Journal of Oceanography

Vol. 51, pp. 375 to 383. 1995

An Automated Precise Winkler Titration for Determining

Dissolved Oxygen on Board Ship
KEN FURUYA and KAZUTOSHI HARADA*
Biological Oceanography Laboratory, Faculty of Bioresources, Miè University,
1515, Kamihama, Tsu 514, Japan

(Received 6 June 1994; in revised form 3 February 1995; accepted 6 February 1995)

An automated titration for the Winkler method is presented for measuring

primary productivity in the ocean. The system is based on a microcomputer-
controlled titration with potentiometric endpoint detection. By the use of 0.005-N
sodium thiosulfate as a titrant and a program designed to shorten the time for
measuring, the method achieves a precision of 0.04% coefficient of variation with
a range of 0.01 to 0.10% for six replicates of samples at oxygen concentration of 70
to 250 µM. It takes about four to five minutes to measure one sample.

1. Introduction
Dissolved oxygen is central in studies on biological productivity and biochemical processes
in the aquatic environments. Various methods are used for determining oxygen concentration.
Because of the highest accuracy and precision, the iodometric Winkler method (Winkler, 1888)
and its modifications have been widely used. Since the classical Winkler titration imposes a
difficulty in visual determination of the endpoint of titration by means of the changes in the color
of the starch-iodine complex, there have been attempts to apply instrumental detection of the
endpoint to increase the precision of the method. Furthermore, automated titration techniques for
the Winkler method have been developed in the last decade. Williams and Jenkinson (1982)
proposed a system based on a photometric endpoint detector. Their system was applied
successfully for measuring plankton primary productivity and community respiration in the
oligotrophic ocean (Williams et al., 1983). Since the photometric endpoint detection can be
affected by particles in the sample solution and bubbles generated during the measurement,
Culberson and Huang (1987) proposed a system by using an amperometric endpoint detector.
They reported that the precision of the method was 0.3 µM. Considering magnitudes of daily
oxygen fluxes of plankton metabolism is in the order of 1 µM in the oligotrophic oceans, a better
precision is required for productivity measurements.
Compared to the photometric and amperometric endpoint detection, little attention has been
paid to the potentiometric detection for use in automated titration probably because the
electrochemical equilibrium at the platinum indicator electrode is considered to be established
slowly (Grasshoff, 1981) and it is difficult to detect iodine at low concentration of 10–6 N (Potter,
1957). However, since potentiometry is simple, convenient and generally precise, it seems worth
while exploiting potentiality of the potentiometric endpoint detection for shipboard analysis.
We have developed an automated titrator using a potentiometric endpoint detection with a
redox potential electrode. The system substantially consists of commercially available components
which are controlled by an improved computer program which facilitates a short measurement

*Present address: Techno-Chubu Co. Ltd., 3-12, Oh-è, Minato-ku, Nagoya 455, Japan.
376 K. Furuya and K. Harada

time. Our primary concern is the use for plankton productivity on board ship. Using the
automated system, we attain a high precision enough for measurements of plankton productivity.

2. Materials and Methods

2.1 Apparatus
The automated system is composed of five major components (Fig. 1): (1) the oxidation-
reduction potential electrode (Toa, PTS-5011C) as an endpoint detector, connected with a
pH/mV meter (Toa, HM-18ET or HM-70V), (2) a DC signal isolator (Watanabe, DCE 21R-1)
which amplifies the electrode output, (3) a 12-bit analog to digital convertor (Canopus, Analog
Pro 1) to interface the isolator with a microcomputer, (4) a microcomputer (NEC, PC9801) to
control the titration, and (5) a piston burette (Metrohm, Dosimat 665) to feed sodium thiosulfate
as titrant and to read volume of the titrant dispensed. The burette is fitted with a non-diffusive
burette tip (Metrohm) which eliminates diffusion of thiosulfate into specimens. Although a
combination of an Ag/AgCl reference and a platinum indicator electrode is also satisfactory in
terms of responses, we prefer the composite electrode because of easy handling of the electrode
and narrow neck of BOD bottles in which a whole titration is processed.
The Dosimat 665 is fitted with a 50-mL cylinder (Metrohm, E552), with which a minimum
amount of titrant in a single stepwise feeding is 5 µL. Since we use lower concentration of sodium
thiosulfate than 0.01 N as used in the usual analysis, we need the large capacity burette cylinder
to enable titration of one sample without refill that is a source of errors. The burette was chosen
because the Dosimat provides the largest cylinder capacity among commercially available
products and offers remote-controllable operation with the computer. The control is made by
serial transmission of command and volume data on titrant via an interface according to RS 232
C in half duplex procedure at baud rate of 600. The whole titration process is controlled by a
program written in BASIC (Furuya and Harada, in preparation).

