Introduction

Geochemical surveys are carried out for various different reasons. Initially, they were used to define the extent of mineralised areas in prospectivity studies (Hawkes and Webb 1962), and often urban areas would have been avoided in these surveys (Johnson and Ander 2008). However, with developments in the understanding of the effects potentially toxic elements (PTEs) have on the environment and human health, geochemical surveys are increasingly being used in investigations to determine land quality and contamination (Salminen and Tarvainen 1997). A fundamental aim of geochemical surveys was often to define PTE concentrations that provide relevant thresholds within spatial element distributions. Originally used as a prospecting tool (Sinclair 1974), threshold values are increasingly employed as a method by which to discriminate “contaminated land” (Rodrigues et al. 2009). In this respect, the threshold is often set to differentiate between concentrations of the element that naturally occur in the soil and concentrations that result from diffuse anthropogenic sources, or even to differentiate between diffuse and point anthropogenic sources. However, there remains little consensus on what the aim of calculating these values is, and how values should be calculated.

Many terms are used in the literature to describe concentrations of elements in the soil, often with conflicting or overlapping definitions. In order to distinguish between geogenic and anthropogenic contaminations, Matschullat et al. (2000) define the geochemical background as a “relative measure to distinguish between natural element or compound concentrations and anthropogenically influenced concentrations”, which is similar to Hawkes and Webb (1962) definition of background as “the normal abundance of an element in barren earth material”. However, British Standards (BS19258) state that the background content of a substance in soil results from both geogenic sources and diffuse source inputs and that the background values should be a “statistical characteristic of the background content” (British Standards 2011), which is therefore similar to Salminen and Tarvainen’s (1997) definition of a geochemical baseline as an element’s average concentration in the Earth’s crust regardless of the source. A discussion by Reimann and Garrett (2005) examines in detail the various terms used to describe these values, including background, threshold, natural background and baseline and the many definitions that exist for these terms.

Salminen and Tarvainen (1997) suggest that baseline values are of “essential importance in environmental legislation” to define limits of PTEs in contaminated land, and recent changes in contaminated land legislation in England and Wales have recognised this by stating that “normal levels of contaminants in soil should not be considered to cause land to qualify as contaminated land, unless there is a particular reason to consider otherwise” (Defra 2012). Similarly, a recent government decree in Finland on the Assessment of Soil Contamination and Remediation Needs (Ministry of the Environment Finland 2007) requires the input of geochemical baseline concentrations in Finnish soils during the assessment process. An investigation of arsenic concentrations at a site of specific interest in Southern Italy led to the development of a statistical methodology for determining the difference between natural and anthropogenic concentrations of metals and metalloids in soils (APAT-ISS 2006). This methodology was retained by the Italian government as it was considered to be not only applicable to this particular site, but also to all other sites of national interest where the same problem was occurring.

Within the research described in this paper, the term TTV is used to refer to a value which gives a characteristic concentration for an element within a defined geographical area known as a domain. Previous work by Ander et al. (2013a, b) has seen the development of a methodology to determine NBCs of contaminants in English soils, supporting the recent changes to the statutory guidance (Defra 2012). Within this methodology, a domain was defined as an area in which a readily distinguishable factor could be identified as controlling the concentration of the element. This approach has been maintained within this investigation, remembering that these areas need to be defined on an element by element basis using initial assessments of the distribution of the elements within the study area. It is important that the methods used to identify domains take all the relevant factors affecting soil element concentrations into account; geogenic factors, diffuse source anthropogenic inputs and point source contamination. In order to be most relevant and useful for environmental legislation, the typical threshold values calculated should define concentrations of PTEs that are typical of the threshold between geogenic and diffuse anthropogenic source contributions to soil and concentrations that are associated with point sources. If point sources of anthropogenic contamination can be identified, they can be more readily assessed to determine whether they pose any risk to the surrounding environment. A number of different industries can make use of definite concentrations, which achieve this differentiation. In particular, contaminated land professionals can more easily determine sites that possibly require further investigations because the TTVs are exceeded. In addition, the agricultural industry may be interested in depleted concentrations of these elements where they are also considered to be essential to animal and plant life, e.g. copper.

Commonly investigated PTEs include arsenic (As), cadmium (Cd), cobalt (Co), copper (Cu), chromium (Cr), iron (Fe), mercury (Hg), manganese (Mn), nickel (Ni), lead (Pb), vanadium (V) and zinc (Zn) (Ajmone-Marsan et al. 2008; Kelepertsis et al. 2006; Palmer et al. 2013; Paterson et al. 2003; Ramos-Miras et al. 2011). Concentrations of PTEs are assessed for a variety of reasons; As, Hg and Pb are examples of elements commonly investigated in urban areas (Chirenje et al. 2003, 2004; Rodrigues et al. 2006; Wong et al. 2006), while Cr, Ni, V and Zn have previously been investigated as geogenically controlled PTEs within Northern Ireland (Cox et al. 2013; Palmer et al. 2013). Previous research has identified concentrations of As, Cd, Cr, Cu, Ni, Pb, V and Zn in Northern Irish soils that exceed relevant generic assessment criteria (GAC)/soil guideline values (SGVs) (Barsby et al. 2012; Martin et al. 2009a, b; Nathanail et al. 2009). Six PTEs have been selected for investigation in this research; As, Cr, Cu, Ni, Pb and V. These elements are expected to be governed by a mixture of geogenic and anthropogenic sources, a necessary factor in order to complete the aims of this study.

