Computers and Electronics in Agriculture 93 (2013) 46–54

Contents lists available at SciVerse ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Automated sensing of hydroponic macronutrients using a

computer-controlled system with an array of ion-selective electrodes
Hak-Jin Kim a,⇑, Won-Kyung Kim a, Mi-Young Roh b, Chang-Ik Kang c, Jong-Min Park d,
Kenneth A. Sudduth e
a
Department of Biosystems Engineering and Biomaterials Science, Seoul National University, Seoul 151-921, Republic of Korea
b
Rural Development Administration, Suwon 441-707, Republic of Korea
c
Science Lab Center Co., Daejeon 305-509, Republic of Korea
d
Department of Bio-Industrial Machinery Engineering, Pusan National University, Miryang 627-706, Republic of Korea
e
Cropping Systems and Water Quality Research Unit, USDA-ARS, Columbia, MO 65211, USA

a r t i c l e i n f o a b s t r a c t

Article history: Automated sensing of macronutrients in hydroponic solutions would allow more efficient management
Received 21 August 2012 of nutrients for crop growth in closed hydroponic systems. Ion-selective electrodes (ISEs) are a promising
Received in revised form 8 January 2013 approach because they can directly measure the analyte with a wide range of sensitivity and are small
Accepted 20 January 2013
and portable. However, signal drift and reduced accuracy over time are major concerns with an in-line
management system based on continuous immersion of ISEs in a solution. An ISE-based, computer-con-
trolled measurement system for automatic sampling, calibration, and electrode rinsing is a viable tech-
Keywords:
nique to improve accuracy and precision in the determination of nutrient concentration. This study
Hydroponic nutrients
PVC ion-selective membrane
reports on development and evaluation of such a system for direct measurement of macronutrients in
Sensor array recirculating hydroponic solutions. The sensitivity and selectivity of PVC membrane-based ISEs fabricated
Greenhouse with TDDA–NPOE, valinomycin, and calcium ionophore II were satisfactory for measuring NO3–N, K, and
Nitrate Ca concentrations in single-ion solutions over ranges typical of hydroponic solutions. However, an
Potassium acceptable Mg membrane was not identified. An ISE array incorporating the selected membranes, in com-
bination with a computer-controlled measurement system, was evaluated for the direct measurement of
NO3–N, K, and Ca concentration in paprika hydroponic nutrient solutions. Using previously developed
baseline correction and two-point normalization methods, ISE-measured NO3–N and K concentrations
in spiked or diluted hydroponic nutrient samples were strongly related to those determined using stan-
dard laboratory instruments (R2 = 0.85). However, the tested Ca electrode did not provide satisfactory
results due to reduced sensitivity and poor selectivity in the hydroponic solution. The approach used
in this research, where a base solution is used for both referencing the baseline and rinsing electrodes
and two calibration solutions are injected to implement the two-point normalization method, could be
used in an automated sensing system for hydroponic nutrients in greenhouses. Additional research is
required to identify ion-selective membranes for Ca and Mg ions in hydroponic solutions.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction
Hydroponics, known as a soilless farming, is a method of grow-
ing plants using mineral nutrient solutions in water, without soil.
The macronutrients, nitrogen (N), phosphorus (P), and potassium
(K) in the hydroponic solution are essential for crop growth. These
nutrients are generally taken into plants in various ionic forms,
2
such as nitrate ðNO 
3 Þ, orthophosphates (H2 PO4 or HPO4 ), and
potassium (K+) through a combination of root interception, mass
⇑ Corresponding author. Tel.: +82 2 880 4604; fax: +82 2 873 2049.
E-mail addresses: kimhj69@snu.ac.kr (H.J. Kim), w-kyung@nate.com (W.K. Kim),
rohmy17@korea.kr (M.Y. Roh), me-gane@hanmail.net (C.I. Kang), parkjssy@
pusan.ac.kr (J.M. Park), ken.sudduth@ars.usda.gov (K.A. Sudduth).
flow and diffusion processes (Havlin et al., 1999). Since the re-
serves of nutrients for the plants are limited in hydroponic green-
house cultivation, an inaccurately balanced nutrient solution may
result in an unbalanced nutrient composition in the root environ-
ment (Savvas, 2002). In closed hydroponic systems that reuse the
drainage solution, it is important to determine the nutrient con-
centrations in the reused solution to regenerate a nutrient solution
of optimal composition (Gutierrez et al., 2007).
Current practices for managing hydroponic nutrients in closed
systems are usually based on automatic control of electrical con-
ductivity (EC) of the nutrient solution. In this method, solution
quality is maintained through the automated injection of nutrient
concentrates into the hydroponic reservoir in proportion to declin-
ing solution EC arising from plant uptake (Bailey et al., 1988;

