1. Introduction
At regional and global scales, weather and climate are significantly influenced by the availability of soil moisture (Betts et al. 1996), and accurate soil-moisture estimates will improve initialization of atmospheric models. At present, the only practical way of providing such estimates is using Land Data Assimilation Systems (LDAS; information available online at http://ldas.gsfc.nasa.gov.). These are two-dimensional arrays of the land surface parameterization used in the relevant atmospheric model and are driven by available observations, rather than output from the atmospheric model. LDAS often give less than perfect simulations of soil-moisture fields because of poor parameterization (e.g., poorly defined soil properties), unknown initial conditions, low-quality forcing data (e.g., poorly defined precipitation fields), or errors in model physics. The assimilation of microwave brightness temperatures (TB) into LDAS potentially provides a way of improving the specification of these soil-moisture fields and may also stimulate improvements in the parameterization and perhaps structure of the land surface models themselves (Galantowicz et al. 1999; Burke et al. 2001b; Reichle et al. 2001). However, any inaccuracies in the parameterization of the relationship between TB and the near-surface soil moisture will render microwave brightness temperatures less valuable for assimilation.
L-band (1.4 GHz) passive microwave remote sensing can be used to estimate the moisture in the top 5 cm of the soil (Schmugge 1998). Many truck- and aircraft-based field experiments (e.g., Burke et al. 1998; Jackson et al. 1999; Laymon et al. 2001) have been carried out to evaluate the relationship between near-surface volumetric soil moisture (θυ) and microwave brightness temperatures (TB). The Soil Moisture and Ocean Salinity mission (SMOS; Wigneron et al. 2000) will provide global measurements of TB at a set of different angles and is proposed for launch in 2006.
The presence of both vegetation and soil surface roughness impacts the relationship between θυ and TB, with TB being less responsive to θυ as both the amount of vegetation and the roughness increase. Simple models have been proposed to account for these effects (Schmugge et al. 1986; Wigneron et al. 1998; Jackson et al. 1999). In addition, methods have been developed to extrapolate surface information to deeper soil layers by assimilating surface soil moisture into a land surface model (Li and Isham 1999; Calvet and Noilhan 2000; Hoeben and Troch 2000; Burke et al. 2001b). Thus, it is now accepted that coupling L-band passive microwave remote sensing with modeling has the potential to improve the representation of both surface and deeper soil moisture in LDAS.
Coupled land surface and microwave emission models (e.g., Burke et al. 1997, 1998; Judge et al. 1999) will allow the direct assimilation of TB into LDAS, enabling predicted soil-moisture fields to be improved by minimizing any error between the modeled and measured TB (Galantowicz et al. 1999; Burke et al. 2001b; Reichle et al. 2001). Forward modeling of TB using a coupled model incorporates model physics that is neglected in simple retrievals of θυ, including the effect of different near-surface profiles of soil moisture. In addition, it ensures consistency in the parameterizations used within the system.
Such coupled models are usually run at one intensively studied point. A distributed array of coupled models requires spatial distributions of model parameters and model forcing, which are less accurately known. Therefore, an assessment of the performance of these coupled land surface microwave emission models at regional scales is an important preliminary step toward assimilating distributed measurements of TB.
This paper evaluates the ability of a coupled land surface and microwave emission model (MICRO-SWEAT) to predict aircraft-based TB data collected during the Southern Great Plains 1997 field experiment (SGP97). On the basis of the results, the potential usefulness and limitations of directly assimilating observed TB into an LDAS to improve the representation of soil moisture are discussed. For this purpose, the concept of an Assimilation Value Index is introduced.
2. Data and methods
a. Study area
A distributed grid focused around the U.S. Department of Agriculture (USDA) Grasslands Research Center at El Reno, Oklahoma, represents the study area. The grid is 48 cells by 24 cells and has a resolution of 200 m. It extends 9.6 km east of 582140E and 4.8 km north of 3932275N (Fig. 1). The modeled domain is relatively flat: the elevation ranges between 395 and 457 m, with a mean elevation of 417 m (Pennsylvania State University database, available at http://eoswww.essc.psu.edu./dbndx/tree/amer_n/us_sc/sgpr.html). Much of the data used in this paper were collected as part of the Southern Great Plains 1997 experiment (information available from the SGP97 Web site at http://daac.gsfc.nasa.gov./CAMPAIGN_DOCS/SGP97/sgp97.html which took place between 18 June and 17 July 1997.
