Approaches to calibrate in-situ capacitance soil moisture 1 sensors and some of their implications 2

Correspondence to: Nicole A. L. Archer (nicarc@bgs.c.uk) 7 Abstract. Capacitance probes are increasingly being used to monitor volumetric water content (VWC) in field 8 conditions and are provided with in-built factory calibrations so they can be deployed at a field site without the 9 requirement for local calibration. These calibrations may not always have acceptable accuracy and therefore to 10 improve the accuracy of such calibrations soil-specific laboratory or field calibrations are required. In some cases, 11 manufacturers suggest calibration is undertaken on soil in which the structure has been removed (through sieving 12 or grinding), whilst in other cases manufacturers suggest structure may be retained. The objectives of this 13 investigation were to i) demonstrate the differences in laboratory calibration of the sensors using both structured 14 and unstructured soils, ii) compare moisture contents at a range of suctions with those predicted from soil moisture 15 release curves for their texture classes iii) compare the magnitude of errors for field measurements of soil moisture 16 based on the original factory calibrations and the laboratory-based calibrations using structured soil. 17


Introduction
Field-based, in situ, automatic sensors and networks formed from them, are increasingly used to monitor soil moisture, and provide unique possibility to investigate spatial and temporal soil water dynamics (Vereecken et al., 2014), which is important for a range of applications related to the hydrological cycle including agricultural production, meteorology and groundwater recharge.
One of the most popular soil moisture sensors is the capacitance probe and is based on functional relationships established between soil moisture content and dielectric permittivity (Ɛ), which can be determined for a small volume of soil using a series of probes (e.g.ML3 ThetaProbe, Delta-T devices Ltd, Cambridge, UK, 5TE sensor Decagon Devices Inc, Pullman, US and CS658, Campbell Scientific Inc., Logan, US).Such sensors are provided SOIL Discuss., doi:10.5194/soil-2016-40, 2016 Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.
with in-built factory calibrations so they can be deployed at a field site providing data on volumetric moisture content, without the requirement for local calibration.The commercial companies who supply the sensors provide uncertainties for these calibrations.For example, Decagon Devices (2014) suggest accuracies ± 3% of volumetric water content (VWC) for the 5TE sensor using a generic calibration in mineral soils that have a solution electrical conductivity <10dS/m, Delta-T Devices(2016) specify that the ML3 ThetaProbe has ± 1% accuracy using three different generic soil calibrations depending on soil type and soil salinity <0.034 m 3 m -3 and Campbell Scientific (2016) state ± 2.5% accuracy for the CS616/625s standard factory calibration when bulk soil EC is ≤0.5 dSm -1 and soil bulk density is ≤1.55 g cm -3 .All three soil moisture sensors have a VWC measurement range between 0% and 50%.
Of the few published studies that have assessed the accuracy of the factory calibrations, Varble and Chávez (2011), conclude that the factory calibrations of capacitance probes, such as the 5TE Decagon Devices, Inc., Pullman, WA and TDT, Acclima, Inc., Meridian, ID are acceptable for the sandy clay loam in applications that do not need high accuracy, however for soils that have more clay/silt content, a field-based calibration is recommended.Luis Gabriel et al. (2010) on the other hand recommends soil-specific calibrations are needed only if users are interested in absolute values, but not necessary for relative differences.
To improve capacitance probe accuracy for soil water measurement, soil-specific laboratory or field calibrations are required (Blonquist et al., 2005;Bogena et al., 2007;Evett et al., 2006;Luis Gabriel et al., 2010;Parvin and Degre, 2016;Varble and Chavez, 2011).To calibrate capacitance sensors it is necessary to understand the relationship between the dielectric constant (Ɛ) and soil moisture.This is done by using standard dielectric liquids e.g.(Bogena et al., 2007;Jones et al., 2005;Rosenbaum et al., 2010) or soil-specific calibrations, e.g.(Chanzy et al., 1998;Luis Gabriel et al., 2010;Varble and Chavez, 2011).Soil-specific calibrations can be achieved in the laboratory or field site by modifying moisture content and using gravimetric approaches to determine a range of soil moisture and associated Ɛ values.In some cases, manufacturers suggest calibration is undertaken on soil in which the structure has been removed (through sieving or grinding) (Campbell Scientific Inc., 2016;Cobos and Chambers, 2010;Parvin and Degre, 2016), whilst in other cases manufacturers suggest structure may be retained (Delta- T Devices Ltd., 2013).There are three reasons we considered that sensor calibrations should ideally be undertaken using structured soils.First, removal of structure could lead to unreliable results when sensors are deployed in structured soils in the field.Second, the importance of soil structure in determining pore size distributions (and associated matric potentials; (Nimmo, 2004)).Third, the so called 'sphere of influence' (the soil volume around the probe's electrodes influencing the probe's measurement), which is typically small for capacitance probes (Chanzy et al., 1998), making them sensitive to small scale variations in soil structure and soil water content near to the probe electrodes (Evett et al., 2006).
To our knowledge, there has been little or no research to date on the magnitude of any error and (or) systematic bias introduced into soil moisture sensor measurements using either: i) factory calibrations; or ii) calibrations based on soils from which the structure has been removed.This has important implications because practitioners may be unaware of these potential errors.
Such practitioners may also rely on pedotransfer functions (PTFs) derived from relatively large databases for soils of differing texture class and bulk density (e.g.ROSETTA (Schaap et al., 2001)) to provide a certain amount of SOIL Discuss., doi:10.5194/soil-2016-40, 2016 Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.validation for the initial measurements from their moisture sensors.However, PTFs may also be prone to error and (or) bias and could provide false confirmation of erroneous sensor outputs.For example, Hodnett and Tomasella (2002) found that PTFs developed from a temperate dataset had a poorly represented clay textural class, which resulted in errors for predicting water retention curves for clay soils.To provide an absolute understanding of soil moisture relationships across a range of hydrological conditions it is often necessary to measure soil moisture release characteristic curves using a pressure plate (Klute, 1986).
In this paper we present our findings from laboratory measurements to determine the accuracy of factory calibrations for a capacitance probe in clay dominated soils at a field site in northern England.We demonstrate the differences in laboratory calibration of the sensors using both structured and unstructured soils.We compare moisture contents at a range of suctions with those predicted from soil moisture release curves for their texture classes.Finally, we compare the magnitude of errors for field measurements of soil moisture based on the original factory calibrations and the laboratory-based calibrations using structured soil.We summarize our findings and discuss their implications for the use of Ɛ-based soil moisture sensors.

