Classification of particles from the farm environment by automated sizing, counting and chemical characterisation with scanning electron microscopy-energy dispersive spectroscopy† Asbjørn Skogstad,* Lene Madsø and Wijnand Eduard National Institute of Occupational Health, P.O. Box 8149 Dep, 0033 Oslo, Norway. E-mail: asbjorn.skogstad@stami.no; Fax: +47 23 19 52 06; Tel:+47 23 19 53 25 Received 13th April 1999, Accepted 28th May 1999 About 60 000 particles in 288 aerosol samples collected during farm work have been characterised with automated particle analysis using scanning electron microscopy (SEM) and energy dispersive X-ray spectrometry (EDS).Based on EDS-analysis of materials with known composition (potato flour, a-quartz, K-feldspar and b-wollastonite), criteria were developed for classification of particles as: (1) organic, (2) silicon-rich (silica), and (3) other inorganic particles.The reproducibility of the relative mass proportions in dust samples collected during farm work was 0.078 when approximately 200 particles were characterised per sample. Field samples from the farm environment showed clear diVerences in composition. Generally, inorganic particles dominated the particle mass.The proportion of the organic particle mass was highest for tending of swine and poultry, 55 and 38% respectively. Silica particles amounted to 10 to 20% of the total mass during handling crops, e.g. grain, straw, hay, potatoes, and onions. It seems likely that the results can be used in etiologic studies, but further validation would be needed for quantitative purposes.automated particle analysis program for SEM-EDS was used Aim of investigation in the farmer study. Such instrumentation has been commer- In a study of farmers’ respiratory health and exposure to cially available for several years. However, a literature search bioaerosols (Melbostad et al.1), exposure to organic dust and revealed only a few studies (Ho� flich3 and Gunst4) on inorganic silica was considered to be of interest.Measurement of these particles collected in the work environment and no studies on components by ashing and X-ray diVraction or infrared spec- organic particles or particles from the farm environment. The trometry was not feasible as sample sizes were expected to be present paper therefore describes classification of particles by small, typically <0.5 mg.A method based on chemical charac- chemical type in samples collected during diVerent types of terisation of particles by combined scanning electron farm work. microscopy and energy dispersive spectroscopy (SEM-EDS) was therefore considered. Materials and methods The analytical scanning electron microscope can provide both chemical and morphological data of individual particles.Sampling It is frequently used for qualitative characterisation of aerosols, The farmers carried battery powered pumps (AFC 123, Casella e.g., in the work environment. However, particle mass esti- Ltd., London, UK) connected to a closed-faced, graphite- mation by microscopy is inaccurate. Particles with surface filled polypropylene filter holder with a 50 mm extension tube structure may not be recognised as single particles but counted (Gelman 25 mm Air Monitoring Cassette, Gelman Sciences, as several particles with smaller size.The recognition of Ann Arbor, MI, USA) loaded with a polycarbonate filter individual particles in complex aggregates can be diYcult and (Poretics, Osmonics, Livermore, CA, USA) with pore size particles may be obscured. The volume of a particle is estimated 0.4 mm.The airflow through the filters was approximately from the microscopic dimensions by assuming that the particle 1 Lmin-1 which was measured with a calibrated Brooks 2-65 height is half of the minimum observed diameter and the mass MM rotameter (Emerson Electric Co., Hatfield, PA, USA).is estimated from the estimated volume and the assumed Sampling time was dependent on the work task, and varied particle density. If these errors are systematic and comparable between 2 to 120 min. for all particle types their influence will be reduced when the relative composition of particle types is estimated and the combined mass is measured by another method, e.g., gra- Preparation vimetry.Finally, the counting process induces an error as a The particle mass on the filters was measured by gravimetry limited number of the particles are analysed. The latter error using an ultra-microbalance (Satorius model S4, Goettingen, and other random errors decrease if the number of analysed Germany) with a detection limit of 3 mg. Samples were resus- particles is increased.Therefore 1000 to 5000 particles have pended in filtered (<0.45 mm) Tween 80 solution (0.05% w/v) been characterised in samples from outdoor air (Hanna et al.,2 by treatment for 3 min in an ultrasonic bath with a frequency and personal communication with the author). of 35 kHz and a high-frequency power of 225W (Sonorex As manual counting would be extremely time