The rationale behind this research is to investigate soil geochemical data for Northern Irish soils by (1) using a variety of techniques to identify the principal controls on the spatial variation of the PTEs and determining which technique is most appropriate for the available data set; (2) identifying domains, i.e. areas of elevated and reduced PTE concentrations; (3) using previously developed statistical methodologies to calculate TTVs of PTEs within the above mentioned areas and (4) critically comparing the values calculated to determine which statistical method is most appropriate for use in differentiating between diffuse and point source element concentrations. It is worth noting that the TTVs do not assess risk, but instead provide an indication of PTE concentrations that are typical at a site.

Materials and methods

Study area

Northern Ireland is part of the United Kingdom which sits in the northeast of the island of Ireland (Fig. 1a) and is home to over 1.8 million people. Despite being less than 14,000 km2 in area, the bedrock in Northern Ireland (Fig. 1b) ranges from Mesoproterozoic to Palaeogene in age and as a result is said to present an “opportunity to study an almost unparalleled variety of geology in such a small area” (Mitchell 2004). The bedrock is often simplified into a series of Caledonian terranes and part of a Palaeogene igneous province with distinct geological characteristics. The psammites in the northwest of Northern Ireland are of Neoproterozoic age. The southeastern terrane is lower Palaeozoic in age and also contains younger igneous intrusions. These Palaeogene igneous intrusions consist of three central complexes; the Mourne Mountains, Slieve Gullion and Carlingford. The southwest comprises of a mixture of sandstones, mudstones and limestones, which are mainly upper Palaeozoic in age with a distinct lower Palaeozoic inlier. The northeast is dominated by a large area of extrusive Palaeogene basalts. In terms of superficial geology, peatlands cover 12 % of land area (Davies and Walker 2013) as shown on Fig. 1c. Two main urban areas exist within the country; Belfast and Londonderry, with populations of approximately 280,000 and 108,000, respectively, (NI Statistics and Research Agency 2013), with other smaller urban centres including towns and villages (Fig. 1c).

Fig. 1
figure 1

Maps showing a location b simplified bedrock geology and c areas of peat substrate (superficial geology), rural and urban areas across Northern Ireland [Bedrock and superficial geology derived from data provided by GSNI (Crown Copyright)]

Full size image

Soil geochemical data

The Tellus project, managed by the Geological Survey of Northern Ireland (GSNI), comprised both geophysical and geochemical surveys. The geochemical survey saw the collection of nearly 30,000 soil, stream-sediment and stream-water samples across Northern Ireland between 2004 and 2006. Urban and regional soil samples were collected at densities of 4 per km2 and 1 per 2 km2, respectively. Two depths were sampled at each location; a shallow sample taken between 5 and 20 cm, and a deeper sample taken between 35 and 50 cm. The sample taken at each location was a composite of auger flights collected at the four corners and the centre of a 20 by 20 m square. Samples were air-dried at the field-base before transport to the sample store where they were oven dried at 30 °C for approximately 2–3 days. The shallow samples were shipped to British Geological Survey (BGS) laboratories in Keyworth, Nottingham for preparation and analysis via x-ray fluorescence (XRF). Sample preparation entailed sieving to a <2 mm fraction, from which a subsample was produced for milling and pressed pellet production.

A number of quality control methods were employed during the XRF analysis. Two duplicate and two replicate samples were analysed per batch of 100 samples. Three secondary reference materials that were collected in Northern Ireland specifically for the Tellus survey, and one material from BGS’s Geochemical Baselines Survey of the Environment (G-BASE) programme, were routinely analysed at a rate of two insertions per batch. Certified reference materials were also analysed before and after each batch. Further details of quality control methods are provided by Smyth (2007).

Domain identification

Known controls over PTEs

In order to calculate TTVs, domains were identified for each element. Domains were selected based on knowledge of the factors shown in Fig. 1, which were identified as the main controls over element concentrations in soils.

Studies have shown that the majority of glacial till in Northern Ireland is found within only a few kilometres of its origin, suggesting that soils usually reflect the character of the underlying geology (Cruickshank 1997; Jordan 2001). Therefore, bedrock geology is expected to provide a strong control over element concentrations in soil. Geochemically, it is likely that the extrusive and intrusive (in particular, the Antrim Basalt formation and Mournes Mountain complex) igneous rocks of Northern Ireland will be of most interest, as previous studies have shown that they contain reduced and elevated concentrations of a number of elements (Barrat and Nesbitt 1996; Green et al. 2010; Hill et al. 2001; Smith and McAlister 1995; Smyth 2007). Previous studies have demonstrated that soils from the basalt area are more homogeneous in their geochemical content than soils from areas of other rock types, suggesting that the basalts are acting as the soil parent material and the main control over geochemistry in that area (Zhang et al. 2007). A simplified representation of bedrock geology in Northern Ireland derived from GSNI’s 1:250,000 bedrock geology map is shown in Fig. 1b, grouping bedrock types of similar composition and age.

The existing literature suggests that areas of peat substrate within Northern Ireland have a control over the distribution of a number of elements (Palmer et al. 2013). Peat bogs that are fed solely by atmospheric deposition (ombrotrophic) can be used as archives of many types of atmospheric constituents (Shotyk 1996), including contamination in the form of PTEs. Topographically elevated areas of peat are more likely to be affected in this way, as increased precipitation is usually associated with elevation (Goodale et al. 1998). Areas of peat were defined using data derived from the GSNI’s 1:250,000 superficial geology map and are shown on Fig. 1c.