2.2 Procedure
The titration protocol substantially follows the manual titration. There are three stages in the
titration protocol, each called initial, absolute and differential modes in sequence (Fig. 2) after

Fig. 1. Diagram of an automated potentiometric titrator.

 An Automated Precise Winkler Titration for Determining Dissolved Oxygen on Board Ship 377

Williams and Jenkinson (1982). First, in the initial mode, approximately 50 to 60% of a rough
estimate of the final volume of titrant is dispensed continuously. Once information on sample and
BOD bottle number has been entered via keyboard, the program begins. The program requests
input of the rough estimate. With the input of the estimate the titrant is fed at a rate of 2.5
mL s–1. The rough estimate is arbitrarily decided according to nature of samples. When the
oxygen content of a sample is difficult to make the rough estimate, a conservative value is
preferable to avoid over-feeding of the titrant. In this mode, the redox potential does not change
much compared to the later modes. For surface water of the Kuroshio, the redox potential of the
acidified fixed sample is around 720 mV, which decreased to around 685 mV during the initial
mode.
Next, in the absolute mode, the titrant is dispensed by a series of pulse addition. With a pulse

Fig. 2. Schematic change in electrode output during the titration. A series of vertical bars indicate
successive addition of thiosulfate. Titration protocol changes from the absolute mode to differential
mode when the output reaches a predetermined value, k1 . The largest shift of the output during the
differential mode is taken as an endpoint. An inset gives an example of a time course of the electrode
output in an actual sample after an addition of thiosulfate. Using 576 successive electrode readings
taken from two seconds after a pulse addition through the serial port, the microcomputer estimates a
plateau value k by the curve fitting to a formula shown here. Then next feeding is prompted at the time
indicated by an arrow in the inset.
378 K. Furuya and K. Harada

addition, the electrode output increases stepwise. At each pulse, the electrode output is referred
to a predetermined value. Once the output reaches the value, the procedure is repeated with a
renewed set of a smaller constant volume of titrant and a smaller electrode-output reference value
than those in the preceded stage. All these sets of titrant volume and reference value are
predetermined in the laboratory, stored in the memory and called up. For the Kuroshio and
subtropical open waters, we use six pulse volumes; 1, 0.5, 0.1, 0.05, 0.025 and 0.01 mL, each
corresponding to the electrode-output reference values of 655, 631, 609, 599, 583, and 565 mV,
respectively. Throughout this mode, each pulse feeding of the titrant is made before the electrode
output reaches a plateau. The plateau is estimated from time course of the electrode output using
an exponential function (Fig. 2):

y = k − a ⋅ bx (1)

where x is time after a single feeding, y is the electrode output, k is estimated plateau value, a and
b are empirical constants that vary according to titration setup, that is, volume of a BOD bottle,
rotation rate of a magnetic stirrer within the BOD bottle, volume of added titrant and relative
positions of the tip of the burette and the electrode. These constants a and b are predetermined
in the preliminary experiments and stored in the memory. Different sets of a and b are needed,
each corresponding to each pulse volume of titrant listed above. The program estimates k by the
curve-fitting to the above formula using 576 successive readings taken from two seconds after
a pulse feeding of titrant. The period of two seconds is for mixing time of the added titrant and
the sample solution. Using the calculated k, next feeding is prompted without the output reaching
a plateau to shorten the titration time.
The absolute mode is continued until an electrode output reaches the provisional value, for
which we use 565 mV. Then the rest of the titration is made in the differential mode. During the
mode a 5-µL titrant is dispensed every six seconds. The largest increase of electrode output on
a single addition is taken as an endpoint. The titration continues for five additional feedings of
the titrant after reaching the endpoint to ensure the endpoint, which is found around 535 mV.

