Environ. Sci. Technol. 2007, 41, 2874-2880
Near-Infrared Spectroscopy (NIRS) of Epilithic Material in Streams has a Potential for Monitoring Impact from Mining JAN PERSSON,† MATS NILSSON,‡ CHRISTIAN BIGLER,† STEPHEN J. BROOKS,§ AND I N G E M A R R E N B E R G * ,† Department of Ecology and Environmental Science, Umeå University, SE-901 87 Umeå, Sweden, Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, SE-901 83 Umeå, Sweden, and Department of Entomology, The Natural History Museum, Cromwell Road, London SW7 5BD, England
There is an increasing demand for cost-effective methods for environmental monitoring, and here we assess the potential of near-infrared spectroscopy (NIRS) on epilithic material from streams (material covering submerged stones) as a new method for monitoring the impact of pollution from mining and mining-related industries. NIRS, a routine technique in industry, registers the chemical properties of organic material on a molecular level and can detect minute alterations in the composition of epilithic material. Epilithic samples from 65 stream sites (42 uncontaminated and 23 contaminated) in northern Sweden were analyzed. The NIRS approach was evaluated by comparing it with the results of chemical analyses and diatom analyses of the same samples. Based on Principal Component Analysis, the NIRS data distinguished contaminated from uncontaminated sites and performed slightly better than chemical analyses and clearly better than diatom analyses. Of the streams designated a priori as contaminated, 74% were identified as contaminated by NIRS, 65% were identified by chemical analysis, and 26% were identified by diatom analysis. Unlike chemical analyses of water samples, NIRS data reflect biological impacts in the streams, and the epilithic material integrates impact over time. Given that, and the simplicity of NIRS-analyses, further studies to assess the use of NIRS of epilithic material as an inexpensive environmental monitoring method are justified.
Introduction Mining and metal-producing industries, active as well as abandoned, are significant contamination sources to surface waters. Mining wastes are a growing environmental problem and contribute a substantial part of the total amount of wastes in Europe, as well as in many other areas of the world (1-3). Legislation and other efforts to reduce emissions and improve * Corresponding author phone: +46 90-7866029; fax: +46 907866705; e-mail:
[email protected]. † Umeå University. ‡ Swedish University of Agricultural Sciences. § The Natural History Museum. 2874
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 8, 2007
surface water quality have been launched, for example, by the European Union (4), and these actions require effective control and monitoring programs. Effluents from mining and metal-producing industries often impact stream ecosystems. These are complex ecosystems and a variety of methods are required to fully assess the environmental conditions of streams and their temporal changes. Current methods include assessment of biological assemblages of fish, benthic macroinvertebrates, macrophytes, bryophytes, and algae, and abiotic variables, such as pH, conductivity, metals, and nutrients (5). Large fluctuations in water quality are characteristic of small streams and may be episodic, diurnal, or seasonal, as well as inter-annual (6). Stream monitoring requires an intense sampling program to cover the different scales of temporal variation, but this is expensive. Development of cost-effective methods complementary to water sampling is therefore important. Epilithic material, i.e., the layer of organic and inorganic material that covers stone surfaces in streams, can be used for alternative stream monitoring approaches. It is composed of material deposited from the water as well as benthic algae and bacteria, the latter responsible for most of the energy and material turnover in the stream. The submerged epilithic material is continuously exposed to and influenced by the water that passes in the stream. The epilithic material is therefore a time-integrating material with potential for environmental