consuming an RK510H, Bandelin Electric, Berlin, Germany).Sub-volumes of the particle suspension containing 20–40 mg were filtered through a second polycarbonate filter with pore size 0.4 mm †Presented at AIRMON ’99, Geilo, Norway, February 10–14, 1999. J. Environ. Monit., 1999, 1, 379–382 379using a funnel with a 15 mm internal diameter. This step was fractions were done in Microsoft Excel for Windows 95 version 7.0.necessary to achieve an appropriate density of particles on the filter without overlap and aggregates were also dispersed to The precision was estimated by one-way analysis of variance (ANOVA) of samples that had been analysed twice. The some extent. Specimens of 8 mm×8 mm were cut from the filter and gently fixed on carbon specimen stubs with double- composition of the dust during diVerent work operations was compared by one-way ANOVA and by the Kruskal-Wallis sided carbon adhesive discs.In addition spots of carbon cement were applied at each corner of the specimen to secure test on the proportions of the chemical types. Statistical analyses were carried out with SystatA 8.0 (SPSS Inc., Chicago, good conductivity to earth.Finally the sample was coated with approximately 30 nm carbon film in a sputter coater IL, USA). (Balzers SCD 050, Balzers, Liechtenstein) equipped with a carbon thread flashing unit. Results Analytical procedure Analysis Inorganic particles and most organic particles were observed The specimens were analysed with a Jeol (Akishima, Tokyo, with good contrast in the backscattered electron image mode.Japan) JSM-6400 scanning electron microscope (SEM) con- However, some organic particles appeared smaller in size or nected to a Series II energy dispersive X-ray spectrometer were counted as several smaller particles, Fig. 1A and B. (EDS)-system (NORAN Instruments, formerly Tracor Northern Middleton, WI, USA) equipped with a thin window Classification criteria detector allowing detection of elements with atomic number As the main focus of the project was to study organic dust 6 (carbon).The automated analysis was performed using exposure, classification into three categories was considered the software package ‘Particle Recognition and suYcient for our purpose: 1, organic, 2, silicon-rich, and Characterisation’ (PRC, NORAN Instruments Inc., 3, other inorganic particles.To define the particle types, Middleton, WI, USA) run on the Series II EDS-system. A classification criteria were developed from PRC analysis of a modified version of the PRC program (made locally at particle mixture of quartz, feldspar, wollastonite and starch. NORAN Instruments B.V., Naarden, The Netherlands) was Fig. 2 shows the number of particles (total number 363) as a used which was similar to a previous version of the program function of silicon and carbon content.Since specimens were (Fritz et al.5) where the spectrum was acquired while the flashed with carbon, the silicon counts were corrected for electron beam was scanning over 8 rotational cords through carbon counts by the formula the centroid of the particle.X-ray counts were more reproducible with this acquisition mode as emical inhomogeneities Sicorr= Sicount Totalcount-Ccount and geometric and absorption eVects are averaged over the entire particle during the scan (Johnson6). The SEM was operated under the following conditions: accelerating voltage 20 keV, working distance 8 mm, electron probe current ca. 3 nA, solid-state backscattered electron detection, magnifications 500×for particles 1–5 mm and 100× for particles >5 mm, and X-ray acquisition time 6 s per particle. The elements selected for chemical characterisation were: C, Na, Mg, Al, Si, P, S, K, Ca, Ti, Cr, Mn, Fe, Cu and Zn. The total net X-ray counts varied between 1000 and 5000. The net peak area for a given element was expressed as the percentage of the total X-ray counts.Classification criteria Criteria for classification of particles were based on PRC analysis data of well defined particle types: K-feldspar (Norfloat NGP 325), a-quartz (Fyle-quartz), b-wollastonite and potato flour (commercial domestic quality), prepared and analysed following the same protocol as the field samples. Quantification of field samples Particle volume was estimated using particle area×smallest diameter/2.A minimum of two frames per magnification were analysed. The analysis was terminated when 100 particles or 4 frames per magnification had been counted. The particle mass was estimated assuming that the particle density of inorganic particles, including quartz, was 3 times the density of organic particles.The proportions of three particle types were computed by division by the total estimated mass. Data analysis The particle data were transferred in ASCII format from the Fig. 1 Scanning electron micrographs of organic particles from the EDS-system to a PC using the terminal-emulator program farm environment (poultry house) as seen in (A), secondary electron Procomm Plus 3.0 (Datastorm