U.S. Climate Reference Network Soil Moisture Observations with Triple Redundancy: Measurement Variability Open Access

The U.S. Climate Reference Network (USCRN) is operated by the NOAA National Climatic Data Center (NCDC) to address questions regarding changes in the United States climate over time (Heim, 2001). Conceived in response to a statement of best climate observing principles declared in the mid-1990s (Karl et al., 1995) and adopted by the National Research Council (NRC, 1999) and Global Climate Observing System (GCOS, 2003), USCRN was deployed between 2001 and 2008 in the conterminous United States. A total of 114 stations covers the lower 48 states and offers the highest quality measurements of temperature and precipitation in stable rural environments; the network is continuing to expand in Alaska (Diamond et al., 2013). The network is in the early part of its planned life span of more than 50 yr but has already produced enough observations to show a strong correlation between annual United States air temperature anomalies as measured by USCRN and those derived from bias and inhomogeneity corrected U.S. Historical Climatology Network stations (Menne et al., 2009), verifying these results (Menne et al., 2010).

The engineering of USCRN stations is quite different from most standard observation networks (Diamond et al., 2013). Key to the design is the use of triplicate configurations of instruments for its mission critical measurements of temperature and precipitation. While the air temperature and precipitation instruments are calibrated to National Institute of Standards and Technology traceable standards, the key advantage of three measurements is the assurance of accuracy through pairwise comparison and continuity through redundancy. The stations also measure ancillary variables that can assist in the quality control of the primary measurements: global solar radiation, surface temperature, 1.5-m wind, and precipitation occurrence (using a wetness sensor). The stations are designed to be extensible, and as new NOAA missions arise, new instruments can be added to the original complement.

Therefore, when the NIDIS Act of 2006 (Public Law no. 109–430 (2006), 15 U.S.C. §313d) brought to the fore concerns about the amount and types of observations available with which to monitor drought, USCRN was in a position to augment its original climate measurements with soil moisture, soil temperature, and relative humidity observations in support of NIDIS.

A panel of experts in soil moisture measurement convened in a workshop at Oak Ridge, TN, in March 2009 to discuss the nature of the instrument package to be added to USCRN stations. The type of instrument chosen by the USCRN program was one capable of measuring soil moisture using existing coaxial impedance dielectric sensor technology, while simultaneously measuring soil temperature. The Hydra Probe II (from Stevens Water Monitoring Systems, Inc.) was used by the USDA NRCS Soil Climate Analysis Network (SCAN, Schaefer et al., 2007). This SCAN experience and the robust scientific analysis of the instrument (Seyfried et al., 2005; Seyfried and Grant, 2007) were important in its choice by the USCRN Program.

The configuration of soil monitoring instruments is the key decision by the USCRN Program that makes this network truly unique among other soil observational networks. Instead of placing one set of probes at standard WMO depths of 5, 10, 20, 50, and 100 cm, the USCRN follows its triplicate instrumentation approach to key observations by placing three sets of soil probes at each station. Each of the three sets was placed in undisturbed soil 3 to 5 m from the instrument tower, usually oriented to the north, southeast, and southwest. The first four depths were installed from a pit at each location, with the probes placed horizontally at the required level. The 100-cm probe was placed through a vertical hole bored into the 100-cm-deep strata. Wire leads are looped so as to avoid any water running directly down wires to the vicinities of the probes. Since the station locations were already established using criteria designed specifically for temperature and precipitation measurements, only 90 sites were located where soil probes could be installed at all five depths targeted. For 23 locations with considerably rocky substrates or near-surface rock or mineral layers, instruments were installed at only the 5- and 10-cm depths. One USCRN station near Torrey, UT, is surrounded by solid rock, and no soil probes were installed at that location. Finally, all stations received an atmospheric relative humidity instrument as part of this installation, as well as an improved datalogger and advanced circuitry to account for the needs of up to 15 soil probes at each station. The distribution of USCRN stations across the conterminous United States is displayed in Fig. 1, with the symbols representing the number of layers installed (5, 2, or 0) at each site. The first 90 d of observations are not used in any analyses so as to allow the soil matrix to settle around the soil probe tines.