2.3 Reagents and fixation of oxygen

Preparation of chemical reagents, fixation of oxygen, standardization of working solution
of sodium thiosulfate, and acidification of the sample were made following Carpenter (1965b).
Reagent blanks were corrected also after Carpenter (1965b). The CSK standard solution of KIO3
was used for the standardization of sodium thiosulfate solution. Precise dispensers with
repeatability of <0.1% coefficient of variation (CV)were used for the addition of alkaline iodide
and manganese (II) chloride into the 200–220-mL BOD bottles. The titration was performed
within the BOD bottles without transferring the whole sample to a titration vessel, as the transfer
is a major source of errors due to volatilization of iodide (Carpenter, 1965a).
Volumetry of BOD bottles was made at 20°C down to two decimal places in mL after
Culberson (1991). This procedure was repeated at least five times, and mean bottle volume was
obtained.

2.4 Comparison of visual starch and potentiometric endpoint indications

The sensitivities of endpoint indication by starch and the potentiometric methods were
compared using a pottasium iodate solution and a seawater sample. Two 10.00-mL portions of
0.0100-N potassium iodate (CSK standard solution, Wako Chemicals) were taken with a
 An Automated Precise Winkler Titration for Determining Dissolved Oxygen on Board Ship 379

calibrated automatic dispenser (ADT-1, Toa Electronics) and to each were added 0.3 g of
potassium iodide and diluted in 90-mL portions of distilled water with 10-N sulfuric acid to adjust
pH of the solution to be 2.0. For the starch endpoint detection 0.5 mL of 1% (w/v) starch solution
was added to one portion. The seawater sample taken from the surface of the Kuroshio was
introduced into 200- to 220-mL BOD bottles and filled after overflow by at least one bottle
volume and the dissolved oxygen was fixed. The potassium iodate solution and the seawater were
titrated with 0.004967-N sodium thiosulfate, and added amount of the titrant was compared.

2.5 Seawater samples and analysis

Field experiments were conducted during the three cruises of T/S Seisui Maru in the
Philippine Sea in summer and a K93-03 cruise of R/V Kaiyo in the East China Sea in October
1993. Oxygen content was also analyzed ashore laboratory for seawater taken from Ise Bay and
mesocosms settled over bottom sediments in an eutrophicated embayment in Gokasho Bay in
July 1993. Seawater samples were taken into the BOD bottles and fixed.
The oxygen content in a part of East China Sea samples and all samples of Gokasho Bay was
determined manually without the computer control by a practiced operator using the same
apparatus and protocol described above. In the absolute mode, successive pulse addition of the
titrant was made before the output reached plateaus, based on the electrode output on a digital
chart recorder. This is referred as the potentiometric manual titration here.

3. Results and Discussion

3.1 Effect of the titrant concentration

Reproducibility of the oxygen determination depended on the concentration of sodium
thiosulfate (Fig. 3). In our system, 0.005 to 0.0067 N of sodium thiosulfate gave lowest CV.
Based on this low variance, we used 0.005 to 0.0067 N of sodium thiosulfate in the present study.
In the differential mode, the endpoint was determined by a series of a 5-µL addition of titrant.
With 0.005-N thiosulfate, each addition contains 0.025 µ normality of thiosulfate that is
equivalent to 0.00625 µmole O2. Since we used about 200-mL BOD bottles, 0.03 µM (0.00625
multiplied by 5) is considered to be lower limit of the precision, provided that other sources of
errors are not taken into account.
The high concentration over 0.01 N yielded significantly higher CV (F-test, p < 0.05). This
suggests that the concentration of the titrant is so high that the amount of the titrant fed in a single
addition is too large to resolve endpoint well. This probably reduces the reproducibility. There
is also another advantage in the use of a diluted titrant. We checked leakage of the titrant from
the end of the tip into the sample by the use of Rhodamin B added in the titrant. Even without
burette dosage, fluorescence was detectable in the sample as determined by fluorometry,
indicating occurrence of the leakage in spite of the use of the anti-diffusion burette tip. A normal
burette tip without the membrane resulted in a more intense fluorescence, suggesting more
leakage than the anti-diffusion tip. Thus, low concentration of the titrant can reduce the erroneous
effect of the leakage, especially in the automated titration with pulsing addition of the titrant.

