Uncertainty assessment of environmental tritium measurements in water

Abstract

The method most widely used nowadays to measure environmental tritium levels in water is electrolytic enrichment followed by liquid scintillation spectrometry. Although these techniques have been in use for many years, there is a gap in systematic analysis of the sources of uncertainty associated with this particular application. The paper presents a comprehensive assessment of the individual uncertainty components of the entire analytical process, starting from sample preparation to radioactivity measurement. Examples of the complete uncertainty budget for typical tritium levels in the analysed water samples are included. The relative importance of individual uncertainty components is given and possible ways of improving the overall quality of analyses are discussed.

Introduction

Tritium is a β-decaying radionuclide (E_max=18.6 keV), with the recently re-evaluated half-life value equal 4500±8 days corresponding to 12.32 years [1]. This half-life value is in agreement with that given in the Evaluated Nuclear Structure Data File (mirror site at IAEA: http://www-nds.iaea.or.at/ensdf/). The formerly recommended half-life value of 12.43 years [2] should not be used anymore in light of new half-life determinations [3] and comprehensive re-evaluation of all published experimental tritium half-life determinations [1].

Large quantities of tritium were released into the atmosphere in the thermonuclear tests performed during the early 1960s. This isotope is used nowadays as a global transient tracer for studying dynamics of the hydrological cycle [4]. In particular, environmental tritium is widely used in hydrology as an age indicator for young groundwaters. With decreasing levels of environmental tritium there is a need for sensitive measurement techniques capable of quantifying these levels with sufficient precision.

The method most widely used nowadays to measure concentrations of environmental tritium in water is electrolytic enrichment followed by liquid scintillation spectrometry. Although both techniques have been in use for many years, there is a gap in systematic analysis of the sources of uncertainty associated with this particular application. This paper presents a comprehensive assessment of the individual uncertainty components for the entire analytical process, starting from the sample preparation up to the radioactivity measurement.

The uncertainty assessment presented in the paper is related to the specific type of electrolytic enrichment cells developed and used at the Isotope Hydrology Laboratory of the International Atomic Energy Agency (IAEA) and in numerous other laboratories around the world. The enrichment procedure and the uncertainties involved in this process may vary with the type of the cells used, e.g. higher initial volume of the samples, higher enrichment factor, different design of the cells, different material used for fabrication of the cells, etc. [5, 6]. The characteristic feature of the enrichment process is that the samples are processed in batches, usually around 20 samples at a time. Technical details of the enrichment system being used at IAEA can be found elsewhere [7, 8, 9].

The IAEA has been organising interlaboratory comparisons for environmental tritium measurements since the late 1960s. Up to now, five intercomparisons of this type have been performed. About 90 laboratories participated in the most recent exercise concluded in December 2000. The results of this exercise clearly showed that a substantial number of the laboratories are underestimating their quoted analytical uncertainty associated with tritium measurements. It is therefore anticipated that the uncertainty assessment procedure outlined in this paper will help tritium laboratories to improve their analytical performance.

Throughout this contribution the tritium concentration in the analysed sample is expressed as tritium activity concentration (Bq kg^–1). In brackets Tritium Units (TU) are reported since this quantity is still widely used. One Tritium Unit is defined as:

$$ 1{\rm{ TU}} \equiv {{\left[ {{}^{\rm{3}}{\rm{H}}} \right]} \over {\left[ {{}^{\rm{1}}{\rm{H}}} \right]}} = {\rm{10}}^{ - {\rm{18}}} $$

One TU corresponds to 0.11919±0.00021 Bq kg⁻¹. The following numerical constants and parameters were used to convert TU to Bq kg⁻¹: Avogadro constant N_A=6.02214199(±0.00000047)×10²³ mol⁻¹ [10]; molar mass of water m_H2O=0.01801528 kg mol⁻¹; tritium half life T_1/2=4500±8 days [1]; tritium decay constant λ=(1.5403±0.0027)×10⁻⁴ d⁻¹.

