Rapid Magnetic Biomonitoring and Differentiation of Atmospheric

Mar 21, 2012 - Cluster analysis of the steelworks leaf receptors and potential sources ... Environmental Science & Technology 2017 51 (12), 6648-6664...
0 downloads 0 Views 3MB Size
Article pubs.acs.org/est

Rapid Magnetic Biomonitoring and Differentiation of Atmospheric Particulate Pollutants at the Roadside and around Two Major Industrial Sites in the U.K. R. Hansard,† B. A. Maher,†,* and R. P. Kinnersley‡ †

Centre for Environmental Magnetism and Paleomagnetism, Lancaster Environment Centre, Lancaster University, Lancaster, United Kingdom, LA1 4YQ ‡ Environment Agency, Evidence Directorate, Olton, Solihull, United Kingdom, B92 7HX S Supporting Information *

ABSTRACT: Emissions of particulate matter (PM) from vehicle and industrial sources constitute a hazard to human health. Here, we apply biomagnetic monitoring to (a) discriminate between potential PM10 sources around a steelworks and (b) examine magnetic source differentiation for a combined, U.K.-based, magnetic data set (steelworks, roadside, power-generating site). Tree leaves (sampled September 2009, as passive PM receptors) and putative sources were subjected to rapid magnetic characterization (magnetic remanence measurements). Fuzzy cluster analysis of the combined data set identified three clusters, showing that particulates emitted from vehicle fleets (e.g., diesel/petrol), and from different industrial processes can be magnetically differentiated. Cluster analysis of the steelworks leaf receptors and potential sources identified seven magnetic groupings. Leaves from one PM “hotspot” showed no affinity with any available source sample, suggesting an as yet untested PM source. These data indicate the value of fast, inexpensive magnetic techniques for particulate source discrimination and indication of “missing” sources.



INTRODUCTION In the U.K. alone, up to ∼50 000 people die prematurely each year due to exposure to air pollution, of which “Particulate matter (PM) is thought to have the most damaging impact on health. It is a “no threshold” pollutant, with adverse effects to health seen at very low concentrations”.1 PM sources can be mobile (e.g., road traffic), point (e.g., combustion plant), and/ or diffuse (e.g., wind-blown dust or long-range transported PM). Identification of PM sources, and spatial variations in PM concentration, can help in quantifying health risks and targeting and assessment of emission control measures. While particulates may consist of a complex mixture of organic and inorganic, and primary and secondary components, urban anthropogenic PM is enriched in metals (including Fe, Pb, Zn, Ba, Mn, Cd, and Cr), and so almost invariably contains magnetic particulates. The bulk Fe content of urban atmospheric PM is typically between ∼5 and 15%, of which magnetic iron oxides and hydroxides comprise ∼10−70%.2,3 The latter derive from abrasion/corrosion processes,4,5 and/or the presence of iron impurities in fuels which on combustion form a nonvolatile residue, often a mix of magnetically ordered iron oxides and oxyhydroxides. In the case of industrial PM © 2012 American Chemical Society

sources, magnetite and hematite are the major magnetic minerals, which arise from a range of processes, including fossil fuel combustion, steel processing, and cement production.6,7 For vehicular sources, particulate ferrites (magnetite/ maghemite) can be combustion-derived,8 and/or arise from abrasion of Fe-bearing materials, such as brake pads.9,10 Magnetic PM may not only be dangerous in itself, but can also become associated with other hazardous pollutants formed during combustion, including lead11 and known carcinogens such as benzo[a]pyrene.12,13 The ubiquity of magnetic particulates within anthropogenic PM, and the sensitivity of magnetic analyses (down to trace levels), makes possible the magnetic characterization and quantification of PM, whether sampled on filters within a range of available instruments (e.g., TEOMs, FDMS, Partisol, BAM) or from widely dispersed natural surfaces, such as tree leaves, bark, or soils. Received: Revised: Accepted: Published: 4403

October 16, 2011 February 28, 2012 March 21, 2012 March 21, 2012 dx.doi.org/10.1021/es203275r | Environ. Sci. Technol. 2012, 46, 4403−4410