Site description
The study area is located 12 km west of Malton, North Yorkshire, UK, and is part of the south facing Hollin Hill escarpment.It provides rough grazing through-out the year for approximately 30 sheep and is a landslide research site monitored by the Automated Time Lapse Electrical Resistivity Tomography (ALERT) system, as described by Wilkinson, et al. (2010) and Chambers et al. (2011).The area is considered as a representative landslide site, typically associated to the Lias Group mudrocks (Hobbs et al., 2005) and is described in detail by Gunn, et al. (2013).
The use of 3D Electrical Resistivity Tomography methods has detailed the contrasting layers of the weathered mudrock sliding above the more resistive, permeable layers of coarser grained silt and sandstones, causing zones of depletion and accumulation of superficial clay, silts and sandy materials, as described by Chambers et al (2011).
The combination of the Lias Group formations and landslide processes creates a slope of contrasting soil textures where the upper part of the slope is dominated by clays relating to the Whitby Mudstone formation and the lower slope is dominated by sand derived from the Straithes Sandstone and Cleveland Ironstone Formations (Hobbs et al., 2005).The zones of rotational landslide failure and zones of accumulation have developed active uneven landslide lobes across the slope.The most recent failures are shown as features (fig. 1) of an along-contour backscar and then below zones of hummocky ground downslope, while broken-up and annealed materials from former older lobe advances are present at the bottom of the slope, as low-lying lobate humps, which extend beyond the lower part of the field.(Gunn et al., 2013).

Methodology
Dominant soil characteristics of the field area were first determined to understand the dominant soil types and where capacitance probes should be located within the field site.As the site was dominated with clay soils, soil shrinkage was measured to determine the significance of soil shrinkage of soil cores taken from the field and used Discuss., doi:10.5194/soil-2016Discuss., doi:10.5194/soil- -40, 2016 Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.