Urban and rural areas were defined using a revised version of the Corine land cover 2006 seamless vector data (European Environment Agency 2012). This approach is different to that taken in the NBC methodology, where the generalised land use database statistics for England 2005 (Communities and Local Government 2007) were used. The Corine land cover data set covers all of Europe and defines 44 land use classes based on the interpretation of satellite images (European Environment Agency 2012). This has been simplified in Fig. 1c to show urban and rural areas in Northern Ireland. The majority of the land use classes were easily defined as either rural or urban, with a few others defined on a site by site basis.

In the NBC methodology, metalliferous mineralisation and mining maps were used to define mineralisation domains throughout the study (Ander et al. 2011). As this information is not available for Northern Ireland, a different approach was taken in using mineral occurrence locations provided by GSNI, alongside the relevant literature (Lusty et al. 2009, 2012; Parnell et al. 2000) to aid in mineralised domain identification.

Method used for domain identification

It is important that the methods used to define domains be robust to nonnormality and the presence of outliers that are common in geochemical data. Three methods to aid in the domain identification process were compared in this study; k-means cluster analysis (Ander et al. 2011), boxplot mapping (Reimann et al. 2008) and empirical cumulative distribution function (ECDF) mapping (Reimann 2005). All statistical analyses of data were completed in the R statistical software package (R Core Team 2013), and all geographical analyses and images were completed using ArcMap 10.0 (ESRI 2009).

The k-means cluster method (Fig. 2a) was used to define domains by Ander et al. (2011) in the NBC methodology. As k-means cluster analysis is of the partitional variety, the number of clusters must be assigned to the technique at the outset (Jain et al. 1999). The most visually acceptable number of clusters, based on an antecedent visual assessment (Templ et al. 2008), was input into the technique, and the data were partitioned into the selected number of clusters by minimising the “average of the squared distances between the observations and their cluster centres” (Reimann et al. 2008). The algorithm constructed by Hartigan and Wong (1979), generally considered to be the most efficient (Ander et al. 2011), was used as the default setting in the R software package (R Core Team 2013). Each data point was classified into a cluster by the technique, allowing the creation of a map of the clusters across Northern Ireland.

Fig. 2
figure 2

Domain identification methods completed for Ni concentrations in the shallow soils of Northern Ireland analysed by XRF; a completed by a k-means cluster analysis, b classes defined by boxplot of log-transformed concentrations as shown and c classes defined by ECDF of log-transformed concentrations as shown, with inverse distance weighting used to map the results (output cell size of 250 m, power of two and a fixed search radius of 1,500 m)

Full size image

Tukey boxplots of the log-transformed data (Fig. 2b) were also used to define the classes for producing maps of the data distribution. Assumptions regarding normal distribution of the data appear in the boxplot construction when the whisker values are calculated, as their calculation (box extended by 1.5 times the length of the box in both directions) assumes data symmetry (Reimann et al. 2008). Log-transformations were applied as geochemical data are often strongly right-skewed, and the log-transformation helps the data distribution to approach symmetry, allowing a better visual demonstration of the data when mapped (Reimann et al. 2008). The boxplot was used to split the element concentrations into five classes for mapping; lower extreme values to lower whisker, lower whisker to lower hinge, lower hinge to upper hinge, i.e. the box, upper hinge to upper whisker and upper whisker to upper extreme values.

The third method applied to map the distribution of the elements used classes based on the empirical cumulative distribution function (ECDF) (Sinclair 1974). The ECDF graph is a discrete step function, which jumps by 1/n at each of the n data points. As shown in Fig. 2c, the ECDF plots have been constructed using the log-transformed concentrations of the element, in this case nickel, in order to make breaks in the distribution more obvious. Breaks in the distribution are demonstrated through changes of gradient in the graph and are likely to be caused by the presence of different subpopulations within the data set, with different underlying factors controlling the concentrations of elements in these populations (Díez et al. 2007; Reimann et al. 2005, 2008). Therefore, breaks in the distribution can be used to distinguish mapping class boundaries.

Domain corroboration

In order to corroborate the results of the domain identification process, a geostatistical approach involving the construction of semi-variograms was used to ensure that the controlling factors over element concentrations in soil were correctly identified. The geostatistics were generated using ArcMap 10.0 (ESRI 2009). A semi-variogram is based on the theory of regionalised variables (Matheron 1965), which permits interpolation of values on a surface by assuming that data points closest to each other spatially will have a greater influence over estimated values than would data points further away from each other. Several important pieces of information can be identified from the semi-variogram:

  • Spatial variation at a finer scale than the sample spacing (Deutsch and Journel 1998) and measurement error (Journel and Huijbregts 1978) is represented by the nugget (C 0). Such small scale sources of variance can be an indication that sampling or analytical error is present or that micro-scale processes are governing geochemistry to a greater degree than was detected by sampling resolutions.

  • The spatially correlated variation is represented by the structured component (C 1) (Lloyd 2007).

  • The sill (C x ), where the semi-variogram levels off, is the distance at which pairs of data points are no longer spatially dependent upon each other.

  • The nugget: sill ratio (C 0/C x ) gives the proportion of random to spatially structured variation at the scale being investigated.

  • Ranges of influence (a) can be statistically inferred from the lag distance at which the sill is reached (McKinley et al. 2004), permitting interpretation of specific environmental factors that may be influencing the mapped element of interest.

  • Depending upon the nature of fitting measured values to a semi-variogram, multiple spatial structures can be identified. This is of particular interest where investigations into multiple environmental factors that may be controlling the element of interest are required.