Environmental Science & Technology

Article

Table 1. Two-Tailed t-Test Results Showing Magnetic Parameters Significant (|t| > 1.984) in Discriminating between Samples at a Significance Level of 0.05 for Three Different Leaf Sample Groupings (Port Talbot, Industrial Site A & Roadside)a

a

Significance is indicated by a tick symbol. The magnetic parameters are defined in the Methods section and/or the SI.

pensive) could be used to identify links between receptors (tree leaves from trees across the industrial site and in the immediate surrounding area) and potential particulate sources. To establish magnetically defined sample groupings objectively, we used multivariate statistical methods; probabilistic [fuzzy c-means (FCM)] cluster analysis27,28 and nonlinear mapping.29 FCM is appropriate for classification of large data sets with potential compositional overlap, since no a priori knowledge regarding sample partitioning or source is required. Rather than forcing a sample to be assigned completely to one cluster, the similarity between a sample and all clusters is calculated. Geochemical applications of FCM have included deep-sea sediment sourcing30 and groundwater tracing.31 Magnetic FCM applications include classification of polluted soils32 and sediment diagenesis.33 Here we used FCM to identify (a) any significant differences between magnetic particulates on leaves from different sites and (b) the similarity of potential source samples to leaf receptor samples. We also used nonlinear mapping (NLM), a multidimensional scaling method, which seeks to illustrate the relationships between samples by 2-dimensional representation of the distance matrix between the data points in the multidimensional variable space.29 Labeling of the FCM-derived cluster affinity of the samples in this 2-D map enables assessment of the coherence of the cluster groupings. Comparison of the NLM with the FCM partition coefficient, F, and the classification entropy, H27, assists identification of the optimal number of sample clusters.

Magnetic analysis can provide a robust means of quantifying ambient PM concentrations both at the roadside and around PM point sources.14−18 Strong correlations, typically explaining ∼40−85% of the variance of street-level PM10 concentrations, have been identified between ambient PM10 and measured magnetic susceptibility and/or remanences, particularly at city center roadsides where a dominant, vehicular source prevails in simple terrain.18,16,19 Because magnetic measurements are fast and inexpensive, PM measurement and source characterization can be achieved for large numbers of samples (e.g., 100s of leaf and/or filter samples), giving high spatial resolution, a prerequisite for PM source attribution. Mapping of proxy PM levels at high spatial resolution has been undertaken using magnetic measurements of tree leaves,18,11,19,14 air filters,16,20 soils,21,22 and roadside dusts.23,24 The nature of the PM/ magnetic correlation depends on the contributing PM sources. The magnetic minerals and concentrations emitted by vehicles differ from those produced by industrial point sources, while different industrial processes and/or sources in turn carry distinctive magnetic “fingerprints”.18,20,15 Magnetic analysis parameters can be selected to give a PM signature which varies spatially and temporally, depending on magnetic particle concentrations and composition. Since magnetic techniques can discriminate between Fe-bearing particles produced at different temperatures and/or redox conditions, this has significance in terms of potential for source attribution; such discrimination is difficult using chemical compositional analyses, where nonunique elemental contributions, e.g., of iron, salt, calcium, are frequently obtained. Magnetic approaches may thus offer a means of identifying and quantifying PM sources even in situations where multiple sources (atmospheric and/or fugitive) have so far been difficult to constrain, such as at the steelworks complex examined here. We report magnetic characterization of tree leaves, as passive PM receptors, around a complex industrial site, together with a range of putative sources sampled from the site (kindly made available by Tata Steel). In comparison with its local area, the town of Port Talbot experiences elevated concentrations of PM10, which have been linked to activities at its dominant industrial complex, a major steelworks.25 Because of the complexity of sources at and around the site (the area also includes a major motorway, and is adjacent to the sea), source attribution for these elevated PM10 concentrations remains unresolved, despite more than a decade of measurement and analysis programmes.26 We examined these data in combination with published data18 for urban roadside leaves, road dusts, and leaves sampled around a single industrial point source (a large combustion plant in the U.K.). Our aim was to examine if routine, room temperature magnetic analyses (rapid, inex-