Two L-band passive microwave radiometers were flown during this measurement period. The Electronically Scanned Thinned Array Radiometer (ESTAR), a one-dimensional synthetic aperture horizontally polarized radiometer, measured TB at a spatial resolution of 800 m during 16 days (Jackson et al. 1999). The Scanning Low Frequency Microwave Radiometer (SLFMR), a vertically polarized push-broom microwave radiometer (Jackson 2001), was flown on three days (29 June, 2 July, and 3 July). The SLFMR was flown at three altitudes, providing TB measurements at three different spatial resolutions (nominally 200, 400, and 600 m). The SLFMR data were cross calibrated with the ESTAR data and, because TB are given at nadir, there should be no significant differences between the two datasets resulting from instrument characteristics (Jackson 2001).
Distributed soil information was derived from the 200-m spatial resolution Map Information Assembly and Display System (MIADS) found on the Pennsylvania State University database. The series name was used to define texture classification, and the USDA texture triangle then used to estimate the particle size distribution. The MIADS description of the distributed grid indicated two soil types: a silty loam covering a majority of the domain, and a small area of loamy sand in the northeast part of the domain (Fig. 1).
Distributed vegetation characteristics were defined from the 30-m SGP97 land cover classification generated from the 25 July Landsat overpass. When resampled to 200 m, the wheat/wheat stubble and rangeland/pasture classes were found to cover 90.5% of the area (Fig. 1). The leaf area index was estimated at 200-m resolution from the Landsat-derived Normalized Difference Vegetation Index (NDVI), which is a measure of the relative vigor of the canopy, using the relationship proposed by Norman et al. (1995) (French et al. 2000). It should be noted that the Landsat overpass occurred 7 days after the last ESTAR flight, leading to possible differences between this image and the actual vegetation present during the 4-week simulation period.
Figure 1 outlines the 15 fields (all with silty loam soil) that were monitored in situ, and highlights the four fields (ER-1, ER-5, ER-13, and ER-15) where there were more intensive measurements. Each field is approximately 800 m × 800 m, that is, one ESTAR pixel. Field-average near-surface volumetric soil water contents were obtained using two different methods. The first method (measuring 0–5 cm) uses 14 gravimetric samples made every 100 m along two transects, each transect separated by 400 m (Jackson et al. 1999). The second method (measuring 0–6 cm for the four highlighted fields only) uses 49 samples collected using Theta Probes (Delta-T Devices) at selected nodes on a 7 × 7, 100-m separation grid (Famiglietti et al. 1999). The value of the field-average volumetric soil water content measured using the first method is highly dependent on the bulk density. Two measurements of bulk density are available, one from the USDA Salinity Laboratory (http://daac.gsfc.nasa.gov/CAMPAIGN_DOCS/SGP97/soil_bd.html), and one from the USDA Hydrology Laboratory (http://daac.gsfc.nasa.gov/CAMPAIGN_DOCS/SGP97/arssl.html). The USDA Salinity Laboratory measurements are typically about 20% larger than the Hydrology Laboratory measurements and hence result in significantly higher values of volumetric soil water content. The USDA Salinity Laboratory measurements were used in this study.
Measurements of the surface energy balance along with relevant surface meteorological variables are available for the duration of SGP97 for three of the four highlighed sites (ER-1, ER-5, and ER-13). In addition, ER-5 contains a Mesonet site (information available at http://okmesonet.ocs.ou.edu.), which provides long-term measurements of surface meteorological variables including wind speed and direction, incoming solar radiation, relative humidity, air temperature, pressure, and precipitation.
b. MICRO-SWEAT
The MICRO-SWEAT model involves sequential coupling of two one-dimensional simulation models. The first is a model of simultaneous heat and water movement in the soil–vegetation–atmosphere system [Soil Water Evaporation and Transpiration (SWEAT); Daamen and Simmonds 1996]. The soil component of SWEAT includes representation of Darcian water flow through the soil matrix, isothermal and thermally driven vapor flow, and the conduction of heat through the soil. When vegetation is present, transpiration and root water uptake are modeled assuming a simple electrical resistance analogue of soil–plant hydraulics. Water flow through the soil–plant system is coupled with the atmosphere via a stomatal resistance that depends on leaf water potential. The link between subsurface and surface processes and the atmosphere is obtained by modeling the latent and sensible heat fluxes from two interacting, evaporating surfaces following Shuttleworth and Wallace (1985).
The outputs from SWEAT directly relevant for modeling microwave emission from the land surface are the vertical distributions of soil temperature and soil water content. These profiles contain information for 24 discrete soil layers extending down to 2.5 m, including 8 layers in the top 5 cm. They are interpolated to 200 layers at the following depth: 0–1 cm, 10 layers at 1 mm; 1–3 cm, 10 layers at 2 mm; 3–6 cm, 10 layers at 3 mm; 6–10 cm, 10 layers at 4 mm; 10–20 cm, 20 layers at 5 mm; and 20–160 cm, 140 layers at 10 mm. The profile of the soil dielectric constant required for input into the emission model is derived from this interpolated soil water content profile using the Wang and Schmugge (1980) semiempirical mixing model. This model takes account of the proportions of free and bound water, air and soil solids within the soil medium, and their relative dielectric constants (real parts ∼80, 4, 1, 3, respectively). The amount of water that is bound within the soil medium is a function of the percentage of clay solids. The Wilheit (1978) model of coherent radiative transfer in stratified media predicts the microwave intensity emergent at the soil surface from the vertical distributions of soil temperature and dielectric constant.