  • An apparent lack of spatial structure also provides important information, such as giving an indication about the suitability of analytical or sampling methods in accurately detecting total element concentrations, which are thoroughly representative of a particular study area.

Calculation of typical threshold values

In response to the legislative requirements discussed in the introduction, different authors have derived methods by which “background” values can be calculated. The NBC methodology (Ander et al. 2013a, b) aims to provide a mechanism for revised legislation that differentiates between levels of contamination from geogenic and diffuse sources and those from point source contamination. In order to take account of spatial variability, domains were defined for each element by comparing the results of a k-means cluster analysis to a soil parent material model, land use classifications and mineralisation and mining geographical mapping. Within the methodology, it is recommended that the domains are based on at least 30 values. The NBC was then calculated for each domain using a statistical methodology that (1) assesses the skewness of the geochemical data by observing a histogram and calculating the skewness and octile skewness of the distribution. Based on the results of that assessment, the method (2) performs either a log-transformation or a box-cox transformation on the data if necessary and then (3) computes percentiles using either parametric, robust or empirical methods depending on the results of the transformation applied. The NBC is then taken to be the upper 95 % confidence limit (UCL) of the 95th percentile. A detailed explanation of how the methodology was constructed and how it should be applied is given in Cave et al. (2012).

Jarva et al. (2010) have developed a methodology to allow the calculation of “baselines” in Finland, which in this instance refer to both the “natural geological background concentrations and the diffuse anthropogenic input of substances at regional scale”. As with the above NBC calculations, Finland was divided into a number of geochemical provinces. A key difference between the two methodologies is the consideration of soil type, with baseline values calculated by soil type within geochemical provinces. The ULBL is based on the upper limit of the upper whisker line of the box and whisker plot. A box and whisker plot identifies any values which fall above the upper whisker line as outliers, which may “represent natural concentrations of an element at the sampling site” (Jarva et al. 2010) but are probably not typical of the geochemical province as a whole. Logarithmic transformed data were not used to plot the box and whisker plots, as the untransformed data led to the highest amount of outliers and therefore were felt to give a more conservative value.

A key difference between the NBC and ULBL methodology is the determination of what a “conservative” value is considered to be. The ULBL methodology aims to identify the maximum number of outliers, therefore generating a lower concentration for the ULBL and the possibility that larger areas of land will be identified as exceeding the ULBL. The NBC methodology supports the English contaminated land regime, which aims to identify sites where “if nothing is done, there is a significant possibility of significant harm, such as death, disease or serious injury” (Ander et al. 2013a). Therefore, by taking the upper 95 % confidence limit of the 95th percentile, the aim seems to be to identify the highest risk sites in order to prioritise further investigation and management of these sites.

Within this research, both the NBC and the ULBL methodologies were applied to the shallow XRF data available for all of Northern Ireland. NBC calculations were carried out using the R-scripts prepared by Cave et al. (2012) while ULBL calculations were undertaken in R using scripts prepared by the authors. Data from shallow soils were selected for analysis, so both anthropogenic and geogenic influences on the element concentrations could be determined. XRF was selected as the most appropriate analytical method as it is said to give total results (Ander et al. 2013a).

Results and discussion

Comparison of domain identification methods

Figure 2 gives a comparison of the three methods used to map the distribution of elements and therefore identify domains using Ni as an example. The k-means map (Fig. 2a) highlights only the basalts, which overlie northeast Northern Ireland. Both the boxplot map (Fig. 2b) and the ECDF map (Fig. 2c) show areas of elevated and reduced concentrations. Both show elevated concentrations over the basalts, with the boxplot method mapping the boundary of the basalts most effectively. Reduced concentrations are more easily identified through the ECDF map and are obviously correlated to the Mourne Mountain complex of southeastern Northern Ireland. By comparing this image with areas of peat substrate (Fig. 1c), an association between areas of peat and reduced concentrations was also identified.

The k-means technique, used in the NBC methodology, produces useful results in the determination of elevated domains; however, it is more commonly used to compare a number of variables and estimate which variables are similar and dissimilar to each other (Romesburg 2004), and the inability of the method to determine domains of reduced concentrations does limit its applicability in practice. In the case of Ni (Fig. 2a), the initial assessment demonstrated that three clusters would be most appropriate; however, a certain amount of prior knowledge regarding the controls over element concentrations is expected in this assessment.

Boxplots allow identification of both elevated and reduced concentrations of the elements, with different sections of the distribution related to separate parts of the boxplot. However, the splits in the distribution are still set at arbitrary values within the data set; meaning actual controlling factors over the element concentrations could be missed.

The ECDF method was superior in terms of spatially identifying both elevated and depleted concentrations of elements as it retains a great deal of information about the distribution of the element in the mapped output. It clearly delineates areas of both elevated and depleted concentrations allowing controls over the PTE concentrations to be determined. However, the method does require a level of interpretation as the individual inspecting the graphs decides where the gradient changes occur. This introduces potential for bias as a level of knowledge of the modelled domains could influence the results. It is, however, important to remember that the outputs from this process are maps, and therefore, even though different individuals will generally produce different results when splitting the ECDF plot by gradient, the same general trends will be obtained. Of each of the 3 methods, the ECDF methodology provides the greatest detail and opens the methodology to applications other than the identification of land contamination.

PTE domains

Within this investigation, the maps produced using the ECDF technique were compared with the main factors known to control the distribution of elements across Northern Ireland in order to identify domains. The majority of domains were easily identified and the results correlated well with the existing literature describing the distribution of elements in Northern Ireland (Young 2014).

