Summary

A new Ammonium Analyser is described which uses well proven Ion-Selective Electrode technology coupled with a unique and highly sophisticated computer-based method of data acquisition and processing to measure directly the concentration of ammonium ions in dilute aqueous solutions. An artificial neural network software is used to interrogate an extensive calibration data-base and compensate for any interference effects from potassium ions, which is inherent in any ISE measurement of ammonium. The true concentrations of both ions are displayed as ppm (mg/l). The system is simple and rapid to use and only needs calibrating with a single measurement of a mixed standard before each analytical session. Replicate analysis of standard solutions indicate a precision and accuracy of better than ±10% (S.D.) over most of the working range (0.1 to 100 ppm NH₄⁺ ).

Contents

1. Apparatus Used
2. Basic Principles
3. Interference Effects
4. Difficulties in ISE Measurements
5. Measuring Procedures.
6. Difficulties in Constructing the Calibration Database
7. Graphical Display of Database to Permit Visual Interpolation.
8. Re-calibration.
9. Concentration Ranges and Detection Limits
10. Results on Test Solutions
11. Conclusions
12. Acknowledgements.

1. Apparatus Used

All apparatus used in this study were provided by Nico2000 Ltd, London, and are marketed under the brand name ELIT. The Ion-Selective Electrodes are an all-solid-state, 8mm diameter, plastic-bodied design which plug in to a special multiple electrode head. The electrode potentials were measured by connecting the head with low noise cables to a unique computer interface which amplifies the signals and converts them to a digital format for direct reading by a conventional desk top PC, via a parallel printer port. Specialized software is then used to control the data acquisition, and process, record and display the data and results, and any analytical details required for archiving.

2. Basic Principles

This system is based on the use of Ion-Selective Electrodes (ISEs) to measure the concentration of ammonium (NH₄⁺) ions in dilute aqueous solutions such as lakes, rivers or drinking water, or other solutions with a total Ionic Strength of less than 0.01 M. The basic principle of ISEs is that, on immersing the electrode in a test solution, the selected ions in solution pass through the ion-selective membrane until an equilibrium is reached between the concentration inside and outside the membrane. This causes an electrical potential to be developed which is proportional to the concentration of the ions in the external solution. This is measured in millivolts (mV) as the difference in potential between the ISE membrane and a stable constant reference voltage produced by a reference electrode which is also immersed in the same solution. In this study, a lithium acetate double junction reference was chosen because it has an equi-transferent filling solution (in order to minimize drift in the liquid junction potential) and does not cause interference with either of the detected ions.

3. Interference Effects

Unfortunately, most ion-selective membranes are not entirely specific for the designated ion but also allow the passage of some other ions to a greater or lesser extent. This increases the charge on the membrane above that which is due solely to the detected ion and causes a spuriously high measurement. This phenomenon is known as Interference and the extent of this interference is expressed as the Selectivity Coefficient. The selectivity coefficient is the ratio of the apparent concentration calculated for the interfering ion to that calculated for an equal amount of the detected ion. NB: this is not the same as the ratio of the mV signals developed by the electrodes in response to these ions, because the voltage is only directly proportional to the logarithm of the concentration. The selectivity coefficient can be determined experimentally to give some idea of the order of magnitude but it is not a constant factor since it depends on several parameters including the concentration of both ions, the presence of other ions in solution, and the temperature. In the past this phenomenon has limited the use of ion-selective electrodes to only those applications where the interfering ion is known to be absent or only in a relatively small concentration compared to the detected ion.

In the case of ammonium analysis, the most significant interference is from the potassium (K⁺) ion. Potassium is a very common element and occurs in relatively high concentrations in most natural waters. The selectivity coefficient for the NH₄ electrode for K is about 0.1. This means that if there are equal concentrations of both elements then the K interference will increase the NH₄ measurement by about 10%. To put it another way, if the K content is ten times more than NH₄ then the measured concentration will be about twice what it should be. A further complication is that the K electrode is also interfered with by the NH₄ ion. However, in this case the selectivity coefficient is only about 0.01 so this has a much smaller effect on the analysis.