Recent papers by Burke et al. (1997, 1998) and Simmonds and Burke (1998, 1999) provide examples of the point verification of MICRO-SWEAT, in which the predicted time courses of TB were successfully compared with truck-mounted radiometer measurements. Typically, the root-mean-square error in predicted TB is less than 5 K. The present paper is concerned with applying MICRO-SWEAT to TB data collected from an aircraft. Consequently, potential errors are larger because the field conditions are much less well controlled, and there is a possibility of georeferencing errors.
c. Derivation of the vegetation optical depth
Knowledge of the optical depth of vegetation cover is crucial in MICRO-SWEAT because it specifies the effect of the vegetation on microwave emission from the soil. Optical depth is usually assumed to be proportional to the vegetation water content, but the constant of proportionality, the opacity coefficient, is uncertain and ranges between 0.05 and 0.15 (Kerr and Wigneron 1994). Since the optical depth and vegetation-specific opacity coefficients were neither measured nor known, the optical depth was derived using the relationship between surface soil moisture and TB (Burke and Simmonds 2001, 2002).
Burke and Simmonds (2001, 2002) show average near-surface (0–5 cm) soil moisture as a simple semiempirical function of TB. This function was derived from multiple simulations of MICRO-SWEAT under a wide range of conditions, and only requires information on the soil particle size distribution, vegetation optical depth, instantaneous air temperature, and mean annual air temperature. It assumes that the effect of soil surface roughness is negligible. Burke and Simmonds (2001) discuss the sensitivity of the TB to the various input parameters. They determined that the accuracy of the optical depth was a crucial component of this function. Therefore, it was assumed that all of the information required by the function, except the vegetation optical depth, was accurately known, and the function was solved for optical depth for the 15 fields outlined in Fig. 1 by minimizing the difference between the time courses of measured and modeled TB.
d. The Assimilation Value Index
One potential application of a coupled land surface and microwave emission model, such as MICRO-SWEAT, is in the assimilation of TB to improve estimates of the soil water content. The probability distribution of likely errors in TB can be used to estimate the probability distribution of equivalent near-surface soil water content estimated using measurements of TB, and hence determine the weight given in the assimilation procedure to the difference between measured and modeled TB. In this case study, the primary and most easily quantifiable source of errors in the calculation of TB is in the estimation of the optical depth, although errors in soil type, surface roughness, and effective temperature are also likely.
3. Results
a. Single-site MICRO-SWEAT model simulations
The purpose of this single-site calibration exercise was to check the model performance using the four more intensively studied field sites, before applying it to the larger domain where there are few measurements.
The four field sites labeled in Fig. 1 were used for the single-site calibration of MICRO-SWEAT. All of these sites have silty loam soil (there were no sites available on the sandy loam soil). Two sites (ER-13 and ER-15) contained senescent, then harvested winter wheat, and the other two (ER-1 and ER-5) contained pasture/rangeland. For the purposes of the modeling, it was assumed that the wheat was senescent throughout the experiment. Consequently, distinguishing between senescent wheat and wheat stubble has minimal effect on the model simulations. Neither senescent wheat nor stubble transpires; senescent wheat is only ∼30 cm taller than stubble and has a small additional shading effect; and both have very low vegetation water content.
Table 1 summarizes the vegetation and soil properties used for these single site simulations and subsequently over the whole domain. The single-site simulations were initialized with uniform soil water potential and temperature profiles on day of year (DOY) 152, 25 days before the first ESTAR overpass. The Mesonet site located in ER-5 provided the model forcing data. The soil hydraulic properties, that is, the parameters describing the soil water release curve and the saturated hydraulic conductivity (ks) in SWEAT, were estimated from the particle size distribution and pedotransfer functions developed by Cosby et al. (1984) and were assumed to be constant for the entire soil profile. During calibration the saturated hydraulic conductivity was decreased by a factor of 10 at all four of the field sites. This improved the comparison of the time courses of modeled surface soil moisture (0–5 cm) with Theta Probe (0–6 cm) and gravimetric measurements (0–5 cm) particularly just after precipitation, by slowing the drying.