3.2 Comparison of visual starch and potentiometric endpoint indications

The starch endpoint occurred before the potentiometric endpoint (Table 1). The difference
in the endpoint between the two methods was significant (t-test at p < 0.005). Similar com-
parisons were repeated five times using surface waters of the Philippine Sea, and the endpoint
380 K. Furuya and K. Harada

Fig. 3. Relationship between concentration of sodium thiosulfate and dissolved oxygen determination.
Error bars denote one standard deviation for six replicates.

Table 1. Comparison of starch and potentiometric endpoints. 10 mL of 0.01-N KIO3 and surface water
of Kuroshio are titrated with 0.004967-N sodium thiosulfate. For the seawater titrant volume was
normalized for 100-mL seawater. n: number of replicates.

Sample Starch endpoint Potentiometric endpoint

mean added titrant CV mean added titrant CV

(mL) (%) (mL) (%)

KIO3 (n = 5) 19.900* 1 0.2531* 2 20.134* 1 0.0111* 2

Seawater (n = 5) 16.406* 3 0.2381* 4 16.522* 3 0.0373* 4

*1, *2 , *3 , *4 : significant at p < 0.005.

bias occurred consistently; more feeding of titrant was necessary in the potentiometric titration
than in the starch method by 0.7 to 0.9% of titrant volume of the latter method. This difference
in the endpoints is in agreement with the fact that the amperometric endpoint detection requires
more feeding of titrant than the starch method (Bradbury and Hambly, 1952; Knowles and
Lowden, 1953; Culberson and Huang, 1987; Culberson et al., 1991). Bradbury and Hambly
(1952) concluded that the visual endpoint with starch is significantly different from the
equivalence points detected by amperometry. However, the difference in the endpoint among the
detection methods does not necessarily yield difference in oxygen concentration as long as the
titrant is calibrated by the same detection method as used for samples. In the comparison shown
in Table 1, we found no significant difference in oxygen concentration between the two methods.
The precision of the potentiometric endpoint detection in terms of CV was significantly
higher than the starch method (Table 1) (F test at p < 0.005). Since precision is the primary
 An Automated Precise Winkler Titration for Determining Dissolved Oxygen on Board Ship 381

requirement for biological production studies, the obtained CV for the surface seawater is
satisfactorily small for our purpose.

3.3 Seawater analysis

Overall the titration procedure achieved a standard deviation of 0.026 to 0.191 µM (mean
0.096 µM) in measurements of dissolved oxygen concentration ranging from 70.91 to 250.01 µM
(Fig. 4). The potentiometric manual titration yielded the standard deviation ranging from 0.040
to 0.161 µM (mean 0.103 µM). The automated titration attained the same precision as the
potentiometric manual titration in the range 0.026 to 0.191 µM (mean 0.087 µM). The standard
deviation of 0.1 µM of oxygen indicates that the automated method is suitable for measuring
plankton primary production and community respiration in the open ocean, where daily oxygen
change is generally in the order of one to several µmole per liter. The precision of the method
expressed as CV is in the range of 0.01 to 0.10% with a mean of 0.049%. This precision is
equivalent to or a little better than that of the automated photometric titration whose CV is in the
range of 0.03–0.1% (Williams and Jenkinson, 1982).
Rough weather tended to produce high standard deviation on board ship analysis: the high
standard deviation above 0.14 µM in the Philippine Sea and the East China Sea were obtained
under rough sea conditions due to a low pressure (Fig. 4). High standard deviation was also
observed during the mesocosm experiment, where one batch was intensely aerated to examine
degradation of organic materials (Yamaguchi and Washio, unpublished). The aerated water
showed significantly higher CV (F-test at p < 0.01) in the oxygen determination than non-aerated

Fig. 4. Dissolved oxygen concentration and its standard deviation of seawater samples analyzed on
shipboard (open and closed circle: East China Sea, closed triangle: Philippine Sea) and in land
laboratory (open square: Ise Bay, cross: Gokasho Bay). The analysis was made for six replicate
measurements in the East China Sea, Ise Bay and the Philippine Sea, and five replicates in Gokasho
Bay. Samples of Gokasho Bay and a part of samples from the East China Sea (open circle) were
analyzed manually without computer control.
382 K. Furuya and K. Harada

Table 2. Comparison of variations in oxygen contents of aerated and non-aerated waters during a
mesocosm experiment.

