GB2443435A - Method of determining whether land is contaminated - Google Patents

Abstract

A method of determining whether the land at a site is contaminated, comprising: taking a calibration set of samples (150) of the land from the site; analysing the calibration set of samples to determine the level of at least one contaminant in each sample using a first, field, method and a second, reference method (152), the analysis being carried out using measurement apparatus; determining, using the levels of contaminant measured using the reference method which of the samples of the calibration set have more than a threshold level of contaminant - the contaminated samples - and which have less than a threshold level of contaminant - the uncontaminated samples (158); deriving, using the levels of contamination measured for the contaminated and uncontaminated samples from the calibration set using the field method, a probability distribution that a sample is contaminated (P(Ct x)) or uncontaminated (P(Ct' x)) given a certain value of level of contaminant measured by the field method (158); taking at least one further sample from the land at the site; measuring the level of at least one contaminant using the field method (164); and by comparing the level of the at least one contaminant thus measured to the probability distribution, determining whether the sample is contaminated. The method may comprise the application of Bayesian analysis to the comparison.

Description

METHOD OF DETERMINING WHETHER LAND IS

CONTAMINATED

This invention relates to a method of determining whether land, for example soil, is contaminated.

As the pressure for building land increases, there has been increasing use of so-called "brown-field" sites, where land is reused, often changing from previous industrial use to uses such as housing. In some territories, there are incentives for building on such land, so as to reduce the amount of building on land on which there has not previously been building -so-

called "green-field" sites.

However, land with a history of industrial usage often has an associated legacy of contamination. Pollutants such as lead, phenols, cyanide and benzene can remain in the land after industrial buildings have been removed and must be removed -so called "remediation" -before the land can be safely reused. Determining whether land is contaminated is therefore very important for a developer wishing to develop a piece of land.

Herein, we include in the term "land" the soil and other constituents making up the land's surface, such as sand or rock. The invention is particularly applicable to soil.

The measurement and analysis uncertainties associated with the measurement of contaminant concentrations in land using accredited laboratories and field tests can be enormous -over 85% in some cases.

This makes reliable decision-making impractical unless the uncertainties are defined and managed.

Often, two tests are used. The first, a field test, can be carried out on site and are relatively quick and simple to perform. However, it is generally desirable, and in some cases (such as in the United Kingdom) required by legislation, that some testing using an accredited reference laboratory be carried out. However, laboratory testing is expensive and time-consuming. A predictive relationship then needs to be developed

between the reference and field testing methods.

As noted above, the errors in both the field and the reference testing methods can be very large. It is desirable to improve the reliability and ease of use of the field test and to enable robust analysis of the errors associated with the field test method. If sufficient confidence in the field test results can be gained, then more field tests as opposed to laboratory tests can be carried Out, leading to quicker and easier analyses.

According to a first aspect of the invention, there is provided a method of determining whether the land at a site is contaminated, comprising: taking a calibration set of samples of the land from the site; analysing the calibration set of samples to determine the level of at least one contaminant in each sample using a first, field, method and a second, reference method, the analysis being carried out using measurement apparatus; determining, using the levels of contaminant measured using the reference method which of the samples of the calibration set have more than a threshold level of contaminant -the contaminated samples -and which have less than a threshold level of contaminant -the uncontaminated samples; deriving, using the levels of contamination measured for the contaminated and uncontaminated samples from the calibration set using the field method, a probability distribution that a sample is contaminated or uncontaminated given a certain value of level of contaminant measured

by the field method;

taking at least one further sample from the land at the site; measuring the level of at least one contaminant using the field method; and by comparing the level of the at least one contaminant thus measured to the probability distribution, determining whether the sample is contaminated.

Accordingly, by calibrating the field method against the reference method, any systematic bias of the field method as against the reference can be eliminated. Furthermore, the calculation of a probability distribution of the probability that a given sample is contaminated given its field measurements allows the user of such a method to estimate the confidence of a resulting decision.

Any or all of the samples of land may typically comprise a sample of soil from the site, although it may comprise a sample of sand or another component of the land at the site.