Figure 2 shows the calibrated time series of modeled and measured near-surface soil moisture. For all four sites, there is reasonably good agreement between the model and measurements. The modeled soil moisture deviates from measurements by significantly less than one standard deviation. Table 2 shows the rmse between model and measurements is approximately 4% water content. There are two exceptions. The Theta Probe measurements in ER-5 are significantly higher than either the model or gravimetric sampling. There might be some differences expected between the Theta Probe measurements (0–6-cm water content) and the gravimetric sampling (0–5-cm water content) because of the different depths sampled, although, under most conditions, these differences will fall within the measurement error. The discrepancies between the gravimetric and Theta Probe measurements in ER-5 are likely either related to poor specification of bulk density or to poor calibration of the Theta Probe data. The rmse between model and gravimetric sampling for ER-13 (Table 2) is also poor (6%). Figure 2 shows these differences mainly during the second half of the experiment after the field was plowed (DOY 186). The plowing likely resulted in modified soil properties that would affect both model and measurements. The remaining 11 sites monitored during SGP97 were used as validation sites. Table 2 shows the rmse between model and gravimetric measurements was 4.9% with a bias of 1.3%. These errors are within acceptable limits.
Figure 3 shows the modeled TB at the four single-point calibration sites compared with observations taken from both the multiresolution SLFMR data and the 800-m-resolution ESTAR data. The rmse and bias between the model and all of the measurements (both SLFMR and ESTAR) are also given in Fig. 3. The optical depths used in these examples were the field-specific values found from the 16-day time courses of ESTAR observed TB using the procedure outlined in section 2c. Considering that aircraft data are prone to geolocation error, Fig. 3 shows excellent agreement for the 800-m ESTAR data. However, the multiresolution SLFMR data agree less well. The SLFMR observed TB are apparently strongly resolution-dependent, particularly in fields ER-1 and ER-15 (Burke et al. 2001a; Burke and Simmonds 2002). This resolution dependence results in a significant degradation of the rmse and bias. Figure 3 also shows an underestimation of the TB during the latter half of the experiment at the ER-13 site. This is likely, attributable to the change in soil properties associated with plowing on DOY 186 at that site—the water content is overestimated and hence TB is underestimated. There is also an increase in surface roughness, which is not accounted for in the model, which leads to an additional underestimation of TB.
b. Distributed MICRO-SWEAT model simulations
MICRO-SWEAT was run for each cell within the 4.8 km by 9.6 km model domain, that is, 1152 times for the 200-m-resolution grid and 72 times for the 800-m-resolution grid. Lateral interactions between processes in each grid cell were assumed to be negligible. This is an LDAS-type approach to modeling distributed land surface processes. Both distributed forcing and input parameters are required for the model simulations.
1) Distributed forcing data
Little distributed forcing data is available. Short-term flux stations monitored several of the sites, but did not measure all of the forcing data required, or for long enough to enable sufficient spinup before the start of the ESTAR flights. However, there is only a small change in elevation (∼62 m) across the model domain, which will have a minimal effect on atmospheric conditions (e.g., <0.5°C change in air temperature). Therefore the measurements at the Mesonet site (ER-5) were taken to be representative of the whole domain. Assuming one measure of precipitation is representative certainly decreases the reliability of the model simulations, but there are only two gauge measurements within the domain and radar-based estimates have a minimum resolution of 4 km (resulting in just ∼two grid squares across the domain).
2) Distributed soil parameters
The model domain mainly comprised a silty loam soil (92%), and all of the 4 calibration and 11 validation sites overlay silty loam soil. Therefore, no information is available on the characteristics (and resulting effect on model performance) of the remaining 8% of the model domain, which comprised sandy loam soil. For lack of better information it was therefore assumed that the whole domain comprised silty loam soil. The initial soil moisture and temperature profiles were assumed to be land cover–dependent, and drier for the wheat sites than for the rangeland sites. The soil parameters given in Table 1 were used in the distributed modeling.
3) Distributed vegetation parameters
The calibration sites are either senescent/harvested wheat or rangeland. These land cover types account for 90.5% of the domain. Consequently, the domain was assumed to comprise wheat (no differences between the characteristics of wheat and wheat stubble were accounted for because time of harvest is unknown for most fields), rangeland, or water. The vegetation heights given in Table 1 were used for the distributed modeling. The distributions of LAI (described in section 2b) and optical depth (described later) at the relevant resolution were obtained from the NDVI, where the linearly averaged Landsat radiances were used to estimate the NDVI.