The method

The method consist of two major parts: (i) pre-concentration of tritium in the analysed water sample using electrolytic enrichment, and (ii) detection of tritium activity in the concentrated sample using liquid scintillation spectrometry. General characteristics of the electrolytic enrichment system and the liquid scintillation spectrometers being used by the IAEA laboratory are listed in Table 1.

Table 1. Basic characteristics of the analytical system for environmental tritium measurements in water used at the IAEA Isotope Hydrology Laboratory

The following formula is used to calculate the tritium activity concentration in the analysed sample:

$$ {A_{T} = {{N_{{SA}} \cdot A_{{ST}} } \over {N_{{ST}} \cdot Z_{I} }} \cdot D{\rm{ }}} $$

(1)

where: N _SA - net count rate of the sample (cpm), N _ST - net count rate of the standard (cpm), A _ST - activity concentration of the standard (Bq kg⁻¹), Z _I tritium enrichment factor for the given sample, D - factor taking into account decay of tritium in the sample from the date of sampling to the date of measurement.

Equation 1 is valid with the assumption that the mass of the sample and the standard used for radioactivity measurement are the same, and both are measured in identical conditions (cf. sections on evaluating uncertainty of the average enrichment parameter P and of the tritium enrichment factor Z _I ).

Gross and net count rates are defined as:

$$ {N_{{SA}} = N_{{GSA}} - N_{B} } $$

(2)

$$ {N_{{ST}} = N_{{GST}} - N_{B} } $$

(3)

$$ {N_{{GSA}} = {\sum\limits_{i = 1}^n {{{N_{{GSA,i}} } \over n}} }{\rm{ }}} $$

(4)

$$ {N_{{GST}} = {\sum\limits_{i = 1}^m {{{N_{{GST,i}} } \over m}} }} $$

(5)

$$ {N_{B} = {\sum\limits_{i = 1}^k {{{N_{{B,i}} } \over k}} }{\rm{ }}} $$

(6)

where: n, m, k - number of individual measurements of the sample, standard and background in the given run, respectively; N _GSA,i , N _GST,i , N _B,i - individual (gross) count rates for the sample, standard and background, respectively, in the given measurement run.

The tritium enrichment factor Z _I is given by the formula [7]:

$$ {Z_{I} = \exp {\left\{ {{\rm{ }}{{P \cdot Q} \over {{\left( {W_{I} - W_{F} } \right)}F}}\ln {{W_{I} } \over {W_{F} }}} \right\}}} $$

(7)

where: Q - number of ampere-hours for the given enrichment run (Ah); W _I - initial mass of water (including OH⁻ ions) in the given electrolytic cell (g); W _F final mass of water (including OH⁻ ions) in the given electrolytic cell (g); F - Faraday constant F=2.975 Ah g⁻¹; P - average enrichment parameter for the given electrolytic cell calculated as:

$$ {P = {1 \over n}{\sum\limits_{i = 1}^n {P_{{\rm{i}}} {\rm{ }}} }} $$

(8)

where: P _i - enrichment parameter obtained during spike run i for the given electrolytic cell; n - number of spike runs for the given cell for which the average enrichment parameter is calculated.

The enrichment parameter is defined by the formula:

$$ {P_{i} = {{F \cdot {\left( {W_{{IS}} - W_{{FS}} } \right)}} \over Q} \cdot {{\ln {\left( {Z_{{IS}} } \right)}} \over {\ln {\left( {{{W_{{IS}} } \over {W_{{FS}} }}} \right)}}}} $$

(9)

where: W _IS - initial mass of water in the given cell during spike run i (g); W _FS - final mass of water in the given cell during spike run i (g); Z _IS enrichment factor for the given cell during spike run i, derived from the following formula:

$$ {Z_{{IS}} = {{N_{{SPF}} } \over {N_{{SPI}} }}} $$

(10)

where: N _SPF - net count rate of the spike water from the given cell, enriched during spike run i; N _SPI net count rate of the spike water (before enrichment). Net count rates of the spike water are calculated in an analogous way to regular samples (Eqs. 2, 4, 6).