Finalised domains are shown in Fig. 3 for the elements under investigation: arsenic, chromium, copper, lead, nickel and vanadium. Similar controlling factors were identified for Cr, Cu, Ni and V, with elevated concentrations of these elements observed over areas of basalt bedrock geology creating a basalt domain. Reduced concentrations are seen in the Mourne Mountains complex, associated with naturally occurring low concentrations of these elements in granites, creating the Mournes domain. Cr, Cu, Ni and V are known to be found at elevated concentrations in the Antrim basalts (Barrat and Nesbitt 1996; Hill et al. 2001; Smith and McAlister 1995) and at reduced concentrations in granites (Wedepohl et al. 1978).

Fig. 3
figure 3

Domains identified for a arsenic b chromium, copper, nickel and vanadium and c lead based on the ECDF maps produced

Full size image

Concentrations of Cr, Cu, Ni and V were generally depleted in peat samples overlying all bedrock geologies except the basalts. Overlying the basalt formation, some areas of peat showed depleted Cr, Cu, Ni and V concentrations, while others (generally at lower topographical elevation) showed higher concentrations of each element in line with the basalt domain. This distribution is probably explained by the type of peatland, and the land use activities taking place on it (Joint Nature Conservation Committee 2011). Lowland peats appear to have less of a control over element concentrations; meaning the basalts remain as the primary controlling factor, and higher concentrations are observed. Upland peats, however, appear to exert a greater control with reduced concentrations being observed. It is also possible that the differing land use on the peat could be affecting its ability to function efficiently; however, this subject would require further exploration. This distribution of Cr, Cu, Ni and V was also observed by Young (2014), and therefore, a peat domain was selected which incorporated all areas of peat that do not overlie the basalts, along with areas of peat overlying the basalts that are associated with reduced concentrations of these elements. Depletion of Cr, Cu, Ni and V in this domain may reflect biogeochemical cycling of PTEs within the peats (Novak et al. 2011) but further research would be required to confirm this.

All the domains shown for lead are associated with elevated concentrations of the element. The Mourne Mountains complex shows elevated concentrations of lead (Mournes domain), fitting with known elevations of lead in granites (Krauskopf 1979). Elevated concentrations of lead were also associated with urban areas across Northern Ireland (urban domain). Pb is well known for its correlation with anthropogenic activity, and therefore urban centres (Albanese et al. 2011; Locutura and Bel-lan 2011). Identified sources of lead in urban environments include historical use of leaded fuel and lead in paint (Chirenje et al. 2004; Mielke and Zahran 2012; Mielke et al. 2011). A strong correlation was observed between areas of elevated topography with a covering of peat and elevated lead concentrations, as previously described by Young (2014). This is not surprising, as peat soils are well known for acting as historical records of atmospheric pollution, with higher heavy metal concentrations common in upper peat layers (Givelet et al. 2004; De Vleeschouwer et al. 2007). Chronologies of Pb deposition have been completed for a raised bog in Ireland (Coggins et al. 2006), showing elevated concentrations within the depth range (5–20 cm) of the shallow Tellus samples. These topographically elevated areas of peat were separated out from the full peat data set to form lead’s peat domain.

Finally, a mineralisation domain was also identified for lead. This area was defined using the elevated concentrations of lead as determined on the ECDF map, and the extent of the domain was corroborated by the strong correlation between lead mineral occurrences as provided by GSNI, shown in Fig. 4a, areas of high to very high prospectivity potential (Lusty et al. 2012) and the mineralised domain identified for Pb. It is worth noting that lead mineral occurrences were identified by GSNI in the southeast of Northern Ireland (Fig. 4a); however, as significantly elevated Pb concentrations were not recorded in that area, a mineralised domain was not created in this region. In contrast, the area of lead mineralisation identified in the northwest of Northern Ireland was in an area of elevated Pb concentrations. However, closer inspection of the spatial distribution of areas of elevated concentrations and comparison with peat maps showed that although mineralisation may be contributing to the elevated concentrations, their spatial distribution suggests that peat has a controlling role in accumulating this element, probably from atmospheric deposition.

Fig. 4
figure 4

Mineral occurrences provided by GSNI shown for a lead on a map showing lead’s mineralisation domain and b zinc, lead, gold and copper on a map showing arsenic’s mineralisation domains (Crown Copyright)

Full size image

The domains for arsenic were more difficult to define as there was greater uncertainty regarding the factors controlling the distribution of this element. The Shanmullagh formation consisting of early Devonian age sandstone and mudstone (Mitchell 2004) contained slightly elevated arsenic concentrations, creating the Shanmullagh domain. Two other areas, thought to be associated with mineralisation, were also shown to contain elevated concentrations. These two mineralisation domains were defined in the same manner as the lead mineralisation domain, using the ECDF map, and were named mineralisation 1 and 2. As there are no shown arsenic occurrences on the mineral occurrences information provided by GSNI, a different approach was used to corroborate the extent of these areas. Arsenic is well known as a pathfinder for gold and is therefore used in prospectivity investigations along with silver, gold, copper, lead, zinc, bismuth and barium (Lusty et al. 2009). A strong correlation is again shown between the mineralised zones identified in this study and mineral occurrences of Ag, Cu, Pb and Zn (Fig. 4b), the only mineral occurrences from the previous list that were available from GSNI. The prospectivity for gold within all the identified As mineralised domains is again shown to be high (Lusty et al. 2009, 2012). An interesting and possibly unexpected finding is the lack of an urban domain for arsenic. However, this was also the case in the calculation of NBCs for England (Ander et al. 2013a).

