Lithological and pedological influences on the ... - KUNDOC.COM

Science of the Total Environment 382 (2007) 351 – 363 www.elsevier.com/locate/scitotenv

Lithological and pedological influences on the magnetic susceptibility of soil: Their consideration in magnetic pollution mapping Monika Hanesch ⁎, Gerd Rantitsch, Sigrid Hemetsberger, Robert Scholger Department of Applied Geosciences and Geophysics, University of Leoben, Peter Tunner Str. 25, 8700 Leoben, Austria Received 22 November 2006; received in revised form 26 March 2007; accepted 3 April 2007 Available online 16 May 2007

Abstract Magnetic susceptibility measurements are widely used to map and monitor the heavy metal pollution of soils. However, the magnetic properties of soils are influenced significantly by the bedrock lithology and soil-forming processes. Therefore, a main challenge in the data interpretation is to filter out the anthropogenic pollution signal. In this study we address this problem by analysing susceptibility values, heavy metal concentrations, as well as pedological parameters in a large soil data set from the eastern segment of Austria, covering a wide range of different lithologies and soil types. The statistic assessment demonstrates an influence of lithology and soil type on the magnetic susceptibility signal. Therefore anomalies are defined in sub sets of different soil types separately. Three different methods were applied to detect susceptibility anomalies: the median absolute deviation method, the boxplot method, and the population modelling method. These methods evaluate topsoil data only and can therefore also be applied to field measurements of magnetic susceptibility. The results were compared to the conventional method of calculating the difference of topsoil and subsoil susceptibility. All three approaches identify the main anomalies in the study area and are successful in circumventing the problem of erroneous anomaly definition due to pedological processes. However, knowledge of the lithological background is still necessary for a meaningful interpretation and can only be substituted by a large amount of data. The tested methods lead to thresholds of different height and therefore act as filters of different strength for the definition of anomalies. © 2007 Elsevier B.V. All rights reserved. Keywords: Environmental magnetism; Soil pollution; Anomaly recognition

1. Introduction Society becomes more and more aware of the necessity to conserve and protect our natural resources. In addition to air and water, soil is an important resource for life on earth. Soil has the particular quality to collect and store substances which are inserted into it over many years. ⁎ Corresponding author. Tel.: +43 3842 402 2608; fax: +43 3842 402 2602. E-mail addresses: [email protected] (M. Hanesch), [email protected] (G. Rantitsch), [email protected] (R. Scholger). 0048-9697/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2007.04.007

Some of these, especially the heavy metals, may harm plants, animals or human beings. To evaluate the degree of pollution, heavy metal concentrations in soil samples are analysed chemically. As these analyses are timeconsuming and expensive, they cannot be repeated regularly and there are limited possibilities to condense the sample grid. However, during recent years, magnetic susceptibility mapping was developed as a fast and cost effective technique to analyse more soil sites and to monitor temporal changes in the environment (Petrovský and Ellwood, 1999). Magnetic susceptibility can be used as a proxy for the accumulation of heavy metals because both, magnetic particles and inorganic pollutants are

352

M. Hanesch et al. / Science of the Total Environment 382 (2007) 351–363

produced together during industrial processes. Their close relationship has been proven by combined analyses of chemical and magnetic data (Heller et al., 1998; Bityukova et al., 1999). Measurements of magnetic susceptibility have been used to assess environmental pollution on a regional scale, e.g. along highways and roads (Hoffmann et al., 1999; Moreno et al., 2003), around power plants (Strzyszcz et al., 1996; Kapička et al., 1999), and steel smelters (Hanesch et al., 2003). Magnetic susceptibility measurements are also useful to map the degree of anthropogenic input in a large scale (Hay et al., 1997; Magiera et al., 2002; Hanesch and Scholger, 2002). However, two problems (among others) arise in the evaluation of large magnetic susceptibility data bases. Firstly, the complex soil system demands a careful evaluation of the factors which control the magnetic properties of the sampled medium. An extensive data base can contribute to a better understanding of this system. However, such data are rarely available for magnetic studies. Secondly, in magnetic susceptibility mapping, more or less subjective detection techniques were applied in most studies. During the last years, several promising approaches were proposed to identify background values and anomalous values of environmental proxies (Matschullat et al., 2000; Reimann et al., 2005). Their usefulness in magnetic pollution mapping is not tested so far. Both problems are addressed in this paper. The Austrian soil survey provides a large scale data base which comprises extensive information about the investigated soil system. To explore the influence of bed