monitoring. Near-infrared spectroscopy (NIRS) is a rapid, nondestructive technique that is commonly used in industry for control of product quality and for process monitoring, for example in paper, food, and the pharmaceutical industries (7). The technique is based on absorption of near-infrared radiation by polar bonds, such as C-H, N-H, and O-H, and thereby describes the molecular composition of the organic material. NIRS has also been used to a limited extent in ecological and limnological studies, but has still not found wide application in environmental monitoring (8-9). NIRS analyses are rapid and require limited sample preparation, and the instrumentation is relatively inexpensive. Our hypothesis is that an impact on the epilithic material, either directly or indirectly, caused by the emissions from mining and metal-producing industries can be detected using NIRS, and furthermore, that the NIRS-technique can become a fast and inexpensive complement to frequent water sampling and methods that require specialists such as most biological analyses. Near-infrared spectroscopy performed on epilithic samples can be used as a screening method to detect environmental impacts, which then can be further assessed by specialized methods. We test this hypothesis using a set of samples of epilithic material from stream sites in northern Sweden comprising both forest streams without any direct contamination source and streams affected by mining-related activities. The concentration levels of heavy metals in the water of the contaminated streams are quite low when compared globally to heavy metal polluted streams. To evaluate the performance of NIRS we compare NIRS with chemical analysis and diatom analysis, made on the same samples of epilithic material. Heavy metal analyses and in particular analyses of stable lead isotopes (206Pb/207Pb ratios) disclose influence from mining-related industries since lead is present is most sulfide ores, and the isotopic signature of the ores and the common bedrock usually is different. The natural geogenic 206Pb/207Pb ratio in bedrock and uncontaminated soils in Sweden is 1.5 ( 0.2 (10), while the ratio of the lead ore in the study area in northern Sweden is 1.021.12 (11), i.e., it has a much lower ratio than the natural 10.1021/es062329b CCC: $37.00
2007 American Chemical Society Published on Web 03/17/2007
TABLE 1. Confusion Matrix Showing Correct/Incorrect Classifications of the Samples number of sites classified type of data set NIRS calibration data sets NIRS validation data sets chemistry calibration data sets chemistry validation data sets diatom calibration data sets diatom validation data sets a
uncont.a (col. A) cont.b (col. B) 21 21 21 21 21 21
15 8 15 8 15 8
Streams without any known connection to mining areas.
b
“uncontaminated” sites (col. A) “contaminated” sites (col. B) classified correct/incorrect (%) classified correct/incorrect (%) correct
incorrect
correct
incorrect
95 86 100 95 86 81
5 14 0 5 14 19
88 37 100 75 88 0
12 63 0 25 12 100
Streams draining mining areas.
FIGURE 1. Coomans plot showing modeled distance (arbitrary units) between each individual stream site and the principal component model of NIRS results from the uncontaminated sites (x-axis) and NIRS results from the contaminated sites (y-axis). The 95% confidence interval of each model is indicated by dashed lines, i.e., stream sites with values lower than the value corresponding to the confidence limit belong to the corresponding PC-model. For example, stream sites with y-values below the confidence limit for the model of contaminated sites and x-values higher than the confidence limit for the uncontaminated sites represent contaminated sites and are represented by filled symbols. Insert shows enlargement of the clustered plot area. Bruba1 cken (2), Vormba1 cken, (V1-V9 1), other mining-related streams (b), and uncontaminated reference streams (O). geogenic lead. Given the large difference in the isotope ratio, a small impact from mining can be detected. Diatoms are unicellular algae that are very common in streams and are sensitive to water quality including mining related elements (12-13).