The ELIT Ammonium Analyser employs a unique solution to the problem of ionic interference by simultaneously measuring the signals from an ammonium and a potassium ISE and interrogating an extensive calibration data base using artificial neural network software to derive the true concentration of both ions.

4. Difficulties in ISE Measurements

Apart from the problem of ionic interference, the main problems with ISE measurements are variable stabilization times, changes in the measured potential during repeated analysis, and the effect of variable ionic strength of the solutions.

A. Effect of Ionic Strength - Activity versus Concentration.

This problem is due to the fact that ISEs only record the activity, or effective concentration of the ions in solution. At low ionic strength the concentration and activity are essentially the same, but as the concentration increases, or if other ions are present, then the number of ions reaching and interacting with the membrane is reduced by inter-ionic interactions in the bulk of the solution. The standard calibration graph for ISE measurement relies on a linear relationship between the potential difference and the logarithm of the activity of the ion in question. In conventional analysis, if the samples are expected to have a high ionic strength (greater than about 0.01Molar), then an ISAB (Ionic Strength Adjustment Buffer) is added to all standards and samples to increase the ionic strength to the same high level, and hence generate a uniform difference between activity and concentration. In this case the potential reading can then be directly related to concentration. The choice of ISAB is critical and varies depending on the ion being analysed and the reference electrode being used. In the case of the ELIT Ammonium Analyser, the calibration does not rely on linear relationships and it is anticipated that the bulk of the samples will have low ionic strength and so the effect of variable ionic strength can be ignored and ISAB is not necessary.

B. Potential Drift

The effect of potential drift can easily be seen if a series of standard solutions are repeatedly measured over a period of time. The results show that the difference between the voltages measured in the different solutions (i.e. the electrode slope) remains the same but the actual value of the mV generally drifts downwards.

This is due to four main effects.

a) Constantly removing and replacing an ISE in different solutions will produce hysteresis or electrode memory effects, the extent of which depends on the relative concentrations of the new and old solutions. Thus if the same solution is re-measured after measuring a different one it will retain a memory of the previous solution and cannot be expected to give exactly the same mV reading the second time. During this study it was found that this effect could be minimized by soaking the electrodes in de-ionized water for a few seconds before each measurement.

b) With prolonged use, the ISE membrane can become hydrated and/or oxidized and this will cause a progressive change in its response to an external solution.

c) Reference electrodes have stable voltages over short periods of time but tend to suffer from slow drift in the liquid junction potential when immersed for long periods. Moreover, it is unlikely that the liquid junction potential will always settle to exactly the same value whenever a reference electrode is immersed in a new solution.

d) The electrode potential is temperature dependent. Thus it is generally recommended that the temperature of the sample and the calibrating solution should not be allowed to differ by more than 2°C.

These last three problems can be minimized by frequent re-calibration.

C. Stabilization Times

After immersing the electrodes in a new solution, the mV reading normally falls rapidly at first by several mV, and then gradually, and increasingly slowly, falls to a stable reading as the ISE membrane equilibrates and the reference electrode liquid junction potential stabilizes. This equilibration may take up to 3 or 4 minutes to reach a completely stable value. Sometimes the reading begins to rise again after a short period of stability and it is important to ensure that the recording is made at the lowest point, before this rise has proceeded to any great extent. In this study it was found that it was not necessary to wait for a completely stable reading but that satisfactory results could be obtained by taking a reading after a pre-set time, so that each measurement was made at the same point of the decay curve. For optimum performance it was found that this delay time should be at least two minutes to ensure that the reading was in the shallower part of the curve.

5. Measuring Procedures.

When measuring samples and standards, it is important to attempt to maintain a uniform effect from the drift and hysteresis factors described above. Thus all samples and standard solutions should always be measured using exactly the same procedure.

That adopted in this study was as follows:

a) Before each new solution, the electrodes were rinsed with a jet of de-ionized water and gently dabbed dry with a low-lint tissue.

b) The electrodes were immersed in pure de-ionized water for about 30 seconds to ensure that the membranes were always in the same state at the beginning of each measurement. After immersion, they were again dabbed dry to avoid any dilution of the next solution.

c) Between 50 and 100 mls of solution were poured into a 100 or 150 ml plastic beaker and the three electrodes were immersed in the solution to a depth of about 1 to 2 cm.

d) The solution was swirled manually for a few seconds to ensure homogeneity and good contact between the solution and the membranes; i.e. no air bubbles.

e) The solution was then left to stand and a reading was taken exactly two minutes after ceasing swirling.