Figure 4 shows the relationship between optical depth derived for each of the 15 outlined sites using the procedure discussed in section 2c and log(1 − NDVI). Figure 4a shows the relationship using the 200-m SLFMR data and Fig. 4b shows that using the 800-m ESTAR data. In each case the data used (the Landsat-TM radiances, TB, and the measured 0–5-cm volumetric soil moisture) were aggregated to the field scale. A linear function can be fitted to these relationships. In the case of the SLFMR TB, 78% of the variation in the optical depth is explained by variation in the NDVI (Fig. 4a). This is a strong dependence despite the assumptions that the relationship between TB and water content is a correct representation of the relationship between surface soil moisture and TB; all other parameters, including the measured near-surface water content, are well defined; and surface roughness is negligible. Figure 4b shows that only 58% of the variation in the optical depth derived from the 800-m ESTAR observed TB is explained by variation in the NDVI. This degradation is likely because of the influence of vegetation outside the fields (Burke and Simmonds 2002). Figures 4a,b also show the 66% confidence intervals of the fitted relationship, corresponding to one standard deviation of the optical depth, approximately ±0.08.
The 15 fields have NDVI ranging between 0.31 and 0.75. This is comparable to the range of NDVI for the entire domain (0.26 to 0.81). All of these fields are classified as either rangeland or wheat—these land cover types represent 90.5% of the grid. Arguably it is reasonable to assume the relationships derived in Fig. 4 can be extrapolated to estimate the optical depth for the entire area (Fig. 5). The distribution of optical depth shown in Fig. 5 agrees well with the land cover classification (see the SGP97 Web site noted earlier). Potential errors in the optical depth at each grid location were estimated by assuming that the true optical depth follows a normal distribution with a mean given by the best-fit line and a standard deviation given by the half-range confidence interval shown in Fig. 4.
4) Distributed model results
Distributions of near-surface soil moisture and TB were simulated at both 200-m (SLFMR) and 800-m (ESTAR) resolution. Table 3 shows the difference between area-average modeled and measured TB is generally less than 2 K. However, the rmse between model and measurements on a point-by-point basis is significantly larger—of the order 15 K—indicating that there are some large discrepancies between the spatial distributions.
The relationship between TB and soil moisture is highly sensitive to the value of the optical depth. For example, given an average value of soil water content (25%) and an average value of the optical depth (0.3), an error of 0.05 in the optical depth results in an error of approximately 18 K in TB (Burke and Simmonds 2001). Thus one of the main reasons for the difference in spatial distributions is the scatter in the relationship between optical depth and NDVI.
Figure 6 shows a qualitative comparison between the spatial distributions of the 200-m-resolution model results and the SLFMR data. The western third of the image, which contains mostly wheat, has relatively lower modeled and measured TB, while the remainder of the image has both higher predictions and measurements. A sewage treatment works to the east of the images was not modeled (white area in Fig. 6). There are features in the measured data not present in the modeled data, and vice versa. For example, an area with low TB in the northeastern corner of the modeled image is not present in the measurements. Thermal/infrared observations indicate that this is an area of wheat with low NDVI (Fig. 5), similar to the ER-15 site, but it is not apparent in TB. Similarly, in the southwestern corner, there is an area of rangeland with high NDVI that is not present in the TB measurements. One possible reason for these discrepancies is that the NDVI image was taken on 27 July, after the field experiment ended, and some NDVI values may have changed over the course of the study period. Other possible reasons include lack of detailed knowledge about the spatial distribution of changes in elevation, topography, soils, and precipitation, and also poor estimation of the optical depth.
4. Discussion
Assuming that the microwave emission model in MICRO-SWEAT adequately represents the relationship between near-surface water content and TB, the discrepancies between the spatial distributions of modeled and measured TB observed in this study are mainly due to inaccurate inputs to the microwave emission model. The microwave emission model is most sensitive to the spatial fields of estimated vegetation optical depth and modeled soil moisture. The purpose of assimilation is to improve the modeled soil-moisture field, but the accuracy to which the soil moisture can be retrieved from TB is highly dependent on the accuracy with which the optical depth can be estimated.
Figure 7 shows the relationship between near-surface soil moisture and TB derived using the function described in section 2c for two representative covers along with the effect of an error of ±0.08 in optical depth. For dense vegetation cover with an optical depth of 0.5 (vegetation water content ∼5 kg m−2), the dynamic range of TB is relatively small (the signal is fairly insensitive to soil moisture). At higher water contents, errors resulting from an imperfect specification of vegetation optical depth are of the order ±8 K, that is, about 10% of the dynamic range in TB, corresponding to an error of 5% in the volumetric water content. For a less dense canopy with an optical depth of 0.1 (vegetation water content ∼1 kg m−2), the errors resulting from imperfect specification of optical depth approach ±25 K at high water contents. This corresponds to an error of ±10% in volumetric water content, again about 10% of the dynamic range in TB. These are significant errors, which emphasize the need for accurate knowledge of vegetation optical depth for effective assimilation of TB. On the basis of the results summarized in Fig. 7, the differences between the modeled and measured TB in Fig. 6 are most likely to be the greatest for the wettest day (top image) and, assuming the error in optical depth is independent of the optical depth itself, the error in the TB will be greater for low optical depths.