EXPERIMENTAL SECTION

First we examined the magnetic properties of pollutant particulates from tree leaves and possible source samples within and around the Port Talbot steelworks. For statistical analysis, we then combined these data with magnetic data for tree leaves taken during two previous U.K.-based sampling campaigns between 2007 and 2009: at the urban roadside34 (Lancaster, U.K.); and around a large, single-source combustion plant18 (“industrial site A”, S. England). To obtain data at greatest spatial resolution, all available tree species were sampled (Supporting Information Table S1). Magnetic deposition velocities (mVd) had been established previously18 for some species. For additional species found here, magnetic remanences were “calibrated” against colocated (i.e., within 2 m radius) tree species. Available fixed instrumental monitoring station data (i.e., 1 at Lancaster city center; 2 north and south of “Site A”; 6 around Port Talbot) indicated no transient peaks in long-range transported particulates at the time of sampling. To minimize any local meteorological effects, sampling was 4404

dx.doi.org/10.1021/es203275r | Environ. Sci. Technol. 2012, 46, 4403−4410

Environmental Science & Technology

Article

remanence to the IRM300 (χARM/IRM300). This is particularly sensitive to the presence of fine-grained ferrites (magnetically single domain, ∼0.05 μm in magnetite/maghemite).38 To avoid disproportionate influence on the clustering,32 extreme outliers were identified (i.e., magnetically very weak samples, with parameter values deviating >3× the standard deviation of the mean) and removed from the data set (11of 398 data points) prior to data entry and FCM. All of the magnetic data were normally distributed, with the exception of the HIRMaf100, which was lognormally distributed and so was then log (base 10) transformed for the FCM. The FCM assigns each sample an affinity to each cluster from 0 (no similarity) to 1 (identical), thus enabling any overlap between samples and their potential affinity to different sources to be examined. The FuzME program39,40 was used first to classify the combined magnetic data set from the three sampling campaigns (Port Talbot, industrial site A and Lancaster roadside, i.e. 387 samples comprising leaves, source samples, and road dusts) into clusters. The algorithm starts with 2 clusters, assigns random cluster membership coefficients to each point and iterates until the difference in coefficients between two iterations is no more than the given sensitivity threshold. The centroid for each cluster and the cluster membership coefficients for each point are then calculated. This is repeated for consecutively increasing numbers of clusters until a set maximum (usually 9). The optimal cluster number is that which minimized spread within the cluster, while maximizing the distance between the cluster centers. The FCM analysis was then run as a nested analysis, including only those samples associated with the Port Talbot leaves and potential sources. To ensure a robust solution, the FCM analysis was run a number of times with different starting samples, and the cluster coherence examined from the NLM.

always done within one or two consecutive days, after several (∼5) rain-free days.24,18 All leaf samples were refrigerated at 5 °C before being taken to the Centre for Environmental Magnetism and Palaeomagnetism (CEMP) at Lancaster University for magnetic analysis, using a routine, room temperature protocol,24 as outlined in the SI. Potential source samples (x20; S.I. Table S1.1) and road dusts (x11) were weighed, and analyzed using the magnetic protocol above. The t tests were undertaken on the measured magnetic signatures (Table 1) of all of the leaf samples to identify which magnetic parameters afforded most discrimination between the sampled leaf groups (Port Talbot, industrial site A; roadside). Magnetic mineral concentrations depend on the flux not only of magnetic particulates but also of magnetic dilutants, such as organic carbon or siliceous material (both diamagnetic). To eliminate any dilution effects, only ratio (i.e., non concentration-dependent) magnetic parameters were used, to enable mineralogical comparison between mass- and surface areanormalized samples exposed to different concentrations of magnetic particulates. The independence of those magnetic parameters which were discriminatory in all cases was then tested (using the Pearson’s coefficient of determination). Where the r2 value exceeded 0.3 (ρ = 0.05), one of the correlated pair was excluded. The resulting independent magnetic properties (Table 2) reflect variations in sample Table 2. Magnetic Properties of the Cluster Centres Identified in Figure 3a



