Long-term data collection of volumetric soil moisture under sod has been conducted in Illinois for more than 25 years. Numerous applied and modeling studies have been undertaken with these data, often relating results to regional conditions under a variety of surface covers. However, the actual level of representation of these data to nearby areas with different surface covers is unknown. In 2006/07, the Soil Moisture under Sod Experiment was conducted at Bondville, Illinois, to increase understanding of soil moisture variability across a very small area of seemingly uniform surface and near-surface conditions. Ten locations were chosen at random within a 5.9-ha sodded field for twice weekly neutron probe soil moisture observations over a period of more than 13 months. Measurements were taken at the surface and at 20-cm intervals down to 2 m, precisely matching the historic Illinois depth observations. A detailed surface terrain analysis was conducted to consider effects on soil moisture attributable to surface slope or ponding potential at each monitoring location across the very low relief surface. The near-surface water table level at the field location was monitored. At the end of observations, soil property heterogeneity (i.e., soil porosity, bulk density, and soil color) was determined by digging trenches and extracting soil cores immediately adjacent to each monitoring site at all observation levels within the predominantly loess soil.
Results indicate a strong temporal consistency in intrasite trends of volumetric soil moisture at all depths throughout the experiment. However, intersite spatial variability increased with depth, indicated by an average standard deviation of all temporal observations of 2.26% in the top 30 cm of soil and 5.19% in the 170–200-cm layer. Differences between the average field soil moisture at all primary randomly selected sites and the historic Bondville site were 2.39% and 6.51%, respectively. In addition, an apparent strong relationship was observed between soil moisture in deeper layers and surface terrain slope, and to a lesser extent with soil porosity and bulk density.
The question of representativeness of soil moisture under sod to adjacent surface covers was not answered with this work, but the large differences measured across this seemingly uniform field suggest that proper use of the historic Illinois dataset by future research related to adjacent areas may need greater attention. Most of Illinois is under an agricultural cover, not sod. Adequate data monitoring of surface terrain slope, soil profiles, and water table climatology under various major surface covers within a region may be necessary prior to the installation of new soil moisture monitoring networks and before useful assumptions concerning spatial representation can be made that attribute individual soil moisture datasets to adjacent areas. These results highlight the importance of a strict globally unified protocol for soil moisture network design and data collection in support of quality in situ global soil moisture assessment, a primary goal of the International Soil Moisture Working Group of the Global Energy and Water Cycle Experiment.
This document presents results from a simple, but noteworthy, experiment to understand the degree of soil moisture variability observed within a relatively small sod-covered area located inside the considerable agricultural region of Illinois. Soil moisture is a key component in the hydrologic cycle. It is useful in numerous settings, such as its relation to the magnitudes of localized floods and regional droughts, modeling research on watershed studies, and projected effects due to climate change. Quality assessments of soil moisture are becoming increasingly important to those involved in assessing water resources via global satellite imagery. Perhaps as much as precipitation, soil moisture holding capacities yield considerable information on the abilities of crops to withstand a dearth in rainfall, project the timing of drought onset, longevity, and recovery, as well as yield early and developing impacts prolonged drying has on other water resources of a particular region.
Determining the representativeness of high-quality in situ soil moisture data is complicated, not only because of precipitation variability but also because of unknown variability in soil characteristics where moisture observations are made. Often, these characteristics are poorly understood because 1) they are not observed easily without disturbing the soils in which moisture data are being collected and 2) the breadth of the variability of local soil characteristics is similarly undefined.
One long-term source of soil moisture data in Illinois has been neutron probe observations collected since 1981 by the Illinois Soil Moisture Network (ISMN) of the Illinois State Water Survey (ISWS), Institute of Natural Resource Sustainability, University of Illinois at Urbana–Champaign. These volumetric data (fractional volume of water in the soil) have been employed by numerous researchers to represent soil moisture conditions under regional crop fields and other surface covers adjacent to monitoring sites. Adegoke and Carleton (2002) found weak correlations of vegetation indices for croplands and forests when comparing Advanced Very High Resolution Radiometer satellite data to Illinois soil moisture data. Vinnikov et al. (1999b) recognized a soil moisture signature over grass or crops by comparing Scanning Multichannel Microwave Radiometer data with the Illinois soil moisture data. Brown and Arnold (1998) used Geostationary Operational Environmental Satellite imagery combined with computed evapotranspiration values from the ISMN data, concluding that significant convective cloud mass development occurs along agricultural, urban, and forest land cover boundaries.
Other researchers have used ISMN data for various determinations. Findell and Eltahir (1997, 1999) quantified a positive correlation between soil moisture and subsequent summertime rainfall. Wu et al. (2002) related the soil moisture profile to long-term precipitation variability. Dirmeyer (2000) used average soil moisture in the top 1 m of soil to help evaluate the degree of climate drift in a land–atmospheric simulation. Gao and Dirmeyer (2006) used Illinois soil moisture data to help evaluate the performance of a combined land surface and land–atmosphere model ensemble analysis that generated global soil wetness products. In addition, ISMN data have been employed, in part, within studies and simulations of the water balance in the Illinois River basin (Niemann and Eltahir 2004), the state of Illinois (Yeh et al. 1998; Rodell and Famiglietti 2001; Yeh and Eltahir 2005; Amenu et al. 2005), the midwestern United States (Zangvil et al. 2004), the conterminous United States (Maurer et al. 2002), and North America (Fan et al. 2006).
