Ground-based investigation of soil moisture variability within remote sensing footprints during the Southern Great Plains 1997 (SGP97) Hydrology Experiment

. Surface soil moisture content is highly variable in both space and time. While remote sensing provides an effective methodology for mapping surface moisture content over large areas, it averages within-pixel variability thereby masking the underlying heterogeneity observed at the land surface. This variability must be better understood in order to rigorously evaluate sensor performance and to enhance the utility of the larger-scale remotely sensed averages by quantifying the underlying variability that remote sensing cannot record explicitly. In support of the Southern Great Plains 1997 (SGP97) Hydrology Experiment (a surface soil moisture mapping mission conducted between June 18 and July 17, 1997, in central Oklahoma) an investigation was conducted to characterize soil moisture variability within remote sensing footprints (approximately 0.64 km 2) with more certainty than would be afforded with conventional gravimetric moisture content sampling. Nearly every day during the experiment period, portable impedance probes were used to intensively monitor volumetric moisture content in the 0- to 6-cm surface soil layer at six footprint-sized fields scattered over the SGP97 study area. A minimum of 49 daily moisture content measurements were made on most fields. Higher-resolution grid and transect data were also collected periodically. In total, more than 11,000 impedance probe measurements of volumetric moisture content were made at the six sites by over 35 SGP97 participants. The wide spatial distribution of the sites, combined with the intensive, near-daily monitoring, provided a unique opportunity (relative to previous smaller-scale and shorter-duration soil moisture studies) to characterize variations in surface moisture content over a range of wetness conditions. In this paper the range and temporal dynamics of the variability in moisture content within each of the six fields are described, as are general relationships between the variability and footprint-mean moisture content. Results indicate that distinct differences in mean moisture content between the six sites are consistent with variations in soil type, vegetation cover, and rainfall gradients. Within fields the standard deviation, coefficient of variation, skewness, and kurtosis increased with decreasing moisture content; the distribution of surface moisture content evolved from negatively skewed/nonnormal under very wet conditions, to normal in the midrange of mean moisture content, to positively skewed/nonnormal under dry conditions; and agricultural practices of row tilling and terracing were shown to exert a major control on observed moisture content variations. Results presented here can be utilized to better evaluate sensor performance, to extrapolate estimates of subgrid-scale variations in moisture content across the entire SGP97 region, and in the parameterization of soil moisture dynamics in hydrological and land surface models.