RESULTS AND DISCUSSION The leaf saturation magnetic remanence (SIRM, imparted in a dc field of 1 T) varied markedly between the three sampling campaigns. The Port Talbot leaf SIRMs range from 35−400 × 10−6 Am2; the roadside leaves from 10−250 × 10−6 Am2;34 leaf SIRMs around Industrial site A were relatively low (2−12 × 10−6 Am2).18 The putative source samples also displayed marked differences, both in IRM acquisition and SIRM, which varied by up to 4 orders of magnitude (Figure 1). Such large differences in magnetic concentration might eventually assist in magnetic source apportionment; low SIRM-sources (e.g., smelting plant output, Figure 1) would require relatively high concentrations compared with higher-SIRM sources (e.g., sinter dust) to contribute similarly to the total receptor magnetic remanence. In terms of remanence acquisition, the Port Talbot leaf samples displayed spatial variability, exhibiting magnetic behavior ranging from “soft” (magnetite-like), to extremely hard (goethite-like). The industrial site A leaf samples were dominated by “hard”, hematite-like, remanence behavior.18 In contrast, the roadside leaves were “soft”, acquiring most remanence at low dc fields.34 Many magnetic properties are grain size-dependent, since grain size determines the number of magnetic domains into which a crystal can subdivide. With decreasing grain size the number of magnetic domains decreases; from multidomain grains (>∼4 μm for magnetite) to single domain (0.03−0.05 μm), and superparamagnetic grains, too small (300 mT) is dominated by “hard” magnetic minerals, like hematite and goethite. The fifth magnetic variable used here, % HIRMaf100, discriminates between these “hard” magnetic minerals; goethite is magnetically harder, and resists alternating field demagnetization.37 The sixth parameter used in the fuzzy c-means analysis was the ratio of the susceptibility of anhysteretic 4405

dx.doi.org/10.1021/es203275r | Environ. Sci. Technol. 2012, 46, 4403−4410

Environmental Science & Technology

Article

Figure 1. Saturation isothermal remanent magnetization (SIRM) values (logarithmic scale) of 20 putative industrial source samples. The bars indicate the remanance acquired at each dc magnetizing step, from 20 mT to 1 T, the “saturating” field.

(ARM) or, normalized by the dc field, the susceptibility of ARM (χARM), is a sensitive grain size parameter.42,38 The dominant magnetic grain size for the leaf particulates from the urban roadside and around industrial site A was ∼0.1 to 1 μm (Figure 2). In contrast, the leaf samples around Port Talbot displayed a larger magnetic grain size, ∼5 to 15 μm.

Figure 3. 2-D nonlinear map indicating fuzzy cluster results of magnetic properties of leaves sampled from areas exposed to particulate pollution from vehicles,34 industrial site A18 Port Talbot and residential areas surrounding Port Talbot (this paper), with 11 road dust samples from Port Talbot and 20 potential source samples and filters. Samples presented as pie charts indicating the affinity to each statistical cluster: Cluster (white), Cluster 2 (gray), and Cluster 3 (black).

campaigns (urban roadside; industrial site A; Port Talbot) fall, respectively, into the three statistically distinct clusters (table 2), indicating that the variation in leaf magnetic properties between these three sampling locations was greater than the variation within any one of the sampling locations. With only four exceptions (out of 92), the leaf samples collected from Port Talbot and the surrounding area had the greatest affinity to Cluster 1 (white). Four of the Port Talbot road dust samples and eight of the potential source samples (coal-fired boiler, graphite plant, smelting plant, coke, olivine sand, coal, basic oxygen steel (BOS) slag, and iron ore) also exhibited an affinity >66% with this cluster. A second cluster (Cluster 2, gray in Figure 3) contains all samples collected from within 10 km of a combustion stack (i.e., collected around Industrial Site A, at distances exceeding 30 m from the roadside). In terms of potential sources, the wood-fired boiler, coal-fired power station, gas/oil boiler, and gas/oil power station source samples all exhibit an affinity >66% with Cluster 2. In contrast, all of the roadside leaf samples fall within a tight third cluster (Cluster 3, black in Figure 3), together with the vehicle car exhaust, biodiesel, and brake pad source filter samples. Five of the putative source samples (oil-fired boiler, gas cattle incinerator, magstone, limestone, and sinter) and seven of the road dust samples around Port Talbot exhibited less than 66% affinity with any of these three statistical clusters (Figure 3; SI Table S1.1). The three clusters displayed marked magnetic differences. For the roadside leaf samples (Cluster 3; Figure 3, Table 2),

Figure 2. Comparison of leaf χARM/IRM ratios and MDFARM values from around industrial site A,18 Port Talbot (this work), and areas with predominantly vehicle-derived PM34 with magnetite particles of known grain sizes.38,43,42Given the variable magnetic composition of the samples, the χARM was normalized with the IRM300mT (rather than the SIRM), as any magnetite present should be saturated at this field.