Domain corroboration

The results of the geostatistical approach employed as corroboration are given in Table 1 for all the PTEs investigated. These show that the extent of Cr, Cu, Ni and V distributions in Northern Ireland is strongly controlled by the presence of basalts in the northeast of the region, with ranges (a) not exceeding the largest spatial extent of this geologic formation of approximately 90 kilometres (Fig. 1b). Based on variography and source domain identification, elevated concentrations of these four trace elements in the region are attributable to geogenic sources on a spatial scale that exceeds other potential influences over the distribution of this element. However, approximately 13–35 % of total variances in Cr, Cu, Ni and V spatial distributions are accounted for by the nugget variance (C 0), suggesting such proportions of variance may be accounted for by smaller scale processes not detected within the soil sampling resolution of the Tellus Survey (Table 1). Arsenic, by comparison, is controlled by a spatial function covering a smaller spatial extent than elements associated with the basalts, with a nugget effect accounting for approximately half of all variance in spatial distribution (48.1 %). Lead exhibits the shortest range spatial function, characteristic of trace elements whose distributions are heavily influenced by small scale processes, such as anthropogenic activity in urban areas. This trend is also supported by the large nugget effect for this element.

Table 1 Experimental semi-variogram modelled parameters for Ni, Cr, V, As, Cu and Pb concentrations as measured by XRF in shallow soils of Northern Ireland where C 0 = nugget effect, C 1 and C 2 = structured component, a 1 and a 2 = ranges of influence, total C x  = sill and C 0/total C x  = proportion of variance accounted for by C 0
Full size table

On the whole, these results confirm the main controlling factors identified for the PTEs investigated, especially the identification of a basalt domain for Cr, Cu, Ni and V. While the semi-variogram is very useful for identifying spatial controls over elevated element concentrations, reduced domain concentrations cannot be identified using this method.

Typical threshold values

Normal background concentrations

In order to assess what is a typical concentration of elements in Northern Irish soils, both the NBC statistical methodology developed by Cave et al. (2012) and the ULBL statistical methodology (Jarva et al. 2010) were applied to data within the defined domains.

Figure 5 gives a visual example of how the NBC methodology was applied. The distribution of Ni within the basalt domain was assessed using a histogram; the values fell within the parameters set by Cave et al. (2012) (Fig. 5a) allowing a Gaussian approach to be adopted for calculating percentiles (Fig. 5b). A bootstrapping method was applied to calculate the uncertainty surrounding the percentile values (Fig. 5b) and the 50th, 75th and 95th percentiles and their associated uncertainty are plotted in Fig. 6.

Fig. 5
figure 5

Outputs derived from NBC methodology using R-scripts developed by Cave et al. (2012) for nickel’s basalt domain showing a histogram and b percentiles and relative uncertainty computed using the empirical, gaussian and robust methods

Full size image
Fig. 6
figure 6

50th, 75th and 95th percentiles of each of the elements’ domains along with their respective 95 % upper and lower confidence limits (vertical lines shown), ULBL concentrations and SGV/GAC where R residential, A allotment, C commercial. Previous SGVs for Pb have been withdrawn; elsewhere SGVs/GACs that are not given are beyond the scale of the graphs

Full size image

Elevation of Cr, Cu, Ni and V in the basalt domain is obvious from Fig. 6, while reduced concentrations of these four elements are seen in the Mournes and peat domains. For Cr, the differences between the domains are maintained throughout the 50th, 75th and 95th percentiles. The upper 95 % confidence limits of the 95th percentiles for the basalt, Mournes, peat and principal domains are 460, 84, 150 and 290 mg/kg, respectively. The Mournes and peat domains contain substantially lower concentrations than those in the principal domain.

Similar results are shown for Cu at the 50th and 75th percentiles, but the 95th percentile shows a slight skew in Cu concentrations in the peat domain, with a higher concentration calculated at the 95th percentile in the peats than in the principal domain. The upper 95 % confidence limits of the 95th percentile for the basalt, Mournes, peat and principal domains are 130, 41, 68 and 59 mg/kg. The distribution of Cu in the peat domain is heavily right-skewed, with a large presence of outliers. However, a log-transformation of the data brought it within the skewness limits set by Cave et al. (2012), and the Gaussian approach was followed for calculating percentiles. Another possible explanation for this distribution is the existence of another controlling factor, other than solely peat substrate, which is contributing to the concentrations of Cu in this domain.

A large degree of uncertainty is associated with the values generated for Ni and V in the Mournes domain. For Ni, the 95th percentile was calculated as 24 mg/kg, with the lower and upper confidence limits calculated as 12 and 170 mg/kg, respectively. The NBC value calculated for the Ni Mournes domain, of 170 mg/kg would therefore appear to be unrealistic, as the maximum value of Ni encountered in this domain was 37 mg/kg. The Mournes domain for Cu, Cr, Ni and V is based on the same 73 data points, making it the smallest domain, but still exceeding the 30 data points recommended in the NBC methodology (Cave et al. 2012). Reasonably, large uncertainty is also calculated for vanadium, with the lower and upper confidence limits of the 95th percentile calculated as 39 and 174 mg/kg, respectively. In comparison, much less uncertainty is shown for chromium and copper, where the differences between the upper and lower confidence limits for the 95th percentile are 40 and 22 mg/kg, respectively. It seems that the distributions of these elements in the Mournes domain are responsible for the degree of uncertainty associated with the percentiles calculated. This may be due to the fact that the Mourne granite complex has several fabrics associated with fractionation of basaltic and crustal rock melts (Meighan et al. 1984; Stevenson and Bennett 2011). For V and Ni, a larger occurrence of outliers means that the box-cox transformation was applied but higher uncertainties were still recorded for the 95th percentiles using this method. Fewer outliers present for Cr and Cu reduced the uncertainty associated with the percentiles, allowing for the calculation of more realistic NBCs.