0168-1699/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.compag.2013.01.011
 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54
Cloutier et al., 1997). A main problem with this practice is that EC
measurements provide no information on the concentrations of
individual ions, and therefore do not allow individual real-time
corrections to each nutrient in response to the demand of the crop
(Cloutier et al., 1997). Improved efficiency of fertilizer utilization
may be possible through accurate measurement and control of
the individual nutrients in the solution in real time.
The need for such fast, continuous monitoring has led to the
application of ion-selective electrode (ISE) technology to measure
hydroponic macronutrients (Bailey et al., 1988; Cloutier et al.,
1997; Gutierrez et al., 2007). Advantages over standard analytical
methods (spectroscopic techniques) include simple methodology,
direct measurement of analyte, sensitivity over a wide concentra-
tion range, low cost, and portability (Hashimoto et al., 1989; Hei-
nen and Harmanny, 1992). The key component of an ISE is the
ion-selective membrane that responds selectively to one analyte
in the presence of other ions in a solution. Recently, significant pro-
gress has been made toward the development of various ion-selec-
tive membranes in the area of analytical chemistry. There are
currently ion-selective membranes available for most of the impor-
tant hydroponic nutrients, including NO + 2+
3 , K , Ca , and Mg
2+
(Niel-
son and Hansen, 1976; Shen et al., 1998). Reliable phosphate
sensing has been more difficult. Although several polymer mem-
brane-based phosphate ISEs have been developed (Carey and Rig-
gan, 1994; Glazier and Arnold, 1991), they have generally
exhibited poor selectivity for phosphate and a limited lifetime. As
an alternative, cobalt-based electrodes have been developed and
have exhibited sensitive responses to phosphate and acceptable
selectivity (Xiao et al., 1995; Kim et al., 2011).
There are several potential disadvantages of ion-selective elec-
trodes, as compared to standard analytical methods. One is chem-
ical interference by other ions because ion-selective electrodes are
not truly specific but respond more or less to a variety of interfer-
ing ions. To overcome interference issues, various data processing
methods have been used. For example, multivariate calibration
models have been proposed to allow cross responses arising from
primary and interfering ions to be decoupled, thus allowing accu-
rate determination of individual ion concentrations within mix-
tures (Forster et al., 1991). In some cases, another compound can
be added to suppress the interference effect. For example, Ag2SO4
can be used to suppress the chloride interference in nitrate sensing
(Van Lierop, 1986).
Another disadvantage is reduced accuracy due to electrode re-
sponse drift and biofilm accumulation caused by the presence of
organic materials in hydroponic solutions (Carey and Riggan,
1994; Cloutier et al., 1997). In particular, signal drift and biofilm
accumulation may be a major concern when considering an in-line
management system that includes continuous immersion of ISEs in
hydroponic solution. Bailey et al. (1988) investigated the perfor-
mance of commercially available ISEs (NO3, Ca2+, K+, H+, and
Cl) in conjunction with a flow cell installed in a nutrient film tech-
nique (NFT) system. They found a decrease in sensitivity over time
for all of the tested electrodes, which they suggested could be ad-
dressed by regular calibration. Heinen and Harmanny (1992)
investigated the use of membrane-based NO3 and Ca ISEs and a
glass-based K ISE to measure hydroponic nutrients in an auto-
mated NFT system. They reported that the electrodes provided
more accurate estimates when temperature of the nutrient solu-
tion was measured or maintained constant due to a temperature
dependency of the electrode response and that frequent calibra-
tions were needed to maintain usability over long periods. Addi-
tionally, the performance of the Ca electrode was not satisfactory
for use in hydroponic solution due to its low sensitivity.
Successful application of ISE technology to real-time sensing of
hydroponic nutrients requires continuous determination of indi-
vidual nutrient concentrations with acceptable sensitivity and
47
stability. In general, stability and repeatability of response are a
concern in the use of an array of multiple ISEs to measure ion con-
centrations in a series of samples because accuracy of the measure-
ment may be limited by drift in electrode potential over time. The
use of a computer-based automatic measurement system would
improve accuracy and precision in the determination of nutrient
concentration because consistent control of sample preparation,
sensor calibration, and data collection can reduce variability
among multiple electrodes during replicate measurements (Dor-
neanu et al., 2005; Kim et al., 2007, 2011). Ideally, an automated
sensing system for hydroponic nutrients would be able to period-
ically calibrate and rinse the electrodes and continuously measure
nutrients in the hydroponic solution, while automatically intro-
ducing solutions for calibration and rinsing as well as
measurement.
In a previous study (Kim et al., 2007), we established the ability
of ISEs and Kelowna extractant (pH = 4.0) to simultaneously quan-
tify NO3–N, P and K ions in a set of 37 soils. We developed methods
(baseline correction and two-point normalization) that standard-
ized ISE response and allowed application of a single calibration
equation across multiple ISEs fabricated simultaneously. A test
stand with an array of ISEs was constructed and used to analyze
soil extracts obtained through a standard Kelowna-based extrac-
tion process in the laboratory.
The overall goal of this research was to modify and apply the
ISE-based macronutrient sensing system of Kim et al. (2007) for
real-time and direct measurement of macronutrients in recirculat-
ing hydroponic solutions based on automatic sampling and elec-
trode rinsing. Specific objectives were to (1) identify ion-selective
membranes suitable for measuring typical ranges of macronutrient
concentrations in hydroponic solutions with respect to their sensi-
tivity, lower detection limits, and selectivity against interferences
of other ions, and (2) evaluate the ability of a sensor array consist-
ing of the selected ion-selective membranes, in combination with a
laboratory-made automated test stand, to directly estimate NO3–N,
K, and Ca concentrations in hydroponic nutrient solutions.
2. Materials and methods
2.1. Preparation of NO3–K–Ca–Mg ion-selective electrodes
Five different PVC-based ion-selective membranes were pre-
pared with different mixtures of chemicals for sensing nitrate
ðNO + 2+ 2+
3 Þ, potassium (K ), calcium (Ca ) and magnesium (Mg ) ions
(Table 1). The chemical compositions of the NO3 and K ion-selec-
tive membranes were the same as those reported by Kim et al.
(2006). One and two chemical compositions for Ca and Mg ion-
selective membranes, respectively, were prepared according to
the procedures described in previous studies (Erne et al., 1980;
Maj-Zurawsk et al., 1989; Müller et al., 1988; Schefer et al., 1986).
The membrane disks with a diameter of 2.5 mm were attached
to the ends of laboratory-made plastic bodies of 44 mm length
using tetrahydrofuran (THF) solvent. Each ISE was filled with an
internal solution consisting of 0.01 M NaNO3 + 0.01 M NaCl,
0.01 M KCl, 0.01 M CaCl2, and 0.01 M MgCl2, for NO3, K, Ca, and
Mg ions, respectively. An Ag/AgCl electrode prepared by coating
silver wire (99%) with a diameter of 1 mm with Ag/AgCl ink (Model
01164, ALS, JAPAN) was immersed as the inner reference electrode.
A double junction glass electrode (Orion 90-02, USA) was used as
the reference electrode.
2.2. Development of computer-controlled test stand
A computer-controlled test apparatus (Fig. 1), based on auto-
matic data collection and electrode rinsing, was developed for
48 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54