The usefulness of assimilating TB into SWEAT is defined using the AVI [Eq. (2)]. In this case, uncertainty in soil moisture (σequiv) is attributed to a poorly determined optical depth–NDVI relationship [see Fig. 4 and section 3b(3)]. Other sources of variance in soil moisture are discussed in section 2d. The probability distribution of the optical depth at each cell within the grid (which can be used to estimate the probability distribution of the surface soil moisture) is obtained from Fig. 4a. The probability distribution is assumed to be normal with the mean represented by the best-fit line and the standard deviation represented by the difference between the best-fit line and the 66% confidence interval. One hundred distributed grids of optical depth were randomaly generated using the defined probability distributions. These, along with measured TB, were then used as inputs into the simple function defined in section 2c and the probability distribution of soil moisture at each grid cell was obtained. Figure 8, column 2, shows the mean of the retrieved soil moisture for each grid cell, for the 3 days when SLFMR measurements were available. Column 1 shows the distribution of modeled soil moisture at the same times. The average modeled and retrieved soil moisture for the entire domain are very similar on all 3 days, the modeled values being 43.4%, 39.3%, and 34.2%, respectively, and retrieved values 45.4%, 37.4%, and 34.2%, respectively.
The AVI reflects the differences in the distributions of modeled and retrieved soil moisture. The proposed accuracy of soil-moisture retrievals from passive microwave radiometers is 3% (Jackson et al. 1999). Therefore, in column 3, grid cells where there is less than 3% difference between modeled and retrieved soil moisture are shown in white. Here, the model already agrees well within the measurement error and the TB data have little worth. Elsewhere, dark colors indicate that data assimilation has a greater usefulness, while light colors reflect it has less value. Two general trends are apparent as the soil dries out. First the proportion of white patches increases from 24% to 35% of the total area, indicating that the model more accurately estimates soil moisture in drier conditions than in wetter conditions. Second, the proportion of dark grid cells increases indicating that, among the shaded pixels, the usefulness of data assimilation increases. In general, even a relatively conservative estimate of the errors in the optical depth results in the data having little use over about 50% of the area. Presumably, a more precise, remotely sensed estimate of optical depth would increase the overall usefulness of assimilating remotely sensed TB.
5. Summary and conclusions
This paper describes a study in which a coupled land surface and microwave emission model, MICRO-SWEAT, is used to predict the microwave brightness temperature for comparison with measurements provided by airborne passive microwave radiometers in the El Reno, Oklahoma, area during the SGP97 experiment. Calibration of MICRO-SWEAT resulted in good agreement between the time courses of modeled and measured near-surface soil moisture and TB for four specific sites.
Estimates of the distributed grids of required soil and vegetation parameters were obtained from soils maps and remote sensing. In particular, a novel method of estimating the vegetation optical depth from NDVI data is discussed. A 2D array of MICRO-SWEAT models was used to predict the distributed fields of TB. The area-average TB are predicted well by the model but the spatial distributions differ noticeably. The discrepancies between model and measurement result from inaccurate parameterization and initialization of the land surface model, errors in the model physics, errors in the distribution of precipitation, and errors in the parameterization of the microwave emission component of MICRO-SWEAT. Assimilation of the measured TB into the land surface model could correct for the first three issues as long as the parameterization of the microwave emission model was correct.
A demonstration of the usefulness of TB for data assimilation is given using the error in the estimated optical depth. This revealed that, despite providing some information, the usefulness of TB is significantly reduced by poor parameterization of the microwave emission model, for example, by imperfect knowledge of optical depth. One way to improve estimates of optical depth is to retrieve simultaneously both soil moisture and vegetation optical depth using multiple measurements of TB for the same location at different angles. This is the approach proposed for the SMOS mission, and should result in TB data being more valuable for the purpose of data assimilation. However, evaluating the potential of this approach is not feasible with the SGP97 datasets. Future field studies should seek to correct this weakness.
Acknowledgments
Primary support for Dr. Eleanor Burke for preparing this paper came from NOAA Project NA96GP0412. Additional support for Prof. James Shuttleworth came from NASA Project NAG5-7554. The authors wish to acknowledge the contribution of the many people who obtained the soil-moisture samples at El Reno. In addition, we thank Dr. Paul Houser (NASA GSFC) for providing the soil-profile data and Dr. Andrew French (USDA Hydrology Laboratory) for providing the leaf area index data. The editorial assistance of Corrie Thies is greatly appreciated. Comments and suggestions from the numerous reviewers are greatly appreciated.
REFERENCES
Betts, A. K., Ball J. H. , Beljaars A. C. M. , Miller M. J. , and Viterbo P. , 1996: The land-surface–atmosphere interaction: A review based on observational and global modeling perspectives. J. Geophys. Res., 101 , 7209–7225.
