Mapping the contaminant legacy of a coking plant, the avenue, chesterfield, UK

Abstract

With 300,000 ha of contaminated land, 1.2% of the Britain land area [13, 14], the UK has a major need for effective environmental risk assessment for land remediation and reclamation [22]. Such a risk assessment is usually based on the characterization of potential site contaminants and analysis of source "pathway" target scenarios [8, 9, 11]. A risk-based contaminant description requires a conceptual model of the site that includes qualitative and quantitative analyses of pollution sources, contaminant pathways and pollutant receptors [4, 22]. This characterization typically has to rely on limited and irregularly distributed point data. In addition, soil and surface material heterogeneity, as well as the quasi-random nature of contamination sources add to the complexity of developing good spatial models of pollutants on old industrial sites [13, 22]. Improved and appropriate geostatistical tools and GIS based analysis can help to overcome some of these problems. This paper tackles the development of such a methodology for a former coking plant by examining the sources and pathways of Polycyclic Aromatic Hydrocarbons (PAHs), as part of an analysis of a wider range of contaminants at the Avenue Coking Works, near Chesterfield, UK (Fig. 1). In the UK, coking works were established alongside the iron and steel plants from the mid 18th century. By the end of the 19th century surplus gas from coking works was sold as town gas, and by 1912 coke ovens were being installed at town gas works [10]. Each works occupied between 0.3 and 200 ha. By 1995, only four of the total of 400 such works were still operating [10]. Tar distillation took place on coal and, or coke works sites, and was the primary source of organic chemicals for different industries until petrochemical products took over in the 1960s [10]. The contamination at former gas and coke works varies with the range of products and by-products manufactured. On such sites, ground contamination arises from by-products, waste products from landfills and lagoons, and ancillary products such as ammoniacal liquor, coal tar, spent oxide and foul lime [12]. The organic contaminants are derived from constituents of coal tar such as aromatic hydrocarbons, polycyclic aromatic hydrocarbons (PAHs), phenils and phenols, nitrogen compounds, and organo sulphur compounds, natural gas processing compounds such as alcohol, glycols, resins, heavy oils, and organic fuels such as petroleum and naphthalene [12]. Several of these coking works were developed on sites with a long industrial history, adding further contamination to that derived from the original activity. This study sets out to construct a conceptual model for the Avenue site that distinguishes local point source PAH16 (Polycyclic Aromatic Hydrocarbon) pollution arising from the coking plant activities, from the general historical diffuse pollution caused by other older industrial operations on the same site. Legacies of later phases of pollution that contribute to the same type of contamination can hide the special pattern of diffusion of contamination due to the earlier phase of industrial activity. One common problem in environmental risk assessment is to determine the value of a continuous attribute at any particular unsampled location; the uncertainty of any unsampled value; and the probability that a regulatory threshold for soil pollution or a criterion for soil quality is exceeded at any unsampled location, when only few sampled values are known [6, 7, 15, 16, 17, 19]. Geostatistics provide the basis for analysing data that vary continuously spatially and permit the inference values of the same variable at unsampled locations through interpolation techniques. Two key assumptions in geostatistical analysis are that (1) sample values are expected to vary continuously from one location to another; (2) at any particular location the value of the variable comprises a fixed component of the variation trend, which is usually unknown, and a random variable following one specific distribution [5, 15, 19] expressed by: z(x) = a Z(x) (1) z(x) = value of the variable z at location x; "a = fixed unknown component of the variation trend" Z(x) = random variable described by: Z(x) = m(x) + (x) + (2) m(x) = deterministic function that describes the structural component with a constant mean or trend [3]; (x) = a random but spatially correlated component, known as the variation of the regionalized variable, and it is the locally varying but spatially dependent residual of m(x); = is residual, spatially independent noise, having a mean of zero and a standard deviation or variance 2. If the assumption that an element varies continuously over a certain area is true, then it is customary to assume that the value at any point will be influenced much more by a closer known value of that element than by one farther away. Interpolation techniques are based on a normal or Gaussian distribution of data, and good spatial correlation. Problems arise when potentially continuous processes have not yet led to a normal spatial distribution, yet overlie an older continuous, but random process, which is spatially correlated. Lark [21] models complex soil properties by assuming that the soil contamination is formed by a continuous but random component combined with a quasi point process. The quasi point process characterizes contamination (or any other process) in a small area of finite extent, which is represented by only one (or very few) soil sample(s) and does not diffuse continuously towards its neighbours. The continuous random processes are representative for the native metal content of the soil parent material and diffuse sources of pollution, while the quasi point process is defined by localized point sources of pollution. This situation may describe the pollution of an industrial site. If we consider that contamination with the same pollutant can result from both diffuse and point sources, we can expect its measured values to show very little spatial correlation, if any. These values, which have the point process values embedded, are called outliers and are considered unusual in their spatial context. The outliers do not belong to the continuous, but random, distribution of the majority of data, and are not necessarily extreme low or high values [2, 23]. In the case of pollution, if the outliers are not statistical anomalies due to errors in measurement or recordings, they indicate different processes superimposed on the same area and affecting the same variable [2, 18, 24]. © 2009 Springer Berlin Heidelberg.