Evaluation of a low cost soil water cont

() Evaluation of a low cost soil water content sensor for wireless network applications H R Bogena *, J A Huisman, C Oberdörster, H Vereecken Research Centre Jülich, Agrosphere Institute, ICG 4, 524[.]

Evaluation of a low-cost soil water content sensor for wireless network applications

Research Centre Ju¨lich, Agrosphere Institute, ICG 4, 52425 Ju¨lich, Germany

Received 2 February 2007; received in revised form 29 May 2007; accepted 15 June 2007

KEYWORDS

Soil water

content sensor;

EM sensor

characterization

method;

Wireless sensor

network

Summary Wireless sensor networks are a promising new in situ measurement technology for monitoring soil water content changes with a high spatial and temporal resolution for large areas However, to realise sensor networks at the small basin scale (e.g 500 sensors for an area of 25 ha), the costs for a single sensor have to be minimised Furthermore, the sensor technique should be robust and operate with a low energy consumption to achieve a long oper-ation time of the network This paper evaluates a low-cost soil water content sensor (ECH2O probe model EC-5, Decagon Devices Inc., Pullman, WA) using laboratory as well as field exper-iments The field experiment features a comparison of water content measurements of a for-est soil at 5 cm depth using TDR and EC-5 sensors The laboratory experiment is based on a standardized sensor characterisation methodology, which uses liquid standards with a known dielectric permittivity The results of the laboratory experiment showed that the EC-5 sensor has good output voltage sensitivity below a permittivity of 40, but is less sensitive when per-mittivity is higher The experiments also revealed a distinct dependence of the sensor reading

on the applied supply voltage Therefore, a function was obtained that allows the permittivity

to be determined from the sensor reading and the supply voltage Due to the higher frequency

of the EC-5 sensor, conductivity effects were less pronounced compared to the older EC-20 sensor (also Decagon Devices Inc.) However, the EC-5 sensor reading was significantly influ-enced by temperature changes The field experiment showed distinct differences between TDR and EC-5 measurements that could be explained to a large degree with the correction functions derived from the laboratory measurements Remaining errors are possibly due to soil variability and discrepancies between measurement volume and installation depth Over-all, we conclude that the EC-5 sensor is suitable for wireless network applications However, the results of this paper also suggest that temperature and electric conductivity effects on the sensor reading have to be compensated using appropriate correction functions

ª2007 Elsevier B.V All rights reserved

0022-1694/$ - see front matter ª 2007 Elsevier B.V All rights reserved.

doi:10.1016/j.jhydrol.2007.06.032

* Corresponding author Tel.: +49 (0)2461 616752; fax: +49 (0)2461 612518.

a v a i l a b l e a t w w w s c i e n c e d i r e c t c o m

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / j h y d r o l

A remaining challenge in hydrology is to explain the

ob-served patterns of hydrological behaviour over multiple

space-time scales as a result of interacting environmental

factors The large spatial and temporal variability of soil

water content is determined by factors like atmospheric

forcing, topography, soil properties and vegetation, which

interact in a complex nonlinear way (e.g Grayson et al.,

1997; Western et al., 2004) For the determination of the

spatiotemporal structure of hydrological state variables,

an enormous measuring effort is necessary In the case of

soil water content remote sensing, technologies like passive

or active radiometry can avoid direct measurements by

pro-viding area-wide indications of surface soil water content

(e.g Wigneron et al., 2003; Lo¨w et al., 2006) However,

the received signal is strongly influenced by the vegetation

and surface structure, and the sampling depth is restricted

to the uppermost soil (2–5 cm) (Walker et al., 2004)

Conse-quently, direct measurements are still indispensable in

areas with significant vegetation and litter cover

In recent years, ground penetrating radar methods have

been developed that allow mapping of the spatial variability

of soil water content profiles (e.g Huisman et al., 2001,

2002; Lambot et al., 2006) However, these methods are

not easily amenable to automation Clearly, there is still a

need for measurement techniques that can assess large

scale three-dimensional soil moisture fields with extremely

high temporal and spatial resolution (Schulz et al., 2006)