The step of calculating the probability distribution typically comprises generating probability distributions of the field method contaminant levels for the contaminated samples and for the uncontaminated samples. This can be regarded as calculating the probability of a given sample containing a given level of contaminant as assessed by the field method, given that it is contaminated or uncontaminated respectively. This may comprise calculating the mean and standard deviation (or equivalently, the variance) of the level of contaminant as measured by the field method for the relevant samples.

The mean and standard deviation may then be used to generate parameterised probability distributions of the field method contaminant levels using a probability distribution such as a normal distribution. The parameterisation of the probability distributions of the field method contaminant levels may therefore integrate any bias or analytical errors in the field method by shifting the mean level relative to the "true" value.

The standard deviation may indicate the overall accuracy of the analysis using the field method, including the sampling uncertainty. Thus, the method can implement a measure of quality control.

Using a mathematical notation, indicating the level of contaminant as determined by the field method as x, the sample being contaminated as C: and not contaminated as Ct', the probability distributions thus generated can be denoted as P(x I Ct) and P(x Ct') respectively. The probability distribution that a sample is contaminated or uncontaminated given a certain measured value of the level of contaminant can be denoted as P(CtIx)or P(C1'Ix).

In order to determine the probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant the method may apply Bayes' Theorem. This states that: P(Ct I x)= P(x I Ct) P(Ct) P(x) where P(Ct) is the probability that any given sample is contaminated and P(x) is the probability of a given sample obtaining a particular value of the level of contaminant by the field method. Conversely, the probability that a given sample is uncontaminated given a certain level of the contaminant may be given by: P(Ct' I x) = P(x I Ct') P(Ct') P(x) where P(Ct)=l-P(Ct) is the probability that any given sample is not contaminated.

Accordingly, the method may comprise the step of calculating a probability distribution for the level of contaminant in both the uncontaminated and contaminated samples aggregated -this can be used to estimate the probability of obtaining a given level of contaminant as measured by the field method. The method may also comprise the step of determining the probability of a sample being contaminated. Assuming that a sample is either contaminated or uncontaminated, this can be calculated by determining the ratio of uncontaminated to contaminated samples in the calibration set.

This use of Bayes' Theorem in determining whether land is contaminated represents a leap forward in the ease and simplicity of the determination of the status of contaminated land and reliable decision making, as well as providing a means for harmonizing uncertainty in measurement methods.

Not only does the method simply result in a determination of the contamination status of the land, but it also indicates how reliable the determination is.

The thresholds for determining whether the samples in the calibration set are contaminated or uncontaminated may be different. This allows for uncertainties in the reference method to be allowed for. The reference method may have a known estimated uncertainty; the thresholds may be set such that it is reasonably certain that a sample is uncontaminated or contaminated. The threshold for indicating that a sample is contaminated may be set at a nominal contamination threshold plus the known error, whilst the threshold for indicating that a sample is uncontaminated may be set at the nominal contamination threshold less the known error. Thus, some samples from the calibration set may be excluded from the contaminated and uncontaminated samples aggregated. This allows for uncertainty in the reference method to be accounted for, by eliminating samples which are not clearly contaminated or uncontaminated.

The known error may comprise an element corresponding to the precision of the analytical method employed and an element corresponding to errors induced by the sampling method used to pick the calibration set. These may be interpreted as a percentage uncertainty of the measured value.

The method may also comprise calculating from the probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant measured by the field method thresholds in measured contaminant level where, to a certain level of probability, a further sample will be contaminated or uncontaminated.

This may be, for example, at 95% probability contaminated or uncontaminated. Thus, the method may give three simple results to a user: * Contaminated (certain to a level greater than the threshold, for

example 95%)

* Uncertain * Uncontaminated (certain to a level greater than the threshold, for

example 95%)

The probability threshold may be varied depending on the importance of correctly determining contamination status.

It is to be noted that the contaminant or contaminants the levels of which are measured by the field and by the reference methods may be different; the method herein described is typically sophisticated enough to use any correlation between different contaminant levels. The level of the correlation will be highlighted in the probability given for a certain value of the measured contaminant level using the field test.