Table 1 and automated control of the system. A data acquisition board

Chemical compositions of NO3, K, Ca, and Mg ion-selective membranes used in the (NI PCI-6221, National Instruments, USA) and an Intel-core i3 com-
study.
puter were used to collect and store ISE voltage outputs and to con-
Target Sensing element Plasticizer PVC Reference trol the motor speed and the on/off operation of solenoid valves
ion (wt%) (Figs. 1 and 2 (left)).
Component wt% Component wt%
NO TDDA 15 NPOE 40 45 Kim et al.
Various test solutions were contained in eight stainless steel
3
(2006) buckets, and were automatically transferred to the sample holder
K+ Valinomycin 2 DOS 64.7 32.8 Kim et al.
by a multi-channel peristaltic pump (Masterflex L/S 7519-06,
(2006) Cole-Parmer, USA) and 3-way solenoid valves (HTV 0301-3, Hyo-
KTpClPB 0.5 shin, Korea). The rotational speed of the sample holder was con-
Ca2+ Calcium 1.0 NPOE 65.6 32.8 Schefer trolled using a servo motor (HC-KFS13, Mitsubishi, Japan), a
ionophore II et al. servo amplifier (MR-J2S-10A, Mitsubishi, Japan), and a motion con-
(1986) troller (NI PCI-7390, National Instruments, USA). The test solution
KTpClPB 0.6
was slowly stirred by rotating the sample holder at approximately
Mg2+ Magnesium 1.4 NPOE 64.5 33.1 Erne et al. 37 rpm during data collection and the rotational speed was in-
Ionophore I (1980)
creased to approximately 290 rpm to expel solutions centrifugally
KTpClPB 1.0
Magnesium 1.0 Chloroparafin 65.5 32.9 Müller from the holder between tests and rinses. These rotational speeds
Ionophore III et al. were determined through preliminary tests conducted to optimize
(1988) performance of the automated test stand. Three rinses were used
KTpClPB 0.6
at each solution exchange to completely remove any residue of
TDDA: tetradodecylammonium nitrate; NPOE: 2-nitrophenyl octylether. the previous solution. An isolation circuit board (NI SCC-AI13, Na-
KTpClPB: potassium tetrakis(4-chloropheny)borate; DOS: bis(2-ethyhexyl) tional Instruments, USA) in voltage follower configuration (Fig. 2,
sebacate. right) was used for buffering the impedance of each electrode.
A program was developed in Labview CVI (ver. 9.0) to control
the simultaneous sampling of electromotive forces (EMFs) from 8 the system and record values obtained from the ISE electrodes
ISE electrodes measured relative to a reference electrode. This test (Fig. 3). Different test sequences were created in the Labview pro-
stand was revised from the system originally developed for ISE- gram for the sensitivity, selectivity, and sample tests. The sensitiv-
based soil macronutrient measurements by Kim et al. (2007). As ity test investigated the response of ion-selective membranes to
shown in Fig. 1, the test stand included a pan-like Teflon sample different ion concentrations through automated sequential intro-
holder in which the test solution was contained, containers to hold duction of six different solutions, in which the concentration of
calibration and rinse solutions, and a data acquisition and control the ion of interest ranged from 106 M to 101 M. The selectivity
system that allowed automatic measurement of the ISE voltages test compared ISE response to the primary ion at 0.1 M with the

Rinsing solution Calibration /Test solutions

Valve Magnetic Relay

Data Acquisition System Computer

Pump

Automatic Introduction Automatic/Manual Motor Driver

Loading
N K Ca
Sample Holder
ISE ISE Ref. ISE ISE

Motor

Fig. 1. Automated test stand for multiple electrode tests (left) and the schematic diagram of the test stand (right, modified from Kim et al., 2007).