A promising new technology for environmental

monitor-ing is the wireless sensor network (Cardell-Oliver et al.,

2005) Environmental sensor networks will play an

impor-tant role in the emerging terrestrial environmental

observa-tories (Bogena et al., 2006), since they bridge the gap

between local (e.g lysimeter) and regional scale

measure-ments (e.g remote sensing) A wireless network may consist

of hundreds of water content sensors that can transmit

information to a main server with wireless communication

technology There are several factors that have to be

con-sidered when selecting a sensor for network applications

In order to maximise the lifetime of such a sensor network,

the sensors have to be very economic concerning energy

de-mand and should be reasonably robust Because of the

mul-titude of soil water content measurements within the sensor

network, the interpretation of the sensor signal has to be

straightforward and unambiguous Last but not least, in

or-der to maximise the number of sensor nodes, the soil water

content sensors have to be as inexpensive as possible Since

capacitance sensors are relatively inexpensive and easy to

operate, they seem to be a promising choice for soil water

content measurements with sensor networks

The aim of this paper is to evaluate a low-cost

capaci-tance sensor (ECH2O probe, model EC-5, Decagon Devices

Inc., Pullman, WA) using laboratory as well as field

experi-ments We chose to use a two-step calibration procedure

in the laboratory experiments In the first step, the sensor

reading is related to permittivity using the standardized

sensor characterization methodology proposed by Jones

et al (2005) In the second step, permittivity is related to

soil water content Such a two-step procedure permits a

more physically based calibration procedure (Kelleners

et al., 2004) Furthermore, knowledge of the sensor

read-ing–permittivity relationship enables a more direct compar-ison with other electromagnetic sensing systems, e.g other capacitance sensor designs, passive and active radiometry, etc Sensor evaluations that directly relate the sensor read-ing to the soil water content are more ambiguous and diffi-cult to compare between sensors because of the uncertainty and systematic deviations introduced by the variable per-mittivity–soil water content relationships for different soil types The field experiment features a comparison of per-mittivity and water content measurements in a forest soil

at 5 cm depth using TDR and EC-5 sensors

Theory

Sensor technology

The capacitance and time domain reflectometry (TDR) method are two widely used electromagnetic (EM) tech-niques for soil water content estimation (Blonquist et al., 2005) Both methods make use of the strong dependence

of EM signal properties on volumetric water content that stems from the high permittivity of water (ew 80) com-pared to mineral soil solids (es 2–9), and air (ea= 1) The capacitance method is already known for a long time (Dean et al., 1987) One of the first workers to use a high frequency capacitance technique for soil water content determination was Thomas (1966) The basic principle of the capacitance method is to incorporate a dielectric med-ium (e.g soil) as part of the dielectric of the sensor capac-itor The equivalent circuit diagram of the ECH2O probe is illustrated inFig 1 The ECH2O sensor circuitry measures the dielectric permittivity of the material surrounding a thin, fiberglass enclosed probe The circuit board includes

an electronic oscillator that generates a repetitive square waveform with a characteristic frequency (e.g EC-5:

70 MHz) The total sensor capacitance is then made up

of the capacitance of medium C and the capacitance Cs

due to stray electric fields (Kelleners et al., 2004) The soil permittivity is determined by measuring the charge time from a starting voltage, Vi, to a voltage V, with

an applied voltage, Vfof a capacitor which uses the soil as a dielectric If the resistance R, Vfand Viare held constant, then the charge time of the capacitor, t, is related to the capacitance according to

t¼ RC ln V Vfþ Vi

Vi Vf

ð1Þ The capacitance is a function of the dielectric permittiv-ity (e) of the medium and a geometrical factor g and can be calculated by

The factor g is associated with the electrode configura-tion and the shape of the electromagnetic field penetrating the medium By assuming that the charge time of the capac-itor is a linear function of the dielectric permittivity of the surrounding medium the dielectric permittivity e can be cal-culated as follows:

1

e¼1

t Rg ln V Vfþ Vi

Vi Vf

ð3Þ

A typical capacitance probe determines V at a high

fre-quency (between 10 and 100 MHz) for a specific pulse length

Dt(seeFig 2) The course of the charging curve depends on

the dielectric permittivity and thus on the soil water

con-tent At a high soil water content, the capacitor will charge

slower and, therefore, the charge curve will be flatter than

at a low soil water content This implies that a capacitor

filled with wet soil will reach a given threshold voltage

(Vt) later than a capacitor filled with dry soil (seeFig 2)

The sensor output is directly related to the average voltage

over the period Dt Consequently, a high soil water content

will result in a high output voltage of the sensor This basic

principle is utilized in all ECH2O soil water content sensors

used in this study To transform the alternating current into

direct current, the ECH2O sensor uses a RMS-value

con-verter The resultant output voltage is then related to

per-mittivity or directly to soil water content by fitting

regression curves

Sensor response to temperature variation

Soil water content sensors may be affected by temperature

variations through effects on the dielectric permittivity of

water, through effects on soil–water interactions, as well

as through direct effects on the sensor circuitry The

dielec-tric permittivity of pure water ewdecreases with increasing

temperature According to the semi-empirical model for the dielectric permittivity of water by Meissner and Wentz (2004), a negative temperature trend of 0.35 K1for fre-quencies <500 MHz can be expected This decreasing trend

is also observed for soil dielectric permittivity of some soils (e.g.Pepin et al., 1995) In contrast, for other soils a change from negative to positive correlations between permittivity and temperature with decreasing soil water content or even

a positive correlation for all soil water contents is observed (Wraith and Or, 1999) Positive correlations between per-mittivity and temperature are often observed for high sur-face area soils with a high cation exchange capacity, in which the relaxation frequency of colloid-bound water is of-ten lower than the measurement frequency of the soil water content probe (e.g.Evett et al., 2006) The positive corre-lation with temperature is explained by the release of bound water with increasing temperatures and the complex interactions between measurement frequency, frequency– dependency of the soil permittivity, temperature and bulk soil electrical conductivity