rock lithology and soil properties on the magnetic susceptibility values measured in soil, this data base is completed here by magnetic susceptibility data. We present a large scale susceptibility map which covers the eastern segment of Austria, comprising the provinces Burgenland, Carinthia, Lower Austria, Upper Austria, and Styria (Fig. 1). This area includes highly industrialised regions as well as regions almost untouched by anthropogenic influences. Within the study area, geological background (Fig. 1), topography and climate vary significantly. The relation between bed rock lithology and soil type is used to interpret maps of the magnetic susceptibility. Alternatives are given by applying different anomaly detection methods which can also be applied to field measurements of magnetic susceptibility. The suitability of these methods is tested by comparing it to the conventional method of subtracting subsoil from topsoil values which has previously been proven to reliable identify anomalies in susceptibility maps (Hanesch and Scholger, 2002). The results provide a procedure to identify polluted areas in a complex environment. 2. Materials and methods 2.1. Samples and analytical methods The soil samples were taken in the years 1990 to 1997 during the soil surveys of the Austrian provinces in a base grid of 3.9 to 3.9 km. In Lower Austria, Upper Austria and Carinthia, additional sites in a grid of 2.75 to

Fig. 1. Geology of the Austrian provinces (redrawn from Ebner, 1997). Sampling sites are shown as dots.


2.75 km were taken. In some areas in Carinthia, the grid spacing is condensed to 1.38 to 1.38 km. Within the five provinces, 3363 sites were sampled (Fig. 1). For each sampling site and depth, a composite sample of 1 kg was produced from four sub samples taken 10, 8, 6 and 2 m from the sampling point towards the cardinal directions. Depth layers were sampled which were not uniform for all provinces. Therefore, a weighted mean for the depths 0–20 cm and 20–50 cm is used as representative top- and subsoil values. For example, the weighted mean for topsoils sampled in four layers was calculated by χtopsoil = (χ0–5 cm + χ5–10 cm + 2χ10–20 cm) / 4. An exception is the data base of Upper Austria where the subsoil value corresponds to a depth of 20–40 cm. The composite sample was dried at a maximum temperature of 30 °C and passed through a 2 mm sieve. Heavy metals were analysed after aqua regia acid digestion by AAS, ICP-OES or ICP-MS (Bundesanstalt für Agrarbiologie, 1993; Bundesanstalt für Bodenwirtschaft Wien, 1994; Bundesamt und Forschungszentrum für Landwirtschaft, 1996; Amt der Steiermärkischen Landesregierung, 1998; Amt der Kärntner Landesregierung, 1999). The pedological information and heavy metal data were taken from the soil information system BORIS of the Umweltbundesamt Wien (Federal Environment Agency, Vienna; http://www.borisdaten.at). Magnetic susceptibility of the dried and sieved samples was measured in the soil archives. The samples were stored in different plastic boxes and, depending on the sample geometry, measured with an Exploranium KT9 instrument or a Bartington MS2C loop sensor for cores. If not enough material was available for these measurements, the sieved sample was filled in 10 cm3 plastic boxes and measured in the KLY2Kappabridge. In the end, all measured values were calibrated to mass specific values measured at the KLY2Kappabridge. The exact procedure to ensure the comparability of the results for the different provinces is described in Hanesch and Scholger (2002). 2.2. Exploring the relationship between the used variables The locations were separated in groups of different bedrock lithology and soil type. Only those locations were included where these variables are well defined. Regional pollution studies usually explore only two or three different lithologies and soil types. To see the impact of a differential diagnosis, we applied the nonparametric Mann–Whitney and Kolmogorov–Smirnov test (Davis, 2002) to test groups of topsoil susceptibility values pair wise for statistical significant differences. We assume a real difference in population only if both tests indicate a difference at the significance level of p b 0.05.

353

2.3. Detection of anomalous values The detection of anomalous values is one of the crucial tasks in the exploration of environmental data bases. Within soils, the simplest way to identify anthropogenic contaminations is the comparison of top- and subsoil samples. However, as a major problem, pedogenic processes within the soil profile can obliterate any indicative differences in a contamination related environmental proxy. In addition, this method is not longer applicable if the soil profile is not sampled in two horizons. Several statistical approaches were reviewed by Reimann et al. (2005). To identify extremes (values in the tails of a statistical distribution) and outlier (values belonging to a different population) they proposed a procedure which is based on an inspection of empirical distribution functions and boxplots, and the construction of fences defined by median and median absolute deviation (Tukey, 1977). However, in the presence of polypopulational data distributions this approach is not longer useful. In this case, a decomposition of empirical into parametrical data distribution functions can provide the base for the identification of the controlling factors and the estimation of background values (Rantitsch, 2004). To investigate the performance of these approaches they are applied here to construct alternative anomaly maps. 2.3.1. Topsoil–subsoil difference Anthropogenic input of magnetic particles leads to their accumulation in the upper part of the soil and, hence, to an enhancement of susceptibility in the topsoil. Usually it is restricted to the uppermost 5 cm (Kapička et al., 2001). Therefore, a higher value in the topsoil (0–20 cm) with respect to the subsoil indicates an anthropogenic anomaly. Following Hanesch and Scholger (2002) anthropogenic anomalies are defined if the susceptibility difference exceeds 20 · 10− 8 m3 kg− 1, and geogenic anomalies if it is below −20 · 10− 8 m3 kg− 1. 2.3.2. Median absolute deviation method The median absolute deviation (MAD) is defined as the median of the absolute deviations from the median of all data (Tukey, 1977). The median value ± 2 MAD defines a fence (Bn) which separates outlier and extremes from a population. 2.3.3. Boxplot method Dependent on the empirical cumulative distribution plots, normal or lognormal box plots are constructed. The box length is the interquartile range. Outliers are values between 1.5 and 3 box lengths from the upper or lower edge of the box. Far outliers are values more than 3 box