Surface soil moisture content is an important hydrological variable which influences a wide range of interactions within
many Earth system processes are nonlinearly dependent upon surface moisture content, this variability and hence the degree to which remote measurements reflect actual moisture conditions within footprints must be better understood in order to enable full utilization of the larger-scale remotely sensed averages by the Earth science community. The purpose of this paper is to describe a field investigation of soil moisture variability within remote sensing footprints during the Southern Great Plains 1997 (SGP97) Hydrology Experiment. SGP97 was the largest airborne L band passive microwave mapping mission of surface soil moisture to date. Located in the 40-km by 250-km strip of central Oklahoma shown in Figure 1, soil moisture was mapped at a 0.8-km ground resolution nearly every day between June 18 and July 17, 1997. The area of the mapped region was an order of magnitude larger than in the previous experiment (Washita '92 [Jackson and Levine, 1996]) which utilized the same aircraftbased remote sensing instrument (the Electronically Scanned Thinned-Array Microwave Radiometer (ESTAR) [LeVine et al., 1994]), and the duration of the experiment was 4 times greater. The objective of SGP97 was to demonstrate the potential of ESTAR for mapping surface soil moisture over large regions, with implications for the potential of a spaceborne L band passive microwave instrument to map surface soil moisture globally. A more detailed description of the experiment is provided by T. J. Jackson (Southern Great Plains 1997 (SGP97) Hydrology Experiment Plan, http://hydrolab.arsusda.gov/sgp97/, 1997).
Ground truth data in support of soil moisture remote sensing experiments have typically been collected using the gravimetric method. Such was the case during SGP97, in which field-mean soil moisture content was determined gravimetrically at 49 sites scattered over the mapped region. Thirty-five of these sites were 0.8-km by 0.8-km (the approximate size of the sensor footprint) winter wheat or rangeland fields typical of the region. In these fields, moisture content was sampled each day at 14 points. In the remaining 14 sites, which were smaller than the other 35 fields, nine soil moisture samples were collected each day. Note that the sampling time required by the gravimetric method, combined with the time required to travel to the 49 sites, limits the number of moisture content measurements that can be made on each field on each day. Recent advances in impedance probe technology have resulted in the development of relatively inexpensive instruments capable of rapid volumetric moisture content determination. In our investigation, which was designed to complement the gravimetric ground truth data collection effort, portable impedance probes (see section 3 and Figure 2) were used to intensively monitor soil moisture content on six of the 35 winter wheat or rangeland fields. Volumetric moisture content in the 0-to 6-cm surface soil layer was measured nearly every day at 49 points on a 7 by 7 100-m grid at five of the six fields and at 27 points on a 3 by 8 100-m grid on the sixth field (see Figure  3). Higher-resolution grid and transect data were also collected periodically. In total, more than 11,000 impedance probe measurements of soil moisture were made at the six sites by over 35 SGP97 participants. Note that during the experiment, related studies of soil moisture variability within remote sensing foot-prints were simultaneously being conducted by other investigators on other fields within the SGP97 region. These activities and additional data sources are described by Famiglietti [1999].
Thus the rapid probe measurements provided a highresolution sampling component that enabled a more accurate characterization of the mean and variability of surface soil moisture content within sensor footprints than would have been possible with the relatively small number of samples collected by the gravimetric method [Bell et al., 1980;Owe et al., 1982]. Additionally, the wide spatial distribution of the sites provided the opportunity to observe soil moisture variability over a range of wetness conditions resulting from regional rainfall gradients and local controls on surface drying. The combination of these two aspects of this study afforded a unique opportunity to observe and quantify the dynamics of surface moisture content variability that has not been possible in previous, smaller-scale and shorter-duration soil moisture studies (several of which are discussed in section 2).
In this paper the range and temporal dynamics of the variability in moisture content within each of the six fields are described, as are general relationships between the variability and field-mean moisture content. Specifically, the means and standard deviations versus time and the standard deviation, coefficient of variation, skewness, and kurtosis versus mean moisture content are presented for each field, and the time variation of frequency distributions and higher-resolution transect data are presented for selected fields and/or dates. By providing well-constrained estimates of the mean, higher-order statistics, and distributions of surface soil moisture within sensor footprints, our work will help quantify both the accuracy of ESTAR soil moisture estimates and the underlying variability that remote sensing cannot record explicitly. Estimates of higher-order statistical information will enable broader utilization of the ESTAR data since many Earth system processes are nonlinearly related to subgrid distributions of soil moisture. A thorough investigation of ESTAR performance, using both the impedance probe data and the gravimetric moisture content data, is the topic of current research and will be published separately at a later date.
Our study also represents an important step toward addressing the ground-based and remotely sensed sampling issues identified as imperative to soil moisture research by Wei [1995]. These include (1) characterizing the underlying spatialtemporal covariance structure of the soil moisture field being sampled, (2) quantifying the errors resulting from discrete, ground-based sampling schemes of these variable soil moisture fields, (3) quantifying the ability of remote sensing to provide accurate, integrated soil moisture of such variable fields at the footprint scale, and (4) characterizing the behavior of soil moisture variability across scales. This present study has important implications for issues 1 and 2 above. Ongoing research mentioned above, in which the impedance probe data are compared to the gravimetric ground truth data and to the ESTAR moisture content estimates, will provide insight into the third of these issues. Because our study has added a highresolution component to the SGP97 ground truth data set, it will enable a more comprehensive study of the fourth issue.