Burke, E. J., and Simmonds L. P. , 2001: A simple physically based soil moisture retrieval algorithm and its sensitivity to measurement errors. Hydrol. Earth Syst. Sci., 5 , 39–48.
Burke, E. J., and Simmonds L. P. , 2002: Passive microwave emission from bare soils: Developing a simple model to predict near surface water content. Int. J. Remote Sens., in press.
Burke, E. J., Gurney R. J. , Simmonds L. P. , and Jackson T. J. , 1997: Calibrating a soil water and energy budget model with remotely sensed data to obtain quantitative information about the soil. Water Resour. Res., 33 , 1689–1697.
Burke, E. J., Gurney R. J. , Simmonds L. P. , and O'Neill P. E. , 1998: Using a modeling approach to predict soil hydraulic properties from passive microwave measurements. IEEE Trans. Geosci. Remote Sens., 36 , 454–462.
Burke, E. J., Gurney R. J. , and French A. N. , 2001a: Vegetation indices for soil moisture retrievals from passive microwave radiometry. Hydrol. Earth Syst. Sci., 5 , 671–677.
Burke, E. J., Shuttleworth W. J. , Lee K. , and Bastidas L. A. , 2001b: Using area average remotely sensed surface soil moisture in multi-patch Land Data Assimilation Systems. IEEE Trans. Geosci. Remote Sens., 39 , 2091–2100.
Calvet, J-C., and Noilhan J. , 2000: From near-surface to root-zone soil moisture using year-round data. J. Hydrometeor., 1 , 393–411.
Cosby, B. J., Hornberger G. M. , Clapp R. B. , and Ginn T. R. , 1984: A statistical exploration of the relationship of soil moisture characteristics to the physical properties of soils. Water Resour. Res., 20 , 682–690.
Daamen, C. C., and Simmonds L. P. , 1996: Measurement of evaporation from bare soil and its estimation using surface resistance. Water Resour. Res., 32 , 1393–1402.
Famiglietti, J. S., and Coauthors. 1999: Ground-based investigation of soil moisture variability within remote sensing footprints during the Southern Great Plains 1997 (SGP97) Hydrology Experiment. Water Resour. Res., 35 , 1839–1851.
French, A. N., Schmugge T. J. , and Kustas W. P. , 2000: Estimating surface fluxes over the SGP site with remotely sensed data. Phys. Chem. Earth, 25B , 167–172.
Galantowicz, J. F., Entekhabi D. , and Njoku E. G. , 1999: Tests of sequential data assimilation for retrieving profile soil moisture and temperature from observed L-band radiobrightness. IEEE Trans. Geosci. Remote Sens., 37 , 1860–1870.
Hoeben, R., and Troch P. A. , 2000: Assimilation of active microwave observation data for soil moisture profile estimation. Water Resour. Res., 36 , 2805–2819.
Jackson, T. J., 2001: Multiple resolution analysis of L-band brightness temperature for soil moisture. IEEE Trans. Geosci. Remote Sens., 39 , 151–164.
Jackson, T. J., Le Vine D. M. , Hsu A. Y. , Oldak A. , Starks P. J. , Swift C. T. , Isham J. D. , and Haken M. , 1999: Soil moisture mapping at regional scales using microwave radiometry: The Southern Great Plains Hydrology Experiment. IEEE Trans. Geosci. Remote Sens., 37 , 2136–2151.
Judge, J., England A. W. , Crosson W. L. , Laymon C. A. , Hornbuckle B. K. , Boprie D. L. , Kim E. J. , and Liou Y. A. , 1999: A growing season Land Surface Process/Radiobrightness model for wheat-stubble in the southern Great Plains. IEEE Trans. Geosci. Remote Sens., 37 , 2152–2158.
Kerr, Y. H., and Wigneron J. P. , 1994: Vegetation models and observations: a review. ESA/NASA International Workshop on Passive Microwave Remote Sensing Research Related to Land–Atmosphere Interactions, B. J. Choudhury et al., Eds., VSP, 317–344.
Laymon, C. A., Crosson W. L. , Jackson T. J. , Manu A. , and Tsegaye T. D. , 2001: Ground-based passive microwave remote sensing observations of soil moisture at S-band and L-band with insight into measurement accuracy. IEEE Trans. Geosci. Remote Sens., 39 , 1844–1858.
Lee, K. H., Burke E. J. , Shuttleworth W. J. , and Harlow R. C. , 2002: Influence of vegetation on SMOS mission retrievals. Hydrol. Earth Syst. Sci., 6 , 153–166.
