sand and clay. In this experiment, the vegetation optical depths were specified a priori as approximately one tenth of the leaf area index, and the surface soil moisture of each patch was retrieved. Typical values of the optical depth were selected for each land-cover type (0.1, 0.3, and 0.7 for short grass, crop, and shrub, respectively), and these values held constant throughout the simulation. Within the SCE-UA algorithm, wide fixed limits (0–50%) were set on the allowed range of retrieved surface soil moisture. The contribution of the water patch to the pixel-averaged brightness temperatures was assumed to be well known. Fig. 1 shows the near-surface soil moisture for the remaining seven patches retrieved every three days over a 90-day period. The area-average soil moisture (excluding the water patch) for the pixel is also shown. In each subfigure, the ten possible sets of parameters given by the SCE-UA algorithm are plotted as individual symbols, with the black line linking the sets with the smallest objective function. These preferred (assumed global minimum) values were compared with the “true” values, i.e., the values used to calculate the original area-average brightness temperatures. It is gratifying to see the significant level of agreement between the two. In all cases, the RMSE between synthetic and retrieved soil moisture is less than 4.5%, ranging from 0.5% for the bare sandy soil to 4.5% for shrub growing on the clay soil. The spread of ten possible solutions shown in Fig. 1 is arguably a measure of the precision, with which soil-moisture information can reliably be derived from the available data. The results are consistent with the fact that the microwave brightness temperatures are less sensitive to the soil moisture at higher optical depths than at lower optical depths. This is to be expected, and if the retrieved values were assimilated into a Land Data Assimilation System (LDAS) [25], appropriately different errors would need to be assigned and included in the assimilation process. In a second experiment, the same synthetic SMOS-like observations for the mixed pixel were used, but random errors with


Fig. 1.

1117

Retrieved soil moisture using the SCE-UA optimization methodology for each nonwater patch within the eight-patch pixel and their area-weighted average.

standard deviation of 0.1 K were added to the pixel-average synthetic microwave brightness temperatures to simulate observation errors. Retrievals were repeated for a whole year to keep the optical depth constant. As might be expected, with these extra random observation errors, there was a general deterioration in the accuracy of the retrieved values, with the difference between the retrieved and “true” in the soil–water content ranging between 1.4% for the bare soil patch and 12.8% for the shrub growing in clay soil. The error in the area-average value is 3.1%. The error in the retrieved water content for the entire year is given in Table I. Values are given for five classes of pixel-average soil moisture, each class corresponding to one fifth of the data set, i.e., 24 of the 120 retrievals. As expected, the RMSE increases with increasing optical depth for all the patches. For the patches with the higher optical depth (i.e., crop and shrub) the RMSE is greater when water content is lower; and for the patches with lower optical depth (i.e., bare soil and short grass) the RMSE is less at lower water content. Notwithstanding the loss of accuracy when random errors are added to the synthetic microwave brightness temperatures, these preliminary results are encouraging given that the number of patches in the pixel is fairly large. Moreover, it would be possible to modify the SCE optimization routine to make it more problem-specific, thereby enhancing the potential for extraction of more precise estimates of soil moisture by, for instance, putting limits on the retrieved soil moisture using particle size distribution information.

TABLE I ROOT MEAN SQUARE ERROR (RMSE) IN WATER CONTENT FOR EACH PATCH. THE ERRORS ARE DETERMINED FOR DIFFERENT PIXEL-AVERAGE WATER CONTENTS. EACH GROUP OF PIXEL-AVERAGE WATER CONTENTS CONTAINS A SAMPLE OF RETRIEVALS ON 24 DIFFERENT DAYS

B. Five-Patch Pixel Retrieval of both soil moisture and vegetation optical depth is only possible if there are fewer patches in the pixel. To investigate the potential for retrieving patch-specific values of both parameters, synthetic data were generated for pixels with uniform sandy loam soil (75% sand, 5% clay) made up of five equal patches of bare soil, water, short grass, crop, and shrub. In