According to a second aspect of the invention, there is provided analysis apparatus for determining the contamination status of land, comprising an input for the concentrations in a set of calibration samples from the land of at least one contaminant measured using a first, field method and a second, reference method; processing means arranged to determine, in use, using the levels of contaminant input to the apparatus measured using the reference method which of the samples of the calibration set have more than a threshold level of contaminant -the contaminated samples -and which have less than a threshold level of contaminant -the uncontaminated samples; and to derive, in use, using the levels of contamination measured for the contaminated and uncontaminated samples from the calibration set using the field method, a probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant

measured by the field method; and

an input for the level of at least one contaminant in a further sample of land from the site measured using the field method; in which the processing means is arranged to, in use, compare the level of the at least one contaminant thus input to the probability distribution, in order to determine whether the sample is contaminated.

According to a third aspect of the invention, we provide use of the apparatus of the second aspect of the invention in carrying out the method of the first aspect of the invention.

There now follows, by way of example only, an embodiment of the invention described with reference to the accompanying drawings, in which: Figure 1 shows deterministic and probabilistic methods for determining whether land is contaminated; Figure 2 shows probability distributions for measured levels of contaminant in uncontaminated and contaminated calibration sample sets; Figure 3 shows probability distributions for contamination or non-contamination given certain levels of measured contaminant level; and Figure 4 is a flowchart showing the operation of a method according to an embodiment of the invention.

As discussed above, in order to determine whether land at a site is contaminated, it is important to consider the errors inherent in the measurement and analysis methods used. Frequently, it is desired to compare data analysed in the field to that analysed by a reference laboratory; even though the results from the reference laboratory may be equally as uncertain this is required by regimes such as that in the United Kingdom.

The uncertainty in the level of contamination measured in a given sample will of course depend on how the sample is measured. Typically, one calculates a range in which one is certain to a certain level of probability that the "true" value lies within the range. More reliable decisions may then be made. To estimate the overall measurement uncertainty it is necessary to allow for errors caused by both the sampling method used and the analytical uncertainty.

The analytical uncertainty involves errors associated with the measurement method itself. There are two sources of error that are of concern.

The first is analytical random error -the precision of the analytical method. It may be considered to express how "repeatable" the measurements are. In the methods generally used in determining whether land is contaminated, the errors follow a normal distribution and can be expressed as a standard deviation relative to a mean value, or as a relative value relative to the mean. By using a coverage factor k multiplying the standard deviation can be expanded such that the range covers the true value with an increased probability; for a value of k of 2, the range will include the "true" result with 95% probability.

The second is analytical bias -that is, systematic error. It is expressed as an estimate of the difference between the mean of a number of measurements () and a "true" value. The bias can be estimated by testing a reference material -commonly a "certified reference material" composed of a standardised matrix with known contaminant concentration. As the "true" value is generally unknowable, the known value of the reference material (cr111) is used to represent it. The bias (-c,,) can be expressed in units of concentration or as a percentage relative to the mean.

A further source of error is the sampling method used to take the samples. These errors apply to both field and lab testing. The level of small-scale heterogeneity at sampling locations can be estimated by taking duplicate samples at a subset of locations and determining the differences between the measured contaminant concentrations. The effects of large-scale heterogeneity are countered by increasing the number of samples taken. An estimate of the random error introduced by the sampling S method can be calculated from the duplicated samples using a robust analysis of variance technique.

Once the overall measurement uncertainty has been estimated, the resultant errors can be used to modify decision thresholds so as to explicitly manage uncertainties in the decision making process. Figure 1 of the accompanying drawings shows, in graph (a), a deterministic classification and, in graph (b), a probabilistic classification. In the deterministic classification, if a measured value of the contaminant concentration is above a safe threshold 100, the sample is deemed contaminated; otherwise it is deemed uncontaminated. Accordingly, samples lOla and lOib are deemed uncontaminated, whereas samples lOic and lOld are deemed contaminated. This does not, however, take into account any uncertainties in the measurements.