SOIL
to calibrate the capacitance probes.Water release curves were also created using the pressure chamber technique to obtain saturation, field capacity and perminant wilting points for each soil type.This also created reference values to verify pedotransfer functions and calibration curves.Finally, calibration equations for each dominant soil type were developed using two laboratory techniques: 1) Disturbed calibration method, which ground and sieved soil taken from the field site and 2) Undisturbed claibration method, which took relatively large undisturbed soil cores from the field area, maintining soil structure.The resulting calibration equations for each dominant soil type and calibration method, were then compared to the measured water release curves and VWC observed in the field at the same time as field capacitance probe measurements.

Soil characteristics and sensor layout
Eight replicate soil samples at 0.10 m were taken from the numbered positions in fig. 1. Dry mass, bulk density, soil texture and organic matter (loss on ignition) were measured.From areas 1 to 4 (shown on fig.1), soil texture at 10cm were all 100 % clay, except one sample in area 1, which had 71 % clay, 20 % silt and 8 % sand.Organic matter by loss-on-ignition (maximum temperature of furnace 375 °C) ranged from 6% to 11% in areas 1 to 4. At the bottom of the slope, the clay content decreased and sand content increased particularly in areas 5 and 6 (fig. 1), whereas in areas 7 and 8, clay and silt content increased and sand decreased.This data was combined and plotted in fig.2, showing the particle size distributions and their texture classes for the study areas A (clay), B (sandy clay loam) and C (sandy clay), which are located in fig. 1.
To find the boundary line where the shallow soil changes (0 to 0.4 m depth), a vegetation survey was undertaken to investigate variations in grass associations that may relate to changes in soil.There were three main changes of grass associations across the site slope: 1) the upper slope was mainly dominated by Poa and Lolium genii; 2) below the landside lobes on the western side of the slope and along the south east edge of the site, was dominated by Poa, Festuca and Holcus lanatus; and 3) Holcus lanatus became sparse on the lower slope below the most eastern landslide lobe and Poa and Festuca continued to dominate.Following these vegetation associations, an auger survey to 0.4 meters depth was undertaken along the boundaries of these changing grass genii associations to determine the main areas where soil texture changed.The resulting soil texture map was developed (fig.1), illustrating that the boundary change of soil texture relates to the landslide lobes.In the lower east region more visible fresh slumps are observed than the west side of the slope, suggesting that the east side of the slope has more recent slumping events compared to the west side of the slope (Merritt et al., 2014).The presence of soils with sandy clay classification on the east side of the slope occurs in the more recently slumped areas, as shown in fig.
1.The soils with larger proportions of sand-sized fraction occur on the older flow deposits, which is attributed to leaching and removal of clays infiltrating with rainfall (Merritt et al., 2014).
Following the site survey three areas were chosen to take into account differences in soil properties (fig.1).These areas were Group A located on clay, Group B sandy loam and Group C that had a mixture of clay and sandy clay loam.Each area was instrumented with 12 soil moisture sensors (5TEs, DECAGON Devices, Inc., Pullman, US) at 0.1 m soil depth, following a spatially optimized nested sampling scheme, as described by Lark (2011).
To test maximum Ɛ values in the field, areas near each site were wetted up at 0.1 m depth until ponding occurred and Ɛ was measured in the saturated or ponded water using 2 5TE sensors.The sandier soils in site B had high infiltration and therefore it was difficult to record ponded water at site B. It was also found that the soils at 0.1m depth in site B even if it had been raining consistently, they remain unsaturated, therefore high in-situ VWC in site B was considered to be near 0.4 m 3 m -3 .During dry conditions in the field, soil core samples were taken from 0.10 m depth and Ɛ was measured by 5TE sensors during the same time.The soil cores were weighed in the field to measure soil moisture gravimetrically and then dried in the lab to measure the VWC at the time of in-situ measurements.