Having three sets of soil moisture and soil temperature measurements at each USCRN station provides many unique opportunities for understanding the nature of soil moisture and measurements in sparse networks. Unfortunately, these opportunities are not like those that exist in measuring temperature three times in the same volume of air or in measuring the weight three times of precipitation in a single bucket. Since both the characteristics of the soil matrix and the impacts of probe installation can cause adjacent soil moisture measurements to be different, this does not allow for quality control of the measurements by direct pairwise comparisons, as is done by the USCRN Program with temperature and precipitation. However, the redundant observations still allow for the development of an improved understanding of soil moisture measurement variance over time and the uncertainty inherent in singular instrument measurement of soil moisture. Of course, observation continuity is also enhanced when there are three independent measurements of each depth.

This paper will describe in detail the USCRN Program effort to provide a new set of enhanced soil moisture and soil temperature observations for the United States. The advantages of having a triplicate configuration of instruments will be shown for observation availability, quality control, and understanding the variability of measurements made in small plots over a variety of conditions. The changes in soil moisture between April and July for 2011 and 2012 in the United States will also be examined to see the response of the network soil moisture measurements to the onset of severe drought in the Midwest. Finally, future NCDC research plans for USCRN soil moisture and temperature data will be discussed, including satellite validation and modeling, and data sources and availability will be given.

The adoption of a triplicate configuration for instruments in an observational network is a logical extension of the evolution of experimental and monitoring work in soil moisture and soil temperature. Where resources are available, most experimental watersheds and small area or state networks (McPherson et al., 2007, Illston et al., 2008) often deploy closely spaced and/or redundant soil probe configurations. However, this was not feasible in most cases 30 or more years ago due to the expense and fragility of early electronic soil moisture probes. By the time the first U.S. federal effort to form a national network started with the USDA in the 1990s, single sets of probes sampling the five WMO depth levels were installed at sparse network locations (SCAN, Schaefer et al., 2007). However, these stations are still vulnerable to single instrument failure and associated quality issues.

An examination of prior USCRN air temperature observations in the western United States (Palecki and Groisman, 2011) illustrates the advantages of a triplicate configuration. While the air temperature instruments are substantially more reliable than soil probes of any variety, the rate at which data passed quality control and resulted in calculated air temperature values was higher by a small margin (several tenths of a percent) for the combination of three instruments than for any one instrument. This percentage of increased continuity is much higher for the soil probe network due to the higher failure rate of underground instruments, as will be discussed later. The completion of the USCRN soil-probe and humidity instrument deployment across the continental United States in August 2011 not only increases the number of soil moisture and soil temperature observations available, but also provides measurements of high continuity and quality.

While in theory the dielectric can be lower than 2.7, this still represents a very small amount of moisture, so the assumption of zero is reasonable (Seyfried et al., 2005).

USCRN quality assurance consists of two main components: real-time quality control during data ingest and post-processing longer data records after the fact. The initial quality control determines if dielectric or temperature values are out of the expected physical range, and these cases are quality flagged at occurrence. Dielectric is accepted in the range of 1 (dielectric of air) to 80 (dielectric of pure water), even though realistic values are well inside these bounds. Temperature is accepted between −30°C and +65°C. A separate flag is thrown for the dielectric data if soil temperature at the same depth is below 0.5°C, a temperature that assures calculation of soil moisture does not take place when there could be ice formation near the soil probes that would contaminate the dielectric measurement.

In addition to the automated real-time quality control, overall quality assurance depends heavily on the examination of data by experts using graphing tools and tabular statistics of questionable data derived from post-processing long data segments. A “bad sensor” list is developed so that time ranges of observations from poorly functioning soil probes can be flagged even if they did pass real-time quality control. Among the conditions examined are the number of missing values, gross errors, spikes, and irregular jumps in the data record, along with an overall assessment of sensor noise. These statistics are derived from 1-mo-long periods of service, and those probes with questionable values are examined graphically to determine the type or types of problems being observed. Often, a poorly operating probe can easily be discerned in comparison to its neighboring probes at the same level. Similar to the soil moisture post-processing, soil temperature statistics are collected for the number of missing values, gross errors, spikes, and irregular jumps in the data record, along with an overall assessment of sensor noise. Many fewer probes are placed on the “bad sensor” list because of overall better performance by the temperature thermistors. Probes are not usually immediately repaired unless the entire set of probes is offline, as can happen if lightning hits the ground nearby the station, or an electronic component goes offline. Most of the time, probes are repaired at annual maintenance visits.