This is corrected in the probabilistic method shown in graph (b) of Figure 1. Error bars 103 indicate the range of values encompassed by the range of values within which there is a, say, 95% chance (the "confidence") of the "true" value occurring. If the error bars are clear of the threshold, as in the case of samples 102a and 102d, then the sample is contaminated or uncontaminated to within the confidence level.

Samples where the error bars cross the threshold may, or may not, be contaminated.

The method set out below aims to utilise a probabilistic classification of errors by using an entire dataset to produce an estimate of uncertainty, rather than single data points. Conventional regression analysis is unsuited for a task such as field analysis of contaminant levels since field tool responses are often sigmoidal in relation to a representative concentration range of contaminants across a given site leading to data clustering outside linear concentration ranges.

As is discussed below in more detail, according to a method forming an embodiment of the invention and depicted in Figure 4 of the accompanying drawings, the reference data is classified according to a decision rule and the errors inherent in the sampling and analytical methods; data points are labelled as either "uncontaminated" or "contaminated". The output enables the allocation of a window of total measurement uncertainty surrounding the decision rule -that is a degree of overall uncertainty accompanying any given "field" result. This provides a site-specific, quantifiable level of confidence for any field tool result.

The method starts at step 150 (Figure 4) during the early stages of a land remediation project before remediation work itself commences. A set of calibration samples, preferably at least 20 for reliability's sake, is selected and taken from across a site, based on the relative sampling densities that will be used in the substantive measurements. This calibration set should be collected such that it can be regarded as representative of the contaminant spread across the site for the decisions to be made using field tools during the remediation works.

In the worked example that follows, the project is the remediation of a former gas works in a UK town centre. The site covers an area of approximately 2 hectares. There is an estimated volume of contaminated soil to excavate and treat of 13 500m3. Contaminants of concern include PAH (Polycyclic aromatic hydrocarbons), phenols and organics. Site Specific Trigger Levels (SSTLs) of the following specific contaminants have been set: Phenol 1.3 parts per million (ppm) Napthalene 4.Oppm Cyanide 20ppm Benzene 2ppm If a representative sample from any MEV (minimum economic volume; 10m3 for this site) of soil contains any one of the above contaminants then it must be treated on-site with a soilwashing system. The method of this embodiment aims to provide a tool for determining whether this criterion is met.

At step 152, the calibration samples are analysed for contaminant level both using a field tool and with an accredited reference laboratory. The field tool chosen was a RaPID BTEX Immunoassay. This test apparatus is available from Cybersense Biosystems Limited, of Abingdon, United Kingdom. The BTEX/TPH RaPID assay is a tool for measuring the Benzene, Toluene, Ethylbenzene & Xylene fraction of the Total Petroleum Hydrocarbons (BTEX/TPH) in soil and water.

The kit applies the principles of enzyme linked immunosorbant assay (EL!SA) to determine BTEX/TPH concentration in parts per million. A BTEX/TPH containing sample is mixed with a BTEX/TPH -enzyme conjugate, which compete for the binding sites of BTEX/TPH specific antibodies attached to paramagnetic beads. A magnetic field separates the antibodies with bound BTEX/TPH or BTEX/TPH-enzyme conjugate from the mixture. Remaining antibody bound enzyme conjugate catalyses a colour reaction directly proportional to enzyme conjugate concentration and inversely proportional to sample BTEX/TPI-! concentration. The colour change is measured using a spectrophotometer and compared to a three-point calibration curve for quantificatation. Conveniently, the apparatus is small enough to fit into a briefcase and can be operated on site.

This field test has been found to produce a low number of false negatives and positives and correlated well with all contaminants of interest. It could not directly detect cyanide, but the presence of cyanide appeared to be correlated with the presence of at least one of the other hydrocarbons to which the immunoassay was sensitive. In the laboratory reference tests, the concentrations of phenols, naphthalene, cyanide and benzene for each sample were determined.