Many studies treated the Illinois in situ soil moisture observations, collected under sod, as representative of soil moisture in adjacent land cover areas of various types. Because of the uniqueness of the ISMN dataset, its use in this manner is clearly understandable. Nevertheless, the predominant surface cover in Illinois is not sod but instead is row crops (primarily corn and soybeans) during the growing season, followed by a bare ground–low-tilled–no-tilled cover in the fallow season. Comparative extraction rates of subsoil water and evapotranspiration by a perennial cover of sod in Illinois, operating on a different (longer) schedule than annual agricultural crops, have not been evaluated.
The validity of the ISMN data to provide high-quality soil moisture information is not contested here. However, questions are raised concerning the appropriate understanding of data collections and potential limitations of their representativeness to other surface covers. The current study attempts to highlight limitations associated with the use of these data for the benefit of subsequent studies.
The ISWS monitored soil moisture data at 18 locations in Illinois using neutron probe technology, beginning at most sites in 1981 and ending at some stations in 2004 (Fig. 1). Data were collected twice monthly during the growing season and monthly during the fallow season (19 observations a year) from the surface to a depth of 2 m at 20-cm intervals. All locations had a surface cover of sod.
The initial siting of neutron probe monitoring locations in the ISMN was somewhat limited by other monitoring efforts of the ISWS. Logistically, there was a desire to collocate with the Illinois Climate Network (ICN), a weather mesonet initiated by the ISWS at the same time. However, no documented criteria exist from the period for locating neutron probe access tubes at each monitoring site (e.g., acceptance protocol for natural surface slopes, distances from different vegetation covers, buildings, roadways, tiled areas). Furthermore, although all sites were installed and remain today under grass, actual surface cover at sites varies substantially. Sod cover ranges from well-maintained lawn-type grasses of a few centimeters in depth at sites in public areas to thick-matted sod, perhaps 25 cm deep, at more remote sites. Although many sites, by choice, are near agricultural fields, this is not uniform across the network. The potential effects of different root zones on soil moisture at the various sites, and especially how they relate to conditions in adjacent bare soil, agricultural grounds, are unmeasured.
Gravimetric readings were obtained at each site in later years (Hollinger and Isard 1989) to help define soil characteristics for these predominantly loess, silt loam, soils and to assess their water-holding potential. However, these data provided no information on their representativeness to soils even a short distance away from the site locations, nor into adjacent, typically agricultural areas of different surface cover. Because of the extensive travel that was required to manually collect the ISMN neutron probe soil moisture data every two or four weeks (along with data collections from several other ISWS monitoring efforts at the time on the same data run), no protocol was employed concerning data collections before or after precipitation events and the temporal effects they would have on soil moisture data. In other words, because data collections occurred over a 2- to 3-day period, observations during some soil moisture data collection runs were taken both pre- and postprecipitation events at even adjacent ISMN sites.
One of the long-term data collection locations is at the Bondville Environmental and Atmospheric Research Site (BEARS), approximately 14 km southwest of Champaign, Illinois. The site is a highly instrumented rural field research area (available online at http://www.sws.uiuc.edu/atmos/bears/projects.asp) made available for research by the state of Illinois and the University of Illinois at Urbana–Champaign, housed within the Department of Electrical and Computer Engineering and managed by the ISWS. The area measures roughly 256 m × 232 m (approximately 5.9 ha). The land surface is typical of the general area—that is, considerably flat and treeless—but it is totally covered in sod. It is surrounded by farm fields of corn and soybeans with only occasional farmsteads and their concomitant tree and shrub vegetation. Thus, the site possesses near-perfect exposure conditions for meteorological observations.
During the summer of 2005, three sites within and near the BEARS field station were instrumented to measure soil moisture continuously at six depths, using Stevens-Vitel Hydra soil-moisture-capacitance-measuring sensors. The three locations included the long-term Bondville ISMN site and two sites in adjacent agricultural fields: one under a crop of soybeans and the other in corn. Site locations were separated by approximately 200 m. Hourly observations of soil moisture were collected at 5, 10, 20, 50, 100, and 150 cm of depth. Observations were initiated in mid-June and continued until the following April 2006 when seasonal planting operations began.
Results at 10 cm (Fig. 2a) showed temporal trends in soil moisture that responded reasonably well to precipitation events and subsequent drying between rains. Values among sites ranged between 25% and 35% during the summer and autumn 2005, but they converged by spring 2006. At 150 cm (Fig. 2b), general drying was observed under soybeans and sod during the first four months of monitoring, followed by moisture recharge but at very different rates between the sites (the corn sensor at 150 cm operated intermittently and is not displayed).
These data suggested that information on the variability of soil moisture from closely spaced sites under different surface covers is incomplete. It raised speculation that large differences may exist because of highly variable and unobserved subsurface soil conditions, controlling percolation and porosity. Taken a step further, it was concluded that similar knowledge on the level of soil moisture variability at closely spaced sites under even the same surface cover was also incomplete. This lack of information potentially affects our historic soil moisture network data when used to relate those observations to adjacent areas, all typically under a different surface cover. The current study was designed to better understand the magnitude of soil moisture variability over a small area to aid in design and site selection protocols of future in situ soil moisture network structures to achieve better regional data representation.
a. Soil moisture data collection
In August 2006, eight sites at the BEARS field station were chosen at random for installation of neutron probe sensors to observe soil moisture variability across the field station, initiating the Soil Moisture under Sod Experiment (SMUSE). Two additional sites were installed, each 1 m north and south from a random choice of one of the eight sites, to provide a site observation cluster of soil moisture variability over a very short distance. Neutron probe access tubes were inserted vertically to a depth of 2 m at each of these 10 locations, and observations commenced on a prescribed schedule. Measurements were also taken at the long-term Bondville soil moisture site.