1118


Fig. 2. Retrieved soil moisture and vegetation optical depth using the SCE-UA optimization methodology for each nonwater patch within the five-patch pixel and their area-weighted average.

this experiment, an attempt was made to simulate annual vegetation growth in the MICRO-SWEAT forward simulations by assuming that the vegetation water content followed the prescribed annual cycles of leaf area index for each land-cover class specified in the LDAS (LDAS homepage) [25], and that optical depth is proportional to vegetation water content with a constant of proportionality (the opacity coefficient) of 0.095, a typical value suggested by Jackson and Schmugge [17]. In the SCE-UA algorithm, soil-moisture retrieval was allowed in the range 0–50%. To speed the optimization by restricting the parameter-spaced sample, the retrieved optical depth was required to increase in value with increasing canopy amount, i.e., assuming that parameter sets were sampled with shrubs having greater optical depth than crops and crops having greater than short grasses. As in the previous experiment, the brightness temperatures of the water patch were assumed to be well known during the optimization. Fig. 2 shows the time series of retrieved soil moisture and vegetation optical depth for the heterogeneous pixel during this experiment. Again, the ten identified parameter values and the preferred values are shown in each case. Both the changing soil moisture and vegetation optical depth are well tracked at lower optical depths. Table II shows that the RMSE for soil moisture

ranges from 1.8% for bare soil (excellent) to 4.5% for crop (acceptable) to 7.6% for shrub (less good). For optical depth, the RMSE between the synthetic and retrieved optical depth ranges from 0.01 for short grass to 0.07 for shrub, with the RMSE for crop falling somewhere between these two. In general, the optical depths are retrieved with more accuracy than the surface soil moistures. As for the eight-patch synthetic pixel discussed above, an additional experiment was made with retrievals with a standard deviation of 0.1 K added to each synthetic microwave brightness temperature. Table II shows that the errors in the retrieved parameters again increase when compared to the original retrievals, and in the case of the shrub cover, little accurate information is obtained from the retrieval. However, the results for both the crop-covered and bare-soil patches show retrieved parameters with acceptable errors. The optical depth of the vegetation varies slowly, and the effect of exploiting this fact was investigated in the retrieval process (Table II). On the first day, the optimization procedure was repeated 100 (rather than ten) times, and it was then assumed that the optical depth was correctly retrieved. The optical depth was then assumed to change by less than 20% on the subsequent day during the retrieval, and this limit on subsequent


TABLE II ERROR IN WATER CONTENT AND ERROR IN OPTICAL DEPTH FOR THE FOUR DIFFERENT RETRIEVAL PROCESSES: (1) WITH NO ERRORS ADDED TO THE BRIGHTNESS TEMPERATURES (SHOWN IN FIG. 2); (2) WITH AN 0.1 K RANDOM ERROR ADDED TO THE MICROWAVE BRIGHTNESS TEMPERATURE; (3) WITH AN 0.1 K RANDOM ERROR ADDED TO THE MICROWAVE BRIGHTNESS TEMPERATURES AND THE OPTICAL DEPTH CONSTRAINED TO BE LESS THAN 20% FROM THE OPTICAL DEPTH RETRIEVED THREE DAYS PREVIOUSLY; (4) WITH AN 0.1 K RANDOM ERROR AND A 2 K BIAS ADDED TO THE MICROWAVE BRIGHTNESS TEMPERATURES AND THE OPTICAL DEPTH CONSTRAINED TO BE LESS THAN 20% FROM THE OPTICAL DEPTH RETRIEVED THREE DAYS PREVIOUSLY

retrieved values was propagated throughout the year. Table II shows that introducing this restriction on the rate of change gave an improvement in the agreement between the retrieved and “true” parameters. Finally, the whole retrieval process was repeated with an additional (fixed) bias of 2 K added to the synthetic microwave brightness temperatures. This fixed bias was estimated as an extra free parameter in the retrieval process. On the average, there was little loss of accuracy when this additional (bias) parameter was included in the retrieval, as shown in Table II. IV. DISCUSSION AND CONCLUSIONS This exploratory investigation of the potential for using multiangle, dual-polarization, area-average observations of microwave brightness temperatures to deduce patch-specific estimates of soil moisture and vegetation optical depth using multiparameter optimization techniques gives encouraging results. The preliminary results suggest that multiple measurements of area-average microwave brightness temperatures contain sufficient information to allow retrieval of patch-specific soil moisture and vegetation optical depth, providing there are a limited number of patches present and represented in the pixel. Limitations of the present study include 1) the low level of noise included in the synthetic brightness temperatures and 2) the use of the same model in the generation of the synthetic data and the retrievals.