The existence of outliers in vanadium’s peat domain saw the application of a log-transformation in order to bring its distribution closer to normal. However, the outliers have still affected the calculation of the 95th percentile (120 mg/kg) and its confidence limits, while the 50th (27 mg/kg) and 75th (49 mg/kg) percentiles appear to remain more representative. Nickel’s peat domain contained even more outliers, meaning in this case, the box-cox transformation was required to bring the data within the necessary skewness limits (Cave et al. 2012). In this instance, the box-cox transformation appears to be more effective in reducing the effect of the outliers, meaning the 50th, 75th and 95th percentiles (7, 14 and 46 mg/kg) seem to give appropriate concentrations when compared with the mapped outputs (Fig. 2).

For As, elevated concentrations are seen in the two mineralisation domains and the Shanmullagh domain when compared with the principal domain. If the 95th percentile is used as a comparison, the mineralisation 2 domain result is approximately five times greater than the 95th percentile for the principal domain. However, a large degree of uncertainty is associated with the 95th percentiles in the mineralisation 2 domain, as the lower and upper confidence limits range between 56 and 85 mg/kg, respectively. The data for this domain were box-cox transformed and theoretical percentiles calculated based on the mean and standard deviation of the data set after an assessment of its distribution. However, the presence of one extreme outlier which lies over 60 mg/kg away from the remainder of the outliers is likely to be causing the extreme skew seen in the calculation of the 95th percentile. For the other three domains, mineralisation1, principal and Shanmullagh, the assessment of the data distribution led to robust percentiles and uncertainties being calculated. Although outliers are also present in these domains, none of them contain outliers as extreme as the one identified for the mineralisation 2 domain. This causes much smaller differences between the upper and lower 95th confidence limits than those identified for the mineralisation 2 domain.

For Pb, elevated concentrations are obvious in the urban domain, followed by the mineralisation, Mournes and peat domains. The lowest concentrations of Pb are found in the principal domain. Within the urban domain, reasonably large differences occur between the upper and lower confidence limits, as they range between 240 and 300 mg/kg for the 95th percentile. This is to be expected in the urban domain as anthropogenic sources of Pb increase the amount of outliers present in the data set, which in turn increases the uncertainty associated with the percentiles calculated. Although these values of lead are high compared with the principal domain for Northern Ireland, where the 95th percentile ranges between 71 and 77 mg/kg, they are not as high as those found in some other urban areas. Chirenje et al. (2004) reported a 95th percentile of Pb concentrations in Miami, USA, as 453 mg/kg.

Figure 6 shows both the strengths and the weaknesses of the NBC methodology. A large presence of outliers within the data set causes issues in the distribution assessment and ultimately in the uncertainty calculations. This stems from the overall distribution of the data, and the effectiveness of the transformation applied and is particularly obvious in the Mournes domain for Ni and V. It is important to note that the NBC methodology is not meant to be used for reduced concentration domains, which the Mournes domains for Ni and V are examples of. However, even for As and Pb where all the domains contain elevated concentrations, the high degree of uncertainty associated with some of the domains causes unrealistic concentrations at the upper 95 % confidence limit of the 95th percentile, which is likely to pose difficulties if attempts are made to use NBCs during risk-based assessment of contaminated sites rather than just identifying potentially contaminated land as legally defined in the UK, as was originally intended. Also, this raises the question as to whether the UCL of the 95th percentile is an effective means of differentiating between diffuse and point contamination from anthropogenic sources.

From Fig. 6, it is clear that the values shown for the 50th percentile provide a more realistic representation of the comparison between the domains, i.e. for Ni, the 50th percentile shows elevated concentrations in the basalt domain (100 mg/kg), widespread concentrations of 27 mg/kg in the principal domain and reduced concentrations of 4.0 and 7.2 mg/kg in the Mournes and peat domains, respectively. However, this median value should not be taken as the typical threshold value as it does not fulfil the aim of TTVs as while median values are a central value for the domain, they do not allow for a differentiation between diffuse and point source anthropogenic contaminations. However, considering the 50th percentile may be effective within sectors other than the contaminated land sector, as the 50th percentile values can provide useful information on reduced concentration zones, where a possible depletion of essential elements, such as copper, could have consequences for industries such as agriculture.

Upper limit of geochemical baseline variation

TTVs calculated using the ULBL methodology are also shown on Fig. 6. The NBC value (upper confidence interval on the 95th percentile) calculated for copper’s peat domain (68 mg/kg) is higher than the value calculated for the principal domain (59 mg/kg) despite outputs from the ECDF domain identification method in Fig. 2, suggesting that the peat is an area of depleted copper concentrations. In comparison, the ULBL method provides a more accurate representation of the values expected in reduced domains, with the peat, Mournes and principal domains containing ULBL concentrations of 47, 27 and 76 mg/kg, respectively.