Site locations are superimposed in Fig. 3 on an aerial photograph of the BEARS site (USDA 2007). In past decades, this entire site was used for crops and was subject to the regular agricultural practices of the time. The site is classified with a mixture of Drummer or Elburn soils, quite typical of the silty loam or silty clay loam loess soils within the highly productive agricultural fields of central Illinois. To achieve conditions as pristine as currently possible, protocols were established to restrict random selection of sites near existing BEARS research stations, adjacent crop fields, and access roadways and walkways, a process that resulted in all selected sites being placed in the northern and eastern portions of the BEARS research area. All sites were under an existing thick mat of sod, kept to a height of 10 cm or less around each soil moisture monitoring location.
For consistency, soil moisture measuring equipment matched that used with the historic data. Observations were collected with a model 3221 Troxler neutron depth probe and a model 3411B Troxler neutron surface probe (Troxler Electronics Laboratories 1980). The same pair of surface and depth sensors was used at all sites on each day of observation.
Soil moisture measurements during SMUSE were scheduled twice weekly at 11 sites during a 3-h period beginning at midday on Mondays and Thursdays. If precipitation was occurring and was expected to continue during the scheduled neutron probe operations, measurements were delayed for one day. If precipitation was occurring or appeared to be likely the next day during the designed period of observations, readings were postponed until the next scheduled date for measurements. This restriction, again only valid during the 3-h data collection window, likely caused a reduction of soil moisture in the top layer of soil had collections been allowed during rain events. Regardless, the intent of the protocol was to balance the days of observations as much as possible to every three to four days, to take readings at the same time each day to avoid differences attributable to any diurnal root extraction pattern of moisture use and to avoid imbalanced readings among the sites due to precipitation occurring during an observation period, as well as to mitigate danger to staff from possible lightning strikes.
A detailed topographical elevation survey of the entire BEARS site was conducted to define regions of potential surface water runoff and accumulation from precipitation and their proximity to neutron probe locations. Typical local conditions within the very low relief topography are for broad areas of standing water subsequent to heavy rain events. Depending on subsurface strata as well, the potential of surface ponding affecting the neutron probe values need definition.
A 210 m × 240 m rectangular grid was constructed over the BEARS site to measure land surface elevations. Determinations were made at regular 30-m intervals across the grid. In addition, two 60-m transects were constructed diagonally (northwest–southeast and southwest–northeast), centered on seven of the eight primary neutron probe locations and the historic Bondville neutron probe site. Two selected sites—B61 and B67—were relatively close to one another. Only one set of transects was made at the midpoint of those sites, as well as only one at the cluster of three sites, B66. Height data were collected along each transect at 3-m intervals. In total, 401 elevations were taken across the regular grid and the expanded transects.
Differential leveling observations were made using a Lietz B1C automatic level. Observations were made optically on leveling rods to 0.003-m precision. Data were analyzed with Surfer version 6 terrain and surface modeling software to generate contours of land surface elevation across the BEARS site.
c. Soil profiles
A series of soil cores were extracted at each neutron probe location for a soil analysis throughout the depth of the soil moisture observation profile. Trenches were dug adjacent (within 0.2 m) of each neutron probe tube to provide access to a vertical wall of undisturbed soil at every tube location. Soil cores were taken in triplicate along a horizontal line 10 cm wide at each neutron probe observation depth, every 20 cm below grade, down to 200 cm. Samples were extracted using a hand corer with a diameter of 3.18 cm to a horizontal distance into the wall of 10.1 cm. Sealed soil tins were used for core storage and transportation to an ISWS laboratory for analyses, where they were weighed and placed into a 105°C ventilated oven for 48 h. Subsequent final weights were taken and averaged by level. Analyses included soil bulk density (the mass of dry soil per unit volume of soil sample) and soil porosity (the amount of pore space), assuming that soil solids have an average particle density of 2.65 g cm−3 (Pierzynski et al. 2005).
Because of the thick mat of sod roots at the BEARS location, no soil cores were extracted at the surface to match with neutron surface probe data. Trenching and subsequent coring was not performed at the Bondville neutron probe site to maintain a pristine soil environment for future observations using that historic location. Within each trench, soil color was recorded as part of the visual analyses of soil profiles using Munsell soil color charts. Lastly, photographic records were taken of the vertical wall where soil cores were extracted to document the visible soil layers, color, and structure, which could add to the soil characterizations.
Information on the local shallow water table assisted in the timing of trenching operations. Automated shallow groundwater depths have been collected hourly at the Bondville ICN site (just east of the ICN tower) since 2001 (Illinois State Water Survey 2008). Trenching and core extractions were scheduled when groundwater at the BEARS site was expected to be near its lowest seasonal level to permit core sampling as deeply as possible to lessen possibilities of soil compression due to high moisture content.
4. Results and discussion
a. Topographical and water table analyses
The surface terrain height variability across the entire BEARS field site was roughly 1 m (Fig. 4). Highest elevations were found in a band that stretched from the east-central boundary of the research area toward the northwest corner. Extracting the highest slope observed along a 12-m section of the northwest–southeast or southwest–northeast cross sections centered at each site, where higher-resolution observations were made, revealed that highest topographical gradients were found at the historic site B71 and sites B68, B63, and B62 (0.0492, 0.0592, 0.0433, and 0.0408, respectively), while the clustered site B66 and sites B65 and B64 showed lower slopes (0.0183, 0.0175, 0.0150, respectively). The combined elevation measurements taken at sites B61 and B67 by far had the smallest slope (0.0050). These last four locations appeared to have the best opportunity for ponding because of very small surface slopes and occasional small depressions observed in the elevation data along transects at the sites. Topographic relief around all sites is shown in the online supplemental pages for this experiment (Scott 2009).