Materials and Methods The study was performed in Va¨sterbotten, a county in northern Sweden (65° N, 19° E). The area is generally sparsely populated and is characterized by boreal forests dominated by Scots pine (Pinus sylvestris) and Norway spruce (Picea abies) and the landscape is rich in streams. Large-scale mining and ore processing started in the so-called Skellefte field (14) in the 1930s. The Skellefte field contains complex sulfide ores, and mining for mainly copper, zinc, lead, silver, and gold has occurred, but the ores contain a wide spectrum of trace elements. This study includes 65 stream sites of which 42 sites are considered uncontaminated (reference sites) and 23 receive water from mining facilities (contaminated sites) with a large variation in expected impact. These 23 sites comprise the following: active mines with tailing dams; abandoned mines
FIGURE 2. Modeled distance between each individual stream site and the principal component model of chemistry results from the uncontaminated sites (x-axis) and chemistry results from the contaminated sites (y-axis). The 95% confidence interval of each model is indicated by dashed lines, i.e., stream sites with values lower than the value corresponding to the confidence limit belong to the corresponding PC-model. For example, stream sites with y-values below the confidence limit for the model of contaminated sites and x-values higher than the confidence limit for the uncontaminated sites represent contaminated sites and are represented by filled symbols. Bruba1 cken (2), Vormba1 cken, (V1V9 1), other mining-related streams (b), and uncontaminated reference streams (O). Chemistry data in figure are based on arsenic (As), cadmium (Cd), cobalt (Co), copper (Cu), iron (Fe), lead (Pb, not isotope ratio), manganese (Mn), nickel (Ni), phosphorus (P), sulfur (S), and selenium (Se). of different age and size, with and without open pits; calciumrich/sulfide-poor mines; recently opened mines; and sites just in the immediate vicinity of a mining area. Concentrations of some common mining-related elements and phosphorus in the streamwater were measured in July 2002 (µg/ L): As 0.8-9, Cd 0.0-1.3, Cu 0.5-71, Ni 0.3-10, P 30-118, Pb 0.02-0.8, and Se 0.1-3.4, and the pH range was 6-8. Two of the contaminated streams were sampled in a downstream gradient: (i) Vormba¨cken (9 sites termed V1-V9), a 30-kmlong stream draining the active Kristineberg mining district, although the ore is transported to Boliden’s beneficiation plant since the tailing dam in Kristineberg was closed in 1991. (ii) Bruba¨cken (4 sites), which drains Boliden’s mining district and has a beneficiation plant where the majority of the excavated ores from local mines in the Skellefte district are enriched. To reduce environmental damage by acidity and mining-related elements, lime is added to the streamwater at the tailing dams in Kristineberg and Boliden. For site identification and data about the stream sites, see the Supporting Information (SI). VOL. 41, NO. 8, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2875
FIGURE 3. 206Pb/207Pb ratio and concentrations of arsenic (As), cadmium (Cd), cobalt (Co), copper (Cu), iron (Fe), lead (Pb), manganese (Mn), nickel (Ni), phosphorus (P), sulfur (S), and selenium (Se) in the epilithic material from the 65 stream sites. Element concentrations are expressed per gram of carbon. Bruba1 cken (2), Vormba1 cken (V1-V9 from left to right,1), other mining-related streams (b), and uncontaminated reference streams (O). Sampling was performed in July 2003 when water levels in the streams were stabilized after the spring flood period. First- to fourth-order streams were sampled; riffles with stones were preferred and pools were avoided. A new sampler for epilithic material, the so-called “Stone Brusher” was used (15). Samples of epilithic material of a defined area (∼28 cm2) from the upper surface of submerged stones were dislodged with a coarse brush and the material was transported by suction into a sample bottle. At every site, epilithic material from 10-12 stones along a distance of 10-20 m at a water depth of 10-50 cm was sampled and combined in a 2-L sample bottle and then frozen. After samples were thawed in the laboratory, about half of the water volume was decanted, the bottle was shaken, and 50-60 mL of the slurry of epilithic material was poured through a sieve (mesh size 2 mm) before filtration onto glass microfiber filters, GF/A Whatman. Three replicate filtrations were made, and filtration with distilled water only (blanks) was made for each 30th 2876
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 8, 2007
filtration to ensure analytical quality. After 3 days in a desiccator, the air-dried filters were scanned with a rotating drawer module using a NIRSystem 6500 instrument (FOSS NIRSystem Inc., Silver Spring, MD). Every alternate wavelength from 400 to 2500 nm was recorded, generating 1050 variables per sample. Multiplicative signal correction (MSC) was applied for baseline correction and to reduce light scattering effects arising from differences in particle size (16). Principal Component Analysis (PCA) was made to ensure that the GF/A filters gave a uniform signal; this signal was not subtracted from the stream samples. A PCA was also made to compare the NIRS results of the triplicates, and since they showed very similar results, only one sample was selected for the further multivariate analysis of the NIRS results, and the same filter was also used for the analyses of metals and carbon. After NIRS-analyses the filters were vacuum-dried at 70 °C to constant weight and prepared for chemical analyses by
ceutical industries (22). The performance of the models is presented in a confusion matrix (Table 1). In the next step, these two subset models were applied to the data from the remaining reference and contaminated sites and the result was visualized using Coomans plots (2324). Each sample is described by a Euclidean distance and a leverage-based model distance. Contaminated samples will have a small deviation from the model of contaminated samples and a larger deviation from the model of uncontaminated samples. Uncontaminated samples will have a large deviation from the model of contaminated samples and a small deviation from the model of uncontaminated samples (see Figure 1).