Sensor response to conductivity variation

The application of EM sensors to conductive media, such as saline soils, certain clay soil and organic soils is hindered due to significant attenuation effects of the desired signal

Vi

Vf

Vt

t

∆ (pulse length)

Time

Applied voltage Charge/discharge curve (low permittivity/low water content) Charge/discharge curve (high permittivity/high water content)

low sensor output voltage

high sensor output voltage

Figure 2 The charge and discharge curves of two capacitors with either high or low permittivity using a repetitive square pulse with a pulse length Dt

RMS R

Sensor reading

Electronic oscillator

RMS-value converter

Cs

Figure 1 Equivalent circuit diagram of a capacitance sensor where R is a resistor, C is the capacitance of the medium, Csis the stray capacitance, G is the energy loss due to relaxation and ionic conductivity and Vinpand Voutare the supply and sensor reading voltage, respectively

(Hook et al., 2004) For example,Sun et al (2000)observed

for TDR sensors that at the same water content, the transit

time in a soil with a high bulk electrical conductivity was

longer than that in a less conductive soil, which led to an

overestimation of volumetric soil water content However,

for reasonable bulk electrical conductivities, a unique

rela-tionship is found between permittivity and water content

using TDR

The poor sensor performance especially for capacitance

sensors is mainly attributed to the inability of providing a

general sensor reading–permittivity relationship that is

va-lid for different bulk electrical conductivities (Jones et al.,

2005) The effects of electrical conductivity are particularly

present at low measurement frequencies (less than

100 MHz) and can be significantly enhanced by Maxwell–

Wagner effects associated with charge migration and build

up at interfaces (Hilhorst, 1998) Since electric conductivity

is also temperature dependent, Maxwell–Wagner effects

may be intensified with increasing temperatures or may be

masked due to decreasing temperatures (Or and Wraith,

1999)

Materials and methods

Soil water content sensors considered

In this study, the ECH2O soil water content probes EC-5 and

EC-20 (Decagon Devices Inc., Pullman, WA) and TDR were

used The EC-5 sensor is a more recent development and

operates at a higher frequency (70 MHz) than the EC-20

sen-sor (5 MHz) Furthermore, the EC-5 sensen-sor features a shorter

two-prong design that eases the probe insertion (Decagon

Devices Inc., 2006) However, this modification further

re-duces the sampling volume of the sensor The older EC-20

sensor was already part of several comparative tests (

Blon-quist et al., 2005; Czarnomski et al., 2005; McMichael and

Lascano, 2003) revealing distinct temperature, conductivity

and soil type dependences of the sensor It was also

recog-nized that the design of the EC series results in a small

sam-pling volume (Blonquist et al., 2005), which makes these

sensors susceptible to small-scale variability in soil water

content and disturbance of natural soil conditions due to

probe insertion Nevertheless,Czarnomski et al (2005)

con-cluded that the EC-20 performed nearly as well as a TDR

probe in a field experiment

Laboratory experiment

EM sensor characterization method

The first aim of the laboratory experiment was to test

whether the standardized EM sensor characterization

meth-od proposed byJones et al (2005)is reproducible by taking

the EC-20 sensor as an example Secondly, the

characteris-tics of the new EC-5 sensor were assessed While assessing

the sensor characteristics, it is important to realize that

in addition to the dielectric permittivity, the EM

measure-ments are sensitive to dielectric relaxation, electrical

con-ductivity, and temperature (Topp et al., 1980; Pepin et al.,

1995) To avoid unwanted noise due to these secondary

fac-tors in the sensor evaluation,Jones et al (2005)used liquids

with known dielectric properties to characterize and

com-pare EM sensing systems Furthermore, the problem of imperfect contact between soil matrix and sensor is avoided

by using liquids This standardized method was used for the evaluation of several EM sensors with 2-isopropoxyethanol (i-C3E1)/water mixtures byBlonquist et al (2005) To obtain liquids with a permittivity range between 10.7 and 78.5, they prepared mixtures of i-C3E1and deionized water Due

to the relatively high permittivity of i-C3E1(10.75,Kaatze

et al., 1996), the full range of environmentally relevant soil permittivities was not obtained in their study Therefore,

we selected the strong solvent dioxane (1,4-diethylene dioxide) with er 2.2 as a second reference liquid Dioxane was successfully used for testing a capacitance sensor by Schwank et al (2006a) However, the high volatility of this carcinogenic chemical requires careful handling, and the low flash point of 11 C limits the experimental temperature range Therefore, we chose not to use dioxane for the tem-perature experiment and for the full range of possible per-mittivities The mixtures used in this study can be considered as non-relaxing media since i-C3E1, dioxane and water do not show significant dielectric relaxation below frequencies of 2 GHz (Jones et al., 2005; Kaatze et al., 1996)