354


Table 1 Measured susceptibility values (10− 8 m3/kg) in the study area Susceptibility

Topsoil (0–20 cm)

Subsoil (20–50/40 cm)

50–70 cm

Number of sites Minimum Lower quartile Median Upper quartile Maximum

3357 2.6 18 26 46 1147

2157 2.2 16 24 45 1286

681 3.9 17 25 47 1220

Subsoil is defined as the depth layer 20–50 cm, an exception is Upper Austria where the layer 20–40 cm was sampled. The values of the layer 50–70 cm are shown for information but are not used in this study.

2.3.5. Quantification of anomalous values by the Geoaccumulation Index For the methods using only topsoil data, a threshold has to be defined beyond which we regard a value as anomalous. We calculate here the Geoaccumulation Index Igeo as defined by Müller (1979): Igeo ¼ log2ðCn=1:5BnÞ where Cn is the measured elemental concentration and Bn is the upper limit of the background concentration as determined by the approaches described above. 3. Susceptibility values

lengths from the edge of the box (Tukey, 1977; see also Reimann et al., 2005). Bn is defined here as the whisker of the boxplot (1.5 box lengths from the box edge). 2.3.4. Population modelling method This approach tries to decompose the observed probability functions in components which can be described by a normal or a lognormal function (e.g. Rantitsch, 2004). If a chi-square test (with a significance level α of 0.05) confirms the statistically significant derivation of the observed data from a parametrical probability function, it is assumed that this function is an outcome of the process which controls the data variability (Rantitsch, 2004). In a first trial, a lognormal probability density function was fitted to the susceptibility values of each soil type. A statistically significant fit indicates the soil type as controlling factor. If the fit was statistically insignificant, the underlying bedrock lithology was used to define sub sets of the sample group. A 2σ-approximation of the background population is used to define Bn as the threshold between anomalous and background samples.

The study area covers a wide range of different lithologies (Fig. 1) and therefore, the susceptibility values are spread over a wide range (2 to 1286 · 10− 8 m3/kg). Table 1 lists some statistics for soil layers in a depth of 0–20 cm, 20–50 cm, and 50–70 cm. It is obvious that the position, shape and dispersion of the data values do not differ between the depth layers (Fig. 2) if we regard the whole data set as one entity. Fig. 3 shows a map of the topsoil susceptibility values (0–20 cm) created by inverse distance weighing of the measured values. 4. Relationship between magnetic susceptibility, bedrock lithology, soil type, and heavy metal accumulation 4.1. Comparison of soils over different bedrock types Groups of soils over different bedrock types were separated from the data set. The boxplots of the susceptibility values for these groups demonstrate the significant effect

Fig. 2. Histogram of topsoil (0–20 cm) and subsoil (20–50 cm) susceptibility (10− 8 m3/kg) for the study area.


355

Fig. 3. Contour plot of magnetic susceptibility values in the topsoil (10− 8 m3/kg) with results of the difference method of anomaly definition: Plus signs mark sites where the difference between topsoil and subsoil exceeds a value of 20 (anthropogenic anomaly), circles mark sites where the difference is smaller than − 20 (geogenic anomaly).

of the parent material lithology on the susceptibility values measured (Fig. 4). The lithology groups relate to the geological units in Fig. 1 in the following way: granitoide (Bohemian Massive — granites), gneiss and granulite (Bohemian Massive — metamorphic rocks;

Alps — Austroalpine Crystalline Complex), phyllite (Austroalpine Crystalline Complex), mica schist (Austroalpine Crystalline Complex), siliceous sandstone (Flyschzone), limestone and dolomite (Mesozoic sediments), loam (Neogene sediments), loess (Neogene

Fig. 4. Boxplots of topsoil susceptibility values (10− 8 m3/kg) for different parent material. Circles mark outlier (1.5 to 3 box lengths from the box edge), asterisks mark far outlier (more than 3 box lengths from the box edge). BM: sites in the Bohemian Massif.