Historic shallow water table data (depth to water) from the ICN well (location shown in Fig. 4) indicated that water levels typically are closest to the surface in late winter and early spring, followed by general drying throughout the subsequent summer and autumn. During autumn 2001–06, the earliest dates for when water table levels fell below 2 m (the lowest planned level of soil moisture observations and core extractions) ranged between 14 August and 7 November, averaging 26 September, with levels never falling to 2 m in autumn 2004. In 2007, the depth-to-water data exceeded 2 m on 1 September (Fig. 5) and was 2.3 m below the surface on the date of trenching, 18 September.
b. Soil moisture analyses
Figure 6 shows volumetric soil moisture (percent of volume) at nine neutron probe sites (B61–B68 and B71) within the BEARS field research area during SMUSE. Because of similar empirical results in adjacent 20-cm layers, observed data were averaged into four broader layers: 0–30, 30–90, 90–170, and 170–200 cm of depth. Precipitation totals were included from daily data at the Bondville National Acid Deposition Program site (also located at BEARS), accumulated between successive soil moisture observation days (midnight–midnight, local time).
Overall, strong temporal consistencies in volumetric soil moisture were observed within each layer at each site. Large increases in moisture were evident near the surface, related to rainfall, followed by decreases during periods when rain events were less frequent or with smaller totals. In deeper layers, temporal variability was relatively low at most sites; however, average soil moisture values among the sites spread further apart with depth. In addition, below 90 cm, sudden increases in soil moisture occurred at a few sites in December, remained high but with a stable trend until May, and then fell to much lower values once again late in the experiment, while at other sites, moisture levels remained relatively constant throughout the entire 13 months of biweekly observations. The strong intrasite temporal consistencies throughout the records provide a measure of quality assurance for sensor performance. However, the increasing intersite spatial variability with depth within such a small domain of measurements was greater than expected.
In the layer closest to the surface (Fig. 6a), volumetric soil moisture decreased at all sites during September 2006, a month with very low precipitation but a time when the sod root zone continued to process surface moisture. The trend reversed in October with heavier precipitation. Being the end of the growing season for sod, near-surface water use and evapotranspiration waned. For the region, this is typically the beginning of the near-surface, seasonal soil moisture recharge period.
From November 2006 to the end of March 2007, soil moisture variability was relatively low at each site with values ranging from 35% to 45% by volume and was maintained by substantial precipitation from December to the middle of January. Surface soils froze at 10 cm of depth from January 30 to March 12 (earlier at higher levels), effectively locking in moisture in the top soil layer and capping downward percolation. Furthermore, much of the precipitation in this period fell as snow and remained on the surface with considerable time to sublimate, melt, and evaporate, or blow into ditches before soils thawed, and thus was unavailable for movement into soils. After soils thawed in mid-March, the resumption of sod growth and lower-than-seasonal precipitation caused volumetric soil moisture to fall sharply.
The lowest near-surface values during the experiment were observed in mid-June (16%–23%). Heavy rainfall in late June and early July increased soil moisture quickly, but this was followed by a gradual decrease in moisture again at all sites through the end of the monitoring period, a reflection of average precipitation and continued high summertime moisture use by sod.
The strong precipitation-induced seasonal variability in soil moisture in the top layer of soil is quantified by intrasite standard deviation values, ranging from 5.59% to 8.02% (Table 1). Nevertheless, averaged volumetric soil moisture values over the experiment showed a close grouping of site data—between 30.0% and 33.5% at most sites—throughout the entire period of observations. However, during spring and summer, a peculiar spreading of site data occurred between two relatively close sites—B71 and B68—maximizing at the end of the experiment with overall volumetric soil moisture averages of 29.1% and 34.2%, respectively. Departures between these two sites were accepted as accurate but with no suggestion as to the cause of the differences other than soil attributes discussed during the soil core analyses.
The average intersite standard deviation of soil moisture on each observation day throughout the experiment was 2.26% and varied from 1.33% to 3.64% on individual days (Table 1). One measure of the representative nature of B71 for the field of observations is suggested by comparing the average volumetric soil moisture at the eight primary experiment sites to the amount at B71, which showed a lower value or an under measure of soil moisture of 2.39%.
Results from 30 to 90 cm deep (Fig. 6b) also revealed a response to rainfall, but considerably less striking than that observed in the layer above. Individual site standard deviation values for the experiment were much less as well, varying from 2.86% to 4.46% (Table 1). The strong temporal consistencies continued at each site, but with slightly greater spread of values among sites; the average standard deviation of intersite volumetric soil moisture was 2.92%. The long-term Bondville site was very near the average soil moisture of the experiment sites (a −0.28% difference).
In the 90–170-cm layer (Fig. 6c), the direct effect from precipitation (and drier periods) on volumetric soil moisture was quite muted, but with considerably spreading of site values. The overall ranges of values were higher from early December 2006 to early June 2007 (32%–53%) and lower (24%–42%) prior to and subsequent to these months. Data from the Bondville site barely differed from the average of the showed significant influence on trends in soil moisture variability, resulting in an average standard deviation across all sites of 4.93% (Table 1), more than twice the near-surface value.