Fig. 2. Block diagram of the test stand control system (left) and circuit diagram for voltage source buffering and impedance transformation (right).
 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54
Fig. 3. Screen view of the Labview program developed to operate the automated test stand (modified from Kim et al., 2011).
response to interfering ions at 0.01 M. In this test sequence up to
five interfering ions were automatically introduced, each of these
alternating with re-introduction of the primary ion. In the sample
measurement test, the electrode rinsing solution was automati-
cally introduced, whereas the test samples were manually loaded.
Manual loading allowed more efficient testing of more samples
than using the automated test stand with the eight buckets.
Each individual test began when the desired volume of test
solution had been delivered to the solution holder. After 60 s, three
EMF measurements, each consisting of the mean of a 0.1 s burst of
1 kHz data, were obtained on a 0.1 s interval by the A/D board. The
three readings were averaged to represent a single EMF output re-
sponse at each concentration level.
2.3. Sensitivity and selectivity tests
The sensor array used in the sensitivity and selectivity tests in-
cluded eight ISEs, two each for NO3, K, Ca, and Mg, plus a reference
electrode. The sensitivity responses of all electrodes were charac-
terized by measuring the EMFs of each ISE in six standard solutions
of single ions with concentrations ranging from 106 to 101 M. In
accordance with the methods described by Kim et al. (2006), the
selectivity factors (Kij) were determined using the separate solu-
tion method (SSM), where the selectivity factors are calculated
based on EMF values obtained with pure single electrolyte solu-
tions of the primary ion (0.01 M) and interference ion (0.1 M).
The selectivity factor is a measure of the preference by the sensor
for the interfering ion (j) relative to the ion (i) to be detected. Obvi-
ously, for ideally selective membranes, all of the Kij values should
be zero. A selectivity factor <1 indicates a preference for the pri-
mary ion (i) relative to the interfering ion (j) (Kim et al., 2006).
For sensitivity testing, solutions were arranged and tested in se-
quence from lowest to highest concentration of the test ion. For
selectivity testing, the test solutions were arranged and tested in
a sequence from lowest to highest selectivity for the primary ion
over the interfering ion. In each instance, three iterations of each
sequence were conducted.
2.4. Calibration of the sensor array and application to hydroponic
samples
The sensor array used to measure hydroponic nutrients in real
samples included nine electrodes, i.e., three NO3 electrodes with
the TDDA (tetradodecylammonium nitrate)–NPOE (2-nitrophenyl
octylether) membranes, three K electrodes with the valinomycin-
DOS (bis(2-ethyhexyl) sebacate) membranes, and two Ca
49
electrodes with the Ca ionophore II-based membranes, and one
reference electrode. The Mg electrodes were not included in the
sensor array, as they were found to have unacceptable sensitivity
(results below).
Mixed solutions containing NO3, K, Ca, Mg, and PO4 ions were
used to develop optimum calibration equations that allowed accu-
rate determination of individual ions (NO3, K, and Ca) in mixtures.
The Mg and PO4 ions were used as interfering ions that might affect
the responses of the NO3, K, and Ca electrodes. The concentration
ranges for each ion were chosen to encompass the typical concen-
tration ranges found in hydroponic solutions used in Korea, i.e., 3,
30, and 700 mg L1 of NO3 (equivalent to 0.68, 6.8, and
158.1 mg L1 of NO3–N) and 3, 30, and 700 mg L1 K using NaNO3
and KCl, respectively; and 3, 30, and 300 mg L1 of Ca, Mg, and
PO4–P using CaCl2, MgCl2, and NaH2PO4, respectively. The concen-
trations for each ion were categorized as low, medium, and high
concentration levels, respectively. The composition of the 27 solu-
tions used for the calibration of the sensor array was designed
accordingly to a fractional factorial design with three levels of con-
centration and five factors (ions).
Measurement of nutrient concentrations using the sensor array
was evaluated using a paprika nutrient solution described by Son-
neveld (2002), which was obtained from an experimental green-
house of the National Institute of Horticultural and Herbal
Science located in Busan, Republic of Korea. This solution (Table 2)
was diluted or spiked to provide samples spanning the ranges of 3–
300 mg L1 NO3–N, 3–700 mg L1 K, and 3–300 mg L1 Ca with
four different concentration levels, i.e., very low, low, high, and
very high. A total of 29 hydroponic samples randomly selected
from 64 solutions that encompass the total concentration ranges
of the three nutrients based on a three-factor, four-level, complete
block design (4  4  4) were prepared by adding highly concen-
trated individual NO3, K, and Ca solutions to the original paprika
solution or diluting the paprika solution with DI water. The con-
centrations of the test solutions were verified by analysis in an
independent chemical testing laboratory (NICEM, Seoul, Korea).
Prior to sample measurement, each test sequence began with
triple rinsing of the electrodes by introducing a solution of
106 M KNO3 into the sample holder. After the rinse sequence,
about 70 mL of sample solution was manually loaded. The operator
pressed a key after sample loading to accurately reference the data
collection time to introduction of the new test solution.
As in Kim et al. (2007), two data processing methods, baseline
correction and two-point normalization, were used to minimize
potential drift that might occur during continuous measurement
and to standardize responses of multiple electrodes for each ion
50 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54

Table 2
Average concentrations of nutrients in the original paprika nutrient solution used in the study.

NO
3 -N NH4+–N PO
4 K+ Ca2+ Mg2+ SO2 Cl Na+
4
1
Concentration (mg L ) 198.5 4.3 54.3 103.7 171.8 3.4 102.9 32.2 23.0

(a) 350 (b)

300
-2

250
EMF (mV)

logKij
-4
200

150
-6
100

50 -8
10-7 10-6 10-5 10-4 10-3 10-2 10-1 100 Cl- SO42- H2PO4-

Ionic Concentration (M) Interfering ion

(c) 150
100
(d)
50 -2
EMF (mV)

0
logKij

-4
-50

-100 -6

-150
-8
10-7 10-6 10-5 10-4 10-3 10-2 10-1 100 NH4+ Na+ Ca2+ Mg2+
Ionic concentration (M) Interfering ion

Fig. 4. Response curves of the TDDA–NPOE-based NO3 ion-selective membrane and valinomycin-based K ion-selective membrane to different NO3 and K concentrations,
respectively (a and c) and selectivity coefficients of the two electrodes measured in DI water (b and d).