Background
In this section, previous field investigations of surface soil moisture variability are reviewed. Studies included are those which have focused on moisture content variations in the near surface (0-15 cm) soil layer and which were conducted either in support of past soil moisture remote sensing experiments [e.g., Rao and Ulaby, 1977;Bell et al., 1980;Owe et al., 1982;Charpentier and Groffman, 1992] or at horizontal scales consistent with these past experiments [Hills and Reynolds, 1969;Reynolds, 1974;Henninger et al., 1976;Hawley et al., 1983;Francis et al., 1986;Loague, 1992;Robinson and Dean, 1993;Nyberg, 1996;Famiglietti et al., 1998]. Studies in which variations in surface moisture content have been characterized statistically are emphasized here. A detailed review of the environmental factors responsible for the observed variations is provided elsewhere [Famiglietti et al., 1998]. Because several hydrological, ecological, biogeochemical, and atmospheric processes are nonlinearly related to surface soil moisture, knowledge of its statistical distribution within remote sensing footprints would greatly increase the utility of remotely sensed soil moisture products within the Earth system science community. Hills 1992] suggested that under conditions of increased variance, remotely sensed footprint means would be less reflective of actual soil moisture conditions on the ground and that since the coefficient of variation increases with decreasing moisture content, soil moisture remote sensing will be more precise under wet conditions than dry.

Methods
A number of important constraints required consideration in the design of the experiment plan for this research. Since this study was a complementary investigation within SGP97, many of these constraints were dictated by the larger experi- bUniversal transverse mercator coordinates of northeast corner of field. All fields are 800 m by 800 m and are aligned on a north-south grid.
CEffective cover type for most of the experiment was row-tilled bare soil for LW21 and ER13 and was wheat stubble for CF04. See Table 2 for cultivation dates.
dFor simplicity we recognize two classes of topography: flat and gently rolling. eTerracing refers to a common regional erosion control practice of building berms along terrain contours to inhibit downslope surface water flow. ment. These included limitations with respect to site access, manpower, time available for sampling, and budget; a desire to reasonably limit travel time to sites and time in the field collecting data; and the large size of the SGP97 region. Given these constraints, the final plan, as outlined in sections 3.1-3.3, maximized the number of fields that could be sufficiently sampled on a daily basis, while providing an adequate distribution of study sites across the SGP97 experimental area.

Site Selection
Our six sites were a subset of the 49 sites which were selected as ground truth locations where soil moisture content was measured by the gravimetric method. These 49 sites were chosen such that the range of topographic, soil, and vegetation cover conditions found throughout the SGP97 region was well represented. Their selection was also influenced by important logistical issues such as the location of in situ or experimentspecific instrumentation, facility support, and site access. As such, the 49 sites for ground-based activities were concentrated in three primary locations: the Little Washita watershed, southwest of Chickasha (23 sites); the U.S. Department of Agriculture Agricultural Research Service Grazinglands Research Laboratory in E1 Reno (16 sites); and the Department of Energy Atmospheric Radiation Measurement Program cloud and radiation test bed (ARM CART) Central Facility, near Lamont (10 sites) (see Figure 1).
From these 49 sites, three fields were selected within the Little Washita (LW) watershed, two were selected at E1 Reno (ER), and one was selected at the ARM CART Central Facility (CF). These fields were identified during the experiment as LW03, LW13, LW21, ER05, ER13, and CF04, and they are listed in Table 1 along with their basic vegetative cover, soil, and topographic attributes. They represent typical combinations of cover and soil types found throughout the SGP97 region (e.g., winter wheat on silty loam and rangeland on loamy sand or loam), and their distribution across the experimental area maximized the likelihood that regional gradients in rainfall would be reflected in soil moisture observations.

Equipment
3.2.1. Soil moisture impedance probes. Central to the success of this investigation was the identification of a durable, portable, accurate, and affordable methodology for rapid measurement of surface soil moisture content. Given these re-

quirements, we chose a new impedance probe, recently described by Gaskin and Miller [1996] and Miller et al. [1997] and now being produced commercially as the Theta Probe soil moisture sensor, type ML1, by Delta-T Devices of Cambridge, England. (The mention of product names does not constitute an endorsement of this product.)
The probe, shown in Figure 2, uses a simplified voltage standing wave method to determine the relative impedance of its sensing head (which consists of four sharpened 6-cm stainless steel wire rods) and thus the dielectric constant of the soil matrix and the volumetric water content of the soil. Gaskin and