356


Table 2 Results of the statistical tests for different bedrock types (BM: sites in the Bohemian Massif) BM: granitoide BM:granitoide (N = 309) BM:gneis, granulite (N = 168) Alps: gneis, granulite (N = 82) Phyllite (N = 37) Mica schist (N = 112) Sandstone (N = 83) Limestone, dolomite (N = 46) Loam (N = 318) Loess (N = 289)

– XX XX XX XX – – XX

BM: gneis, granulite

Alps: gneis, granulite

XX XX XX XX – XX XX

XX – XX – XX XX

Phyllite

XX XX XX XX –

Mica schist

XX – XX XX

Sandstone

XX XX XX

Limestone, dolomite

– XX

Loam

Loess

XX

XX indicates a significant difference between the two tested groups. The results of the Mann–Whitney and the Kolmogorov–Smirnov test were identical. N is the number of samples in the group.

sediments). In Table 2 the results of the statistical tests are shown. The susceptibility values on sampling sites over sandstones are clearly distinguished from all other groups. The two groups with the highest susceptibility values (phyllite and loess) also differ significantly from those with lower values. But no significant difference is found between phyllite and loess. The two groups within the Bohemian Massif behave similarly. They are well distinguished from most of the other groups. It is striking that the susceptibility values of gneis and granulite in the Alps differ from the same lithology in the Bohemian Massif. Sites on limestone and dolomite do not form a distinguishable group. These materials have a very low susceptibility of their own. Therefore, the susceptibility in the topsoil might depend more on soil-forming factors than on the parent material. Some inhomogeneous sediment groups (e.g. fluvial sediments, moraine) are not shown in this comparison, because their susceptibility is strongly dependent on the original material from which they were formed. 4.2. Comparison of different soil types To evaluate the soil type influence on the susceptibility values the data set is divided by the soil types (Table 3, Fig. 5). High susceptibility values in colluvium, chernozem and para-chernozem differ significantly from all other groups. These soils were all sampled in Neogene sediments at the eastern margin of the Alps. Half-bog soils and podzols, the two groups with the lowest susceptibility values, are distinct from all other groups. They form a group of soils with low pH-values and a high amount of organic matter in the topsoil. The presence of more organic complexing agents leads to release and relocation of iron. Stagnosol cannot be distinguished from gleysol and from stagnic phaeozem. Also the two latter groups are clearly distinct from each

other. These three soil types have relatively low susceptibility values. They show reductomorphic characteristics caused by the influence of water. Fluvisol is also influenced by water but it lacks reductomorphic characteristics. It belongs to groups with intermediate susceptibility values which cannot be distinguished well. Some of them are inhomogenous by definition because they form on a wide variety of parent materials, e.g. fluvisol, relocated material and cambisol. Table 3 Groups of different soil types and their German classification in italics Soil type

Number of Number of topsoil samples subsoil samples

Half-bog soil, Anmoor Fluvisol, Auböden Gleysol, Gley Leptosol, Rendsina und Ranker Podzol, Podsol Stagnosol, Pseudogley Relic soil, Reliktböden Chernozem, Tschernosem Para-chernozem, Paratschernosem Stagnic phaeozem, Feuchtschwarzerde Cambisol on rock, Felsbraunerde Cambisol on loose sediments, Lockersedimentbraunerde Luvisol, Parabraunerde Anthrosol, Kulturrohboden Relocated material, Halden- und Planieboden Colluvium, Kolluvium

14 161 199 129 24 234 98 334 32

12 107 151 79 16 152 58 183 23

96

62

847 924

508 638

30 109 19

18 69 13

65

39

Relic soils are soils which were formed in a different climate; partly they have undergone a recent soil formation. Para-chernozem shows the same profile as chernozem but was formed on non-calcareous sediments. Anthrosol is a soil changed substantially due to extensive agricultural use over a long time span. Relocated material in this paper is defined as a material transported by human beings. If material changed its place due to gravity, it is called colluvium.


357

Fig. 5. Boxplots of topsoil susceptibility values (10− 8 m3/kg) for different soil types.