5TE Sensor operation principles and acquisition
Like all capacitance probes, according to the 5TE manual (Decagon Devices, 2014) the 5TE sensor uses an electromagnetic field to measure the dielectric permittivity (ε) of the surrounding soil.It supplies a 70 Mhz oscillating wave to the sensor prongs, which then charges depending on the dielectric of the surrounding soil.The resulting charge is proportional to the soil dielectric and soil volumetric water content and the output value from the 5TE microprocessor is a value of ε from the sensor.The 5TE sensor also measures temperature and electrical permittivity.The sensor dimensions are 10 x 3.2 x 0.7 cm, active measurement length is 5.2cm.It has volumetric water content (VWC) accuracy ±3% using the Topp equation in typical mineral soils that have electrical conductivity <10dS/m.VWC accuracy improves from ±1 to 2% in any porous medium using a soil-specific calibration.
During field measurements within each area A, B and C (locations shown in fig.1), the sensors were measured every 15 minutes for 10 months (January 2012 to October 2013) and were logged using an Adcon telemetry system (Klosterneuburg, Austria) which was linked to individual sensors using the SDI-12 serial communications standard.Measurement data was stored locally at each of the eight nodes before being transferred over a radio link to a local coordinator node where it was relayed to a GSM link for storage and processing within a relational database hosted on a server at the British Geologcial Survey offices in Keyworth, UK.
During calibration an Em50 logger was used to acquire data from three 5TE probes via a Stereo to USB port, using the ECH20 Utility software (Decagon Devices, 2016).

Soil shrinkage
Soil shrinkage of clay soils was measured using a technique developed by the British Geological Survey (BGS), which uses a laser rangefinder to measure the height and diameter of a soil core, as it dries on a motorised rotating platform (Hobbs et al., 2014).The laser scans up to 3600 points around the soil core periphery and then weighs the soil after each scan.The volume reduction of the soil core is graphed against the core soil water content and the shrinkage limit is taken at the last measurement, as described by Head (1992) and particle density was set at 2.65 mg/m 3 .At the start of each measurement the clay cores were approximately at field capacity.Replicate cores taken from Groups A and C were measured for shrinkage, but the sandy clay loam (Group B) was not measured, because it loses structure once it begins to dry.Soil shrinkage was also measured for all soil cores that were oven dried to provide percentage shrinkage from saturated to oven dried cores.

Water release curves
Three replicate soil cores taken from Groups A to C were saturated with deionised de-aired water for 2 to 10 days and then equilibrated in a pressure plate chamber at metric potentials 0, -33, -50, -100, -250, -500, -1000 and finally at -1500 kPa.Soil core mass was measured after each equilibration and used to calculate volumetric water content.After each applied suction, shrinkage of core volumes was estimated by measuring the core volumes.
The model proposed by Van Genuchten (1980) (VG) was used as a basis for determining the shape of the water release curves.The VG curve (Eq.( 1) describes how VWC,  changes as a function of the suction pressure, Where   and   are the residual and saturated VWC respectively and  and  are dimensionless empirical parameters.These four parameters were defined in two ways: Firstly using the information provided from the soil texture information (fig.2) to determine the parameters from the widely used ROSETTA pedotransfer function software (Schaap et al., 2001).Secondly, we optimized the parameters to achieve the best fit to the observed data using the unconstrained non-linear minimization procedure in MATLAB (fminsearch function).

Calibration of in-situ sensors in contrasting soil
We tested two methods to calibrate the Decagon 5TEs: a disturbed method and undisturbed method, which measured soil from the three soil type areas A, B and C (shown in fig.1).

Disturbed calibration method
The disturbed method consisted of taking five litres of soil from the field at 0.1m soil depth, air drying it and then sieving the soil through a 2mm sieve.The clay and sandy clay were ground using a grinder to break down the soil aggregates to less than 2 mm.The air-dried sieved material was then put into a bucket of known volume and compressed to bulk densities that ranged from 0.8 to 1.1 cm 2 /cm 3 , depending on soil type at 0.1 m depth.The soil moisture was then measured separately by two 5TE sensors (fig.3a) to provide two values: the dielectric constant of the soil (Ɛ) and volumetric soil water content (VWC) estimated from the DECAGON Devices factory calibration for mineral soil, which uses the Topp equation, Eq. 2 (Topp et al., 1980).
θ is volumetric water content, Ɛ is dielectric constant A small core of soil was taken from the bucket to measure gravimetrically the VWC of the air dried soil.The bucket was then emptied into a larger container and a quantity of water was added to make up approximately 10% VWC.This water was mixed evenly into the dry soil and the bucket was repacked to a similar bulk density and the Ɛ and VWC were again measured using a 5TE sensor.Another small core of known volume was again taken to measure gravimetrically the VWC of the new soil water content.These steps were repeated four times, until the water content approached saturation.This method provided five data points of: gravimetrically measured VWC, Ɛ and VWC measured by the 5TE probes and is a typical method for calibrating soil moisture sensors by Decagon