Miller [1996] and Miller et al. [1997] provide further details on
probe operation. It is accurate to within _+0.02 cm3/cm 3 with site-specific calibration, and probe measurements of moisture content compare well with those of the neutron probe [Gaskin and Miller, 1996]. Site-specific calibration efforts by several of the coauthors yielded calibration curves similar to that of Gaskin and Miller [1996], so that the Gaskin and Miller calibration curve was adopted for use in this study.
The probe is compact, and it proved to be field durable in our study, though the 6-cm stainless steel wire rods tended to bend and break under the very dry conditions encountered in some fields. In most cases, however, bent or broken rods were easily repaired or replaced in the field. While several probes may have been used on each field (typically three or four for the duration of the experiment), comparisons done at each site showed that differences in probe responses were negligible.

Differential Global Positioning System (DGPS).
Differential Global Positioning System was used to accurately geolocate sampling locations within fields and when real-time navigation was required in the field. DGPS functions by correcting for most of the natural and man-made errors that are a component of normal GPS measurements. Corrections are transmitted from a "reference" receiver, which is fixed in position, to the roving receivers in the field so that horizontal position can be determined to within 1-5 m.
Our DGPS system was composed of a 12-channel hand-held GPS receiver, a radio beacon receiver to receive the correction, a 2.6-m whip antenna, which was attached to the radio beacon receiver, and a 12-volt battery. In the field the radio beacon receiver, the whip antenna, and the battery were carried in a small backpack. The correction signal was transmitted by radio beacon from a reference station in Sallisaw, Okla-homa, which is part of a network maintained by the U.S. Coast Guard for navigational purposes. The system described above was relatively inexpensive and field durable and performed well for the purposes of our experiment.

Sampling Plan
Volumetric moisture content in the 0-to 6-cm surface soil layer was measured nearly every day at 49 points on a 7 by 7 100-m grid at five of the six fields sites (LW03, LW13, LW21, ER05, and CF04) and at 27 points on a 3 by 8 100-m grid on the sixth field (ER13, see Figure 3). Additionally, higherresolution data were collected at several of the sites. These included 25-m north-south and east-west transects (LW03 and LW21), higher-resolution transects along which samples were collected at distinct intervals of variable length (e.g., at tops and bottoms of tilled soil rows (LW21) or terraced hillslopes (LW13)), and on higher-resolution grids (ER05, ER13, and CF04). Table 2 lists the additional data collected at each of the six sites.
Once sampling grids were established at each of the six fields, moisture content sampling was conducted on each day possible. Sampling was suspended during rain events or when agricultural activity (e.g., cultivating and fertilizer or pesticide spreading) posed a significant safety concern. In general, two 2-person teams were assigned to each field. One person operated the DGPS and recorded the data, while the second person sampled moisture content with the impedance probe. Sampling was routinely conducted between the hours of 1 and 3 P.M. CST. The dates on which sampling was conducted are listed in Table 2 for each field along with a statistical summary of the daily measurements. In total, more than 11,000 impedance probe measurements of soil moisture were made on the six fields during the course of the experiment. All data are available through the NASA Goddard Distributed Active Archive Center at http://daac.gsfc.nasa.gov. Figure 4 shows the time series of precipitation and the mean and standard deviation of surface moisture content for each of the six fields. The mean moisture content responds predictably to rainfall, increasing after storm events and decreasing thereafter. The E1 Reno and Central Facility fields were often wetter than those in the Little Washira watershed, in particular during the second and third weeks of the experiment, owing to the greater depth of precipitation falling in the central and northern parts of the study region. Field ER05 was the wettest site, with mean moisture content values ranging between 0.48 cm3/cm 3 and 0.28 cm3/cm 3. The driest site was LW03, in which the mean moisture content varied between 0.22 cm3/cm 3 and 0.05 cm3/cm 3. Close inspection of Figure 4 and Table 2 reveals distinct differences in field-mean moisture content resulting from differences in soil types, vegetation cover, and rainfall gradients. For example, fields LW03 and LW13 are both covered by rangeland vegetation and received similar amounts of rainfall during the study period. However, the more sandy soil at LW03 resulted in a lower mean moisture content than at LW13 on all but one day. Fields ER05 and ER13 received comparable amounts of precipitation during the experiment and differed primarily in their type of vegetation cover. The winter wheat grown in ER13 was harvested early in the experiment so that its effective cover type was bare soil for most of the study period. Comparing field-mean moisture contents for the two fields shows that the bare field (ER13) was consistently drier than its rangeland neighbor (ER05). Finally, differences in mean moisture content due to differences in rainfall are evident from comparing field LW21 with ER13: both are winter wheat fields on silty loam soil which differed primarily in the depth of precipitation falling during the course of the experiment. As mentioned above, during the second and third weeks of the experiment, more precipitation fell over the central (ER) and northern (CF) sites, and thus field-mean moisture contents at ER13 were much greater than those at LW21 during this phase of the study. , and Charpentier and Groffman [1992]. The observed decrease is largely controlled by increasing mean moisture content rather than decreasing standard deviation, since the range of the observed mean moisture content is nearly 6 times greater than the range of the standard deviation.