4.3. Influence of bedrock lithology and soil type on magnetic susceptibility The results of the previous sections show that it is not easy to separate the influence of geology and soil type, or to decide which of them is more important. There are cases where the bed rock lithology is the leading influence, others where soil type is clearly dominating, and there also exist cases where the two influences cannot be separated at all. If loess is the parent material, usually the soil shows a high susceptibility. Loess favours the formation of chernozem which has a high humus content, leading to the retention of magnetic particles (Hanesch and Scholger, 2005). The cambisol on loess also shows higher susceptibility values than cambisol on granite, for example. But as soon as a soil develops reductomorphic features, this influence dominates — even on bed rock material which generally favours high susceptibility values. An example is the stagnic phaeozem on loess which develops due to water logging. The sites sampled on sandstone bedrocks provide an example for the close intertwining of the two influencing factors soil type and bed rock lithology. Obviously, sandstone bedrocks favour water logging: 46% of the sites are gleysol and stagnosol, 52% are cambisol, but 51% of these cambisol sites show reductomorphic features. This means that 73% of all sites on sandstone show reductomorphic features (Fig. 6). This explains the low susceptibility values measured on this bedrock

material. The soils on sandstone in this study are all located in the Austroalpine Flysch Zone (Fig. 1) where alternating layers of sandstone, claystone and marl were deposited. The clay and marl layers are responsible for the water logging. The leptosol group shows a large range of susceptibility values (Fig. 5). Therefore it was divided in three sub groups: rendzic leptosol (on calcareous bed rock), eutric leptosol (on marl with 2–70% CaCO3, “Pararendzina”) and ranker (on bed rock with less than 2% carbonate). No significant difference was found between the susceptibility distributions of these groups. This implies that carbonate content only has a negligible influence on susceptibility values.

Fig. 6. Soil types with sandstone as parent material. 73% show reductomorphic features leading to the low susceptibility values within this group of soils.

358


Fig. 7. Correlation of susceptibility with Cd and Zn in the Linz area. R is the Pearson correlation coefficient.

On the other hand, chernozem and para-chernozem display significantly differing susceptibility distributions. The only obvious difference between those soil types is the content of carbonate in the bedrock material. In our study area, both soil types occur in Neogene sediments at the eastern margin of the Alps. They experience the same climate. The difference between these two soil types needs a thorough investigation of additional pedological parameters. 4.4. Correlation between magnetic susceptibility and concentration of heavy metals In a previous study, significant correlations between susceptibility values and heavy metal values were found in a geogenic anomalous site, an industrial site and the city of Vienna (Hanesch and Scholger, 2002). Correlations can only be calculated regionally because they are controlled by the processes which trigger the dispersion of magnetic particles. Magnetic susceptibility values can be used as a proxy for pollutants but the specific heavy metal signature which is represented by the susceptibility values has to be identified for each area separately. The data of this study enable us to investigate the correlations for additional areas. Two examples are shown: the anthropogenic anomaly around the steel smelter in Linz and the geogenic anomaly in the northwestern part of Carinthia. In the Linz area, 5 sampling sites were chosen. All sampled depth layers were included. Therefore, 13 samples were available for the correlation analyses. Interestingly, the magnetic susceptibility values as well as the heavy metal values are normally distributed within this sample, probably due to the fact that anthropogenic input is the only source in this area and therefore the only factor determining these values. Zn, Cd, Pb, and Cr are correlated significantly

to susceptibility (Fig. 7). This finding corresponds to the data of Hanesch and Scholger (2002), which describe a correlation in an area around another steel smelter. In the northwestern part of Carinthia, some samples show rather high susceptibility values. We analysed the relationship between the magnetic susceptibility and heavy metal values for soils formed on metamorphic rocks in this area. These are mainly schists and gneisses and basic rocks. Six sites with a total of 36 samples yield Spearman correlation coefficients above 0.69 for the correlation with Co, Cr, and Ni, respectively (Fig. 8). For three sites with low susceptibility values, a linear relationship between Co and susceptibility exists (Fig. 8b). For these three sites, again, the values are normally distributed and the Pearson correlation coefficient is 0.56. The other three sites form distinct, linearly correlated groups. Consequently, in this area, susceptibility values can be used to discriminate different rock types. 5. Anomalies 5.1. Difference method Anomalies detected by the difference method are marked on the susceptibility plot in Fig. 3. The anomalies in the provinces Styria, Burgenland and Lower Austria were already identified and interpreted in the study of Hanesch and Scholger (2002). The largest anomaly is located in the mining and iron processing district north of Leoben. Leoben itself houses a large steel smelter. Noticeable features are the elevated values along the rivers Mur and Enns where industry and traffic are clustered in the valleys of the rivers. The large geogenic anomaly in southwestern Styria is continued in Carinthia. Some high values north of the lake Wörthersee are caused


359

Fig. 8. Relationship between susceptibility and Co for metamorphic rocks in northwestern Carinthia. a) The samples form distinct groups as shown by the different symbols. Note the lognormal scale. The non-parametric correlation coefficient (Spearman) is 0.69. b) A clear trend is discernible in the group with the lowest susceptibility values. The Pearson correlation coefficient is 0.56 (p b 0.05).