so that a single calibration equation could be applied across all 2002). These ranges of interest correspond to 1.2  102 and
electrodes of given type. Baseline-corrected EMF measurements 8.9  103 M NO3 and 5.7  103 and 4.4  103 M K, respectively,
were calculated by subtraction of the baseline EMF from each sen- which would be measurable with the NO3 and K electrodes.
sor reading and the two-point normalization consisted of a sensi- The mean log(Kij) for Cl, SO2 
4 , and H2 PO4 obtained with the
tivity adjustment followed by an offset adjustment. As a result, TDDA–NPOE-based NO3 ion-selective membrane were 2.49,
the EMF data measured with all two or three sensors for each 3.30, and 4.48 (Fig. 4b), showing that the highest interference
ion were normalized by first applying the baseline correction to was caused by Cl. However, the membrane was still over 300 times
each reading and then by applying the two-point normalization more sensitive to NO3 than to Cl. The valinomycin-based K ion-
approach separately to each replication. selective membrane showed selectivity for K in the following or-
der: NH4+  Na+ < Mg2+  Ca2+ (Fig. 4d). When considering typical
ranges of cation concentrations in hydroponic solutions, it was ex-
3. Results and discussion pected that the valinomycin-based K membrane would determine
K concentrations with acceptable selectivity levels.
3.1. Response of ion-selective membranes to nutrients The Ca ion-selective membrane showed sensitive responses to
varying Ca concentrations ranging from 106 to 101 M (Fig. 5a).
The responses of the NO3 and K ion-selective electrodes based However, since the Ca ion has a double charge, a lower sensitivity
on TDDA–NPOE and valinomycin, respectively, are shown in slope of approximately 26 mV/decade was observed as compared
Fig. 4 when single ion concentrations of NO3 (Fig. 4a) and K to those measured with the NO3 and K electrodes of +50 to
(Fig. 4c) ranged from 106 to 101 M. The EMF values obtained 55 mV/decade. Selectivity levels for Ca over four interfering cat-
with the NO3 and K membranes were nearly linearly proportional ions, Na+, NH4+, K+, and Mg2+, were satisfactory, with mean log
to the logarithm of NO3 and K concentrations, respectively, in the selectivity factors for Na+, NH4+, K+, and Mg2+ of 5.56, 5.79,
range of 105 to 101 M. However, there was little change in volt- 6.97, and 3.5, respectively (Fig. 5b).
age readings in the range of 106 to 105 M concentrations. The re- The two tested Mg electrodes gave sensitive responses to vary-
sults are comparable to those reported by Kim et al. (2006). The ing Mg concentrations in the range of 106 to 101 M (Fig. 5c). The
typical concentrations of interest in hydroponic solutions for papri- Mg electrode based on Mg ionophore I showed greater sensitivity
ka and lettuce crops, which are commonly cultivated in Korean for Mg than did the electrode based on ionophore III. In particular,
greenhouses, have been reported to be 178.5 and 125 mg L1 the sensitivity of the Mg ionophore III based electrode was not sat-
NO3–N and 225 and 175 mg L1 K, respectively (Sonneveld, isfactory in the concentration range of Mg of 104 to 103 M
 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54 51

200
(a) (b)
150

-2
100

EMF (mV)

logKij
50 -4

0
-6
-50

-100 -8
10-7 10-6 10-5 10-4 10-3 10-2 10-1 100 Na+ NH+ K+ Mg2+
Ionic Concentration (M) Interfering ion

200
(c) (d)
2
150

100 0
EMF (mV)

logKij
50
-2

0 Mg(I) ISE
Mg(III) ISE -4
-50

-100 -6
10-7 10-6 10-5 10-4 10-3 10-2 10-1 100 Na+ NH+ K+ Ca2+
Interfering ion
Ionic Concentration (M)

Fig. 5. Response curves of a Ca ion-selective membrane (a) and two different Mg ion-selective membranes (c) to different Ca and Mg concentrations, respectively and
selectivity coefficients of the Ca (b) and Mg (d) electrodes over interfering cations measured in DI water.

0
(a) (b)
250 N-01 N-01
N-02 -50 N-02
N ISE- EMF (mV)

N ISE- EMF (mV)

200 -100

150 -150

-200
100
-250
50
-300
0.1 1 10 100 1000 0.1 1 10 100 1000
NO3-N concentration (mg/L) NO3-Nconcentration (mg/L)

Fig. 6. Responses of two NO3 electrodes to different NO3–N concentrations obtained with (a) original EMF reading data and (b) EMF data modified based on the baseline
correction method.

present in hydroponic solutions, with an average sensitivity slope EMF values at the same NO3 concentrations. However, as shown in
of 7 mV/decade. In the selectivity tests (Fig. 5d), Mg ion-selective Fig. 6b, use of the baseline correction method reduced the variabil-
membrane (I) was 10 times more sensitive to Ca than to Mg, as ity of response between the two electrodes as compared to that
indicated by the mean log selectivity factor for Ca of 1.01. The measured with the direct EMF data. However, high standard devi-
selectivity of the Mg (III) membrane was similar to that of the ations in EMF values were still observed at low NO3 concentrations
Mg (I) membrane (data not shown). Therefore, neither of the tested and the 2-point normalization method was used to reduce the re-
Mg ion-selective membranes was acceptable for use in the hydro- sponse variability (Fig. 7a). Similarly, the EMF responses of the K
ponic solution. It would be necessary to identify a different mem- electrodes were modified using the baseline correction and 2-point
brane that could determine concentrations of Mg in the presence of normalization methods. As a result, two representative response
Ca with acceptable sensitivity and good selectivity. curves for the NO3 and K electrodes were obtained (Fig. 7), with
linearly sensitive responses to the logarithm of NO3–N and K con-
3.2. Development of mixed solution calibration equations centrations ranging from 0.6 to 180 mg L1 and 3 to 700 mg L1,
respectively.
The NO3 electrodes in mixed solution containing NO3, K, Ca, Mg, In contrast to responses measured with the NO3 and K elec-
and PO4 ions were sensitive to different NO3 concentrations trodes, the two Ca electrodes did not show a consistent response
regardless of the presence of other ions (Fig. 6). The direct EMF re- to different Ca concentrations and high variations in EMF values
sponses of the two NO3 electrodes to different NO3 concentrations were observed at the same Ca concentrations (Fig. 8). It appeared
differed across the range of NO3 solution concentration levels that the Ca electrodes were more sensitive to K than to Ca when
tested (Fig. 6a), showing that the two NO3 electrodes had different mixed solutions containing both ions in a concentration range of
52 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54