Li, J., and Isham S. , 1999: On the estimation of soil moisture profile and surface fluxes partitioning from sequential assimilation of surface layer soil moisture. J. Hydrol., 220 , 86–103.
Norman, J. M., Divakarla M. , and Goel N. S. , 1995: Algorithms for extracting information from remote thermal-IR observations of the earth's surface. Remote Sens. Environ., 51 , 157–168.
Reichle, R., Entekhabi D. , and McLaughlin D. B. , 2001: Downscaling of radiobrightness measurements for soil moisture estimation: A four-dimensional variational data assimilation approach. Water Resour. Res., 37 , 2353–2364.
Schmugge, T. J., 1998: Applications of passive microwave observations of surface soil moisture. J. Hydrol., 213 , 188–197.
Schmugge, T. J., O'Neill P. E. , and Wang J. R. , 1986: Passive soil moisture research. IEEE Trans. Geosci. Remote Sens., 24 , 12–22.
Shuttleworth, W. J., and Wallace J. S. , 1985: Evaporation from sparse crops—An energy combination theory. Quart. J. Roy. Meteor. Soc., 111 , 839–855.
Simmonds, L. P., and Burke E. J. , 1998: Estimating near-surface soil water content from passive microwave remote sensing—An application of MICRO-SWEAT. Hydrol. Sci. J., 43 , 521–534.
Simmonds, L. P., and Burke E. J. , 1999: 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., 3 , 31–38.
Ulaby, F. T., Moore R. K. , and Fung A. K. , 1986: Microwave Remote Sensing—Active and Passive. Addison-Wesley, 1100 pp.
Wang, J. R., and Schmugge T. J. , 1980: An empirical model for the complex dielectric permittivity of soils as a function of water content. IEEE Trans. Geosci. Remote Sens., 18 , 288–295.
Wigneron, J-P., Schmugge T. J. , Chanzy A. C. , Calvet J-C. , and Kerr Y. H. , 1998: Use of passive microwave remote sensing to monitor soil moisture. Agronomie, 18 , 27–43.
Wigneron, J-P., Waldteufel P. , Chanzy A. , Calvet J-C. , and Kerr Y. H. , 2000: Two-D microwave interferometer retrieval capabilities of over land surfaces (SMOS mission). Remote Sens. Environ., 73 , 270–282.
Wilheit, T. T., 1978: Radiative transfer in a plane stratified dielectric. IEEE Trans. Geosci. Electron., 16 , 138–143.
(a) The location of the distributed grid in the SGP97 study area. (b) Schematic of the distributed grid. The 15 field sites are outlined; with the 4 more intensively monitored sites highlighted and labeled. Shading represents the different land cover classifications. The black dots in the northwestern part of the grid represent the loamy sand soil. The rest of the grid has silty loam soil
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
The time series of predicted 0–5-cm soil water content compared with measurements made using Theta Probes (0–6 cm) and gravimetric sampling (0–5 cm) for the four highlighted sites, two wheat (ER-13 and ER-15) and two rangeland (ER-1 and ER-5). Standard deviation of ± one is shown for the gravimetric and Theta Probe measurements. ER-13 was plowed on DOY 186
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
The time series of predicted TB for four selected sites, two wheat (ER-13 and ER-15) and two rangeland (ER-1 and ER-5), compared with the 800-m ESTAR data and the multiresolution SLFMR data. ER-13 was plowed on DOY 186
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
The relationship between optical depth and log(1-NDVI) for the 15 sites (outlined in Fig. 1) monitored at El Reno during SGP97 for (a) the 200-m-resolution SLFMR data and (b) the 800-m-resolution ESTAR data. The best fit and the 66% confidence intervals (equivalent to an error in the optical depth of ∼0.08) are also shown
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
The 200-m-resolution spatial distribution of optical depth predicted using the fitted relationship in Fig. 4a, which was derived using the SLFMR data
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
Distributions of modeled and measured 200-m-resolution SLFMR TB for the 3 days for which there are measurements
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
Predicted relationships between TB and top 5-cm soil moisture, for optical depths of 0.1 and 0.5. The error bars represent a standard error of 0.08 in optical depth
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
Modeled and retrieved soil moisture for the 3 days where there are SLFMR measurements at 200 m. Also shown is the distribution of the Assimilation Value Index (AVI). White areas indicate less than 3% difference between modeled and measured soil moisture.
Citation: Journal of Hydrometeorology 4, 2; 10.1175/1525-7541(2003)4<460:UMTMMB>2.0.CO;2
Parameters required for the four calibration sites. All parameters except leaf area index and optical depth were used for the distributed modeling
Bias and root-mean-square error between modeled and measured near-surface soil moisture for the 4 highlighted calibration sites and the 11 other validation sites
Bias and root-mean-square error between modeled and measured brightness temperatures for the El Reno area