The permittivities of the liquid standards were calcu-lated with dielectric mixing models (Bogena et al., 2007) The volume fractions and associated permittivity of each

of the 16 mixture used in this study are listed inTable 1 Although the standardized sensor evaluation method of Jones et al (2005) relies on interpretation of the sensor reading–permittivity relationships, we decided to also pres-ent results where the sensor reading is related to (equiva-lent) soil water content because soil water content is easier to interpret from a hydrological point of view The equivalent soil water content Hvin [m3/m3] was calculated using the empirical relation derived byTopp et al (1980)

Hv¼ 5:3  102þ e  2:92  102 e2 5:5  104

The equivalent soil water content of each mixture is also reported inTable 1 The selected mixture ratios cover a large part of the environmentally relevant range of soil water contents (from 0.9 to 51.8 vol.%)

Experimental setup The experimental setup consisted of a stable DC power sup-ply (Agilent, E3646A, 60W Dual Output Power Supsup-ply) en-abling different supply voltages (from 2.0 to 5.0 V) and a high precision digital multimeter (Escort, 99 TRUE TMS, accuracy: 0.025%) for the determination of the sensor out-put voltage Since we noticed slight variations in voltage re-sponse between single EC-5 sensors (approx 1–2% of the output voltage), we used two sensors in parallel with a dis-tance of about 2 cm During the permittivity determination, the liquids were thoroughly mixed using a magnetic stirrer

to avoid separation The EC-5 and the EC-20 sensor were compared using two supply voltages (2.5 and 5 V) to evalu-ate whether differences in the sensor performance are volt-age dependent

As discussed earlier, temperature effects arise from the temperature dependence of the medium as well as from the temperature dependent response of the sensor itself The

experiment proposed here intends to characterise the

temperature response of the EC-5 sensor itself using a

0.6 i-C3E1–water mixture (e = 40, equivalent soil water

con-tent of 51 vol.%) The temperature dependence of the

i-C3E1–water mixture itself was described by a linear model,

e(T) =0.083T + 41.765 (Jones et al., 2005) The

tempera-ture of the liquid standard was varied using an automated

cryostat with circulating water bath To reveal the

sensitiv-ity of the sensor electronics to temperature variation, the

EC-5 sensor head was totally immersed in the liquid and

the sensor electronics were completely equilibrated with

the liquid temperature For the conductivity experiment,

we also used the 0.6 i-C3E1–water solution with increasing

salt (NaCl) additions The electrical conductivities of the

li-quid standards were directly measured using a calibrated

conductivity handheld meter (WTW, cond 340 i)

Field experiment

A forest site located on the premises of the

Forschungszen-trum Ju¨lich was used to compare four EC-5 sensors with two

permanently installed TDR probes The soil is classified as a

Stagnic Luvisol The soil texture of the investigated topsoil

is loamy silt, the porosity amounts to 65.1%, and the

satu-rated hydraulic conductivity was determined to be

34 cm d1 The test site is equipped with an extensive setup

of instrumentation (ERT probes, TDR probe trench,

temper-ature sensors, tensiometers, and suction samplers)

The capacitance sensors were installed close to a TDR

trench The TDR probes and temperature sensors were

in-stalled horizontally in a depth of 5 cm, whereas the EC-5

sensors were inserted from the soil surface with an angle

of inclination of 45 to avoid ponding of water on top of

the sensors Thus, the resulting sampling depth reaches

from the soil surface to a depth of 3.5 cm The TDR probes

consist of three steel rods, 5 mm in diameter, 25 mm apart

and 300 mm long The TDR as well as the EC-5 sensors were

connected to a Campbell Scientific datalogger CR10X The

TDR measurements were performed using a Campbell

Scien-tific TDR100 cable tester system The transit time (t) for a TDR pulse to travel the length of the 3-wire probe and re-turn is related to the apparent permittivity eTDRof the soil

by the following equation:

eTDR¼ ct 2L

 2

ð5Þ where L is the length of the probe wires and c is the speed of light The transit time was determined with the internal algorithm of the TDR100 cable tester For more information

on TDR, the reader is referred to the review of Robinson

et al (2003)