Distributions, Skewness, and Kurtosis
Frequency distributions of surface moisture content are shown in Figure 6 for each field on selected days during drydown sequences within the study period. Distinct differences are evident between the drier fields within the Little Washira watershed and the wetter fields at the E1 Reno and Central Facility locations. In the Little Washira fields the distributions appear normal following rain events but become positively skewed as the soil dries with increasing time into the interstorm period. This observation is supported by the results of normality testing with the Shapiro-Wilk statistic, which indicated ( Table 2) that all of the LW03 distributions shown in Figure 6, the June 28, July 1, and July 3 distributions at LW13 and the July 6 and July 8 distributions at LW21, have low probabilities of being normal. It is further supported by the increase in skewness reported in Table 2.
The heavier rainfall in the central and northern parts of the SGP97 study region provided an opportunity to observe surface moisture content distributions under wetter conditions than those observed in the Little Washira watershed. The high-   est daily value of mean moisture content at any of the six fields was observed on June 19 at ER05. Figure 6 and Table 2 show that the corresponding moisture content distribution under these very wet conditions (0.48 cm3/cm 3 mean moisture content) was negatively skewed and was not normal. As the soil dried in the days following, positive skewness increased and the distributions tended toward normaIcy. The distributions observed at ER13 remained normal for the interstorm period shown in Figure 6, and as in the case of the other fields shown in Figure 6, positive skewness increased as the soil dried. The heavy rains of June 26 and June 29 at CF04 and the l 1-day dry-down period which followed allowed for direct observation of the temporal dynamics of a surface moisture content distribution during a relatively long interstorm period, in which moisture conditions changed from very wet to very dry. Though not visually apparent in Figure 6, Table 2 shows that the distribution began the drying cycle as negatively skewed and nonnormal. Then, Figure 6 and Table 2 clearly show that the distribution evolves to normal on July 2 and 7 and to positively skewed and nonnormal on July 9. Figure 7a shows skewness versus mean moisture content for each field on each day that data were collected. Combining  Moisture Content (cm3/cm 3)   Figure 4). The distinct differences apparent in Figure 4 suggest that variations in surface moisture content due to large-scale variations in these controls should be readily detectable in the 0.8-km remotely sensed soil moisture data and as such should serve as a qualitative check on ESTAR performance. Higher-resolution results from the previous Washita '92 experiment [Jackson and Levine, 1996] suggest that this will, in fact, be the case.
A second point for discussion is the factors responsible for the observed moisture content variability within fields and, in particular, the evolution of moisture content distributions during the transition from wet to dry conditions. While a comprehensive study of the influence of within-field variations in process controls (e.g., topographic attributes, soil texture, vegetation cover, and precipitation) was not plausible in the context of this large-scale experiment, Figure 8 suggests that the widespread agricultural practices of row tilling and terracing played a major role in controlling subfootprint-scale soil moisture variability during the period of observation. Figure 8 (and similar transect results not shown) indicates further that their influence was continuous through interstorm periods and hence not limited to wet or dry conditions. Row tilling and terracing influence the distribution of surface moisture content by imposing unnatural microtopographic (ridge-furrow rows on the LW21 and ER13 winter wheat fields and regularly spaced berms on the LW03, LW13, and ER05 rangeland fields) and porosity (tilled ridges on winter wheat fields were openly porous) structures on the six fields. Compaction due to grazing and agricultural equipment traffic was another factor that affected both soil structure and created topographic depressions. Under wet conditions, water drained rapidly from plowed ridges on the winter wheat fields and tended to accumulate in microtopographic lows in all fields (furrows on the winter wheat fields, behind berms on the rangeland fields, and in compacted areas on both field types). Together these variations in porosity and microtopography jointly controlled the distribution of surface moisture content under wet conditions, and since the wetter areas tended to persist during interstorm periods, these factors may also explain the increase in variance and positive skewness as fieldmean moisture content decreased.