by local occurrences of meta-tuffs, the anomaly at the northwestern edge of Carinthia is due to the presence of magnetite bearing green schists (see Section 4.4). Generally, the geogenic values in the metamorphic core of the Alps are too high to clearly identify anthropogenic anomalies by this method. Where susceptibility values exceed 200· 10− 8 m3 kg− 1, a topsoil–subsoil difference of 20· 10− 8 m3 kg− 1 can easily occur only due to geological background value. This limitation of the magnetic mapping method was already described by Fialová et al. (2006) for basaltic rocks in Central Bohemia. Slightly elevated values are observed east of Vienna. The predominant chernozem in loess is a soil type with elevated susceptibility values and generally high topsoil–subsoil differences. For these soils, Hanesch and Scholger (2005) found a median susceptibility value of 77 · 10− 8 m3 kg− 1 and a median topsoil to subsoil susceptibility ratio of 1.5. The city of Linz, where a huge steel processing factory is located, is clearly identified from the surroundings by higher susceptibility values. The two smaller anomalies northeast of Linz are due to local limonite occurrences. The precedent discussion shows that only a thorough knowledge of geology and soil types in the investigated region can lead to a reliable interpretation of magnetic susceptibility maps by the difference method. The method finds its limitation where susceptibility values of the underlying lithologies are high. A further drawback of the method is the identification of anomalies in regions where pedological enhancement of magnetic suscepti-

bility takes place. These drawbacks could be circumvented by adjusting the threshold to the lithology and soil type present. But it would be more convenient to find a method of anomaly detection which is applicable to field measurements, i.e. which evaluates topsoil data only. Consequently three promising approaches are tested in the following sections. 5.2. Median absolute deviation method The application of median absolute deviation (MAD) fences for anomaly identification leads to a high number of elevated values (Fig. 9). Sites defined as polluted by the method coincide with the most conspicuous values on the susceptibility map. Generally, the MAD method has a tendency to define too many anomalous values, especially in the case of smaller data sets (Reimann et al., 2005). In this study, some soil type sub data sets contain less than 50 samples. Therefore it seems more appropriate to use a stricter anomaly definition. 5.3. Boxplot method The results of the boxplot method for anomaly detection are shown in Fig. 10. Basically, this map identifies all anomalies of Fig. 3. The method sets the limiting values lower than the MAD method. Therefore, it yields a more probable number of anomalies. Due to the separation of the data into sub data sets for the different soil types, the chernozem and parachernozem area in Lower Austria and Burgenland is

360


Fig. 9. Results of the median absolute deviation method for anomaly detection. High IGEO-values (larger than 1.5) mark sites which are suspect of being polluted.

now inconspicuous. On the other hand, some anomalous values in Upper Austria are identified which are due to local limonite occurrences. In this region, the difference method is more appropriate to identify the anthropogenic influence because it eliminates some of the geogenic anomalies.

5.4. Population modelling method The results demonstrate that most of the samples, which were characterized by soil type and bedrock lithology, can be fitted by a single lognormal probability density function. Therefore, it is concluded, that these

Fig. 10. Results of the boxplot method for anomaly detection. Sites which are not population members are suspect of being polluted. IGEO-values larger than 1.5 point to strongly elevated values.


361

Fig. 11. Results of the population modelling method for anomaly detection. Values higher than 1.5 mark sites which should be closer investigated due to elevated values. IGEO-values of more than 2.5 are only found for two sites close to iron mines.

attributes control the variability of the susceptibility values. The resulting map is presented in Fig. 11. Obviously, this method sets the strictest threshold values and therefore limits the number of defined anomalies to the really extreme values. But like the MAD and the boxplot method it suffers from a lack of differentiation between geogenic and anthropogenic values. A good knowledge of the geology of the region is therefore necessary. 5.5. Comparison of the methods The most extreme anomalies, geogenic and anthropogenic, are clearly identified by all methods: Leoben, Linz and the large geogenic anomalies in Carinthia. The difference method (Fig. 3) emphasizes large anomalies which comprise many different soil types. Local geological anomalies in a region with generally low values are only seen when the values are extreme. For example, in the Bohemian Massif, the susceptibility map shows two of the limonite occurrences (high susceptibility values) but they are not marked as anomalies by the difference method as the values are rather uniform throughout the vertical profile. The analyses of the soil type sub data sets indicate more anomalous values in this region. As far as fluvisols are concerned, no anomalous values are defined along the river Danube, but only at the Mur and the Enns. This coincides with the observation that along the Enns and the Mur, susceptibility values are generally elevated. This retainment of pol-

lutants in the valleys, however, is best visible in the conventional susceptibility map (Fig. 3). The population modelling method bears the great advantage from the elimination of effects caused by pedological processes. This procedure yields very good results especially for soil types depleted in magnetic minerals and for those enhanced in magnetic minerals. For the discrimination of geogenic and anthropogenic anomalies, however, it is better to regard the difference between topsoil and subsoil. Especially if subsoil data is not available, the results of magnetic susceptibility mapping can only be reliably interpreted by an expert with good knowledge of the local geology. The anomaly definition approaches of this study indicate the MAD method, boxplot method and population modelling method as increasingly strong filters. The results show that the MAD method tends to define too many anomalies, whereas the population modelling method restricts itself to the most extreme values. The threshold defined by the three methods for the different soil type groups are compared in Table 4. 5.6. Recommendations for the interpretation of field measurements The previous sections show that the three tested methods for the interpretation of topsoil values yield different results. The choice of method should therefore be done according to the data on hand and the aim of the