Results

Laboratory experiment

EC-5 sensor response to supply voltage variation

To evaluate the effects of supply voltage variation, seven different voltages between 2 V and 5 V were used The re-sults for 2 V, 3.5 V and 5 V are shown inFig 3.Fig 3a pre-sents the relationship between sensor reading and permittivity and Fig 3b presents the relationship between sensor reading and equivalent soil water content It is obvi-ous that there is a significant dependency of the EC-5 sensor reading on the supply voltage Therefore, it is necessary to consider this influence in battery operating sensor network applications in which the supply voltage may not be stable Furthermore, it has to be tested whether the supply voltage influences the sensitivity of the EC-5 sensor to changes in permittivity For this purpose, we fitted the following func-tion (SRP model) to the data for each supply voltage:

where m is the output voltage The supply voltage-specific fitting parameters a, b and c are listed inTable 2

FromFig 3andTable 2, it becomes apparent that the quality of the SRP model differs for the various supply volt-ages used in this study The model provided the best fit for

Table 1 Calculated permittivities of the standard liquids and the equivalent soil water contents according toTopp et al (1980)

the 2.5 V and 3.0 V supply voltages, whereas for 3.5 V and

4.0 V the largest deviations were observed The mean RMSE

of all SRP models expressed in equivalent soil water

con-tents is 1.35 vol.% It should be noted that this error is in

addition to the error in the permittivity–soil water content

relation ofTopp et al (1980), which also has to be

consid-ered in the evaluation of the total uncertainty

Although precautions were taken, the results of the

experiment may be affected by inaccuracies in the

mixture ratio, temperature variations, air bubbles in the

solution due to stirring, or demixing of the solutions

These effects are inherent in the method proposed by

Jones et al (2005) and thus can only be minimized

Alto-gether the SRP-models were able to match the sensor

readings sufficiently well for permittivity values that

oc-cur in natural soils and thus can be used for further

investigations

A sensitivity analysis was carried out to evaluate the

influence of the supply voltage on the performance of the

EC-5 sensor For this analysis, the derivative dm/dH (D

sen-sor reading per unit soil water content) was computed using

the following approximation:

dmn

dHn

 mnþ1 mn1

Fig 4shows that dm/dH and thus the sensor sensitivity decreases strongly with increasing permittivity For exam-ple, at H  5 vol.% the average value of dm/dH is

20 mV/vol.%, whereas at H  50 vol.% this value is re-duced to 7.3 mV/vol.% Furthermore, dm/dH increases with the applied voltage, which indicates that the sensitiv-ity of the EC-5 sensor to soil water content changes in-creases with supply voltage x For example, at H  5 the value of dm/dH is 13.6 mV/vol.% for x = 2.5 V and

29.0 mV/vol.% for x = 5 V The reason for this effect is re-lated to the measuring principle of the ECH2O sensors (see Sensor technology section) When a capacitor is charged, the voltage gradient decreases until the maximum voltage

is reached (Fig 2) The higher the applied voltage on the capacitor, the higher will be the effective charging gradi-ent, which is correlated with dm/dH

Wireless networks may be associated with problems con-cerning stable power supply, especially in the case of long-term multi-year measurements Therefore, it is convenient

to define a model similar to the SRP model that directly ac-counts for variations in supply voltage This was achieved by fitting fifth order polynomial functions to describe a  x and b  x relationships This sensor reading–permittivity-supply voltage (SRPS) model is obtained by substituting a with a(x) and b with b(x) in Eq.(6):

0

200

400

600

800

1000

1200

1400

Permittivity [-]

0 200 400 600 800 1000 1200 1400

Equivalent soil water content [Vol.%]

5.0 V 3.5 V 2.0 V

SRP model (5

3.5 V) 0 V)

.0 V) SRP model ( SRP model (2

5.0 V 3.5 V 2.0 V

SRP model (5

3.5 V) 0 V)

.0 V) SRP model ( SRP model (2

Figure 3 Permittivities (left) and equivalent soil water contents (right) of the dioxane- and i-C3E1–water mixtures plotted against the EC-5 output voltage for different supply voltages (2.0 V, 3.5 V and 5 V) and the fitted SRP models