Undisturbed calibration method.
The undisturbed method uses a soil core of diameter 19.5 cm and height 23 cm, where a 5TE sensor is placed at 0.10 m below the ground surface and the soil core is cut out of the ground and placed directly into a PVC core (fig.3b), ensuring that the sensor remains undisturbed.The 19.5 ×23 cm core size ensures there are no edge effects since the sensor is further than 5 cm from the core sides.The core was placed on a sand table and water was added to allow the core to saturate.After five days of saturation, the core was then weighed and the Ɛ and VWC were measured from the sensor placed in the core.The core was then allowed to air dry and every 5 to 10 days the core was weighed, the volume estimated and the Ɛ and VWC were measured from the sensor in the core.After two months of air drying, the core was put into a 30 °C oven to dry for a month.During this time measurements of core weight and volume were taken and the sensor in the core continued to monitor Ɛ and sensor VWC.In this way a drying curve of gravimetric VWC and sensor Ɛ and VWC were measured to provide a calibration curve.A final measurement was taken in ponded water, by removing the 5TE probe and creating a depression in the centre of the soil core.The depression was then filled with water and Ɛ and VWC were measured by inserting the 5TE probe in the ponded water.The ponded water was also sampled to estimate the percentage sediments in the water.

Soil characteristics
The electrical conductivity of saturated soil within the field had conductivities at 0.1 m depth from 0.001 to 0.12 dS/m in clay soils, 0.001 to 0.01 dS/m in sandy loam clay and 0.001 to 0.7 mdS/m in sandy clay soils.These (ranging from 85 to 100%) and positions 5 to 7 (at the bottom of the slope, fig. 1) have relatively large sand contents (between 20% to 67%), while position 8 (lower east side of the site slope) has the largest average amount of silt (26%), a relatively small amount of sand (51 to 10%) and a larger clay content (up to 64%).The decreasing clay content from the upper to lower slope, coincides with the Whitby Mudstone Formation and the lower Staithes Sandstone Formation (Gunn et al., 2013).Bulk density ranged from 0.93 to 1.22 g cm -3 in areas 1 to 4, 1.22 to 1.34 g cm -3 in areas 5 and 6 and 1.36 to 1.51 g cm -3 in areas 7 and 8.

Soil Shrinkage
Using the BGS method to estimate soil shrinkage limit for Group A was 13% and 16% and for Group C it was 9% and 14%.Shrinkage for these cores was drying from field capacity to oven drying at 30°C.Measuring the change of clay volumes from saturated soil water content to oven drying gave much higher shrinkage limit ranges such as 35% for Group A, and 4% for Group B and 23% for Group B. The shrinkage of clay soils for group A are comparable to some Dutch clay soils, which have been measured to have 42% shrinkage from saturation to a pressure head of -16000cm (Bronswijk and Eversvermeer, 1990).
During dry field conditions, the clay soils were observed to crack and create fissures up to 0.10 m in width and during prolonged wet conditions the soil was observed to expand and these fissures closed.During wet winter intervals rainfall was observed to pond within these fissures on the upper hillslope in the clay size-fraction dominated soils.

Water release curves and PTF
The maximum, minimum and average values for saturated VWC, for permanent wilting point (taken at -1500 kPa), field capacity (-33 kPa), and saturated VWC using the pressure chamber technique are given in Table 1.
The water release curves for the three different soil groups are shown in fig. 4.Here the VG curves using the ROSETTA parameters (green lines) underestimate the mean observed VWC (black dots) significantly, with biases between -0.12 and -0.22 cm 3 /cm 3 (Table 2).The fitted VG curves (red lines in fig.4) are much more reflective of the soil moisture characteristics of all of the soil types and show considerably smaller biases of 0.001 cm 3 /cm 3 and root mean quadratic errors between 0.01 and 0.02 cm 3 /cm 3 .A comparison of the ROSETTA and fitted parameters in Table 1 shows that the ROSETTA pedotransfer function underestimates both the   and   parameters for all soil types resulting in large negative biases.