For statistical analysis of the leaf receptor data, t tests were first used to identify which magnetic parameters afforded most discrimination between leaf sample groups (Port Talbot, industrial site A, roadside; Table 1). When all samplesleaves, source filters, and source powderswere included in the fuzzy cluster algorithm, three distinct (and statistically optimal) groups were identified by both their cluster affinities and the sample locations, as represented in 2-dimensional space (the nonlinear map shown in Figure 3) Samples from the three leaf sampling 4406

dx.doi.org/10.1021/es203275r | Environ. Sci. Technol. 2012, 46, 4403−4410

Environmental Science & Technology

Article

Figure 4. (a) An inverse distance weighted interpolation of the SIRM values with spatially plotted pollution roses for 6 of the local air quality monitors, presenting maximum hourly (white) and mean daily (black) PM10 concentrations. Black dots represent leaf sample locations. (b) Spatially plotted sample fuzzy cluster affinities presented as pie charts, together with putative sources. Suspected particulate source area26 enclosed within dashed line on both (a) and (b). Crown Copyright OS 1:25,000 Colour Raster 2010. An Ordnance Survey/Edina supplied service. Lancaster University is licensed to use this service.

∼75% of the total IRM was acquired by 100 mT, increasing to ∼98% by 300 mT, and 300 mT) retained upon AF demagnetization at 100 mT (HIRMAF 100). Around the two industrial sites, the IRM100mT and IRM300mT values were lower, and the HIRMAF 100 higher, than at the roadside. Around industrial site A (Cluster 2; Figure 3, Table 2), the IRM100mT and IRM300mTvalues were 60% and 71%, respectively, with an HIRMAF 100 of 34%. In and around the Port Talbot site, 68% of the IRM was acquired by 100 mT, increasing to 90% by 300 mT; but the HIRMAF 100 slightly lower, 29%. The greater resistance of HIRM demagnetization in the industrial samples indicates a greater contribution to the magnetic signal from the magnetically ‘harder’ minerals, hematite and/or goethite.37 Because hematite has much lower magnetization than magnetite, it must be present in very high concentrations, relative to magnetite, in order to contribute the HIRMs measured here (e.g., ∼0.01% hematite compared with ∼0.0001% magnetite at industrial site A). These differences in leaf and source magnetic mineralogy between the three clusters are likely to be due to different fuels, combustion temperatures, redox status, and abrasion processes occurring at the three locations. For example, at temperatures >850 °C44 (achieved during industrial combustion processes45 such as dry bottom boiler combustion (900−1200 °C), wet bottom boiler combustion (>1400 °C), fluidized bed combustion (750−950 °C) and grate firing (1000 and 1300 °C), a mixture of magnetite and hematite is produced, but hematite is stable to higher temperatures. In contrast, vehicle

internal combustion engines operate at substantially lower temperatures (400 × 10−6 A) also corresponds with the 4407

dx.doi.org/10.1021/es203275r | Environ. Sci. Technol. 2012, 46, 4403−4410

Environmental Science & Technology

Article

Table 3. Magnetic Properties of the Cluster Centroids for the 7 Clusters Identified in Figure 4ba

Figures in brackets = no. of samples assigned to each cluster. The “hard” magnetic behavior of Cluster 1 (little remanence acquired at low fields, ∼75% HIRM retention upon af demagnetization) contrasts markedly with the much “softer” behavior of Cluster 5, for example. SI Table S2 summarizes indicative magnetic mineralogies for the cluster centroids.