Sample Mean oxygen contents Average of standard Average of coefficient

(µM) deviation (µM) of variation (%)

Aerated water (n = 6) 241.27* 1 0.214* 2 0.089* 3

Non-aerated water (n = 12) 154.41* 1 0.090* 2 0.060* 3

*1, *2 , *3 : significant at p < 0.01.

water in the control mesocosm (Table 2). Although the high standard deviation during the rough
sea conditions is not ascribed to either errors in the titration procedure, or sampling errors due
to bad sea conditions, the observation made during the mesocosm experiment suggests the high
standard deviation is a manifestation of variance of oxygen content itself among the BOD bottles.
We applied the potentiometric endpoint detection in very turbid waters near the mouth of
Changjiang River, where the extinction coefficient of the photosynthetically available radiation
exceeded 0.65 m–1 in February 1993. While samples were rich in small particles, in our
measurements we observed high precision of mean CV of 0.079%, although turbid waters can
produce low reproducibility in the precise photometry and in amperometry (Culberson and
Huang, 1987). Thus, applicability to wide range of waters and simplicity are major advantages
of the potentiometric titration. The improvement of our titration protocol in estimating k in the
absolute mode shortens the measurement time. With this procedure, analysis time per bottle is
about four to five minutes. A well-experienced operator can attain a shorter measurement time
using the same titration protocol without the computer control. For a more rapid titration,
maximization of magnetic stirrer speed control and rates of titrant feeding seems to be important.

Acknowledgements
We are grateful to Dr. E. Nakayama for his criticisms on the earlier version of the manuscript.
We thank Dr. T. Ishimaru for providing the A/D board, Messrs. H. Yamaguchi and Y. Washio
for offering facilities during the mesocosm experiment, and scientific members and crew on
board T/S Seisui Maru and R/V Kaiyo for their cooperation at sea. Comments of the aonymous
reviewers were very helpful. This work was supported by Ministry of Education, Science and
Culture, and Science and Technology Agency of Japan.

References
Bradbury, J. H. and A. N. Hambly (1952): An investigation of errors in the amperometric and starch indicator
methods for the titration of millinormal solutions of iodine and thiosulfate. Austral. J. Sci. Res., Ser. A, 5, 541–
554.
Carpenter, J. H. (1965a): The accuracy of the Winkler method for dissolved oxygen analysis. Limnol. Oceanogr.,
10, 135–143.
Carpenter, J. H. (1965b): The Chesapeake Bay Institute technique for the Winkler dissolved oxygen method. Limnol.
Oceanogr., 10, 141–143.
Culberson, C. H. (1991): Dissolved Oxygen. WHP Operations and Methods, Unpublished manuscript, 15 pp.
Culberson, C. H. and S. Huang (1987): Automated amperometric oxygen titration. Deep-Sea Res., 34, 875–880.
Culberson, C. H., G. Knapp, M. C. Stalcup, R. T. Williams and F. Zemlyak (1991): A comparison of methods for
the determination of dissolved oxygen in seawater. WOCE Report 73/91, 77 pp.
Grasshoff, K. (1981): The electrochemical determination of oxygen. p. 329–420. In Marine Electrochemistry, ed.
 An Automated Precise Winkler Titration for Determining Dissolved Oxygen on Board Ship 383

by M. Whitfield and D. Jagner, John Wiley and Sons, New York.

Knowles, G. and G. F. Lowden (1953): Methods for detecting the end-point in the titration of iodine with thiosulfate.
Analyst, 78, 159–164.
Potter, E. C. (1957): Microdetermination of dissolved oxygen in water. I. Nature of the problem. J. Appl. Chem., 7,
285–297.
Williams, P. J. LeB, and N. W. Jenkinson (1982): A transportable microprocessor-controlled precise Winkler
titration suitable for field station and shipboard use. Limnol. Oceanogr., 27, 576–584.
Williams, P. J. LeB., K. R. Heinemann, J. Marra and D. A. Purdie (1983): Comparison of 14 C and O2 measurements
of phytoplankton production in oligotrophic waters. Nature, 305, 49–50.
Winkler, L. W. (1888): Die Bestimmung des im Wasser gelösten Sauerstoffen. Ber. Dtsche. Chem. Ges., 21, 2843–
2855.