The errors for the reference laboratory testing are then determined (step 154). In order to determine the sampling error, duplicate samples are taken over the site and analysed with the field tool. The contaminant levels thereby determined for eight pairs of samples were as follows:

Sample ref RaPID assay field test result

Duplicate 1 Duplicate 2 1 2.6 3.4 2 1.1 2.3 3 9 6 4 17 11 2.8 0.6 6 170 190 7 58.4 53.1 8 26 19 The results were analysed with a robust Analysis of Variance (ANOVA) method so as to split the variances for sampling and non-sampling effects.

The standard deviation for the sampling precision was therefore found to be 4.027, with a mean contaminant concentration of 20.51.

Expressing the uncertainty as a percentage relative to the mean concentration at a desired (95%; k=2) level of probability as U, we can define: u 200Smeas = 200-2 + where s,/ is the sampling variance, S,,,,a12 the analytical variance and is the mean concentration (20.51).

A maximum analytical uncertainty of 10% at k =2 is provided by the reference laboratory. To convert to a standard deviation, we apply I 0%i s = k In the samples taken for this purpose, is therefore 1.025. The value of s5 has been found above to be 4.027. Therefore, applying the formula above, U is 40.5%. This is the range about the measured value of the reference results that the "true" value of contaminant level is likely to be.

This range can then be used at step 156 to set bounds for the reference laboratory measurements of the calibration set. By adding and subtracting U% from the SSTLs for each contaminant, levels at which it can be ascertained reasonably certainly that a given sample is either contaminated or uncontaminated can be generated as follows: Level (ppm) Contaminant Phenol Napthalene Cyanide Benzene SSTL 1.3 40 20 2 Lower 0.78 24 12 1.2 boundary Upper 1.82 56 28 2.8 Boundary Accordingly, if the contaminant levels as measured by. the reference laboratory are less than the lower boundary, it is most likely uncontaminated. If the contaminant levels are above the upper boundary, the sample is most likely contaminated. Anything falling between the two boundaries is uncertain.

The boundaries are used at step 158 to classify each of the calibration samples as contaminated (at least one contaminant level above the relevant upper boundary), uncontaminated (all contaminant levels below the relevant lower boundaries) and uncertain (the remaining samples).

The results of this classification can be seen in the following table: Sample Phenol Napthalene Cyanide Benzene RaPID Classification (ppm) (ppm) (ppm) (ppm) BTEX 1 13 12 4.3 0.10 147 Contaminated 2 5 37 1 0.10 211.4 Contaminated 3 2.5 11 12.0 0.10 58.4 Contaminated 4 3.1 22 8.9 0.10 53.1 Contaminated 1.1 5.4 4.0 0.10 16 Uncertain 6 9.4 43 6.7 0.10 55.2 Contaminated 7 7.6 82 1.0 0.20 170 Contaminated 8 8.1 200 1.0 1.7 190 Contaminated 9 1.2 19 3.6 0.1 69 Uncertain Sample Phenol Napthalene Cyanide Benzene RaPID Classification (ppm) (ppm) (ppm) (ppm) BTEX 4.4 61 1.0 0.10 70 Contaminated 11 <0.50 <1.00 2.4 61 4.5 Uncontaminated 12 1.2 7.3 1.0 0.10 85 Uncertain 13 <0.50 <1.00 <1.0 55 1.4 Uncontaminated 14 0.50 1.00 5.0 0.10 12.65 Uncontaminated 0.50 1.00 6.4 0.10 17.1 Uncontaminated 16 0.50 1.0 5.0 0.10 5.98 Uncontaminated 17 0.95 15 5.7 0.10 26 Uncontaminated 18 0.50 1.00 3 0.10 9 Uncontaminated 19 0. 50 1.00 2.4 0.10 2.8 Uncontaminated 0.50 1.00 1.0 0.10 0.6 Uncontaminated Note that the field tool results (RaPID BTEX) are not included in the classification of samples as contaminated or uncontaminated but are set out in the table for convenience.