The decay correction factor D is calculated as:

$$ {D = \exp {\left( { - \lambda \cdot t} \right)}} $$

(11)

where: λ - decay constant for tritium equal to (1.5403±0.0027)×10⁻⁴ d⁻¹; t - time elapsed between the sampling and the measurement (d).

During routine operation of the electrolytic enrichment system, each electrolytic run (24 cells connected in series) usually consists of 19 samples, three spike waters and two tritium-free waters to check the contamination level (cf. the section on "Quantifying the contamination level"). For each run three cells (different in each consecutive run) are filled with spike water to determine the actual tritium enrichment. In addition, spike runs with all 24 cells filled with spike water are carried out after 16 normal sample runs (about 4 months) to monitor the performance of the whole enrichment process for all cells. Each measurement run in a liquid scintillation spectrometer usually consists of 19 enriched samples, three enriched spike samples, two enriched background samples (tritium-free water), three background samples and three standard samples.

Each part of the analytical procedure can be broken down into individual steps (Table 2).

Table 2. Functional steps of the analytical process of tritium assay in water samples using electrolytic enrichment and liquid scintillation spectrometry

Evaluating uncertainty components

Initial and final distillation.

These steps do not contribute directly to the combined standard uncertainty of the measured quantity A _T . However, they may contribute to the overall contamination level of the measured samples (cf. the section on Quantifying the contamination level).

Weighing of cells

Weighing of the cells is performed at the beginning and at the end of the enrichment process. The uncertainty components related to the determination of the initial and final mass of water in the enrichment cell are due to the variability of weighing by difference in the appropriate region, the uncertainty associated with the calibration of the balance and the uncertainty related to buoyancy forces. The combined uncertainty associated with weighing of cells is in the order of 0.14 g for the discussed system. Details of the weighing process and the associated uncertainty assessment can be found in [11].

Electrolysis

The measured quantity during the electrolysis is the electrical charge flowing through the cells. The instrument used to measure the charge is an electronic amperehour meter. The certified accuracy of the instrument is 0.5% of the measured charge. For the charge applied in a typical enrichment run (697 Ah) the corresponding relative uncertainty is in the order of 0.005.

Preparation of the scintillation mixture

For the measurement of the tritium activity in a liquid scintillation spectrometer, a mixture consisting of water and scintillation cocktail is prepared in a special polyethylene counting vial for each enriched sample, background and standard. To transfer water to the counting vials a set of calibrated pipettes is used. The quoted accuracy is equal ±0.015 ml. Thus, the corresponding standard uncertainty will be equal 0.0l5/√3=0.0087 ml [12]. The standard uncertainty associated with the variations of temperature during preparation of the scintillation mixture (within 3 °C of the stated operating temperature), is equal to around 0.003 ml. The standard uncertainty associated with repeated deliveries of the same volume of water sample was determined experimentally and is equal 0.0091 ml. Thus, the total standard uncertainty associated with transfer of water (measured samples, standard, background) to the counting vials is 0.013 ml.

Radioactivity measurement

A typical measurement run consists of 19 enriched samples, three enriched spikes, two enriched tritium-free waters, three tritium-free waters and three standards. All samples are measured sequentially in 20-minutes intervals. The spectrometer usually accumulates readings of 23 intervals per sample within a week, 460 minutes being the typical duration for a single measurement run to achieve an acceptable uncertainty of the final result. At the IAEA, no original non-enriched spike samples are included in the normal measurement run, since the tritium activity of the spike is initially determined with high precision in relation to the used tritium standard.