It is likely that a seasonal invasion by the local water table occurred at some sites in this layer. Data at sites B63 and B65 recorded a flat trend in soil moisture at the beginning of the experiment with values of approximately 25% and 35%, respectively. In early December, soil moisture increased rapidly at both locations to approximately 40% and 50%, respectively, which were maintained well into spring. By late summer, soil moisture levels had returned near their original values. These patterns closely coincide with the shallow groundwater table depth-to-water data at the field site (Fig. 5) and generated site standard deviation values at these locations of 7.12% and 8.63%, respectively. Similar trends but with smaller ranges were observed at several other sites. Conversely, data values from sites B61, B66, and B67 were uniformly high in this layer with standard deviation values ranging between 1.21% and 1.90%, suggesting that water was more permanent in this layer at these locations throughout the experiment, a fact supported by the trenching and soil core analyses reported in the next section.
A second pattern observed in this layer was an apparent stratification of sites within these same above groups by surface terrain slope. Sites with low surface slope reported higher continuous soil moisture values, whereas sites with relatively higher slopes had periods of drier soils but with a seasonal invasion of the water table. This may imply that local differences exist in other soil attributes, which also may play a part.
Trends in the deepest layer, 170–200 cm (Fig. 6d), were consistent with relatively flat temporal trends in volumetric soil moisture at nearly all sites. Intrasite standard deviation values were the lowest of all layers and varied between 0.43% and 1.71% during the experiment. One exception occurred at site B63 (5.01%), again showing probable water table invasion, or perhaps viewed another way, better drainage may be possible at this site. Conversely, a considerable range of average soil moisture values was observed among the various sites within the layer, ranging between 31.73% at B71 and 44.84% at B61. Comparison of the former to the average volumetric soil moisture across the rest of the network revealed a substantial under measure of soil moisture during the experiment in this layer at the historic neutron probe site by 6.51%.
Earlier research by Vinnikov et al. (1999b) using 14 of the Illinois soil moisture sites (Fig. 1) over a 16-yr period with 19 observations each year found a standard deviation by volume in the top 10 cm soil layer of 8.5%, and 4.0% in the top 1 m of soil. Corresponding values in these layers during this experiment were 3.41% and 2.02%, respectively. The differences between the two experiments are not surprising because of the huge difference in temporal and spatial scales of observations. Nevertheless, most of the Illinois long-term stations boast the same loess, silt loam soil texture with total porosity in the top meter of soil ranging between 417 and 544 mm (Hollinger and Isard 1994). Thus, strong similarities appear to exist; however, as the results from SMUSE show, substantial differences can occur on the very localized scale.
More comparisons of volumetric soil moisture values at the historic Bondville site and each of the eight randomly selected sites within the summed layers are shown in Table 1. The square of the Pearson product moment correlation coefficient values (R2) ranged between 0.85429 and 0.94486 within each station–B71 pair in the 0–30-cm layer but decreased considerably with most pairs in deeper layers, ranging between 0.009 33 and 0.666 78 in the 170–200-cm layer. The spread of data is consistent with earlier results, and it suggests the importance of a pre-evaluation process in future soil moisture site selection protocols.
One method to assess representativeness of sites to the analysis domain is with temporal stability analyses developed by Vachaud et al. (1985). This analysis compares soil moisture data from each neutron probe site to the average soil moisture across the network (Fig. 7). As described by Cosh et al. (2008), sites with small mean relative differences (MRD) reflect the field average, while sites with relatively small standard deviations can be good candidates as regional representative sites. An examination of results in the near-surface layer shows that most sites matched the network average quite well and with small standard deviations (Fig. 7a). The historic site (B71) appeared to be the worst performer, under measuring soil moisture, but was likely was an acceptable site. Site B68 was the next worse, over measuring moisture. A comparison of these two sites quantifies their differences observed in Fig. 6a.
Similar analyses in deeper layers revealed a gradual spreading of MRD values from the network mean at many sites, and some with relatively high standard deviations. For example, confirmation is seen for sites B61 and B67 (Figs. 7c and 7d) that are holding the most moisture in the deeper layers of Fig. 6. Likewise, the likely invasion (and drainage) of the water table at sites B63 and B65 documents how poorly representative these sites are for the data domain. Overall, sites B62 and B64 may be considered to be the best representative locations through all layers.
Of strong interest is a temporal stability plot for the top soil moisture layer, 0–10 cm (Fig. 7e). Most sites show the MRD near zero but with relatively large standard deviations values compared to the 0–30 cm layer (Fig. 7a). Certainly, this layer would be expected to have greater soil moisture variability than other layers: residual moisture from heavy rainfall, direct surface evaporation, higher plant mass, and percolation. The similarities among sites regardless of a selection process provide interest to those seeking ground truth data in support of soil moisture measurements from satellite imagery. Five of the nine sites reported near-zero MRD values and very similar standard deviations. Thus, these five sites would be equally representative in connection to satellite imagery soil moisture observations.
A matrix of spatial correlation coefficients for each layer (Table 2) provides similar information, showing high correlations in the 0–30-cm layer with all sites: 0.9 or better at all sites. Correlations decrease with considerable depth and are especially poor in the lowest layer. In relative terms, it appears site B65 reported the highest consistent performance across all layers. In individual layers, B63 and B65 correlated well to other sites in the 90–170-cm layer that showed some soil moisture increase on the same schedule as a rising water table, just not as high. Strangely, B64 was poorly correlated to all sites in the 170–200-cm layer. Data from the B71 indicate that it may be a less worthy representative of the sod-covered region as a whole than other sites during this short 13-month experiment. Regardless, the question of any site’s long-term representativeness to adjacent crop fields remains unanswered.