In the next step, step 160. probability distributions for the probability of obtaining a given field test results for the contaminated and uncontaminated calibration samples respectively is generated. The mean and standard deviation of the field test results for the contaminated and uncontaminated samples are calculated separately and then each fitted to a normal distribution. In the present example, the results are as follows: Mean Standard deviation Contaminated 119.38 67.03 Uncontaminated 8.89 8. 39 The probability distributions thus generated can be seen in Figure 2 of the accompanying drawings. Denoting the result of a field measurement as x, a sample being contaminated as Ct and a sample being uncontaminated as Ct', the two probability distributions can therefore be considered as P(x Ct) and P(x I Ct'). Note that the "uncertain" samples are not used at this stage. Furthermore, the validity of the parameterisation thus performed can be checked by determining whether all of the data in each of the contaminated and uncontaminated sets falls within two or three standard deviations of the mean.

It is at this stage that the method accounts for analytical errors in the field tool, in the fact that it classifies the whole data set. The mean determined reflects any analytical bias, whereas the standard deviations represent the precision of the tool.

Using the notation from above, Bayes' Theorem states that: P(x Ct) P(Ct) , P(x J C:') P(Cz') P(Ct fx)= and conversely P(C: (x)= P(x) P(x) P(Ct) and P(Ct') can be calculated as the ratio of contaminated to uncontaminated samples in the calibration set; here P(Ct) = and P(Ct') = j. P(x) can be calculated using the assumption that a sample can only be contaminated or uncontaminated -the method is a binary classifier: P(x) = P(x Ct)P(Ct) + P(x I CtP(Ct') Using these relations, it is possible, at step 162, to calculate P(CtIx) and P(Ct'Ix) -the probability that, for a given field tool result x, the sample is contaminated or uncontaminated respectively. The two distributions generated in this example are shown in Figure 3 of the accompanying drawings.

In step 164. substantive testing can begin. As and when is necessary.

samples of soil are taken and analysed using the field tool. The output of these analyses are then compared to the P(CtIx) and P(Ct'Ix) distributions to determine with what probability it can be said that each sample is contaminated or uncontaminated.

Indeed, levels of the field tool result at which there is 95%, 97%, 99% (or any desired value) probability that the sample is contaminated or uncontaminated. Any sample taken with a field tool result less than the uncontaminated level or higher than the contaminated level can be seen to be uncontaminated or contaminated respectively. With respect to the example project, if any field tool result is less than l9ppm, there is greater than 95% certainty that the sample is uncontaminated with respect to the SSTLs. If any field result is greater than 4Oppm, there is greater than 95% certainty that the sample is contaminated. If any field tool result falls between the boundaries -here between l9ppm and 40ppm -there is uncertainty whether the sample is contaminated or uncontaminated and so further analysis is necessary; typically, this sample could be sent for fixed (reference) laboratory analysis. By using the method, the number of samples for which laboratory analysis is required can be reduced.

Accordingly, this method accounts for many if not all major sources of error in both lab and field data. It facilitates a field tool to be used which can be used on-site to make rapid, reliable decisions as to the status of any given soil sample. This provides opportunities for savings in project time, volumes of material going for treatment or disposal and analytical costs, whilst improving accuracy of remediation work because more samples can be analysed cost-effectively and more rapidly than previously possible.

Claims (19)

1. A method of determining whether the land at a site is contaminated, comprising: taking a calibration set of samples of the land from the site; analysing the calibration set of samples to determine the level of at least one contaminant in each sample using a first, field, method and a second, reference method, the analysis being carried out using measurement apparatus; determining, using the levels of contaminant measured using the reference method which of the samples of the calibration set have more than a threshold level of contaminant -the contaminated samples -and which have less than a threshold level of contaminant -the uncontaminated samples; deriving, using the levels of contamination measured for the contaminated and uncontaminated samples from the calibration set using the field method, a probability distribution that a sample is contaminated or uncontaminated given a certain value of level of contaminant measured

by the field method;

taking at least one further sample from the land at the site; measuring the level of at least one contaminant using the field method; and by comparing the level of the at least one contaminant thus measured to the probability distribution, determining whether the sample is contaminated.