Dielectric constant, factory calibration and in-situ soil conditions
The data measured by the 12 5TE sensors in each area A, B and C were pooled together for each site and maximum, minimum and average VWC was estimated using the generic Topp equation and are plotted in fig. 5.The clay (site A) and sandy clay (site C), had the largest maximum and minimum Ɛ in comparison to the sand clay loam (site B), which had the smallest maximum and minimum Ɛ.
To test the range of VWC calculated by the Topp equation, we compared in-situ observations and measured VWC for each site using gravimetric methods.During dry field conditions, the clay and sandy clay sites were observed to form cracks up to 0.15 m wide and 0.20m deep.After prolonged heavy rainfall, some cracks were observed to be filled with water, which coincided with maximum in-situ Ɛ values for sensors in areas of clay cracking.Similar maximum Ɛ values were reached when 5TE sensors were in a slurry of ponded soil water both in disturbed and undisturbed soil cores (Table 3).Sandy clay soils (site B) however, did not crack and were observed to freely drain, causing no ponding water; this created lower in-situ maximum Ɛ values in site B, as shown in Table 3.

Calibration of the soil moisture sensors
Calibration curves using cubic, square and linear regression statistics are shown in fig.6, where the gravimetric volumetric water content is shown on the y-axis and the Ɛ is measured by the 5TE sensors are on the x-axis.The results of the regression statistics are given in Table 5, where the cubic model is (Eq.3): the quadratic model is (Eq.4): And the linear model is (Eq.5): where, Ɛ is the dielectric constant, measured by the 5TE sensors at 0.1 m soil depth.
According to the adjusted R 2 (Miles, 2005), the cubic model was the best fit for all soil types and calibration methods.
Fig. 7 shows the Ɛ measured by the 5TE sensors converted to VWC using the best-fit models for each disturbed, undisturbed calibration methods and the Topp equation (eq. 1) for each soil type.The saturated VWC (at 0 kPa), field capacity (at -33 kPa) and permanent wilting points (at -1500 kPa), and known in-situ VWC (given in Table 3) were superimposed on fig. 5 and 7 The largest differences between the calibration models occur for wet clay soils, where the disturbed calibration method estimates maximum VWC values to be 0.31, 1.73 and 0.22 larger than the VWC values estimated by the Topp equation using the same Ɛ in-situ values (shown in fig.7) for clay, sandy clay loam and sandy clay respectively.Whereas, the undisturbed method estimates maximum VWC values to be 0.38, 0.10 and 0.43 larger than the VWC values estimated using Topp equation.The known in-situ VWC values for all soils (shown as circles) coincide with the best-fit calibrations, showing that the undisturbed method of calibration increases the VWC into the expected range of saturated and permanent VWC.