Non-statistical effects can contribute to the observed count rates. These include fluorescence which might be induced by excessive exposure to UV radiation (e.g. direct sunlight) during preparation of the scintillation mixture, static charges (during transport of counting vials into the counting chamber) and/or pulse surges in the power supply network. The distribution of individual readings for each sample around the mean value is therefore assessed statistically through calculation of the sample standard deviation:

$$ {u_{x} = {\sqrt {{{{\sum\limits_{i = 1}^n {{\left( {N_{i} - N} \right)}^{2} } }} \over {{\left( {n - 1} \right)}}}} }{\rm{ }}} $$

(12)

where: N _i - individual gross count rates recorded for the given type of sample (N _GSA,i; N _{GST,i ;} N _B,i ); N - average count rate for a sample; n - number of individual readings.

The outlier rejection procedure is then applied—all individual readings of the given sample for which N _i differs from N by more than 2.8×u _x are rejected. The final N and u _x for each measured sample are then calculated from the remaining accepted readings using Eqs.2, 3, 4, 5, 6 and Eq. 12, respectively.

The calculated uncertainties of the average (gross) count rates for the samples, standard and the background resulting from counting statistics (obeying Poisson distribution) and the stability of the spectrometer are then calculated as a standard deviation of the mean for the population of individual readings after outlier rejection:

$$ {u_{{\bar{x}}} = {\sqrt {{{{\sum\limits_{i = 1}^n {{\left( {N_{i} - N} \right)}^{2} } }} \over {n{\left( {n - 1} \right)}}}} }{\rm{ }}} $$

(13)

The uncertainty values for those count rates at the IAEA Isotope Hydrology Laboratory are derived from routine measurements with Packard ultra-low level counters in its specially designed underground laboratory (cf. Table 1). At typical counter locations with less extensive shielding the background count rates and the resulting uncertainties may be higher. The typical uncertainties for sample, standard and background count rates, for the used counter type¹Packard 2770TR are listed in Table 1.

The standard uncertainties of the net count rates of the sample and standard are then calculated according to the law of uncertainty propagation. As an additional component, the uncertainty associated with preparation of the scintillation mixture is included (cf. the section on Preparation of the scintillation mixture). For N _SA and N _ST the combined standard uncertainty equals 0.157 cpm and 1.25 cpm, respectively. It should be noted that the above uncertainties are related only to the counting statistics and to the long-term stability of the spectrometer during the given counting run.

Evaluating uncertainty of the average enrichment parameter P

The actual average enrichment parameter for the given cell is calculated as the arithmetic mean of the values of the enrichment parameter obtained during the electrolytic runs in which the cell contained spike water. This value is modified each time spike water is loaded in the given cell (the new average is calculated). The standard uncertainty of the actual value of the enrichment parameter is calculated as sample standard deviation of the population of individual enrichment parameters available for the given cell.

An alternative method can be used to derive the average enrichment parameter P in cases when electrolytic cells reveal relatively small differences in the individual enrichment parameters. The average P value for the cells at the IAEA laboratory is 0.964±0.006. The average enrichment parameter for the given run can then be calculated from individual values derived from three or more different spike cells. In this way, the performance of the enrichment system is re-determined for each run individually. Equation 8 is used to calculate the average enrichment parameter for the given run. This method is mandatory during the initial period of the operation of an enrichment system, when not enough information concerning performance of the system is available from only few spike runs.

Evaluating uncertainty of the tritium enrichment factor Z_I

The standard uncertainty of the enrichment factor Z _I for the given cell is calculated from Eq. 7 using the law of uncertainty propagation by calculating partial derivatives of Z _I and inserting the appropriate values of the parameters and their estimated standard uncertainties. For the discussed system the calculated standard uncertainty u(Z _I ) is equal to 0.41. The largest contribution to u(Z _I ) stems from the uncertainty of the enrichment parameter (around 75%). Details of the calculations are given in [11].

Evaluating uncertainty of the decay correction factor D

The standard uncertainty of the decay correction factor is calculated using the law of uncertainty propagation applied to Eq. 11. For typical delay times between the sampling and the measurement equal six months and the maximum uncertainty attached to that period equal one day, the calculated standard uncertainty u(D) amounts to 1.6×10⁻⁴.