Results
FIGURE 4. Modeled distance between each individual stream site and the principal component model of diatom results from the uncontaminated sites (x-axis) and diatom results from the contaminated sites (y-axis). The 95% confidence interval of each model is indicated by dashed lines, i.e., stream sites with values lower than the value corresponding to the confidence limit belong to the corresponding PC-model. Bruba1 cken (2), Vormba1 cken, (V1-V9,1), other mining-related streams (b), and uncontaminated reference streams (O). wet dissolution in open Teflon vessels with HNO3 + HClO4 [10:1] (17). Arsenic (As), cadmium (Cd), cobalt (Co), copper (Cu), iron (Fe), manganese (Mn), nickel (Ni), phosphorus (P), lead (Pb), stable lead isotopes (206Pb and 207Pb), sulfur (S), and selenium (Se) were analyzed using ICP-MS (PerkinElmer PE 6100 DRC) at the Soil Science Laboratory, Department of Forest Ecology, Swedish University of Agricultural Sciences, Umeå. Carbon analyses were performed using a Perkin-Elmer Series II CHNS/O Analyzer 2400, at the same laboratory. Chemistry data including lead isotope ratios were validated using certified reference material, SRM 981 (National Institute of Standards and Technology) and internal reference sediment from a lake (17-18). For each element a mean was calculated for the blanks and this was subtracted from the samples. Chemical analytical data are given as µg element/g carbon in the epilithic material to normalize for the variable content of inert mineral grains and diatom valves in the epilithic material. Diatom slides for microscopy were prepared according to standard methods (19). The samples were taken from the shaken sample bottles at the same time as the filtration for NIRS was performed. A phase contrast light microscope with 1000× magnification was used and 300-400 diatom valves per slide were counted, percentage values for taxa were calculated, and all data were used for the assessment. Diatom taxonomy followed Krammer & Lange-Bertalot (20). The same approaches were separately applied for evaluation of the NIRS, chemistry, and diatom data sets. Evaluation of the data involved modeling with Principal Component Analysis (PCA), and compilation with Coomans plot using Simca software, Umetrics Inc (21). Diatom and metal data are log transformed and NIRS data centered only. The first step was a PCA analysis to select a subset of 21 representative samples from the 42 uncontaminated sites including low, median, and high score values on each of the significant principal components to cover all variation in the reference dataset. This subset was used to make a model (calibration dataset) for the uncontaminated sites. Development of a subset of the contaminated sites followed the same procedure. Development of a sub-dataset for calibration follows recognized procedures applied in, e.g., food and pharma-
The PC-models based on different calibration data sets (i.e., NIRS, chemistry, and diatoms) all performed rather well, as indicated by the high proportion (>80%) of correctly classified samples, both for contaminated and uncontaminated sites (Table 1). The PC-model validation, involving classification of remaining samples that were not used for calibration, reveals comparably accurate classification for chemistry data (75%) and moderate classification for NIRS (37%). In contrast, the PC-model validation based on diatom data failed completely, and none of the validation samples from contaminated sites was correctly classified (Table 1). Below we present and discuss the calibration and validation samples together, based on Coomans plots. Of the 23 sites from contaminated streams, those streams with a more or less clear connection to mining industries, 17 sites (74%) were classified by a PC-model using NIRS-data