Discussion
The under-estimation of VWC of the clay-rich soils (groups A and C) at different matric potentials estimated by the ROSETTA PTF in comparison to the water release curves developed from the pressure chamber method (fig.4) is likely due to the volumetric expansion of the clay-rich soil, under wet conditions, which is not taken into account by the ROSETTA PTF.The high saturated VWC of the clay soil in particular for site A is possible because of the relatively low bulk density and swelling of these soils under wet conditions, while soil pores remain water filled at high matric potentials (-1500kPa), causing the fitted curves in fig. 4 to be positioned at larger water contents than ROSETTA predicted PFTs.Such importance of characteristics of clay-rich soils to develop PTFs are discussed by Hodnett and Tomasella (2002) and Gaiser et al. (2000).We therefore suggest that it is important to use lab techniques to estimate water release curves of soil cores taken from the different soil types to develop site specific PTFs for these clay-rich soils.This study corroborates the results from a study by Patil, et al. (2010) who evaluated ROSETTA and concluded that it had limited capability to determine water retention functions for shrink-swell soils in India and recommended region-specific PFTs to predict available water capacity.
The combination of the water release curves (Table 1 and fig.4) and the observed in-situ VWC for wet and dry intervals of three sites (Table 3) are important indicators to verify the generic factory calibration (in this case the Topp equation) for the 5TE sensors.Investigation of Figure 5, shows that the generic factory calibration, consistently underestimates the maximum, mean and minimum ranges of VWC in particular for sites A and C, because the wettest VWC points using the generic factory calibration remained at field capacity, when the these recorded wet VWC points were observed to be saturated in the field and the driest VWC are too dry, when in-situ observations estimate soils to be near permanent wilting point.The sandy clay loam however, was not observed to reach saturated VWC because these soils drained quickly and therefore the generic factory calibration was considered sufficient to estimate VWC for site B.
Figure 6 shows that the undisturbed calibration provides the best calibration for all soils, because it is consistent with observed in-situ field data and soil water release curves of soil cores measured from the three soil types.This calibration is able to represent the large range of VWC occurring through wet and dry intervals.
The Topp equation and the disturbed calibration estimated smaller ranges of VWC, but the disturbed calibration over-estimated VWC, while the Topp equation under-estimated VWC (figs.7, A and C).The change in soil structure when adding approximately 50% water to the ground and sieved sample during the disturbed calibration method, is likely to cause the over-estimation of VWC.The grinding and sieving to < 2mm of the clay-rich samples removed the structure from the soil cores, creating large void ratios and a weak bonded structure, which is typical of "quick" clays, which have a propensity to liquefy and flow (Gauthier and Hutchinson, 2012).This change in soil structure is likely to cause lower Ɛ values at VWC >50% (fig.6), creating calibration curves that over-estimate VWC in comparison to the soil structure of the Undisturbed calibration method as shown in fig. 3.

Conclusions
The volumetric expansion of clay-rich soils under wet conditions is an important characteristic that is not always taken into account using PTF, which can cause under-estimation of VWC during wet conditions.This study shows that grinding and sieving clay soils to <2mm and then repacking the clay to bulk densities similar to in-situ field bulk densities does not represent the same field conditions for accurate calibrations to convert Ɛ to VWC.When adding >50% water to the ground and sieved soil samples, Ɛ values to VWC >50% were observed to be lower than SOIL Discuss., doi:10.5194/soil-2016-40, 2016 Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.
using undisturbed soil cores taken from the field.Further studies are needed to understand the causes of the loss of structure in this soil and its relationship with Ɛ.
Generic factory calibrations for most soil sensors have a range of measurement from 0 to 50%, which is not appropriate for the studied clay-rich soil, where ponding can occur during persistent rain events, which are common in temperate regions.The range of VWC from saturated (0 kPa) to permanent wilting point (-1500 kPa) were 36 to 72%, 12 to 50% and 31 to 69% for clay, sandy clay loam and sandy clay respectively.In-situ observations did not always reach saturation, such as in the sandy clay loam site, but saturation was reached in the clay and sandy clay sites.Other studies have also concluded that the precision of capacitance sensors, worsen in saturated soils (Evett et al 2006).Therefore it is important to know the range of Ɛ values that the sensor is measuring in field conditions to ensure that the conversion to VWC effectively provides actual VWC, rather than simply taking soil from the field and repacking it to similar field bulk density in lab conditions.