0 150
(a) (b)
-50 100

N ISE - EMF (mV)

K ISE- EMF (mV)

-100 50

-150 0

-200 -50

-250 -100

-300 -150
0.1 1 10 100 1000 0.1 1 10 100 1000
NO3-N concentration (mg/L) K concentration (mg/ L)

Fig. 7. Normalized responses of NO3 and K electrodes to different NO3–N (a) and K (b) concentrations, respectively.

(a) 250 (b) 250

200 200
Ca ISE- EMF (mV)

Ca ISE- EMF (mV)

150 150

100 100

50 50

0 0

-50 -50
0.1 1 10 100 1000 0.1 1 10 100 1000
Ca concentration (mg/L) K concentration (mg/L)

Fig. 8. Normalized responses of Ca electrodes to different Ca (a) and K (b) concentrations.

resulting in a linear equation relating NO3–N concentrations to

Table 3
NO3, K, and Ca calibration equations obtained with SAS multiple linear regression the EMF values with a coefficient of determination (R2) of 0.87.
analysis. Similarly, a calibration equation for the K electrode only included
the effects of the K ion (R2 = 0.98). On the other hand, since both
Ion Concentration (X1 Calibration equation R2 Std.
and X2) vs. EMF (Y) error Ca and K ions significantly affected the response of the Ca elec-
(mg/L) trode, both Ca and K effects were included in a multiple linear
NO X1 = NO3–N Y = 39.3logX1-86.5 0.87 15.5 regression equation using the SAS REG procedure. In spite of the
3
K+ X1 = K Y = 60.8logX1-62.7 0.98 6.8 use of the additional variable, a linear regression equation with a
Ca2+ X1 = Ca, X2 = K Y = 8.1logX1 + 25.5logX2 + 35.7 0.58 20.8 low coefficient of determination (R2 = 0.58) was obtained, due to
high variation in EMF values measured at the same concentrations
(Fig. 8).

3–700 mg L1 were used. These results were different from those 3.3. Application to hydroponic solution
measured in the selectivity tests for Ca ions using pure single elec-
trolyte solutions of the primary ion (0.01 M) and interference ion Fig. 9 compares NO3–N and K concentrations in hydroponic
(0.1 M), where the Ca electrodes showed a much more sensitive re- solutions determined by NO3 and K electrodes with those obtained
sponse to Ca than to K (Fig. 5b). The cause of the reduced selectivity by standard laboratory analysis, i.e., ion chromatography for ni-
for Ca in the mixed solutions is difficult to ascertain. However, trate and ICP spectrophotometer for K measurement. On average,
according to Bedlechowicz-Sliwakowska et al. (2006) and Schefer hydroponic NO3–N levels determined by the ISE electrodes were
et al. (1986), this problem might be related to not using a buffered 28% higher than those obtained with the standard analyzer. How-
inner filling solution such as 0.01 M CaCl2 + 0.05 M EDTA (Ethyl- ever, a strong relationship (R2 = 0.85) was observed between the
enediaminetetraacetic acid) adjusted to pH 8.5 and a Ca-buffered two methods. Such high NO3–N estimates might be due to the fact
solution containing a constant K concentration. Another possible that the hydroponic samples with NO3–N levels of >150 mg L1
explanation might be a relatively rapid decrease in sensitivity of were over-estimated with the NO3 electrodes. The overall standard
the Ca membrane over time as compared to membranes for NO3, error of prediction for NO3–N concentration was high at
K, and NH4 ions (Cloutier et al., 1997). More work is needed to 40.5 mg L1. A strong relationship (R2 = 0.85) existed between
identify a Ca ion-selective membrane that is not affected by other hydroponic K concentrations determined by ISE and those by the
interfering ions in mixed solutions. ICP spectrometer, but the ISE method measured an average of
The regression results relating NO3–K–Ca concentrations mea- 26% less K with a standard error of prediction of 74.1 mg L1. As
sured in the mixed solution and the EMF data of each electrode compared to results of soil extract tests for NO3 and K ions in
are shown in Table 3. These results were obtained by SAS GLM Kim et al. (2007), more variable ISE measurements were obtained.
analysis of the significance of the three ions on sensing responses In particular, relatively high standard deviations of 20–100 mg L1
of the NO3, K, and Ca electrodes. The NO3 electrode was signifi- were measured at concentrations >100 mg L1. Such reduced
cantly affected by the primary ion, NO3, whereas the effects of K repeatability might be attributed to lack of a buffer solution. The
and Ca ions on the NO3 electrode were not significant, thereby buffer solution keeps the ionic strength of the solution constant
 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54 53

(a) (b)
500 700 1:1 line

Hydroponic NO3-N concentration (mg/ L)

1:1 line

Hydroponic K concentration (mg/ L)