Data collected from the clustered site (B66) was designed to observe changes in soil moisture over a very short distance, with three neutron probe tubes installed along a 2-m north–south line. Output was subdivided as before and shows similar results in the top two layers of soil (Figs. 8a and 8b) to that observed at other sites. Volumetric soil moisture values generally were within a few percent of each other. Comparisons with the historic monitoring site revealed values that were similar. However, unexpected results came from deeper layers. At 90–170 cm deep (Fig. 8c), larger soil moisture variability existed at the southern site of the cluster. Values at B66S were the lowest of the three stations in early December, highest in winter and early spring, then lowest again at the end of the experiment. Despite the close proximity of these sites, soil moisture trends at site B66S were more similar to sites B63 and B65, while soil moisture at B66 and B66N closely paralleled site B71. Again, moisture trends in the 170–200-cm layer (Fig. 8d) mirrored patterns observed among all sites in this layer (Fig. 6d)—relatively flat temporal trends—but with unexpectedly high spatial variability among the three sites. Soil moisture at sites B66N and B66, located 1 m apart, measured roughly 33% and 41%, respectively. If the differences observed in the lower two layers at the site cluster were due to soil properties, results here suggest that the choice of a representative site for regional soil moisture could be complex.
One further analysis was made with these data: a subjective division of output into upper and lower root zones used in some previous studies, 0–100 and 100–200 cm, respectively (Fig. 9). Volumetric soil moisture appeared relatively similar at all sites in the upper root zone, with an overall average standard deviation of 2.02%. Similarly, the large spread in intersite values indicated in Fig. 6 at depths below 90 cm was very evident in the lower root zone (Fig. 9b), generating an average standard deviation of 4.57%. To a large degree, sites with the smallest terrain slopes had the highest moisture with nearly static trends throughout the experiment, and those with a greater terrain slope displayed periods of both relatively high and low soil moisture. During parts of the experiment, the level of soil moisture observed at site B61 in the lower root zone was 80% higher than the value measured at site B63. Site B71 was located in the middle of the set of sites. Nevertheless, the question of impacts these revelations would have had on previous research if either B61 or B63 had been selected as the location of our original neutron probe location remain unanswered.
c. Soil core analyses
Although the observed water table level from the site’s shallow water well on the day of core extractions was 0.3 m below our deepest planned core sampling, soil core compression due to wetness occurred in lower levels at some sites. The four sites with the lowest terrain slopes were also, coincidently, sites with the wettest soils at deeper levels. Most of these displayed muddy conditions at 2 m; B67 had standing water at 1.7 m. At B64, water pulsed out of a core hole at 1.4 m as the attempt was made to extract a core at 1.6 m, suggesting a trapped water lens within that particular soil profile. Saturated conditions were also observed at the base of the trench at sites B61 and B66. Core compression occurred at all of these sites, beginning as high as 1.6 m. Soil cores were not collected at levels below where compression occurred. The remaining sites were relatively dry at all levels with no compression, except at 180 cm at site B65. Photographs of all excavation walls and trenches can be viewed at Scott (2009).
Results of the soil core laboratory analyses are presented in Table 3. The data indicate an apparent relationship with surface slope. Regardless that the variability in slopes across the entire 5.9-ha field area may be considered as slight (only 1 m of total relief), locations with the lowest terrain slopes were also sites that 1) reported the most consistent and relatively high soil moisture values in the deeper layers throughout the experiment (section 4b) and 2) reported the highest soil porosity and lowest bulk density values (Table 3). Sites with greater terrain slopes generally reported lower soil porosity and higher bulk densities, as well as higher temporal soil moisture variability.
Field and laboratory soil analyses indicated that animal burrows, crotovinas (former burrows filled with A horizon material), partially filled burrows, and mottling (varying soil colors due to the presence of water) existed in all soil profiles (Scott 2009). The mottling, created by the fluctuating water table, is indicative of heterogeneity of soils processing moisture. The parent materials here are 2+ m of loess soil. Despite these heterogeneities, some general trends are present. Soils tended to become more compressed (more dense, less pore space) with depth, with one profile (B65) having a density maximum in the B horizon at 80–100-cm depth (Table 3). The sites with very low surface slope profiles had thicker A horizons and yellower subsoils, features diagnostic of being wetter than the other soil profiles with greater surface slopes. In general, water content in cores tended to increase with depth (Table 3). However, in three drier profiles (B62, B63, B65), this trend was interrupted deep in the profile. This latter trend is consistent with rewetting of previously dried soil profiles. Overall, these differences in soil genesis may suggest that the topographic differences affecting soil moisture were retained from presettlement prairie conditions.
d. Combined analyses
Results from above have shown similar near-surface volumetric soil moisture at each monitoring site during SMUSE and growing levels of variability among sites in deeper layers with empirical connections to site terrain slope. This is summarized in Fig. 10, where monitoring sites are arranged in order of terrain slope at each location. Those sites with a relatively low surface terrain slope are shown using dotted columns, and those sites with larger slopes are striped. The historic Bondville site, which possessed the highest slope, is the solid black column.
Soil moisture in the 0–30-cm layer was quite similar at all sites, showing an approximate range of soil moisture from 29% to 34%. In this layer, no obvious effect was evident from surface ponding at sites with lowest slope. Average soil moisture was essentially identical from 31% at low-sloped sites and 32% at the moderate-to-high sloped sites. The experiment protocol restricting measurements during precipitation events, as well as the thick mat of sod cover everywhere at the surface, may have given sufficient opportunity for uniform water levels at all sites in this layer subsequent to precipitation events.