as contaminated, i.e., having x-values outside the 95% confidence interval of the model based on uncontaminated sites (Figure 1). The remaining 6 sites (26%) are placed within the 95% confidence interval, and they cannot be separated from the uncontaminated reference sites. The sites in Vormba¨cken (1) appear sequentially along the x-axis according to the expected degree of mining impact, where the sites closest to the emission source have scores furthest from the model of uncontaminated sites. The samples V5, V8, and V9 from Vormba¨cken which are classified as not different from the uncontaminated reference sites (within the 95% confidence interval), also had low X-scores in the chemistry data (Figure 2). The samples that deviated most from the NIRS-model of uncontaminated sites (having highest x-scores, Figure 1) were from Bruba¨cken (2), a stream not only contaminated by mining-related elements but also having high phosphorus levels (Figure 3). Some of the reference sites (O) deviate more along the y-axis than mining-related streams deviate along the x-axis, which means that NIRS reflects a larger variation in the organic chemical composition of the epilithic material within reference streams than within the mining-related streams (Figure 1). The PC-model based on chemistry data from uncontaminated sites (Figure 2) classified 16 sites as significantly different from the model of uncontaminated sites, of which 15 sites (65%) were sites in streams that receive water from mining industries. V8, V9, and six other mining-related streams (b) are to the left of the confidence line (Figure 2) and cannot be separated from the reference streams. Chemistry data are also presented in a more conventional manner as concentrations and as lead isotope ratios (Figure 3). The lead isotope ratio is lower in contaminated sites than in reference streams and contaminated sites generally have higher concentrations of metals, especially copper, lead and selenium. The PC-model showing diatom data, which is based on relative abundances, classified only six (26%) of the mining sites correctly, i.e., as deviating from the model of uncontaminated streams. In addition, seven samples of the VOL. 41, NO. 8, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2877
FIGURE 5. Number of diatom taxa and relative abundances (%) of the most common taxa (g10% in g30 sites) in the epilithic material from the 65 streams sites. Bruba1 cken (2), Vormba1 cken (V1-V9 from left to right,1), other mining-related streams (b), and uncontaminated reference streams (O). uncontaminated reference streams were also classified as deviating from the model of uncontaminated streams. The sample deviation on the y-axis is larger than the sample deviation on the x-axis, i.e., here again the uncontaminated reference sites deviate more from the model based on contaminated sites than the contaminated sites deviate from the reference sites model (Figure 4). The number of diatom taxa found in each sample varied from 3 to 43. The most frequent taxa were Achnanthes minutissima, Brachysira neoexilis, Fragilaria nanana, Tabellaria flocculosa, and Fragilaria capucina var. gracilis (Figure 5). Low numbers of taxa were found in the contaminated Vormba¨cken; the sites V1 and V2 were totally dominated by Achnanthes minutissima and Brachysira neoexilis. Low numbers of taxa were also found in Bruba¨cken which was dominated by Achnanthes minutissima and Nitzschia spp. together with some taxa in low total abundance such as Pinnularia subcapitata and Eunotia intermedia (which did not match the abundance criteria for Figure 5). In two of the reference streams Achnanthes minutissima was 2878
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 8, 2007
the dominant species and other taxa occurred only in low abundance.