1119

Future research must focus on improving the multipatch retrieval process itself and better determining the conditions and instrumental sampling scheme where retrieval is potentially feasible. For example, when improvements were made in the specification of the potential parameter space within which retrieval was made (e.g., assuming there is a change of less than 20% in the optical depth from one day to the next), more accurate retrievals were made because the chance of finding the global minimum increased. Such restrictions also speed up the optimization procedure. Additional restrictions might be imposed, such as assuming there is a more restricted range of plausible soil–water content related to specific soil type, assuming that the direction of change (increase/decrease) in retrieved soil moisture for all patches is the same as for the patch with least canopy, and assuming the direction of change in retrieved soil moisture since the last retrieval can be predicted from additional ancillary data, e.g., precipitation. Obviously, running the retrieval process more frequently at each time step would also be beneficial. In the present study, for example, the mean value of the retrieved soil–water content for the entire year for each patch (equivalent to carrying out 1200 separate retrievals) is essentially identical to the “true” mean value. However, other additional research questions remain. For example, what accuracy is required in ancillary data such as the soil and vegetation temperatures, the soil particle size distributions, the fractional area of each patch, etc. (all of which are assumed to be exactly known in the current study)? How large can the observation error be before the ability to retrieve multipatch information is lost? How does the accuracy of retrieval degrade with more patches and less look angles? How can the effect of increased sampled area at larger look angles be incorporated into the retrieval process? There are also technical questions relating to the efficiency of retrieval. Multiparameter optimization techniques are computationally expensive: can more efficient artificial neural network methods be devised to extract the same information on the multipatch status? Notwithstanding these critical issues, the central result of the present study remains that there is significant potential to retrieve multipatch estimates of surface soil moisture and vegetation optical depth from L-band radiometric observations, providing these are made at many look angles and both polarizations for the same pixel. ACKNOWLEDGMENT The 1997 Oklahoma Mesonet data used for model forcing came from the SGP97 database. The authors appreciate the editorial assistance provided by C. Thies. REFERENCES [1] T. J. Schmugge, “Applications of passive microwave observations of surface soil moisture,” J. Hydrol., vol. 213, pp. 188–197, 1998. [2] T. J. Jackson, D. M. Le Vine, A. Y. Hsu, A. Oldak, P. J. Starks, C. T. Swift, J. D. Isham, and M. Haken, “Soil moisture mapping at regional scales using microwave radiometry: The Southern Great Plains hydrology experiment,” IEEE Trans. Geosci. Remote Sensing, vol. 37, pp. 2136–2151, Sept. 1999. [3] E. G. Njoku and L. Li, “Retrieval of land surface parameters using passive microwave measurements at 6–18 GHz,” IEEE Trans. Geosci. Remote Sensing, vol. 37, pp. 79–93, Jan. 1999.