Data from deeper layers, however, revealed a record of relatively high moisture values from sites with low surface slope and lower levels of soil moisture at sites with the highest slope. Average volumetric soil moisture at high- versus low-sloped locations was 34% and 37% in the 30–90-cm layer, 35% and 40% in the 90–170-cm layer, and 34% and 42% in the 170–200-cm layer, respectively. High moisture values at the sites with lowest slopes may suggest the occurrence of surface water ponding, yielding more available water to percolate downward.
From a strict “best representative” standpoint concerning the location of the historic Bondville soil moisture site, Fig. 10 suggests that from this 13-month period, site B62, whose data lie closest to the middle of all sites in each layer, may have been a more fortuitous selection, whereas site B61 would have resulted in generally higher values within the 2-m profile. The actual long-term Bondville site location (B71) reported the lowest averaged moisture values in the 0–30- and 170–200-cm layers and was in the middle-to-upper range in the 30–90- and 90–170-cm layers. A study of the soil structure underneath site B71 would be necessary to help describe the low value in the 170–200-cm layer. However, maintaining this site as a pristine research location prohibited nearby trenching.
Lastly, a contemporary indication of soil moisture variability at the Bondville site can be presented. During 2000–04, all ICN sites were converted to Stevens-Vitel Hydra continuous soil moisture sensors. The differences between data obtained by this instrument and those of the neutron probe are substantial. As described in section 2, these new instruments were buried under sod at depths of 5, 10, 20, 50, 100, and 150 cm. They measure soil moisture within a set of sensor prongs forming a regular triangle, 2.2 cm on a side and 5.7 cm long, yielding a volume from which soil moisture data are measured of approximately 12 cm3. This contrasts with the neutron probe observations, which are taken within a spherical volume of soil with a diameter of at least 10 cm (which varies in size with moisture content), yielding approximately 4189 cm3. Data from the continuous sensors are measured hourly, whereas data from the neutron probe observations were collected twice weekly during the field experiment, and 19 times a year in the historic dataset. For these reasons, neutron probe observations would be expected to display a highly smoothed temporal and spatial dataset with a substantially muted reaction to precipitation.
Both soil moisture observation platforms collected during SMUSE are shown in Fig. 11. The Bondville neutron probe site (B71) is located approximately 21.5 m south of the ICN tower (Fig. 2). The continuous sensors, attached to the ICN datalogger, are located about 3.7 m south-southeast of the tower. Soil moisture is presented for each dataset at a depth of 10 cm. Hourly precipitation observations have been superimposed.
Again, it should be remembered that the experiment protocol did not allow neutron probe readings on days with precipitation, reducing the opportunity for neutron probe sensors to observe the same effects of rainfall as was allowed with the continuous probes. In addition, output from the capacitance sensors were unusable during times of frozen soils at this level (February–mid-March) and were removed from the figure.
Soil moisture values from both the continuous and neutron probe data sources are fairly similar between rain events, within a few percent. However, the continuous data reveal substantially higher soil moisture values during rain events, and they show the speed at which water moves through the soil profile at the BEARS site at the 10-cm depth, and then quickly returns toward a base seasonal value. The restriction set in our experiment protocol caused our observations to miss these events; however, the authors contend the restriction, as described earlier, was necessary.
During SMUSE, soil moisture in the 0–10-cm layer often showed less of an effect to precipitation than at 10–30 cm deep (Scott 2009), suggesting a rapid percolation, evapotranspiration, or root use of precipitation within the top 10 cm of sodded soil. Seasonal trends in Fig. 11 are noteworthy with heavier rainfall events in summer, being concomitant with high sinks of soil moisture (root extraction and evapotranspiration) compared to the winter season, a time of lower precipitation amounts and minimal root extraction of moisture when much of the rain that falls is available for soil recharge.
5. Summary and conclusions
Long-term data collections of volumetric soil moisture under sod have been conducted in Illinois for more than 25 years by the Illinois State Water Survey’s Illinois Soil Moisture Network, primarily using neutron probe observations. Numerous prior studies have applied these data to various regional scenarios. However, true representativeness of these soil moisture data to adjacent areas of various surface covers is unknown. This work indicates that care needs to be taken with how these in situ data are applied.
A 13.5-month field study, the Soil Moisture under Sod Experiment, was conducted at the ISMN site at Bondville, Illinois, to increase understanding of soil moisture variability across a relatively small area under seemingly uniform conditions. Eight locations for neutron probe observations were chosen at random for monitoring across a 5.9-ha sodded field, as well as two additional stations near one of these random sites for a cluster site analysis.
Observations of soil moisture were taken twice weekly at these 10 sites and the Bondville ISMN station using a neutron probe from the surface and at 20-cm intervals to 2 m below ground on days without afternoon precipitation from August 2006 to September 2007. A detailed surface terrain analysis was conducted across the field site to determine the potential for ponding near each monitoring site from rainfall runoff. At the end of the period, trenches were dug adjacent to the randomly selected stations to extract soil cores at the same levels of the neutron probe observations to conduct analyses on soil porosity and bulk density, as well as soil color, and for a determination of heterogeneity of soil conditions.