The availability of the soil moisture and temperature data is enhanced by the triplicate instrument configuration, although the overall failure rate for all three instruments at one level simultaneously is much higher than for the temperature and precipitation measurement systems. For example, during a sample hour on 3 Aug. 2012, 93 individual probe soil moisture measurements were missing or flagged (Table 1), which is 6.3% of the total number for the network. Of these, 20 single probes failed, leaving 2 probes at each depth functioning; 5 pairs of probes at a given depth failed, leaving 1 probe at each depth functioning; and 21 full sets of probes at a given depth failed, leaving no probes functioning at those depths. Therefore, comprising the 6.3% loss of probes, 4.3% involved the total loss of signal for a depth and location, while the other 2.0% represented no loss of information due to redundant probes being available at the depth and location in question.

Comparisons of soil moisture results to the temperature results yield interesting information on probe failure, given that each probe is measuring both soil moisture and soil temperature. Only 5 single probes, 2 probe pairs, and 1 probe triplet failed quality control of soil temperature on this day and hour, mostly due to observations missing. This means that only 0.2% of the probes, at most, have completely failed. By comparison, most of the soil probes with moisture data flagged are still making observations but are failing quality control. The primary cause of soil moisture observation failures is the incompatibility of the coaxial impedance dielectric sensor operating at 50 MHz with the conditions in the ground. These cases also explain the increased likelihood of bad probes being found at 100-cm and 50-cm depths, where the impacts of clay cations and water table height can become pronounced in determining soil moisture but not in observing soil temperature (Bell et al., 2013). Soil salinity levels may also impact readings, as they do at Coos Bay, OR, where the current USCRN soil moisture measurements are strongly affected by a saline-rich tidal flat near the ocean. Overall, the new failure rate for probes measuring soil moisture was approximately 4.6 per month during the first full year of the network in 2012. Probes subject to systematic environmental incompatibilities make up the remainder of the bad probe list.

Probes that are either malfunctioning or not working correctly due to environmental causes are placed on the “bad sensor” list. At the time of the analysis of the sample above, 81 of the 93 bad sensors were already on the “bad sensor” list and had been noted to be problematic by expert review, as described earlier. Some of the probes not already on the “bad sensor” list may be placed there in the next monthly review cycle by the expert, or may return to proper range as conditions change or probes are replaced.

About 80 of the USCRN stations had received soil moisture and temperature instruments by the end of 2010. This representative proportion of the deployment allowed for network-wide analysis of the resulting soil moisture measurements during the 2011 and 2012 growing seasons from April through July. While climate clearly varies across the United States, these bulk statistics can be used to examine the general behavior of the soil moisture as measured by USCRN. Means were calculated at an individual station for every hour and depth using the three probes (excluding flagged probes) and averaged for the month. Standard deviations for each station were calculated hourly from the cases where three valid soil moisture values were available for that hour and depth, and averaged for the month as in the case of the hourly means. The monthly values were then averaged for all available USCRN stations to identify changes across time and depth common to the network as a whole. All analyses were conducted only on stations that had monthly values for both 2011 and 2012 and were limited to the 5-, 10-, 20-, and 50-cm depths. The 100-cm depth was excluded from this examination; locations with clay soil cation exchange issues and/or seasonal water tables above 100 cm resulted in a smaller numbers of valid cases being available.

In addition to examining the means and standard deviations, the cofficients of variation (CVs = standard deviation/mean) were calculated for each station-month and depth and subjected to an aggregate examination to estimate the local noise/signal ratio of USCRN stations with three soil observation plots. Unlike spatial averages of means or standard deviations evaluated in the past with this method (Choi et al., 2007), equations were fit representing the relationships between monthly average CVs and monthly means at stations with multiple measurements for individual depths and all cases in time. A simple logarithmic equation was determined to be the best fit of all the equations evaluated for the analysis. A total of eight scatterplots and equations were used in the Results and Discussion section below to evaluate the relationship of CV to mean soil moisture at different soil depths for April to July 2011 and April to July 2012. Due to occasional instrument failure and the proportion of stations having soil observations only at two depths, the number of station-months used in each analysis varied by depth.

Figure 2 shows two common cases, one being full agreement among the soil moisture probes at the 5-cm depth at Champaign, IL, and the other being parallel agreement with a nearly constant offset for Probe 1 at the 10-cm depth at Mercury, NV. The Champaign case indicates that all three soils being sampled have similar moisture holding characteristics. The Mercury case indicates all samples have received similar amounts of precipitation and have similar soil moisture holding characteristics but with perhaps a differing installation environment that incorporated fewer completely dry rock elements near Probe 1. In both of these cases, all three probes can be used for layer averaging and inclusion in products.