1120


[4] T. J. Schmugge, P. E. O’Neill, and J. R. Wang, “Passive soil moisture research,” IEEE Trans. Geosci. Remote Sensing, vol. 24, pp. 12–22, Jan. 1986. [5] J.-P. Wigneron, T. J. Schmugge, A. C. Chanzy, J.-C. Calvet, and Y. H. Kerr, “Use of passive microwave remote sensing to monitor soil moisture,” Agronomie, vol. 18, pp. 27–43, 1998. [6] AMSR Homepage. [Online]. Available: http://www.ghcc.msfc.nasa. gov/AMSR/. [7] SMOS Homepage. [Online]. Available: http://www.cesbio.ups-tlse.fr/ indexsmos.html. [8] J.-P. Wigneron, P. Waldteufel, A. C. Chanzy, J.-C. Calvet, and Y. H. Kerr, “Two-dimensional microwave interferometer retrieval capabilities over land surfaces (SMOS mission),” Remote Sens. Environ., vol. 73, pp. 270–282, 2000. [9] R. H. Reichle, D. Entekhabi, and D. B. McLaughlin, “Downscaling of radio brightness measurements for soil moisture estimation: A fourdimensional variational data assimilation approach,” Water Resources Res., vol. 37, pp. 2353–2364, 2001. [10] E. J. Burke, W. J. Shuttleworth, K. Lee, and L. A. Bastidas, “Using areaaverage remotely sensed surface soil moisture in multi-patch land data assimilation systems,” IEEE Trans. Geosci. Remote Sensing, vol. 39, pp. 2091–2100, Oct. 2001. [11] E. J. Burke, R. J. Gurney, L. P. Simmonds, and T. J. Jackson, “Calibrating a soil water and energy budget model with remotely sensed data to obtain quantitative information about the soil,” Water Resources Res., vol. 33, pp. 1689–1697, 1997. [12] E. J. Burke, R. J. Gurney, L. P. Simmonds, and P. E. O’Neill, “Using a modeling approach to predict soil hydraulic properties from passive microwave measurements,” IEEE Trans. Geosci. Remote Sensing, vol. 36, pp. 454–462, Mar. 1998. [13] C. C. Daamen and L. P. Simmonds, “Measurement of evaporation from bare soil and its estimation using surface-resistance,” Water Resour. Res., vol. 32, pp. 1393–1402, 1996. [14] G. S. Campbell, Soil Physics With BASIC, Transport Models for SoilPlant Systems. New York: Elsevier, 1985. [15] J. R. Wang and T. J. Schmugge, “An empirical model for the complex dielectric permittivity of soils as a function of water content,” IEEE Trans. Geosci. Remote Sensing, vol. GE-18, pp. 288–295, 1980. [16] T. T. Wilheit, “Radiative transfer in a plane stratified dielectric,” IEEE Trans. Geosci. Electron., vol. 16, pp. 138–143, 1978.

[17] T. J. Jackson and T. J. Schmugge, “Vegetation effects on the microwave emission of soils,” Remote Sens. Environ., vol. 36, pp. 203–212, 1991. [18] L. P. Simmonds and E. J. Burke, “Estimating near-surface soil water content from passive microwave remote sensing—An application of MICRO-SWEAT,” Hydrol. Sci. J., vol. 43, pp. 521–534, 1998. [19] , “Application of a coupled microwave, energy and water transfer model to relate passive microwave emission from bare soils to near-surface water content and evaporation,” Hydrol. Earth Syst. Sci., vol. 3, pp. 31–38, 1999. [20] E. J. Burke, W. J. Shuttleworth, and R. C. Harlow, “Modeling microwave brightness temperatures measured during SGP97 using MICRO-SWEAT,” J. Hydrometeorol., 2002. [21] Q. Y. Duan, S. Sorooshian, and V. K. Gupta, “Effective and efficient global optimization for conceptual rainfall-runoff models,” Water Resources Res., vol. 28, pp. 1015–1031, 1992. , “Optimal use of the SCE-UA global optimization method for cal[22] ibrating watershed models,” J. Hydrol., vol. 158, pp. 265–284, 1994. [23] J. A. Nelder and R. Mead, “A simplex method for function minimization,” Comput. J., vol. 7, pp. 308–313, 1965. [24] SGP97 Homepage. The Southern Great Plains ’97 (SGP97) Data Archive. [Online]. Available: http://daac.gsfc.nasa.gov/CAMPAIGN_DOCS/SGP97/sgp97.html. [25] LDAS Homepage. [Online]. Available: http://ldas.gsfc.nasa.gov.

Eleanor J. Burke, photograph and biography not available at the time of publication.

Luis A. Bastidas, photograph and biography not available at the time of publication.

W. James Shuttleworth, photograph and biography not available at the time of publication.