Implications for Remote Sensing and Hydrological Modeling
This study has provided six locations within the SGP97 region at which the mean moisture content is known with a high degree of certainty, relative to the remaining 43 locations, at which the mean moisture content was determined gravimetrically using only 9 to 14 samples. Figure 9 shows that when the sampling density is increased (e.g., from 14 to 49 at most of the sites), the error with which the daily mean moisture content can be determined is reduced by as much as a factor of 2 (using observed standard deviations, standard limit of error equations, and a = 95%). Consequently, this investigation will allow for a more rigorous evaluation of ESTAR performance at these six sites. Because the different combinations of soil types, vegetation cover, and rainfall amounts at the six fields were broadly representative of those throughout the SGP97 of, and just below each berm. Moisture content measurements were made at 7-m intervals between berms on the north-south transect and just once on the east-west transect. Table 2 lists the dates on which these transect data were collected. Figure  8b for 3 days (July 11, 12, and 15) during a brief interstorm period that occurred in the last week of the experiment. Also shown are the elevations of the 42 sampling locations along the transect. It is evident from Figure 8b that there is no correlation between surface moisture content and the significant downslope decrease in elevation. Rather, the most striking feature apparent in Figure 8b is that the presence of berms controls variability in surface moisture content along the transect: on each day, moisture content is consistently highest above each berm. Similar patterns were observed along the 230-m north-south transect (not shown).

Discussion
In this section the environmental factors responsible for the observed variations in surface moisture content, both within and between the six sites, are briefly discussed, as are the implications for this work with respect to understanding ES-TAR accuracy, enhancing the utility of the SGP97 remotely sensed soil moisture imagery, and the parameterization of soil moisture dynamics in land surface models. Although the primary emphasis of this study was on the statistical characterization of variations in surface moisture content rather than on a detailed investigation of the responsible land surface properties and environmental processes, at least two aspects of the previously presented work warrant further discussion in this regard. First is the degree to which variations in soil type, vegetation cover, and rainfall gradients influenced variations in the mean moisture content between the six fields (see Figure 4). The distinct differences apparent in Figure 4 suggest that variations in surface moisture content due to large-scale variations in these controls should be readily detectable in the 0.8-km remotely sensed soil moisture data and as such should serve as a qualitative check on ESTAR performance. Higher-resolution results from the previous Washita '92 experiment [Jackson and Levine, 1996] suggest that this will, in fact, be the case.
A second point for discussion is the factors responsible for the observed moisture content variability within fields and, in particular, the evolution of moisture content distributions during the transition from wet to dry conditions. While a comprehensive study of the influence of within-field variations in process controls (e.g., topographic attributes, soil texture, vegetation cover, and precipitation) was not plausible in the context of this large-scale experiment, Figure 8 suggests that the widespread agricultural practices of row tilling and terracing played a major role in controlling subfootprint-scale soil moisture variability during the period of observation. Figure 8 (and similar transect results not shown) indicates further that their influence was continuous through interstorm periods and hence not limited to wet or dry conditions. Row tilling and terracing influence the distribution of surface moisture content by imposing unnatural microtopographic (ridge-furrow rows on the LW21 and ER13 winter wheat fields and regularly spaced berms on the LW03, LW13, and ER05 rangeland fields) and porosity (tilled ridges on winter wheat fields were openly porous) structures on the six fields. Compaction due to grazing and agricultural equipment traffic was another factor that affected both soil structure and created topographic depressions. Under wet conditions, water drained rapidly from plowed ridges on the winter wheat fields and tended to accumulate in microtopographic lows in all fields (furrows on the winter wheat fields, behind berms on the rangeland fields, and in compacted areas on both field types). Together these variations in porosity and microtopography jointly controlled the distribution of surface moisture content under wet conditions, and since the wetter areas tended to persist during interstorm periods, these factors may also explain the increase in variance and positive skewness as fieldmean moisture content decreased.