2. The method of claim 1, in which the step of calculating the probability distribution comprises generating probability distributions of the field method contaminant levels for the contaminated samples and for the uncontaminated samples.

3. The method of claim 2, in which the step of generating probability distributions of the field method contaminant levels for each of the contaminated samples and for the uncontaminated samples comprises calculating the probability of a given sample receiving a given level of contaminant by the field method given that it is contaminated or uncontaminated respectively.

4. The method of claim 2 or claim 3 in which the step of generating probability distributions of the field method contaminant levels for each of the contaminated samples and for the uncontaminated samples comprises calculating the mean and standard deviation (or equivalently, the variance) of the level of contaminant as measured by the field method for the relevant samples.

5. The method of claim 4 in which the mean and standard deviation is used to generate parameterised probability distributions of the field method contaminant levels using a probability distribution.

6. The method of any of claims 2 to 5, in which notating the level of contaminant as determined by the field method as x, the sample being contaminated as Cf and not contaminated as Ct', the probability distributions thus generated as denoted as P(x I Ct) and P(x I Ct') respectively and the probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant as denoted as P(CtIx)and P(Ct'Ix), in which the method comprises calculating P(Ct Jx) and P(Ct' J x) according to P(Ct I x)= P(x (Ct) P(Ct) P(x) where P(Ct) is the probability that any given sample is contaminated and P(x) is the probability of a given sample obtaining a particular value of

the level of contaminant by the field method.

7. The method of claim 6, in which the probability that a given sample is uncontaminated given a certain level of the contaminant is calculated as: P(Ct' I x)= P(x I Ct') P(Ct') P(x) where P(Ct')=l-P(Ct) is the probability that any given sample is not contaminated.

8. The method of claim 6 or claim 7 in which the method comprises the step of calculating a probability distribution for the level of contaminant in both the uncontaminated and contaminated samples aggregated.

9. The method of any of claims 6 to 8 in which the method also comprises the step of determining the probability of a sample being contaminated.

10. The method of any preceding claim in which the thresholds for determining whether the samples in the calibration set are contaminated or uncontaminated are different.

11. The method of claim 10 in which the reference method has a known estimated uncertainty; the thresholds being set such that it is reasonably certain chat a sample is uncontaminated or contaminated.

12. The method of claim 11 in which the threshold for indicating that a sample is contaminated is set at a nominal contamination threshold plus the known error.

13. The method of claim 12 in which the threshold for indicating that a sample is uncontaminated is set at the nominal contamination threshold less the known error.

14. The method of any preceding claim in which the method also comprise calculating. from the probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant measured by the field method, thresholds in measured contaminant level where, to a certain level of probability, a further sample will be contaminated or uncontaminated.

15. The method of claim 14 in which the method gives three results to a user for a given further sample: * Contaminated (certain to a level greater than the threshold) * Uncertain * Uncontaminated (certain to a level greater than the threshold)

16. The method of any preceding claim in which any or all of the samples of land comprise a sample of soil from the site.

17. Analysis apparatus for determining the contamination status of land, comprising an input for the concentrations in a set of calibration samples from the land of at least one contaminant measured using a first,

field method and a second, reference method;

processing means arranged to determine, in use, using the levels of contaminant input to the apparatus measured using the reference method which of the samples of the calibration set have more than a threshold level of contaminant -the contaminated samples -and which have less than a threshold level of contaminant -the uncontaminated samples; and to derive, in use, using the levels of contamination measured for the contaminated and uncontaminated samples from the calibration set using the field method, a probability distribution that a sample is contaminated or uncontaminated given a certain value of the level of contaminant

measured by the field method; and

an input for the level of at least one contaminant in a further sample of land from the site measured using the field method; in which the processing means is arranged to, in use, compare the level of the at least one contaminant thus input to the probability distribution, in order to determine whether the sample is contaminated.

18. Use of the apparatus of claim 17 in carrying out the method of any of claims 1 to 15.

19. A method of determining whether land is contaminated substantially as described herein with reference to and as illustrated in the accompanying drawings.