Table 2 Fitting parameters and associated RMSE values of different SRP models (Eq.(9)) and the SRPS model (Eq.(12)) for the EC-5 and EC-20 sensors (RMSE of all data pairs and RMSE of data pairs e < 30)

eðm; xÞ ¼ c þ 1=½aðxÞ þ ðbðxÞ=ðm2Þ ð8Þ

where

aðxÞ ¼ 0:00837x5þ 0:148x4 1:0111x3þ 3:318x2

 5:182x þ 3:009 RMSE ¼ 0:0031 ð9Þ

and

bðxÞ ¼ 0:00462x5 0:0835x4þ 0:584x3 1:953x2

þ 3:139x  1:904 RMSE ¼ 0:0006 ð10Þ

Fig 5 demonstrates that the SRPS-model was able to

reproduce the permittivities of most liquid standards

satis-factorily Naturally, the SRPS-model did not fit the

measured permittivity-sensor reading data as well as the

SRP-model, mainly because the value of the parameter c

in Eq.(8)is constant for different supply voltages

Conse-quently, the mean RMSE of all data pairs expressed in

equiv-alent soil water content increased from 1.35 to 1.78 vol.%

(seeTable 2) Nevertheless, the accuracy of the SRPS-model

is sufficient for EC-5 sensor applications where changes in

voltage supply are expected Especially for the

recom-mended supply voltage (2.5 V, Decagon Devices Inc., 2006), the performance of the SRPS model was still very good (RMSE of 1.12 vol.%) More information on the deriva-tion of SRPS models is available inBogena et al (2007) Comparison of sensor types EC-5 and EC-20

In this section, the EC-5 sensor operating at a frequency of

70 MHz and the EC-20 sensor (5 MHz) are compared The SRP-model (Eq (6)) was used to relate the EC-20 sensor reading to permittivity Since dioxane solutions were not available for this experiment, the model could only be fitted

to i-C3-E1data.Table 2indicates that the SRP-model fitted the experimental data very well Comparing our data with the findings ofBlonquist et al (2005), we observed a very similar response of the EC-20 sensor indicating that method

ofJones et al (2005)leads to reproducible results The derivatives of the SRP models for the two types of sensors indicate that the sensitivity of the EC-5 sensor is slightly lower than that of the EC-20 sensor (Fig 4) At a supply voltage of 5 V, mean sensitivities of 25 mV/vol.% for the EC-20 and of 16.2 mV/vol.% for the EC-5 sensor were observed This difference is similar at a supply voltage

0

200

400

600

800

1000

1200

1400

Permittivity [-]

0 200 400 600 800 1000 1200 1400

Equivalent soil water content [Vol.%]

5.0 V 3.5 V 2.0 V

SRPS model (5

3.5 V) 0 V)

.0 V) SRPS model ( SRPS model (2

5.0 V 3.5 V 2.0 V

SRPS model (5

3.5 V) 0 V)

.0 V) SRPS model ( SRPS model (2

Figure 5 Permittivities (left) and equivalent soil water contents (right) of the dioxane- and i-C3E1–water mixtures plotted against the EC-5 output voltage for different supply voltages (2.0 V, 3.5 V and 5 V) and the fitted SRPS models

0

5

10

15

20

25

30

35

40

45

Equivalent soil water content [Vol.%]

5.0V 4.5V 4.0V 3.5V 3.0V 2.5V 2.0V

0 5 10 15 20 25 30 35 40 45

Equivalent soil water content [Vol.%]

2.5V 5.0V

Figure 4 Derivative dt/dH for different parameterizations of the SRP-model for the EC-5 sensor (left) and the EC-20 sensor (right)

of 2.5 V (EC-20: 10.9 mV/vol.% and EC-5: 8.1 mV/vol.%).

The largest difference in sensitivity occurs for low soil water

content, where the sensitivity of the older EC-20 sensor is

considerably higher than that of the EC-5 sensor These

re-sults indicate that the sensitivity of both sensor types is

comparable, except for dry soils

Sensor response to temperature and conductivity

variation

Fig 6a displays the deviation of the measured permittivity

from the known liquid permittivity for a temperature range

of 5–40 C The measured permittivity was derived with the

SRP models presented inTable 2.Fig 6b presents the same

data expressed as equivalent soil water content It should

be noted that the raw data did not cross the zero deviation

line at 20 C as would be expected because the SRP models

were derived at this temperature This is attributed to

uncertainty in the mixture ratio, which affects the

permit-tivity of the liquid standards, and uncertainty in the SRP

model, which can be quite significant at high permittivities

because of the low sensitivity of the probe in this range

(Fig 4) For a more consistent interpretation, the data were

referenced to room temperature Our experiment shows

that the EC-5 sensor generally shows an increase in

mea-sured soil water content with increasing temperature

(Fig 6) Interestingly, Blonquist et al (2005)found a

de-crease in measured soil water with increasing temperature

for the EC-20 sensor This discrepancy is surprising since

similar behaviour was expected for both sensors The

max-imum error in soil water content due to temperature effects

on the sensor circuitry is an overestimation of 1.8 vol.% at a

temperature of 40 C At temperatures below 20 C, the

EC-5 sensor underestimates soil water content up to a

maxi-mum underestimation of 1.1 vol.% at 5 C

Capacitance sensors are often sensitive to bulk electrical

conductivity For example,Blonquist et al (2005)found a

significant overestimation of soil water content in liquid

standard media (e = 40, equivalent soil water content of

51 vol.%) with electrical conductivities >0.5 dS m1 using

the EC-20 sensor Therefore, the EC-5 sensor was used to

evaluate to which extent the 70 MHz technology of the EC-5 sensor improves the conductivity sensitivity Fig 7 shows that the EC-5 sensor underestimates the actual soil water content in the presence of significant bulk electrical conductivity The maximum underestimation is 5 vol.% at