Implications for Remote Sensing and Hydrological Modeling
This study has provided six locations within the SGP97 region at which the mean moisture content is known with a high degree of certainty, relative to the remaining 43 locations, at which the mean moisture content was determined gravimetrically using only 9 to 14 samples. Figure 9 shows that when the sampling density is increased (e.g., from 14 to 49 at most of the sites), the error with which the daily mean moisture content can be determined is reduced by as much as a factor of 2 (using observed standard deviations, standard limit of error equations, and a = 95%). Consequently, this investigation will allow for a more rigorous evaluation of ESTAR performance at these six sites. Because the different combinations of soil types, vegetation cover, and rainfall amounts at the six fields were broadly representative of those throughout the SGP97 region, understanding sensor performance at these locations should also provide insight into its capabilities across the entire study area. A second implication of this work is that the observed relationships between variability and mean moisture content provide a means for extrapolating this information beyond the six sites. For example, spatial fields of the coefficient of variation and the standard deviation could be derived by fitting an exponential function to the coefficient of variation-mean moisture content relationship shown in Figure 5b and solving for either the coefficient of variation or the standard deviation at each 0.8-km grid cell on land given its remotely sensed mean moisture content. Such higher-order statistical information will enhance the utility of remotely sensed soil moisture to the modeling community, for example, by providing coarse estimates of soil moisture variability for models of Ea•rth system processes operating at subgrid scales.
Finally, the dynamics of surface moisture content distributions characterized by this investigation will aid in the development of hydrological and land surface models, which often parameterize the distribution of surface moisture content statistically [e.g., Entekhabi and Eagleson, 1989;Farniglietti andWood, 1991, 1994;Wetzel and Boone, 1995;Bonan, 1996;Stieglitz et al., 1997]. Until now, however, little field evidence has been available to guide developers in the choice of dist•ribution form and to provide an understanding of its evolution through time. While modelers often choose one distribution (e.g., normal) and assume that it represents spatial variability in moisture content across a full range of wetness conditions, our work indicates that both the form of the distribution and its p•rameters change systematically with average moisture conten,t. On the basis of field observations described in this paper, it appears that appropriate choices for modeling soil moisture distributions may include a normal distribution evolving into a three-parameter gamma distribution as the soil dries between storm events or a flexible distribution such as a beta distribution that can change form from negatively skewed to positively skewed with a corresponding change in its parameters.

Summary
The purpose of this investigation was to characterize variability in surface moisture content within six 0.8-km remote sensing footprint-sized fields during SGP97. Volumetric mQisture content in the 0-to 6-cm surface soil layer was measured nearly every day at 49 points on a 7 by 7 100-m grid at five of the six fields and at 27 points on a 3 by 8 100-m grid on the sixth field. On selected days, additional higher-resolution grid or transect measurements were made on most fields. The wide spatial distribution of the sites, combined with the intensive, near-daily monitoring, provided a unique opportunity, relative to previous smaller-scale and shorter-duration soil moisture studies, to characterize variations in surface moisture content over a range of wetness conditions. Results indicated that distinct differences in mean moisture content between the six sites were consistent with variations in soil type, vegetation cover, and rainfall gradients. Within fields the standard deviation, coefficient of variation, skewness, and kurtosis increased with decreasing moisture content; the distribution of surface moisture content evolved from negatively skewed/nonnormal under very wet conditions, to normal in the midrange of mean moisture content, to positively skewed/nonnormal under dry conditions; agricultural practices of row tilling and terracing were shown to exert a major control on observed moisture content variations. Results presented here can be utilized to better evaluate ESTAR performance, to estimate subgrid-scale variations in moisture content across the entire SGP97 region, and in the parameterization of soil moisture dynamics in hydrological and land surface models.