Uncertainty of the tritium activity concentration of the used standard A_ST

Most laboratories prepare their tritium laboratory standards from calibrated standards, which are available from several national institutions. At the IAEA Isotope Hydrology Laboratory the Certified Reference Material SRM 4927F, being distributed by the US National Institute of Standards and Technology (NIST), was used to prepare the working standard and the spike water. According to the certificate, SRM 4927F had a tritium activity concentration of 634.7 kBq g⁻¹±0.72% (with coverage factor k=2) on September 3, 1998. The dilution of this hot standard was done gravimetrically. The resulting relative standard uncertainty of the working standard and spike water being used in the IAEA is in the order of 0.004 [11]. The ratio of laboratory standard to laboratory spike was calculated from the dilution procedure and verified by careful, high-precision measurements. This ratio was then used in the subsequent enrichment runs, without actually measuring the spike water for each run.

Quantifying the contamination level

Specific characteristics of tritium (gas, relatively easily oxidised to HTO), combined with omnipotent presence of water vapour, make problems of contamination during the assay of low-levels of tritium particularly critical. Tritium may originate from various sources: (i) some types of watches with fluorescent dials, (ii) tritium sources used in gas chromatography, (iii) targets used in neutron generators, (iv) medical experiments where tritium is used as a tracer [4].

Measurements of tritium require consideration of various types of contamination ("blanks"), e.g. sampling, storage and analytical blank [13]. The magnitude of the blank will depend strongly on specific procedures of sample handling, on the ambient tritium concentrations during sampling, storage and analysis and on the type of containers used to store the samples before the analysis. Here, only the problems related to analytical blank will be addressed.

The analytical blank is defined as the amount of tritium picked up by the sample during the analytical process in the laboratory (cf. Table 2). The magnitude of this blank should be assessed and regularly controlled in each laboratory. The determination of a blank is straightforward if the concentration of tritium in "tritium-free" water used in the laboratory is much below the detection limit of the given analytical set-up: the samples of "tritium-free" water are treated as normal samples during all steps of the analytical procedure. If the enriched "tritium-free" samples are then used as background samples in the radioactivity measurement, the analytical blank is corrected even without knowing its absolute value. Alternatively, the radioactivity of enriched "tritium-free" water can be measured against a normal (not enriched) background sample and the amount of tritium picked up during the measurement process can be determined. In the analytical system described above, two samples of "tritium-free" water are added to each electrolytic run.

In cases when the tritium content in the laboratory "tritium-free" water is comparable with the detection limit, the analytical blank can be estimated by linear extrapolation of the measured tritium concentrations in this water, corresponding to the different degree of enrichment, to the value representing zero enrichment [13].

In addition to samples of "tritium-free" water processed in parallel to normal samples, each laboratory should introduce a system for monitoring the tritium levels in the laboratory atmosphere. The simplest way of doing this involves exposing ca. 100 ml of tap water in a beaker to the laboratory atmosphere for a prolonged time period (ca. two weeks is needed to reach isotopic equilibrium with the ambient moisture). After this time period, the remaining water in the beaker will have a tritium content close to that of the laboratory atmosphere [14]. Any unexpected increase in the tritium level in these samples points to possible contamination problems in the laboratory. At the IAEA, a laboratory air moisture sample is taken on a weekly basis for contamination monitoring by freezing out air moisture on liquid nitrogen cooled condenser plates.

Evaluation of combined standard uncertainty

In order to illustrate the influence of the tritium activity concentration in the given sample on the combined standard uncertainty, two cases will be considered below, bracketing the range of natural concentrations of tritium in meteoric waters.

Case A:

The sample tritium activity concentration A _T =0.119 Bq kg⁻¹ (1 TU) is near to the detection limit of even highly sensitive analytical procedures, providing, after the enrichment process (Z _I =17), a sample net count rate N _SA =0.304 cpm with u(N _SA )=0.070 cpm.