Discussion Model PerformancesComparison of Models Based on Contaminated and Uncontaminated Sites, Respectively. For monitoring purposes, it is desirable to be able to identify samples (sites) that are different than samples from unaffected streams, using fast, simple, and inexpensive methods. When deviating conditions are found, further examination with specialized methods can be performed. Assessing and classifying unknown samples with NIRS can be done using either a model based on samples from uncontaminated sites or one based on contaminated sites. This study suggests that the model based on uncontaminated samples performs better, because it “correctly” identifies samples that deviate from the “class of uncontaminated samples” as being contaminated. The x-axis in Figures 1 and 2 shows that the model based on uncontaminated sites correctly classified (i.e., within the 95% confidence limit) the samples from
uncontaminated streams, both those used in the calibration data set and the “unknown” samples from uncontaminated streams. The majority of the samples from the contaminated sites were correctly classified as not belonging to the class of uncontaminated samples. In contrast, the models based on contaminated samples (the scores plotted on the y-axis in Figures 1 and 2) failed to include many samples from contaminated streams and also included a large proportion of the samples from uncontaminated sites. Using the contaminated sites as the model reference means that only sites that are contaminated in the same way will be identified as contaminated. Because it is difficult to build up a data set that includes all types of contaminated sites, it is preferable to define a population of natural sitessalthough the measured variation in this class is larger than the variation in the class of contaminated sites. Advantages of Using an Arbitrary NIRS-Scale. The most common use of NIRS is to calibrate the NIR-spectra against some measured constituent of the analyzed material, e.g., protein or water content of products in food industry (7), or against properties such as lake water pH or TOC in environmental contexts, and then use these calibrations to model the parameter of interest in unknown samples based on their NIR spectra (25-27). This approach utilizes the simplicity, robustness, and cost efficiency of NIRS, but it also requires analysis of some constituent that carries the wanted information. Calibration against a particular constituent also risks utilizing only a fraction of the total environmental pressure that is contained in the analyzed material. The approach that we have used to calibrate the model against a sample population or against reference objects (e.g., uncontaminated stream samples as in this study) utilizes more fully the power of NIRS to register systematic variation in the organic chemical composition of the sample, such as the epilithic material in this study. This is advantageous in comparison to modeling against only one selected chemical element in order to discover environmental pressure on biota by diffuse and complex environmental contamination. Model PerformancesComparison of NIRS, Chemistry, and Diatom Models. The NIRS analyses of epilithic material successfully classified most of the sites in the streams that were a priori classified as contaminated according to geographical data, i.e., NIRS detected deviations in the composition of the epilithic material in comparison to the material in the uncontaminated reference streams. In the majority of cases (75%) the NIRS model classified the same sites as contaminated as the chemistry model did. These were in most cases also the sites that had low 206Pb/207Pb ratios, which is a clear signal of emissions from ore processing in the studied region. The NIRS-based model not only identified the same samples as the chemical-based model, but the resolution was significantly better. The relation in x-scale units within and outside the 95% confidence limit was 8.9 for the NIRS based model and 2.1 for the chemical-based model. Our interpretation of this is that relatively small variations in chemical contents are magnified by effects on the biota and manifested in changed composition of both the living and dead organic material that NIRS registers. It further indicates that NIRS actually reveals the impact from metals on the stream ecosystem while chemistry data just reflect concentrations without any information on biological implications. This makes NIRS a more sensitive tool than chemical analyses to monitor and detect environmental changes in streams from mining impact. In contrast, the diatom-based model identified only half of the sites that were identified by the NIRS and the chemical data models as being contaminated. In addition, the diatom
model also classified several “uncontaminated” sites as deviating from the model of uncontaminated sites. Internal Relationships Among the Contaminated Sites. The internal relationship between the NIRS-based predictions of the sites sampled downstream in Vormba¨cken illustrates the performance of the NIRS methodology. The distance from the contamination source in Vormba¨cken increases with sample number, which is reflected in the Coomans plot (Figure 1). Site V1 is situated about 1 km downstream of the tailing dam and is classified as the most contaminated sample among the Vormba¨cken sites. Most of the samples downstream of the tailing dam appear sequentially in the plot (Figure 1). The three samples V5, V8, and V9 are not distinguished from the uncontaminated samples; V8 and V9 are the most distant sites and are actually situated downstream of an 8-km2 lake (20 km from the emission source). The change in the composition of the epilithic material to one that more closely resembles uncontaminated sites is likely to be due to a dilution effect and chemical coprecipitation of heavy metals and iron from tributaries along Vormba¨cken (28). Of the 23 stream sites originally designated as miningrelated sites, the NIRS-based model classified six as uncontaminated. Besides the above-mentioned V5, V8, and V9, two are located near but not downstream of abandoned mines and one is downstream of a 50-year-old test-mining area; i.e., these are sites with minor contamination. In general, the levels of contamination in the studied streams are very low. The results of the NIRS analyses are encouraging because they show the method is sensitive to very low concentrations of metal pollutants. Further studies to assess the use of NIRS-analyses of epilithic material as an inexpensive environmental monitoring method are therefore justified.