362


Table 4 Thresholds of the background values (Bn) of magnetic susceptibility (10− 8 m3 kg− 1) calculated with the different methods MAD

Fluvisol (brown) Fluvisol (grey) Gleysol Leptosol Stagnosol Chernozem on loess Stagnic phaeozem Cambisol on granite Cambisol on gneiss and granulite Cambisol on mica schist Cambisol on sandstone Cambisol on loose rocks Cambisol on coarse sediment Cambisol on loam Cambisol on loess Cambisol on sediments — mixed Relic soil Atypical soil

Boxplot

Population modelling Min

Min

Max

Min

15.86 15.98 9.99 11.77 9.76 53.78

59.75 46.78 31.65 101.11 34.37 114.63

8.24 9.00 5.91 3.32 5.36 31.31

Max

Max

112.99 10.95 76.33 106.89 10.66 80.87 55.14 6.67 45.54 451.20 6.71 216.50 68.09 7.03 51.02 176.69 17.05 108.10 58.53 123.51 9.64 38.68 5.48 77.02 7.13 46.08 14.70 34.34 9.83 50.61 11.65 43.36 15.46 37.04 10.09 56.97 11.66 49.45 8.90 97.57 2.78 314.04 10.57 21.85 7.52

5.76 161.26

35.86 –

7.65 62.64 2.48 211.15

–

4.66 119.85

16.21 111.62 6.07 348.87 11.41 178.68 16.05 37.29 10.04 63.27 12.21 50.92 19.75 73.38 10.71 142.92 15.40 104.47 20.74 73.64 10.77 135.43 – – 11.66 48.49 5.54 110.83 8.41 16.97 66.92 9.13 135.21 12.99

72.83 89.61

Atypical soil comprises colluvium and relocated material.

study. For all methods it will be helpful to form groups of values with uniform soil type and lithology. If this leads to groups which are too small, data should be grouped by soil type and interpretation has to be done by an experienced person with good knowledge of the lithology of the region. The median absolute deviation method detects too many outliers if small data sets are used. But it may be the method of choice in high resolution mapping for the purpose of early detection of changes in the environment. The boxplot methods reliably detected all known anomalies in this study without “false positives”. We recommend it as a standard method because it yields a useful overview of possibly problematic regions. The population modelling method sets the highest threshold. It is an appropriate method for the rapid detection of “hot spots” of pollution where an immediate intervention might be necessary. 6. Conclusions Both, soil type and lithology evidently influence the magnetic susceptibility measured in soils. Although the

variability of susceptibility values discriminates different soil types better than different bedrock lithologies, the two factors are closely interrelated and in most cases cannot be separated. A dominating influence is water logging. As soon as reductomorphic processes in a soil become dominant, its susceptibility is low, independent from the bedrock lithology. Lithologies with a high content of ferrimagnetic materials complicate the detection of anthropogenic anomalies and require additional methods to delineate the anthropogenic influence, e.g. additional rock magnetic analyses in the laboratory or investigation of the magnetic components by scanning electron microscopy. The decomposition of data distribution functions for the soil type sub data sets enabled us to exclude the influence of pedological processes. This is especially useful in regions where pedological processes lead to a depletion or enhancement of magnetic material in the soil. The three methods applied for anomaly detection (MAD, boxplot and population modelling method) were all successful in identifying the main anomalies in the region. The choice of the method influences the height of the threshold set for an anomaly definition. In this sense, the three methods can be regarded as increasingly strong filters which can be applied for anomaly detection in any univariate environmental data base. For a thorough interpretation of all the results, expert knowledge is always necessary. Acknowledgements The authors are thankful to all the people who provided help during the data acquisition: W. Krainer and his team from the Agricultural Research Centre Styria, K. Aichberger from AGES (Österreichische Agentur für Gesundheit und Ernährungssicherheit) in Linz, The Federal Agricultural Laboratory Vienna (now: AGES Vienna), J. Kölblinger from LUA (Lebensmitteluntersuchungsanstalt) Kärnten and W. Thöny from the Paleomagnetic Laboratory in Gams. This study was financed by FWF-grant P16314. References Amt der Kärntner Landesregierung, Abteilung 15 – Umweltschutz und Technik. Bodenzustandsinventur Kärnten 1999. Druckzentrum GmbH, Klagenfurt – Villach – St. Veit, 1999, 217 pp. Amt der Steiermärkischen Landesregierung RA 8. Bodenschutzbericht 1998. Graz, 1998, 149 pp. Bityukova L, Scholger R, Birke M. Magnetic susceptibility as indicator of environmental pollution of soils in Tallinn. Phys Chem Earth A 1999;24:829–35.