6. Difficulties in Constructing the Calibration Database

The calibration data required are the measured mV for NH4 and K from a series of standard solutions containing known concentrations of both ions. The main problem with this method is that a large number of calibration points are required to permit an accurate interpolation of unknown samples. This could be exceedingly time consuming and would not be a viable proposition if it were necessary to construct a separate calibration database for every set of electrodes. Fortunately, early tests demonstrated that the relative response of the ELIT electrodes to the NH4 and K ions is very uniform and reproducible and thus the shape of the calibration graph is the same for all. Thus it was worth investing considerable time and effort in building up a large database. In practice, however this is a difficult task because of the ever-present problem of potential drift.

The method adopted here was to gradually build up the database by measuring small batches of standards, generally ten or twelve at a time. For each batch, a common "normalizing" standard (N) containing 5 ppm NH4 and 10 ppm K was always measured first, several times to get a good stable reading and allow the electrodes to become fully wetted and operational. Then the calibration standards were measured in the order 1, 2, 3,...10, N, 10,...3, 2, 1, N. The average of these values was then taken in order to compensate for any drift during the measurement period. The next batch of standards were then made up and measured in the same way, usually on a different day and possibly at a different temperature. It was then possible to compensate for any periodic variation in the electrode responses by using the difference in the average value of N to normalize the new batch to the old so that they would all be compatible. The validity of this normalization was confirmed by periodically re-analysing other solutions from previous runs. Having created a coarse calibration with measured standards, it was then necessary to interpolate this in order to provide intermediate data to fill in the gaps.

7. Graphical Display of Database to Permit Visual Interpolation.

Conventional ISE calibration graphs are constructed by plotting the log of the concentration on the X-axis against mV on Y. In the case of NH4/K interference corrections, it is necessary to have two graphs, each with a series of curves representing different concentrations of the other ion. Thus one curve can be drawn showing NH4 ranging from 0.1 to 100 ppm, with all the samples containing 1 ppm K, another with the same range of NH4 but all containing 5 ppm K and so on (Fig.1).

Similarly for the K calibration, but in this case there is less spread and more of the curves are essentially linear and overlapping because the NH4 content only has a significant effect at high NH4/K ratios. These graphs were then used firstly to smooth the experimental data and remove any obviously spurious results, and secondly to find the mV values for intermediate concentrations, at every point where the Log scale crossed the appropriate calibration curve. In this way, the calibration data base was expanded considerably.

Unfortunately these graphs are not very satisfactory for finding the concentration of unknown samples. When an unknown sample is measured, the K concentration must be read from the K calibration using the measured mV from the K electrode. Then NH4 mV must be traced across the NH4 calibration graph until it meets the appropriate curve representing the known K content and thus the NH4 concentration can be determined by dropping a perpendicular to the X-axis. This method proved to be very tedious and time consuming and rather inaccurate, particularly at high NH4 to K ratios. It was found that a far more satisfactory interpolation could be made if the calibration data were plotted on a linear graph of mV NH4 versus mV K and then contoured for concentration (Fig.2). Nevertheless, this method was still rather inaccurate, subjective and prone to operator error and thus the artificial neural network software was developed in order to make a more objective and reproducible interpolation of the calibration data base.

8. Re-calibration.

Before use, the database must be re-calibrated to compensate for any differences between the responses of the electrodes used for constructing the database and the current operating set. Also for any periodic changes in electrode response due to ageing, hydration of the membranes or drift in the liquid junction potential of the reference electrode, and for any temperature differences.