Acknowledgments Financial support was provided by Georange in Malå. Special thanks to Birgitta Olsson, SLU, Umeå, for chemical analyses and to Julieta Massaferro, Ben Williamson, and Malcolm Grant from Natural History Museum in London for assistance with field work. Thanks also to Matilda Guhre´n, Juha Salonsaari, and other colleagues for assistance with analyses at the Department of Ecology and Environmental Science, Umeå University.
Supporting Information Available Information about the stream sites. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Dudka, S.; Adriano, D. C. Environmental impacts of metal ore mining and processing. J. Environ. Qual. 1997, 26 (3), 590-602. (2) Aswathanarayana, U. Mineral Resources Management and the Environment; A. A. Balkema: Tokyo, 2003. (3) Sheoran, A. S.; Sheoran, V. Heavy metal removal mechanism of acid mine drainage in wetlands: A critical review. Miner. Eng. 2006, 19, 105-116. (4) European Union. Directive 2000/60/EC of the European Parliament and of the Council establishing a framework for community action in the field of water policy. Off. J. Eur. Communities: Legis. 2000, L327, 1-72. (5) Hauer, R. F., Lamberti, G. A., Eds. Methods in Stream Ecology; Academic Press: San Diego, CA, 1996. (6) Laudon, H.; Bishop, K. Episodic stream water pH decline during autumn storms following a summer drought in northern Sweden. Hydrol. Processes 2002, 16, 1725-1733. (7) Siesler, H. W., Ozaki, Y., Kawata, S., Heise, H. M., Eds. NearInfrared Spectroscopy: Principles Instruments Applications; Wiley-VCH: New York, 2002. (8) Foley, W. J.; McIlwee, A.; Lawler, I.; Aragones, L.; Woolnough, A. P.; Berding, N. Ecological applications of near infrared reflectance spectroscopy - a tool for rapid, cost-effective VOL. 41, NO. 8, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2879
(9)
(10)
(11) (12) (13)
(14)
(15) (16) (17)
(18)
(19)
prediction of the composition of plant and animal tissues and aspects of animal performance. Oecologia 1998, 116, 293-305. Korsman, T.; Dåbakk, E.; Nilsson, M. B.; Renberg, I. Near-infrared Spectrometry (NIRS) in Palaeolimnology. In Tracking Environmental Changes Using Lake Sediments Vol. 2; Last, W. M., Smol, J. P., Eds.; Kluwer Academic Publishers: Dordrecht, The Netherlands 2001. Bra¨nnvall, M.; Kurkio, H.; Bindler, R.; Emteryd, O.; Renberg, I. The role of pollution versus natural geological sources for lead enrichment in recent lake sediments and surface forest soils. Environ. Geol. 2001, 40, 1057-1065. Billstro¨m, K. Age and provenance of host rocks and ores in the paleoproterozoic Skellefte district, Northern Sweden. Econ. Geol. 1996, 91, 1054-1072. Medley, C. N.; Clements, W. H. Responses of diatom communities to heavy metals in streams: The influence of longitudinal variation. Ecol. Appl. 1998, 3, 631-644. Niyogi, D. K.; Lewis, W. M.; McKnight, D. M. Effects of stress from mine drainage on diversity, biomass, and function of primary producers in mountain streams. Ecosystems 2002, 5, 554-567. Allen, R. L.; Weihed, P.; Svenson, S. A. Setting of Zn-Cu-Au-Ag massive sulfide deposits in the evolution and facies architecture of a 1.9 Ga marine volcanic arc, Skellefte district, Sweden. Econ. Geol. Bull. Soc. 1996, 91, 1022-1053. Persson, J.; Hansson, H.; Renberg, I. The Stone Brusher, a new sampler for submerged epilithic material in shallow streams and lakes. Hydrobiologia 2006, 560, 385-392. Geladi, P.; MacDougall, D.; Martens, H. Linearization and scatter correction for near infrared reflectance spectra of meat. Appl. Spectrosc. 