M. Hanesch et al. / Science of the Total Environment 382 (2007) 351–363 Bundesamt und Forschungszentrum für Landwirtschaft. Burgenländische Bodenzustandsinventur. Amt der Burgenländischen Landesregierung, Eisenstadt; 1996. 209 pp. Bundesanstalt für Agrarbiologie Linz. Oberösterreichischer Bodenkataster — Bodenzustandsinventur. Amt der o.ö.Landesregierung, Agrar- und Forstrechtsabteilung; 1993. 268 pp. Bundesanstalt für Bodenwirtschaft Wien. Niederösterreichische Bodenzustandsinventur. Amt der Niederösterreichischen Landesregierung, Wien; 1994. 220 pp. Davis JC. Statistics and data analysis in geology. 3rd edition. New York: John Wiley & Sons; 2002. 638 pp. Ebner F. Metallogenetische Karte von Österreich. 1:500000. Geologische Bundesanstalt Vienna; 1997. Fialová H, Maier G, Petrovský E, Kapička A, Boyko T, Scholger R. Magnetic properties of soils from sites with different geological and environmental settings. J Appl Geophys 2006;59:273–83. Hanesch M, Scholger R. Mapping of heavy metal loadings in soils by means of magnetic susceptibility measurements. Environ Geol 2002;42:857–70. Hanesch M, Scholger R. The influence of soil type on the magnetic susceptibility measured throughout soil profiles. Geophys J Int 2005;161:50–6. Hanesch M, Scholger R, Rey D. Mapping dust distribution around an industrial site by measuring magnetic parameters of tree leaves. Atmos Environ 2003;37:5125–33. Hay KL, Dearing JA, Baban SMJ, Loveland PJ. A preliminary attempt to identify atmospherically-derived pollution particles in English topsoils from magnetic susceptibility measurements. Phys Chem Earth 1997;22:207–10. Heller F, Strzyszcz Z, Magiera T. Magnetic record of industrial pollution in forest soils of Upper Silesia, Poland. J Geophys Res 1998;103:17 767–74. Hoffmann V, Knab M, Appel E. Magnetic susceptibility mapping of roadside pollution. J Geochem Explor 1999;66:313–26.

363

Kapička A, Petrovský E, Ustjak S, Macháčková K. Proxy mapping of fly-ash pollution of soils around a coal-burning power plant: a case study in the Czech Republic. J Geochem Explor 1999;66:291–7. Kapička A, Petrovský E, Jordanova N, Podrázský V. Magnetic parameters of forest top soil in Krkonose Mountains, Czech Republic. Phys Chem Earth A 2001;26:917–22. Magiera T, Lis J, Nawrocki J, Strzyszcz Z. Podatność Magnetyczna Gleb Polski (Magnetic susceptibility of soils in Poland). IPIŚ PAN Zabrze — Warszawa; 2002. Matschullat J, Ottenstein R, Reimann C. Geochemical background — can we calculate it? Environ Geol 2000;9:990–1000. Moreno E, Sagnotti L, Dinarès-Turell J, Winkler A, Cascella A. Biomonitoring of traffic air pollution in Rome using magnetic properties of tree leaves. Atmos Environ 2003;37:2967–77. Müller G. Schwermetalle in den Sedimenten des Rheines – Veränderungen seit 1971. Umschau 1979;79:778–85. Petrovský E, Ellwood BB. Magnetic monitoring of air-, land-, and water-pollution. In: Maher BA, Thompson R, editors. Quaternary climates, environments and magnetism. Cambridge: University press; 1999. p. 279–322. Rantitsch G. Geochemical exploration in a mountainous area by statistical modeling of polypopulational data distributions. J Geochem Explor 2004;82:79–95. Reimann C, Filzmoser P, Garrett RG. Background and threshold: critical comparison of methods of determination. Sci Total Environ 2005;346:1–16. Strzyszcz Z, Magiera T, Heller F. The influence of industrial immissions on the magnetic susceptibility of soils in Upper Silesia. Stud Geophys Geod 1996;40:276–86. Tukey JW. Exploratory data analysis. Mass.: Addison-Wesley, Reading; 1977. 503 pp.