Case B:

The sample tritium activity concentration A _T =3.58 Bq kg⁻¹ (30 TU) is at the higher side of environmental tritium levels, providing after the enrichment process a sample net count rate N _SA =9.12 cpm with u(N _SA )=0.16 cpm.

The parameters appearing in Eq. 1, for both cases discussed, are listed in Table 3 together with their estimated standard uncertainties. From these values, the combined standard uncertainty of the tritium activity concentration A _T of a sample can be calculated.

Table 3. The summary of uncertainty components for the measured tritium activity concentration A _T, derived from Eq. 1. Two cases are reported: case A—A_T=0.119 Bq kg⁻¹ (1 TU) and case B—A_T=3.58 Bq kg⁻¹ (30 TU). The third column contains the calculated combined standard uncertainties for the variables appearing in Eq. 1. Sensitivity factors being the partial derivatives of Eq. 1 for the given variable are stated in the fourth column. The squares of the sensitivity factors multiplied by the square of the combined standard uncertainties of each variable in Eq. <equationcite>16</equationcite> is used to calculate the percent contribution to u²(A _T ) of the uncertainty associated with the given variable (last column)

The combined standard uncertainty of the measured quantity A _T is expressed using the law of uncertainty propagation by calculating the squares of the partial derivatives of A _T for the individual parameters, called sensitivity factors [15], and multiplying them by the squares of the individual uncertainties of the parameters. The square root of the sum of these products provides the exact derivation of the combined standard uncertainty:

$$ \matrix{ {u\left( {A_T } \right) = } \hfill \cr {\sqrt {\left( {{{\partial A_T } \over {\partial N_{SA} }}} \right)^2 \left( {u\left( {N_{SA} } \right)} \right)^2 + \left( {{{\partial A_T } \over {\partial N_{ST} }}} \right)^2 \left( {u\left( {N_{ST} } \right)} \right)^2 + \left( {{{\partial A_T } \over {\partial A_{ST} }}} \right)^2 \left( {u\left( {A_{ST} } \right)} \right)^2 + \left( {{{\partial A_T } \over {\partial Z_I }}} \right)^2 \left( {u\left( {Z_I } \right)} \right)^2 + \left( {{{\partial A_T } \over {\partial D}}} \right)^2 \left( {u\left( D \right)} \right)^2 } } \hfill \cr } $$

(14)

This approach is valid for any functional relationship (also for non-linear equations as in Eq. 7 for Z _I ), but is limited to uncertainties being reasonably small compared to the measured values [16]. In practice, the law of uncertainty propagation can be safely applied for standard uncertainties below 10% of the measured values.

In cases of linear equations of a purely multiplicative form, the law of uncertainty propagation can be expressed in a simpler form which for u(A _T ) is expressed as:

$$ {u(A_{T} ) = A_{T} \cdot {\sqrt {{\left( {{{u{\left( {N_{{SA}} } \right)}} \over {N_{{SA}} }}} \right)}^{2} + {\left( {{{u{\left( {N_{{ST}} } \right)}} \over {N_{{ST}} }}} \right)}^{2} + {\left( {{{u{\left( {A_{{ST}} } \right)}} \over {A_{{ST}} }}} \right)}^{2} + {\left( {{{u{\left( {Z_{I} } \right)}} \over {Z_{I} }}} \right)}^{2} + {\left( {{{u{\left( D \right)}} \over D}} \right)}^{2} } }{\rm{ }}} $$

(15)

The uncertainty u(A _T ) of the tritium activity is calculated from the relative standard uncertainties of the five parameters in Eq. 1. This special form provides an easy insight in the relative importance of the different sources of uncertainty.