In the case of the Asheville, NC, site ∼21 km (∼13 miles) south of town, all three measurements of soil moisture disagree to some extent (Fig. 3). When precipitation events occur, more soil moisture is measured by Probe 2 than by Probe 3, although they do respond in parallel. On the other hand, Probe 1 has negligible reaction in comparison to the same precipitation events. Probes 2 and 3 appear to reflect real differences in the soil matrix around their sensors, with Probe 2 surrounded by soils with more water holding capacity. Probe 1 behavior seems to be defective, and it would be placed under consideration for inclusion in the bad sensor list.

The previous cases have shown interprobe relationships that were consistent over time. There are cases, however, where the relative probe relationships change over time. At the USCRN site 11 miles southwest of Newton, GA, the 5-cm-depth Probe 1 is consistently wetter than the other probes at that depth (Fig. 4). However, during October 2011, Probe 3 measures the second wettest soil moisture levels, while Probe 2 is second wettest by far during late November and December, despite similarly sandy soils.

The relative switch is long lasting, and any attempt to establish a relationship between sensors would need to be segmented over time. More importantly, it is likely that any attempt to conduct field calibration with gravimetric sampling may garner inconsistent relationships depending on when and where samples are taken on the plot. This is a major finding with the new network, leading to the conclusion that relationships between soil moisture measurements over relatively small areas (just several meters in this case) can change and even reverse as a seasonal cycle progresses. Using gravimetric sampling techniques to calibrate in situ soil moisture instruments may produce less than definitive results where these dynamics exist.

USCRN mean monthly soil moisture values for April, May, June, and July increase with depth in all cases in 2012 (Fig. 5b) and in all cases except for the April 20-cm average in 2011 (Fig. 5a). The values also decrease with a monthly succession in general, except for the increase that occurred in May 2011 at the 20-cm and 50-cm depth (Fig. 5a). These relationships are well understood, with most of the signal for the United States being dominated by semiarid to humid climates with active plant growth preferentially stripping moisture from the top soil layers as the growing season progresses. There are also differences between the 2 yr at all depths and months, with values for 2011 (Fig. 5a) being higher than the values for 2012 (Fig. 5b). More than 70% of the country was involved in drought during the 2012 growing season (Showstack, 2012), while the extent of drought in 2011 was largely confined to two areas in the south-central and southeast United States.

The average hourly standard deviations of USCRN soil moisture values during April, May, June, and July serve as measures of the levels of agreement between the three individual hourly soil moisture measurements during each month (Fig. 5c and 5d). The average 10-cm soil moisture standard deviations demonstrate a strong correlation with means during April to July, falling off in magnitude as the mean soil moisture does the same. However, the other depths are clearly impacted by processes beyond the magnitude of the mean soil moisture. While the standard deviations of triplicate measurements decline at the 5-, 10-, and 20-cm level from 2011 to 2012, as might be expected with less moisture overall, 50 cm standard deviations run contrary to this, increasing with decreasing mean soil moisture values. This hints at a more complex relationship between measurement variance and mean.

While all the results were statistically significant at p < 0.0001, the highest variance explained was consistently associated with the 5-cm soil depth. The lowest variance explained for individual years was the 50-cm soil depth. The R² values for 2012 decrease and the logarithmic equation coefficient become less negative (flatter curve) with increasing depth. However, in 2011 the 20-cm depth R² value and logarithmic coefficient reversed this tendency. Breaking this down by month (not shown), in 2011 this reversal in coefficients and R² at 20-cm depth occurred in April, May, and June, and at 50-cm depth in July.

Previous studies have found similar patterns in the relationship of soil moisture CV to mean soil moisture (Famiglietti et al., 1999; Choi and Jacobs, 2007; Choi et al., 2007). Unlike the previous analysis of Choi et al. (2007) that selected an exponential equation as the best fit between these two parameters, our analysis of the USCRN observations determined that a greater statistical significance is found with the logarithmic equation. However, previous research has focused on individual stations or watershed-scale analysis and this is the first study that could identify patterns between CV and mean soil moisture at the continental scale. These results demonstrate that the variability between sensors is greatest when the conditions are drier. Thus, the representativeness of a single point measurement of soil moisture is less likely to depict the surrounding area in drier regions and occasions. Because of USCRN’s triplicate configuration, researchers can use the variability between the three sensors to approximate the spatial representativeness of the individual station for modeling purposes and satellite validation. These large-scale analyses are possible with the unique design of USCRN soil moisture observations and the relatively uniform spatial distribution of the stations across CONUS (Bell et al., 2013). As previous research focused on relationships with varying spatial resolutions (Rodriguez-Iturbe et al., 1995), our research is able to show that a consistent pattern can be identified at a continental scale. The possibilities of using USCRN soil measurements for further analysis on the spatial and temporal variability of soil moisture will only increase as the network continues to produce more years of data.