List of tables
Table 1.Minimum, average and maximum saturated (0 kPa) water content, field capacity (-33 kPa) and wilting point (-1500 kPa), estimated from pressure chamber data, for s A, B and C. Standard deviation of averages for three replicates are given in brackets.Table 5. Linear (L), quadratic (Q) and cubic (C) models for predicting SWC using the dielectric constant measured by the 5TE sensors."Undist" is the Undisturbed calibration method and "Dist" is the Disturbed calibration method.
The R 2 and Adjusted (Adj.)R 2 for each model are included and the greyed lines indicate the model with the highest Adj R 2 .These values were used to estimate VWC in fig. 7.
SOIL Discuss., doi:10.5194/soil-2016-40,2016   Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.Devices, Inc. (Cobos and Chambers, 2010).A final VWC close to 100 % was taken by creating a small depression in the middle of the bucket in saturated soil.Water was poured into the depression to create ponded water and the 5TE probe was placed in the ponded water to take readings of the Ɛ and VWC.A sample of the ponded water was then taken to measure gravimetrically the percent of sediments in the ponded water.It was observed that once VWC was over 50%, aggregate structures disappeared and its internal structure disappeared to form a soup-like structure.This change in structure is shown in fig.3A.
values are below the conductivity threshold of <10dS/m, which 5TE probes are considered to accurately measure VMC (Decagon Devices, 2014).The particle size distributions and associated soil texture classes for sites A to C are shown in fig. 2. Site positions 1 to 4 (on the upper part of the slope, fig. 1) have very large clay contents
SOIL Discuss., doi:10.5194/soil-2016-40,2016 Manuscript under review for journal SOIL Published: 22 June 2016 c Author(s) 2016.CC-BY 3.0 License.Using the Topp equation, the Ɛ values were converted to VWC for the three soil types, A, B and C and the maximum, minimum and average VWC are shown in fig. 5.The saturated VWC (at 0 kPa), field capacity (at -33 kPa) and permanent wilting points (at -1500 kPa) measured by the water release curves, including known points of in-situ VWC given in Table 3 were superimposed on fig. 5. Aligning the known in-situ VWC and the saturated, field capacity and permanent wilting points, shows that the Topp equation under-estimates the range of measured Ɛ for the three clay soil types.

Fig. 1 .
Fig. 1.Dominant soil texture classes of the study area.Red dots indicate positions of soil moisture sensors.Hatched yellow lines are recent flow deposits and blue hatched lines are relict flow deposits (Merritt et al., 2014) Fig. 2.UK Soil Survey of England and Wales texture triangle, which considers a siltsand limit of 60 µm.Abbreviations: Cl is Clay, SaCl is sandy clay, ClLo is Clay loam, SiClLo is Silty clay loam, SaClLO is Sand clay loam, SaLo is Sandy Loam, SaSiLo is Sandy silt loam, SiLo is Silt loam, LoSa is Loamy sand, Sa is Sand.The letters show the soil texture for three soil texture groups relating to fig.1.Fig.3.A) An aerial view of the approach in which soil structure is removed.Upper image is VWC near 30% and the lower image is VWC > 55%.B) The undisturbed method.The soil core taken is out of the PVC core.

Fig. 4 )
Fig.4) Soil moisture release curves predicted using the VG model with parameters from the ROSETTA pedotransfer function (green curves) and parameters optimised to the observed data (red lines).Mean observed VWC (black dots) were calculated from three replicate cores (grey dots).

Fig. 5 )
Fig. 5) Maximum, minimum and mean VWC calculated by generic factory calibration Topp equation) and hourly rainfall intensity (grey bars).Maximum and minimum saturated VWC, average field capacity and permanent wilting point values estimated from water release curves are plotted.The blue circles indicate maximum in-situ VWC when soil was saturated and the red circles indicate known minimum in-situ VWC A is clay soil, B is sandy loamy clay soil and C is sandy clay.

Fig. 6 )
Fig. 6) Estimated linear, quadratic and cubic models showing the relationship against VWC and dielectric constant for (a) Group A (clay), (b) Group B (sandy clay loam) and (c) Group C (sandy clay).

Fig. 7 )
Fig. 7) Estimated VWC for in-situ Ɛ data measured during 2012 for three different soil types (a) Group A (clay), (b) Group B (sandy clay loam) and (c) Group C (sandy clay).The three calibration models are from disturbed and undisturbed methods and the factory calibration is using the Topp equation.

Figure 1 SOIL
Figure 1

Figure 2 SOIL
Figure 2

Table 2 .
Soil release curve parameters for the VG model estimated by fitting (bold) and using the ROSETTA pedotransfer function (square brackets).

Table 3 .
Maximum (Max.) and minimum (Min.)dielectric constant (Ɛ) measured by the 5TE sensors in-situ for three groups, A, B and C. Gravimetric volumetric water content (VWC) is the soil water fraction measured gravimetrically in relation to maximum and minimum Ɛ.

Table 4 .
Maximum (Max.)andminimum (Min.)dielectric constant (Ɛ) measured by the 5TE for the disturbed and undisturbed calibration methods for three soil types, A, B and C. Gravimetric volumetric water content (VWC) is the soil water fraction measured gravimetrically in relation to maximum and minimum Ɛ.