At elevated concentrations, both the ULBL and NBC methods appear to calculate similar values, with the ULBL method generally calculating slightly lower values for Pb. With regard to the elevated concentrations in the basalt domain for Cr, Cu, Ni and V, the ULBL method calculates concentrations that are at least 20 % higher than the respective NBC. This is probably due to the fact that the distribution of these elements in the basalt domain is relatively homogeneous and therefore closer to a normal distribution. When the boxplot is used to identify outlying values for a more normal distribution, fewer values will be identified, and so, a higher typical threshold value will be set using this method. Although set methods are used in the NBC methodology depending on the distribution of the data, a major strength of the boxplot is its resistance to different types of distribution.

Comparison with relevant criteria

Figure 6 also provides details of SGVs and GACs where they are available for the elements. SGVs and GACs are used to “represent cautious estimates of levels of contaminants in soil at which there is considered to be no risk to health or, at most, a minimal risk to health” (Defra 2012). Therefore, they are based on a different approach than that behind the NBC methodology where the aim was to identify sites where “if nothing is done, there is a significant possibility of significant harm” (Ander et al. 2013a). However, a comparison can still be drawn between the values. Figure 6 highlights certain domains for a number of the elements where the typical threshold values are higher than the reference values for residential, and in some cases allotment end uses. Therefore, depending on the size of the exceedance, surpassing these SGV values could suggest a possible risk to human health. For As, the residential SGV of 32 mg/kg (Martin et al. 2009a) is narrowly exceeded in the mineralisation 2 domain and is therefore unlikely to pose significant risks to human health. GAC for Cr-VI are shown on Fig. 6 (Nathanail et al. 2009), with exceedance of these concentrations shown in all domains. However, no distinction was drawn between Cr-III and Cr-VI in the Tellus survey. For Ni, the basalt domain shows a significant exceedance of the residential SGV (130 mg/kg) (Martin et al. 2009b) using both the NBC (200 mg/kg) and the ULBL (250 mg/kg) methods; however, recent studies by Barsby et al. (2012), Cox et al. (2013) and Palmer et al. (2013) indicate the oral bioaccessibility of Ni in these soils is relatively low (1–44 %). The NBC for Ni in the Mournes domain, while high (170 mg/kg), is not a representative value when compared with mapped outputs (Fig. 2), and all values for Cu fall within the GAC (Nathanail et al. 2009) making them unlikely to pose significant risks to human health. The NBC for vanadium in the Mournes domain again is unrepresentatively high, with the ULBL method calculating a value of 46 mg/kg, which appears to be more representative of mapped outputs. All vanadium results exceed the allotment GAC with values for the basalt, peat and principal domains also exceeding the residential GAC (Nathanail et al. 2009). However, bioaccessibility testing reported in Barsby et al. (2012) and Palmer et al. (2013) suggests that only a small fraction of total V in these areas is bioaccessible (8 %).

Conclusions

In terms of domain identification, the three methods exhibit specific advantages and disadvantages. The k-means technique provided useful results in the determination of elevated domains but its applicability in practice would be limited as it cannot be used to define reduced concentration domains. The boxplot and ECDF methods both allowed identification of elevated and reduced concentration domains. However, the boxplot method splits the distribution at arbitrary values, whereas changes in gradient linked to different data distributions within the overall data set are used to divide the ECDF graph. Splitting the ECDF graph requires a level of interpretation by the individual completing the work, which introduces potential for bias as a level of knowledge of the modelled domains could influence the results. Of the three methods, the ECDF methodology provides the greatest amount of detail and opens the methodology to other practical applications rather than just identification of land contamination. However, choice of method may ultimately lie with the decision maker as whatever method they choose may depend on the original goals behind the use of this methodology.

The NBC methodology has been developed to sit within a specific legislative framework. By defining the NBC as the upper 95 % confidence limit of the 95th percentile, it generates a question as to how conservative the approach taken is. In contrast to this, the ULBL methodology generates the maximum number of outliers using a boxplot of nontransformed data, in order to generate the lowest ULBL concentration. This is demonstrated in Fig. 6, where generally the ULBL values calculated are slightly lower than the NBC concentrations. A notable exception to this is for Cr, Cu, Ni and V in the basalts domain, where on all four occasions, the ULBL method calculated higher concentrations than the NBC methodology. The largest difference was for Cu, where the ULBL method calculated a concentration 28 % higher than the NBC. As well as this, the distribution of the data within each of the domains seems to have a large control over the amount of uncertainty calculated for the relevant percentiles in the NBC method, particularly where a large amount of outliers are present. The transformations applied mainly account for the skewed distributions, but examples remain where the data transformation does not seem to be fully effective (vanadium’s peat domain). It is clear from the previous discussion that both methods have their strengths; however, in general, the ULBL concentrations provide more realistic concentrations for typical threshold values as defined in this study, across the area and elements shown.

An interesting investigation, following on from this work, would be to consider the geographical location of each of the outliers identified using the ULBL method. If specific sources of elements could be identified as causing the elevated concentrations of the outlying values, then clarification of how effectively the method discriminates between anthropogenic diffuse and point source concentrations of elements could be gained.

One of the primary aims of this research was to investigate soil geochemical data for Northern Ireland and determine an output in the form of relevant TTVs, which define the boundary between geogenic and diffuse anthropogenic source contributions to soil, and those associated with point sources. In this respect, the following is suggested for use;

  • ECDF mapping method for establishing the main controls over PTE concentration distributions and allowing the identification of domains,

  • Calculation of TTVs using the ULBL method (currently employed in Finland) within each of the defined domains.

These values will be of interest to a number of parties, as they indicate what a “typical” concentration of an element would be within a defined geographical area. These values should be considered alongside the risk that each of the PTEs poses in these areas, in order to determine potential risk to receptors.