600
400
500

determined by K ISE
determined by N ISE
300 400

300
200
200
100
100
Y = 1.28X - 28.2 Y = 0.73X + 60.2
(R2 = 0.85, SEP=40.5) 0 (R2 = 0.85, SEP=74.1)
0

0 100 200 300 400 500 0 100 200 300 400 500 600 700
Hydroponic NO3-N concentration (mg/L) Hydroponic K concentration (mg/L)
determined by ion chromatography determined by ICP

Fig. 9. Relationships between (a) NO3–N and (b) K concentrations determined by individual NO3 and K ISEs and by standard analysis methods.

regardless of ions present in the solution, causing the ISEs to of EMF from eight ISE electrodes. A sensor array consisting of the se-
exhibit a linear response to concentration. Without a buffer, the lected ion-selective membranes, in conjunction with previously
ISEs would respond to ionic activity, requiring the use of the De- developed baseline correction and two-point normalization meth-
bye–Hückel formula to calculate the ion activity coefficient in or- ods, was evaluated for accuracy in directly analyzing NO3–N, K,
der to directly relate electrode response to the concentration. and Ca ions in hydroponic paprika nutrient solutions. The following
Thus, a buffer solution such as Tris or acetate would be necessary conclusions can be drawn from these tests:
to obtain more accurate estimates of NO3 and K concentrations
in mixed hydroponic solutions when using the ISE sensing method.  The sensitivity and selectivity of PVC membrane-based ion-
In hydroponic solution there was no significant relationship be- selective electrodes with TDDA–NPOE, valinomycin, and cal-
tween the Ca levels determined by the Ca ISEs and those obtained cium ionophore II were satisfactory for measuring NO3, K, and
with the ICP spectrometer (data not shown). Such poor prediction Ca ions in single ion solutions over typical ranges of hydroponic
capability for Ca might be because other ions, such as K, had a large concentrations. However, performance of Mg ion-selective
interference effect on the determination of Ca in mixed solution membranes based on Mg (I) and Mg (III) ionophores was not
(Fig. 8b). One approach to solve this problem might be to identify acceptable due to low sensitivity and poor selectivity.
a different type of Ca sensing element less sensitive to other ions  In mixed solutions of NO3, K, Ca, H2PO4, and Mg ions, responses
present in hydroponic solution. of the TDDA–NPOE-based NO3 and valinomycin-DOS-based K
This research showed the feasibility of using a custom designed ion-selective electrodes were significantly influenced only by
and fabricated test stand based on automatic electrode rinsing and their primary ions, providing linear calibration models with
data collection for obtaining data from an ISE array. Test stand data high coefficients of determination (R2 P 0.87). However, the
from NO3 and K membrane-based electrodes, adjusted with base- Ca ionophore II-based electrodes were significantly affected by
line correction and two-point normalization methods, provided both Ca and K ions and the calibration had a relatively low coef-
good estimates of NO3 and K in hydroponic solution. Results with ficient of determination (R2 = 0.58).
the Ca electrode were not satisfactory due to its reduced sensitivity  The NO3 and K electrodes measured NO3–N and K ions in spiked
and poor selectivity for Ca in the mixed hydroponic solution. or diluted hydroponic paprika nutrient samples independently
The approach used in this research, where a base solution is with 1.28:1 and 0.73:1 relationships between the concentra-
used for both referencing the baseline and rinsing electrodes and tions determined by ISEs and by standard laboratory instru-
two solutions are contained in buckets for the two-point normali- ments (R2 = 0.85 in both cases). Compared to results of soil
zation, could be used in an automated sensing system for hydro- extract tests for NO3 and K ions by Kim et al. (2007), more var-
ponic nutrients in greenhouses. Modification of the existing iable ISE measurements were obtained. This reduced repeat-
system for automated sample introduction, including periodic ability might be attributed to variation in the ionic strength of
sampling of hydroponic solution from the flow line of the recircu- the solution because no buffer solution was used in the current
lating system as well as the addition of temperature measurement study.
and compensation, would minimize the problems with electrode  Results with the Ca electrode were not satisfactory due to lower
drift and sensitivity changes reported for previous sensing systems sensitivity and poor selectivity for Ca in the mixed hydroponic
(Bailey et al., 1988; Cloutier et al., 1997). However, one major nutrient solution as compared to results obtained in a pure Ca
drawback to this ISE system is the lack of acceptable Ca and Mg solution. Future research should address this issue, perhaps
membranes. Future research should be directed toward identifying through identification of a new Ca membrane with acceptable
detection elements for Ca and Mg ions in hydroponic solutions. In sensitivity and selectivity for Ca in mixed solutions of other ions
addition, the use of a buffer solution should be investigated to ob- such as K, or through the use of a Ca-buffered solution that sup-
tain more accurate concentration estimates of NO3 and K. presses the activities of other ions in the mixed solution.