a

The iron ore pellets appeared to be magnetically distinct from any of the leaf samples collected from and around Port Talbot (Figure 4b). The leaf samples, which were magnetically similar to the coal/olivine sand samples, were predominantly located to the south of the site, in close proximity to the coal stockyard (SI Figure 1.1). Those leaf samples with the strongest similarity to the coke source sample were primarily located to the southeast of the site, close to the hot and cold mills. The leaves with similar cluster affinities to the coal-fired boiler filter samples were mostly located near, and downwind of, the coke ovens and slag handling facilities. However, it is notable that many of the leaf samples at the north of the site, including those occurring at the peak SIRM location, showed no affinity to any of the putative source samples measured in this study. This suggests that a key particulate source/s remains untested and thus unattributed. Further progress toward magnetic source apportionment around Port Talbot may be made once all potential sources have been magnetically analyzed. In summary, the magnetic properties of particulates emitted from different sources significantly vary. Even a limited range of rapid, inexpensive, room temperature magnetic measurements of tree leaves (widely distributed and acting as passive receptors) is capable of differentiating between PM emitted by vehicles and by different industrial processes. More detailed magnetic characterization (e.g., of a selected subset of natural and/or filter samples) may provide even greater potential for source differentiation. The nested clustering and resultant spatial mapping achieved by rapid magnetic analysis of leaves around the complex steelworks site show that the sources tested so far cannot account for the large PM impact identified around the northeast of the site. The magnetic grain size data indicate that this PM is relatively coarse, between ∼5 and 15 μm, suggesting fugitive sources as the most likely contributors. When all sources have been tested and identified, further sequences of leaf sampling and magnetic analysis can be used to assess the effectiveness of any subsequent mitigation measures.

predominant source directions during a number of such episodes, as identified by pollution roses for 5 of the surrounding conventional fixed monitoring stations (CFMS; stations 1−5; Figure 4a) located in nearby residential areas. The secondary leaf SIRM peak (350−400 × 10−6 A) corresponded with the predominant source direction on exceedence days for the sixth CFMS, with some potential contribution to CFMS 5. The strong correspondence between the leaf magnetic “hotspots” and problematic sources identified independently both by the AQEG report and via the CFMS data highlights the potential of these rapid and cost-effective magnetic techniques for analysis of ambient PM10 concentrations and sources at high spatial-resolution. Such magnetic discrimination can also enable more effective site selection for implementation of independent, more expensive and time-consuming sampling and analytical techniques. To examine the discriminatory value of the leaf magnetic properties at Port Talbot in greater detail, the fuzzy cluster analysis was rerun for just the original Cluster 1 samples, i.e., the Port Talbot leaf samples, 8 putative sources and the 4 Cluster 1 road dust samples (Figure 3). Here, 7 clusters (Figure 4b; Table 3) formed the optimal statistical solution (i.e., greatest distance between the cluster centroids, while minimizing the within-cluster spread). When the resultant cluster affinities were plotted spatially (superimposed on the leaf SIRM interpolation), statistical and spatial differentiation were evident between the leaf samples. The peak SIRM values were associated with magnetically distinct samples showing strong affinity with Cluster 6 (Figure 4 a; Table 3). The Cluster 6 samples exhibited magnetically very “hard”, hematite-like, behavior (73% IRM100, 46% HIRMAF100), in strong contrast to the majority of the remaining clusters (e.g., Clusters 2, 4 and 5; < 25% HIRMAF100). Significant magnetic discrimination was also evident between the possible source samples. The coal and olivine sand samples showed most affinity with Cluster 5 (>75%), the cluster which displayed the magnetically “softest” behavior (74% IRM100, 6% HIRMAF100). Conversely, the iron ore pellets displayed 100% affinity with the magnetically hard, goethite-like Cluster 1 (40% IRM100, 75% HIRMAF100). All of the other samples exhibited intermediate magnetic “hardness”; the coal-fired boiler and smelting plant samples exhibiting strongest affinity (>75% and 40%, respectively) to Cluster 7. The smelting plant filter also showed some affinity (33%) to Cluster 2. Coke had the greatest affinity (43%) to hematite-like Cluster 3 with a secondary affinity (27%) to Cluster 2. Despite showing more affinity to the Port Talbot samples than to those from industrial site A or the roadside, the graphite plant and magstone samples displayed