Although the distribution of USCRN stations was not optimized for observing soil moisture in the conterminous United States (Diamond et al., 2013), the soil moisture response to the widespread and intense 2012 drought in the mid-latitudes can be seen in the network-wide behavior of means and coefficients of variation. However, the unique variations evident at many individual stations serve to point out the combined effects of local soil properties and installation-related issues that must be taken into account when using soil moisture data from sparse networks. Even with the likelihood of installation inconsistencies and other potential measurement errors, the triplicate design of USCRN helps identify irregularities that result from local issues in a manner not possible with networks deployed as single instrument sets. The USCRN should be considered to be an exemplar for future in situ network deployments.

Because the USCRN is sparsely distributed, further work must be done to study the spatial representativeness and calibration of key sites to provide observations with greater utility for satellite and climate modeling validation. Plans are in place for select locations to be visited at times of differing moisture conditions to take gravimetric sample sets at 5-cm depth and relate volumetric water content derived from the electronic probes to these standard weight-based measurements. In addition, initially two USCRN sites and then more will be surrounded by a number of satellite soil moisture observation systems to determine the representativeness of the station observations for the surrounding several kilometers. The unique triplicate observations at the stations themselves can then be used to better characterize the area around sites which are not subject to special study, simply by relating the results of these special studies to measurement variance among the three station probe sets at those locations.

The USCRN daily and hourly soil moisture and soil temperature values can be accessed at http://www.ncdc.noaa.gov/crn/qcdatasets.html in the Daily01 and Hourly02 subdirectories. README.txt files are available to explain the format of each file type, which includes not only layer average soil moisture and temperature but also other important surface climate variables measured at USCRN sites, including air temperature, surface temperature, precipitation, relative humidity, and solar radiation. These data are also being copied on a regular basis in the International Soil Moisture Network database (Dorigo et al., 2011a, 2011b).

USCRN soil moisture measurements still face some challenges in locations where the currently deployed probe technology does not work well, such as in the saline sediments of the tidal flats at Coos Bay, OR. Alternate technologies are being tested and will be deployed in these locations eventually. As continuity over time is one of the key requirements for observations in this network, the USCRN Program is also participating in other intercomparison studies to always maintain alternatives if primary instrument manufacturers were ever to cease production. It is also anticipated that a program of gravimetric sampling will commence soon to verify surface measurements at a subset of stations for use in model and satellite soil moisture verification. Work is ongoing to address soil moisture instrument issues where needed.

While the primary goal of USCRN soil moisture and soil temperature observations is understanding long-term variability and trends associated with a changing U.S. climate, USCRN soil observations are already proving useful for applications specifically targeting drought detection and impacts. While drought can clearly be seen in absolute volumetric water content values, the nuances of drought classification rely on generating relative variables such as departures from estimated normals or percent plant available water, both of which are under development. USCRN soil moisture and soil temperature measurements are also being used to characterize seasonal phenology, and future applications are anticipated in the fields of ecology, agriculture, hydrology, and meteorology. The USCRN Program plans to maintain high quality, continuous, homogeneous measurements of soil moisture and temperature over many decades for monitoring climate change while serving the needs of user communities benefitting from the availability of these data in real time.

We greatly appreciate the work of the USCRN staff at NOAA NCDC and ATDD for ensuring the continual success of the program. This work was supported by NOAA through the Cooperative Institute for Climate and Satellites—North Carolina under Cooperative Agreement NA09NES4400006. The USCRN soil moisture network is supported by NOAA Climate Program Office and the National Integrated Drought Information System Program. We especially thank Scott Embler, Diana Kantor, and Rocky Bilotta for technical assistance, and Michael Brewer, Scott Applequist, Jay Lawrimore, Tom Peterson, and the external reviewers for their editorial suggestions. The views, opinions, and findings contained in this report are those of the authors and should not be construed as an official NOAA or U.S. government position, policy, or decision.