The combined standard uncertainty of the measured quantity A _T , calculated for the two discussed cases using Eq. 14 reads as follows:

Case A: Sample with tritium activity concentration A_T=0.119 Bq kg⁻¹ (1 TU)

$$ \eqalign{ & u\left( {A_T } \right) = \sqrt {{\rm{7}}{\rm{.4}} \times 10^{ - 4} + {\rm{2}}{\rm{.3}} \times 10^{ - 7} + {\rm{2}}{\rm{.3}} \times 10^{ - 7} + 8.4 \times 10^{ - 6} + 3.7 \times 10^{ - 10} } {\rm{ }} \cr & u\left( {A_T } \right) = 0.0274\;{\rm{ Bq\;kg}}^{{\rm{ - 1}}} \cr} $$

Case B: Sample with tritium activity concentration A_T=3.58 Bq kg⁻¹ (30 TU)

$$ \eqalign{ & u\left( {A_T } \right) = \sqrt {{\rm{3}}{\rm{.8}} \times {\rm{10}}^{{\rm{ - 3}}} + 2.0 \times 10^{ - 4} + 2.1 \times 10^{ - 4} + 7.5 \times 10^{ - 3} + 3.4 \times 10^{ - 7} } {\rm{ }} \cr & u\left( {A_T } \right){\rm{ }} = {\rm{ 0}}{\rm{.108 \;Bq\; kg}}^{{\rm{ - 1}}} \cr} $$

Reporting the results

According to recommendations from CITAC and EURACHEM [15], the expanded uncertainty should be reported along with the analytical result. The expanded uncertainty is obtained by multiplying the combined standard uncertainty by a numerical factor, called the coverage factor, which in most cases will be equal to two. This corresponds to a confidence limit equal to ca. 95%, provided that the results of analysis have a normal distribution (for details see [12]).

For the cases discussed above, the expanded uncertainty equals 0.055 Bq kg⁻¹ and 0.22 Bq kg⁻¹, for sample A and B, respectively. The final result of the analysis should be then reported as:

Case A:

Tritium activity concentration: 0.119±0.055 Bq kg⁻¹ (1.00±0.46 TU)

Case B:

Tritium activity concentration: 3.57±0.22 Bq kg⁻¹ (30.0±1.8 TU)

It should be noted, however, that in nearly all cases of tritium data reporting for environmental applications, only the combined standard uncertainty is reported instead of the expanded uncertainty with a coverage factor.

Concluding remarks

The two cases A and B discussed above clearly show that for samples with low tritium concentrations, the uncertainty of the net count rate of the sample is a dominating component (>99%) of the overall combined uncertainty of A _T (cf. Table 3). For samples with relatively high tritium concentration, the most important component is the uncertainty of the enrichment parameter Z _I (80%), followed by the uncertainty of the net count rate uncertainty (19%).

The preferred strategy for improving the quality of measurements for samples near the detection limit should be to gain better precision of the sample and background measurements. This may be achieved by increasing the counting time of the sample and the background. Introducing an efficient outlier identification routine (by splitting the total measurement time into intervals) may reduce the influence of static charge effects.

For higher tritium activities, a careful check and optimisation of the enrichment parameters for the electrolytic cells should be performed. At the IAEA Isotope Hydrology Laboratory, such an analysis and corresponding optimisation a few years ago resulted in the reduction of that uncertainty component u(Z _I ) by about 40% of its former value.

Periodic checking and optimisation of the counting parameters for the liquid scintillation analyser used via series of test measurements for background and standard samples should be considered, to find the optimal working point for minimal background count rate and highest tritium detection efficiency, adapted for each laboratory according to their special requirements.

It is hoped that the procedure outlined in this paper will encourage tritium laboratories to make the effort to estimate the combined standard uncertainty for their environmental tritium analyses and discontinue the unsatisfactory approach of reporting just the statistical standard deviation of the counting procedure. A detailed, extensive discussion of the uncertainty assessment for environmental tritium measurements in water samples, supplemented by appropriate spreadsheet files to calculate the combined standard uncertainty can be found on the Web page of the IAEA Isotope Hydrology Laboratory (http://www.iaea.org/programmes/rial/pci/isotopehydrology/).