1985, 39, 491-500. Bindler, R.; Bra¨nnvall, M.-L.; Renberg, I.; Emteryd, O.; Grip, H. Natural lead concentrations in pristine boreal forest soils and past pollution trends: A reference for critical load models. Environ. Sci. Technol. 1999, 33, 3362-3367. Klaminder, J.; Bindler, R.; Emteryd, O.; Renberg, I. Uptake and recycling of lead by boreal forest plants: Quantitative estimates from a site in northern Sweden. Geochim. Cosmochim. Acta 2005, 69, 2485-2496. Renberg, I. A procedure for preparing large sets of diatom slides from sediment cores. J. Paleolimnol. 1990, 4, 87-90.
2880
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 8, 2007
(20) Krammer, K.; Lange-Bertalot, H. Bacillariophyceae Band 2/1-4. In Su ¨ sswasserflora von Mitteleuropa; Ettl, H., Gerloff, J., Heyning, H., Mollenhauser, D., Eds.; Gustav Fischer Verlag: Stuttgart, 1986-1991. (21) SIMCA-P Version 10.5, Umetrics AB Computer software, www.umetrics.com, Umetrics AB, Box 7960, SE-907 19 Umeå, Sweden, 2005. (22) Filho, H. A. D.; Galva˜o, R. K. H.; Arau ´ jo, M. C. U.; da Silva, E. C.; Saldanha, T. C. B.; Jose´, G. E.; Pasquini, C.; Raimundo, I. M.; Rohwedder, J. J. R., Jr. A strategy for selecting calibration samples for multivariate modelling. Chemom. Intell. Lab. Syst. 2004, 72, 83-91. (23) Wold, S.; Albano, C.; Dunn, W. J., III; Esbensen, K.; Hellberg, S.; Johansson, E.; Sjo¨stro¨m, M. Pattern Recognition: Finding and Using Regularities in Multivariate Data. In Food Research and Data Analysis; Martens, H.; Russwurm, H., Jr., Eds. Applied Science Publishers: London, 1983. (24) Coomans, D.; Broeckhaert, I.; Derde, M. P.; Tassin, A.; Massart, D. L.; Wold, S. Use of a microcomputer for the definition of multivariate confidence regions in medical diagnosis based on clinical laboratory profiles. Comp. Biomed. Res. 1984, 17, 1-14. (25) Korsman, T.; Nilsson, M.; O ¨ hman, J.; Renberg, I. Near-infrared reflectance spectroscopy of sediments: a potential method to infer the past pH of lakes. Environ. Sci. Technol. 1992, 26, 21222126. (26) Malley, D. F.; Williams, P. C. Use of near-infrared reflectance spectroscopy in prediction of heavy metals in freshwater sediment by their association with organic matter. Environ. Sci. Technol. 1997, 31, 3461-3467. (27) Dåbakk, E.; Nilsson, M.; Geladi, P.; Wold, S.; Renberg, I. Inferring lake water chemistry from filtered seston using NIR spectrometry. Water Res. 2000, 34, 1666-1672. (28) Sjo¨blom, A.; Håkansson, K.; Allard, B. Riverwater metal speciation in a mining region - The influence of wetlands, liming, tributaries, and groundwater. Water Air Soil Pollut. 2004, 152, 173-194.
Received for review September 29, 2006. Revised manuscript received January 26, 2007. Accepted February 7, 2007. ES062329B