a conductivity of 0.8 dS m1 Interestingly, at a conductiv-ity of 2.5 dS m1the underestimation switches to an over-estimation of soil water content The EC-20 sensor already provided unrealistic soil water content measurements at a bulk electrical conductivity of 0.5 dS m1 To put these numbers in a hydrological context, a saturated sandy soil may have a bulk electrical conductivity of 0.1 dS m1, which results in a soil water content underestimation of

1.3 vol.% A loamy soil will typically have a higher bulk electrical conductivity For example, a typical bulk electri-cal conductivity of 0.5 dS m1results in a soil water content underestimation of 4.5 vol.% Even higher bulk electrical conductivities of 1 dS m1 are typically observed in 2:1 clays, which results in the maximum soil water content underestimation of 5.4 vol.% Overall, it can be concluded that the EC-5 sensor is much less sensitive to bulk electrical conductivity than the older EC-20 sensor It should be noted that the numbers provided above are valid for a reference permittivity of 40, which corresponds with an equivalent soil water content of 51 vol.% Since the actual voltage dif-ferences associated with the permittivity and equivalent soil water content estimates inFig 7were relatively small,

it is expected that the influence of electrical conductivity is even less for lower soil water content

To verify whether temperature and conductivity correc-tion funccorrec-tions are feasible for adjusting the field data, pre-liminary temperature and electric conductivity correction functions (third- and fourth-order polynomial equations) were fitted through the EC-5 measurements shown inFigs

6 and 7:

ect¼ 0:0002217T3 0:01442T2þ 0:1175T þ 1:6403

ecc¼ 0:4180r4þ 4:5804r3 18:335r2þ 23:393r

-4

-2

0

2

4

T [C°]

EC-5 (5.0 V) EC-5 (3.0 V)

-2 -1 0 1 2

T [C°]

EC-5 (5.0 V) EC-5 (3.0 V) Temp corr model

Figure 6 Deviations of the permittivity (left) and equivalent soil water content (right) predictions using the EC-5 sensor with two supply voltages (3.0 V and 5.0 V) in a 0.6 i-C3E1–water solution with a temperature range of 5–40 C

where ectand eccare temperature and electric conductivity

correction factors, T is the soil temperature in C and r is

the soil bulk electrical conductivity in dS m1 The

cor-rected dielectric permittivity ec is then calculated as

follows:

It has to be noted that Eqs.(11) and (12)are only

preli-minary correction functions since more measurements for a

larger range of reference permittivity; temperature and

electric conductivity have to be collected and to be used

to develop more universal functions that can be used for

correcting the EC-5 measurements

Field experiment

Fig 8 presents the dielectric permittivity and soil water

content time series hourly measured with TDR and the

EC-5 sensors, and the topsoil temperature and precipitation

during the period from 1 June to 31 December 2006 The

permittivity measured with the EC-5 sensor was calculated

with the appropriate SRP model derived earlier The

corre-sponding soil water content was calculated using Eq.(5)for

both the EC-5 sensor and the TDR probe Although Eq.(5)

might not be the most appropriate permittivity–soil water

content relationship for the loamy silt soil, the comparison

between the two sensors is unaffected by this choice

It can be seen inFig 8that the mean of the EC-5 and the

TDR sensors show similar soil water content dynamics in

re-sponse to precipitation However, there is a systematic

deviation between both sensor types TDR resulted in a

mean soil water content of 31.0 vol.%, whereas the EC-5

sensors only measured 26.6 vol.% There are three possible

explanations for the lower soil water content measured with

the EC-5 sensor First, the electrical conductivity of the

investigated soil is moderate with an average bulk

conduc-tivity of 0.06 dS m1, as measured with TDR.Fig 7shows

that this mean bulk electrical conductivity will result in an

underestimation of 0.8 vol.% for e = 40 Due to the

nonlin-earity of Eq.(4), the correction will be higher for lower soil

water contents Second, there might be a temperature

ef-fect on either the soil dielectric permittivity or the sensor electronics Since the maximum topsoil temperature was 19.76 C the temperature effect on the sensor electronics will lead to underestimation of the soil water content (see Fig 6) The temperature effect on the soil dielectric per-mittivity has not been yet investigated at this experimental site, and the impact of these changes can therefore not be investigated in this study