4. Conclusions
Acknowledgments
In this study, sensitivity and selectivity tests for various PVC-
based ion-selective membranes were conducted to identify those This research was supported in part by funding from the Coop-
suitable for measuring typical concentration ranges of macronutri- erative Research Programs of the Rural Development Administra-
ents, i.e., NO3–N, K, Ca, and Mg, in hydroponic solutions. A custom tion, Republic of Korea in 2012, and from the Korean Institute of
designed and fabricated test stand based on automatic electrode Planning and Evaluation for Technology of Food, Agriculture,
rinsing and data collection was used for simultaneous measurement Forestry and Fishery, Republic of Korea in 2011. The authors
54 H.J. Kim et al. / Computers and Electronics in Agriculture 93 (2013) 46–54
acknowledge Scott T. Drummond, Information Technology Special-
ist, USDA-ARS Cropping Systems and Water Quality Research Unit,
Columbia, Missouri, USA, for valuable technical advice in fabrica-
tion of the test stand and for comments on the manuscript.
References
Bailey, B.J., Haggett, B.G.D., Hunter, A., Albery, W.J., Svanberg, L.R., 1988. Monitoring
nutrient film solutions using ion-selective electrodes. J. Agr. Eng. Res. 40, 129–
142.
Bedlechowicz-Sliwakowska, I., Lingenfelter, P., Sokalski, T., Lewenstam, A., Maj-
Zurawsk, M., 2006. Ion-selective electrode for measuring low Ca2+
concentrations in the presence of high K, Na and Mg background. Anal.
Bioanal. Chem. 385, 1477–1482.
Carey, C.M., Riggan, W.B., 1994. Cyclic polyamine ionophores for use in a dibasic-
phosphate-selective electrode. Anal. Chem. 66, 3587–3591.
Cloutier, G.R., Dixon, M.A., Arnold, K.E., 1997. Evaluation of sensor technologies for
automated control of nutrient solutions in life support systems using higher
plants. In: Proceedings of the Sixth European Symposium on Space
Environmental Control Systems, Noordwijk, The Netherlands.
Dorneanu, S.A., Coman, V., Popescu, I.C., Fabry, P., 2005. Computer-controlled
system for ISEs automatic calibration. Sensor. Actuator. B 105, 521–531.
Erne, D., Stojanac, N., Ammann, D., Hofstetter, P., Pretsch, E., Simon, W., 1980.
Lipophilic di- and triamides as ionophores for alkaline earth metal cations. Helv.
Chim. Acta 63, 2271–2279.
Forster, R.J., Regan, F., Diamond, D., 1991. Modeling of potentiometric electrode
arrays for multicomponent analysis. Anal. Chem. 63, 876–882.
Glazier, S.A., Arnold, M.A., 1991. Selectivity of membrane electrodes based on
derivatives of dibenzyltin dichloride. Anal. Chem. 63, 754–759.
Gutierrez, M., Alegret, S., Caceres, R., Casadesus, J., Marfa, O., del Valle, M., 2007.
Application of a potentiometric electronic tongue to fertigation strategy in
greenhouse cultivation. Comput. Electron. Agr. 57, 12–22.
Hashimoto, Y., Morimoto, T., Fukuyama, T., Watake, H., Yamaguchi, S., Kikuchi, H.,
1989. Identification and control of hydroponic system using ion sensors. Acta
Hortic. 245, 490–497.
Havlin, J.L., Beaton, J.D., Tisdale, S.L., Nelson, W.L., 1999. Soil Fertility and Fertilizers:
An Introduction to Nutrient Management. Prentice Hall, Inc., NJ.
Heinen, M., Harmanny, K., 1992. Evaluation of the performance of ion-selective
electrodes in an automated NFT system. Acta Hortic. 304, 273–280.
Kim, H.J., Hummel, J.W., Birrell, S.J., 2006. Evaluation of nitrate and potassium ion-
selective membranes for soil macronutrient sensing. Trans. ASABE 49, 597–
606.
Kim, H.J., Hummel, J.W., Sudduth, K.A., Motavalli, P.P., 2007. Simultaneous analysis
of soil macronutrients using ion-selective electrodes. Soil Sci. Soc. Am. J. 71,
1867–1877.
Kim, H.J., Son, D.W., Kwon, S.G., Roh, M.Y., Kang, C.I., Jung, H.S., 2011. Determination
of inorganic phosphate in paprika hydroponic solution using a laboratory-made
automated test stand with cobalt-based electrodes. J. Biosyst. Eng. 36, 326–
333.
Maj-Zurawsk, M., Rouilly, M., Morf, W.E., Simon, W., 1989. Determination of
magnesium and calcium in water with ion-selective electrodes. Anal. Chim.
Acta 218, 47–59.
Müller, M., Rouilly, M., Rusterholz, B., Maj-Zurawska, M., Hu, Z., Simon, W., 1988.
Magnesium selective electrodes for blood serum studies and water hardness
measurement. Mikrochim. Acta III, 283–290.
Nielson, H.J., Hansen, E.H., 1976. New nitrate ion-selective electrodes based on
quaternary ammonium compounds in nonporous polymer membranes. Anal.
Chim. Acta 85, 1–16.
Savvas, D., 2002. Automated replenishment of recycled greenhouse effluents with
individual nutrients in hydroponics by means of two alternative models.
Biosyst. Eng. 83, 225–236.
Schefer, U., Ammann, D., Pretsch, E., Oesch, U., Simon, W., 1986. Neutral carrier
based calcium(2+)-selective electrode with detection limit in the sub-
nanomolar range. Anal. Chem. 58, 2282–2285.
Shen, H.J., Cardwell, T.W., Cattrall, R., 1998. The application of a chemical sensor
array detector in ion chromatography for the determination of Na+, NH4+, K+,
Mg2+ and Ca2+ in water samples. Analyst 123, 2181–2184.
Sonneveld, C., 2002. Composition of nutrient solutions. In: Savvas, D., Passam, H.C.
(Eds.), Hydroponic Production of Vegetables and Ornamentals. Embryo
publications, Athens, Greece, pp. 179–210.
Van Lierop, W., 1986. Soil nitrate determination using the Kelowna multiple
element extractant. Commun. Soil Sci. Plant Anal. 17, 1311–1329.
Xiao, D., Yuan, H.Y., Li, J., Yu, R.Q., 1995. Surface-modified cobalt-based sensor as a
phosphate-sensitive electrode. Anal. Chem. 67, 288–291.