After obtaining a display of the database, one point must be selected for re-calibration. If the approximate range of NH4 and K concentrations of the samples to be analysed is known then the re-calibration point should be selected nearest to the middle of this range. Otherwise, any convenient point from somewhere in the middle of the data can be chosen; e.g. 5 ppm NH4 and 10 ppm K. The calibrating solution must be made fresh by the analyst, preferably on a daily basis, to prevent errors due to deterioration of dilute solutions. This solution is then measured and all succeeding sample measurements are normalized to compensate for the difference in the original and new values for the mV of the standard. The normalized mV data can then be compared with the database for interpolation of the ppm values. Once the calibrating solution has been made, the re-calibration procedure need only take two or three minutes, depending on the stabilization time for the electrodes.

For optimum accuracy and precision, it is recommended that the re-calibration procedure is carried out relatively frequently, particularly if there are any significant temperature changes during analysis. Ideally, the temperature of the calibrating solution and any sample solutions should not differ by more than about ±1°C. For the most precise results it may even be beneficial to recalibrate between every sample measurement.

9. Concentration Ranges and Detection Limits

The calibration data ranges from 0.1 to 100 ppm NH4 and 1 to 100 ppm K. However, the reliability of the interference correction depends on the ratio of the two ions in the solution and on which particular area of the calibration is being used. Thus if the K content is less than 10 ppm then reliable NH4 data can be obtained down to about 0.1 ppm. As the K content increases, however, the lower detection limit for NH4 increases so that at around 30 ppm K it may be difficult to distinguish between 0.5 and 1 ppm NH4. At around 70 ppm K it becomes difficult to distinguish between 2 and 3 ppm NH4 and at 100 ppm K only results above about 3 ppm NH4 can be considered reliable.

Samples which lie outside the range of the calibration database cannot be analysed accurately with this system. Nevertheless, if the voltage is lower than the calibration range then it can be confidently predicted that the concentration is less than 0.1ppm NH4 or less than 1 ppm K. If the voltage is above the range for K then this indicates that the K concentration is greater than 100 ppm, but if the NH4 is over-range then more caution must be used in interpreting this result because it will be partially dependent on the K content.

10. Results on Test Solutions

In order to test the reproducibility of the analysis, five pure test solutions were measured five times each. These solutions were made by diluting and mixing appropriate amounts of two bulk standard solutions: 1,000 ppm NH4 as NH4Cl and 10,000 ppm K as KCl. Note that in all ISE work, concentrations are expressed as the concentration of the ion in solution, not the parent molecule. Before each set of measurements the system was re-calibrated using a standard with 5 ppm NH4 and 10 ppm K.

In order to further minimize any drift effects, the solutions were measured in the following order:

re-calibration, A,B,C,D,E, re-calibration, E,D,C,B,A, re-calibration, A,…etc.

The results in Table 1 show a standard deviation on the mean of better than ±10% in all cases and an accuracy (deviation from true - %Error in table) of less than 10% in all but the lowest concentration.

Table 1. Reproducibility and accuracy of measurements of test solutions (ppm).

11. Conclusions

The ELIT Ammonium Analyser provides a quick, easy and inexpensive method for determining the concentration of ammonium ions in low-ionic-strength aqueous solutions and can give a precision and accuracy of better than ± 10% over a concentration range from 0.5 to 100 ppm NH4. It is now ready for field trials and comparison with other methods for measuring NH4 in natural waters and other aqueous solutions such as fruit juices and brewing liquors, to evaluate the effect of complex sample matrices and biological contamination on these analyses.

12. Acknowledgements.

Several people have been involved in bringing this project to a successful conclusion.

Heinz Kreuzberg was the main mastermind and driving force. This project would not have been possible without his vision and dedication in developing the computer interface and data processing software. Dr. Alexander Kapustin developed the solid state ISEs and gave CCR basic training in the theory and practice of ISE measurements.

Special mention must be made of the inspired and dedicated work of Yiying Cui and Feng Xiao in developing and writing the computer programs, which were essential to permit any further advances in the techniques and data processing of ISE measurements. Their patience and understanding in the face of many months of bombardment with interminable requests for changing and re-writing the software cannot be praised too highly. Neil Lawrence and Dr Niranjan (formerly of University of Cambridge) are thanked for their invaluable help in the initial development of the neural network for ISE data processing.

Chris C Rundle, BSc, PhD. Nico2000 Ltd. London, UK. July 2000.