Preliminary investigations with correction functions for bulk electrical conductivity and temperature using Eqs (11) and (12) showed that the systematic deviations can

be reduced (seeFig 8c) The corrected sensor signal using all installed EC-5 sensors led to a mean soil water content

of 29.5 vol.%, which is only 0.6 vol.% lower than the mean

of the TDR sensors Furthermore, the Nash-Sutcliffe effi-ciency coefficient (Nash and Sutcliffe, 1970) increased from 0.23 to 0.74 indicating that the correction functions were able to improve the EC-5 measurements However, it has

to be noted that given the complex relation between volt-age reading and permittivity, it seems likely that the correc-tion funccorrec-tions will vary with dielectric permittivity Therefore, the results achieved with Eqs.(11) and (12)are only a first indication that correction functions are able to improve soil water content measurement with capacitance sensors

A further explanation for the observed differences is soil variability Although we considered the mean of two TDR probes and four EC-5 probes within a relatively small soil volume, this is no guarantee that all variability is captured

by the mean values The significance of vegetation patterns and soil variability at the test site is indicated by significant standard deviations of the measured soil water content by four EC-5 sensors (see Fig 8) Also, the two TDR sensors showed distinct deviations in the measured soil water con-tent (mean: 2.2 vol.%, max.: 7.9 vol.%) Finally, there is a sampling depth issue because the TDR probes were installed horizontally at 5 cm depth and the EC-5 sensors measure the soil water content of the upper 3.5 cm Clearly, a dryer upper soil could also partly explain the observed deviations These sampling volume and soil heterogeneity issues, which are typical for field evaluations of soil water content sen-sors, motivated the development of the standardized sensor

-20

-15

-10

-5

0

5

10

15

20

Conductivity [dS/m]

EC-5 (3.0 V) EC-5 (5.0 V) Conductivity correction model

-10 -8 -6 -4 -2 0 2 4 6 8 10

Conductivity [dS/m]

EC-5 (3.0 V) EC-5 (5.0 V) Conductivity correction model , 3.0 V

Figure 7 Deviations of the permittivity (left) and the equivalent soil water content predictions (right) of a 0.6 i-C3E1–water solution using EC-5 sensors with two supply voltages, and the conductivity correction model of Eq.(12)

testing methodologies (Blonquist et al., 2005; Jones et al.,

2005)

Conclusions and outlook

The laboratory and field evaluation of the EC-5 sensor have

led us to the following conclusions:

•The standardized EM sensor characterization method

proposed by Jones et al (2005) leads to reproducible

results Therefore, it is recommended that future sensor

evaluations also adhere to this characterization method

•The sensor sensitivity test showed that the sensitivity of

the sensor reading to soil water content decreased

strongly with increasing soil water content and to some

extent with decreasing supply voltage Models that relate

the sensor reading to soil permittivity for a given supply

voltage were successfully derived In addition, a model

that relates the sensor reading and the supply voltage

to soil permittivity was also derived successfully

•The electronics of the EC-5 sensor are sensitive to

tem-perature variations A maximum error of 1.8 vol.%

occurred at a soil temperature of 40 C for a medium

with a permittivity of 40, which corresponds to an

equiv-alent soil water content of 51 vol.%

•The EC-5 sensor is sensitive to bulk electrical conductiv-ity, although to a lesser extent than the older EC-20 sen-sor The maximum error of 6 vol.% occurred at a bulk electrical conductivity of 1 dS m1for a permittivity of

40 (equivalent water content of 51 vol.%)

•The field evaluation showed distinct differences between TDR and EC-5 measurements that could be explained to a large degree with the correction functions derived from the laboratory measurements Remaining errors are pos-sibly due to soil variability and discrepancies between measurement volume and installation depth

Overall, we conclude that the EC-5 sensor is suitable for wireless network applications However, the results of this paper also suggest that temperature and electric conductiv-ity effects on the sensor reading have to be compensated using appropriate correction functions The use of temper-ature and conductivity correction functions requires the continuous measurement of these properties This could

be realized using the ECH2O TE sensor (Decagon Devices Inc.), which uses the same technology for measuring soil water content as the EC-5 sensor Therefore, future work will focus on the evaluation of this sensor We will also eval-uate temperature and conductivity effects in more detail to derive more general correction functions

0

10

20

a

b

c

0 10 20

3 cm

0.2

0.3

0.4

0.5

date

3 cm

0.1

0.2

0.3

0.4

0.5

EC5

mean

Figure 8 Rainfall intensities and temperatures for the observed period (a), as well as the volumetric water content obtained from both TDR sensors and mean values and standard deviations of all EC-5 sensors (b), respectively The EC-5 permittivities were corrected using Eq.(13)(c)