Results indicate 1) strong, intrasite, temporal consistencies in volumetric soil moisture at all levels and 2) increasing intersite soil moisture variability with depth. Values in the top 30 cm of soil responded quickly to rainfall and subsequent drying. Intersite soil moisture variability on a given observation day in the near-surface layer was relatively uniform, with a standard deviation of 2.26% during the experiment, but it increased to 5.19% in the 170–200-cm layer. Although surface topography varied by only 1 m across the entire 5.9-ha field, a connection was observed in deeper layers at sites with low surface terrain slope having high soil moisture content and vice versa. Variability in soil porosity and bulk density supported a similar relationship between site terrain slope and soil moisture. A noticeable level of mottling was observed visually and in soil cores, indicative of heterogeneity in the soils’ processing of moisture. Data from the clustered group of stations, with a separation of just 1 m, revealed larger than expected and inconsistent soil moisture variability at such closely spaced locations.
Results suggest a high level of soil moisture variability, typically undetected without a measure of soil profiling. An apparent water lens observed in the 140–160-cm region at site 67, a large seasonal fluctuation of water at sites B63 and B65 apparently due to water table intrusion, a high amount of soil moisture with very low variability at sites B61 and B67, and the conflicting results observed in the clustered site data (site B66) are all indicative of the existence of localized soil conditions in this field location that created a wide variety of water flow patterns under surface conditions that visually appear to be quite uniform. Sites with the driest conditions in trenches at 2 m were also generally sites with the lowest soil porosity. Perhaps the lower porosities at sites B63 and B65 forced a higher water table in their site profiles. Likewise, the larger porosity of soils at sites B61 and B67 allowed more water to be stored in the middle soil moisture layers and thus not invade to the levels closer to the surface. Unmeasured was permeability, which surely affected all sites, allowing surface water to percolate differently as well as affecting upward water movement from below. Furthermore, soil characteristics below 2 m of depth are unknown anywhere across the region, as well as their impacts on the soil layers above. These data indicate the importance of soil characteristics to the results of this experiment and add further to the complexity of proper interpretation of soil moisture trends as well as the selection process of future sites.
Vinnikov et al. (1999a) used root-mean-square errors to conclude that 10 monitoring stations would be sufficient to determine average soil moisture within the top 1 m of soil for the state of Illinois. Famiglietti et al. (2008) suggested that “a maximum of 18 samples would be required to measure the 800-m mean soil moisture to within 3%.” Our conclusions add to these statements, suggesting from temporal stability analyses that a small number of sites may be needed at each regional site to properly define local soil moisture conditions to select the most representative site for a given area.
The experiment raises concerns that the use of just one site to monitor soil moisture within a broad area, without additional analyses for its selection, may have less regional or even local representation than previously accepted. Multiple site monitoring was not conducted at any of the current ISMN sites during the siting selection process. It is anticipated that similarly strong soil moisture variability exists at all ISMN sites over distances as short as those observed at Bondville during SMUSE.
Nevertheless, since the beginning of our network, especially with its limited historic observation schedule, and collections of data without regard to ongoing precipitation events, our advice on the best of the ISMN data has been with the observed changes between successive observations and departures from developed norms. We took this position because of the unknown association between our data collected under sod and the soil moisture community’s greater interest: soil moisture under adjacent areas. The current work, finding high variability across a single sodded field, has not changed that viewpoint.
The future of near-surface soil moisture data collection is moving largely in the direction of satellite imagery and numerical simulation to monitor and predict soil moisture globally, a plan with substantial current efforts by the International Soil Moisture Working Group of the Global Energy and Water Cycle Experiment (Leese et al. 2001). One of our results here is that near-surface soil moisture observations all under sod in loess soils over one small region in Illinois are similar. This is a fortuitous finding for near-surface remote sensing observations. But similar analyses as was done here may be required at each ISMN site to determine the local representativeness of in situ data for adequate data validation and model parameterization. This is even more important for quality soil moisture observations in deeper layers. The relationship between the historic Illinois soil moisture data under sod versus that in adjacent areas under different surface covers was not addressed here, except to point out that arbitrary acceptance of such representation without additional analyses may be unwarranted.
Protocols for high-quality soil moisture data are needed that require multiple data analyses within a region prior to the installation of permanent sensors, including measurements within all major surface covers, to determine a representative location for soil moisture monitoring. Adequate soil core analyses at numerous locations within a small area, a surface slope analysis, a climatology of the near-surface water table, as well as precipitation and perhaps evapotranspiration is advisable. Pitfalls will include, among other items, an impact assessment of local land use, for example, attempting to define a representative soil moisture environment in agricultural areas that are traversed constantly by agricultural vehicles and the magnitude of tiling. Construction of universal protocols of soil moisture data collections and a unified global development strategy for monitoring could address many issues noted here, as they apply to current and future soil moisture networks.
This work was conducted by the Illinois State Water Survey. The authors sincerely appreciate various ISWS staff members’ assistance during the experiment: Charles R. Mitdarfer, for neutron probe operations and soil core preparations; Paul Nelson, for numerous areas of training, data collection, and sensor quality assurance; and Mary LeFaivre, for serving as our safety officer associated with the trenching activities. Special thanks are given to Momcilo Markus of ISWS, for statistical analyses and advice. Appreciation is extended to Lisa Sheppard, for editing expertise, and to Sara Olson, for figure preparation and photographic documentation of trenching activities. Special recognition is given to Derek Winstanley, chief of the Illinois State Water Survey during the experiment, for his direction and guidance.
Corresponding author address: Robert W. Scott, Water and Atmospheric Resources Monitoring Program, Illinois State Water Survey, Institute of Natural Resource Sustainability, University of Illinois at Urbana–Champaign, 2204 Griffith Drive, Champaign, IL 61820. Email: firstname.lastname@example.org