Technologies, Inc., Columbia, image mode, SEI, and (B), backscattered electron image mode, BEI.MO, USA). A tailor-made Basic-program further made the Particles which are underestimated are indicated by arrows. data accessible for Microsoft-Excel. Computations of the mass 380 J. Environ. Monit., 1999, 1, 379–382combinations as the independent variable, Table 1.The overall precision of the proportions was 0.078. The precision of observations where both replicate analyses had found proportions <0.25 or >0.75 was 0.045, whereas the precision for observations with at least one proportion between 0.25 and 0.75 was 0.10. The diVerence was statistically significant, F36,39=5.14 and p<0.001. There was a tendency to a better precision when more than 200 particles were classified (arithmetic mean 270) compared to samples with less than 200 classified particles (arithmetic mean 127), 0.071 and 0.090 respectively, F24,51=1.61 and p=0.08.Characterisation of field samples Results from the classification of organic, silica and other inorganic particles in samples from diVerent farm work are Fig. 2 Distributions of relative X-ray counts from carbon (CcountM) shown in Table 2.One-way ANOVA showed that the pro- and silicon corrected for carbon (SicorrU) of 363 particles from a mixture of potato starch, quartz, feldspar and wollastonite. portions between tasks were significantly diVerent for all particle types, F213,9=4.4 to 15.8, p=0.01 to<0.001. However, the conditions for ANOVA are not satisfied as the proportions The particles were classified by the following criteria.First were not normally distributed and the variances were not were particles with Ccount80 classified as organic. Of the homogeneous. The non-parametric Kruskal-Wallis test was remaining particles were particles with Sicorr>90 classified as therefore also used, which also showed that the diVerences quartz.The following particles were classified as other inorbetween tasks were significant for all particle types, p<0.001. ganic particles: 1, particles 5 mm with Sicorr>35 and Sicorr90; 2, particles >5 mm with Sicorr>40 and Sicorr90; and 3, particles with at least one element count >10 except Discussion carbon. When these criteria were applied approximately 65% Detection of particles with the PRC program relies on a high of the particles were classified.The remaining particles were contrast, preferably binary image, with the particles as white classified as organic particles as the Sicorr and the counts of structures against a black background. As backscattered elec- single inorganic elements were too low for positive recognition tron production is strongly correlated with atomic number, as an inorganic particle.Of these particles 60–80% had inorganic particles are easily recognised using the signal from Ccount70 and 90–96% had Sicount10 for particles >5 and the back-scattered electron detector. Organic particles have a 5 mm respectively. high content of light elements and their average atomic number is close to that of the filter.It can therefore be diYcult to Reproducibility obtain suYcient contrast between organic particles and background. A satisfactory contrast was achieved by fixing the Twenty five specimens were reanalysed. A total of 50 to 500 particles were classified per analysis. The standard deviation backside of the filter to the specimen stub using double-sided carbon adhesive discs.Problems with static electric charging was estimated by one-way ANOVA with sample-particle type of the specimen were absent and a smooth background with a good signal-to-noise ratio was obtained. Although it is likely Table 1 One-way ANOVA of replicate counting of 25 samples colthat some particles are not correctly sized or recognised as lected during farm work.The precision was estimated from the residual variance two or more smaller particles, the underestimation of organic particles was minimised by the improved preparation Source of variance df Mean square F-test p technique. The PRC program classifies particles with low X-ray counts Between all combinations of as non-integratable and does not store the element counts for sample and particle type 74 0.142 23.5 %0.001 such particles.As non-integratable particles from the farm Residual 75 0.00604a environment are likely to be of organic origin, carbon was aPrecision=Ó0.00604=0.078. included in the X-ray acquisition. Very few non-integratable Table 2 Composition of the dust that farmers are exposed to during diVerent tasks Particle type Organic Silica Other inorganic Task Na AM b Median sc AM Median s AM Median s Threshing 21 0.26 0.24 0.21 0.17 0.15 0.12 0.58 0.56 0.28 Handling of hay 28 0.18 0.14 0.15 0.12 0.08 0.13 0.71 0.79 0.20 Handling of silage 15 0.12 0.08 0.10 0.06 0.03 0.07 0.82 0.87 0.13 Sorting of potatoes 20 0.02 0.01 0.02 0.11 0.10 0.06 0.88 0.88 0.06 Sorting of onions 16 0.03 0.02 0.02 0.12 0.11 0.03 0.85 0.86 0.04 Tending of— Dairy cows and cattle 23 0.32 0.31 0.20 0.08 0.02 0.14 0.61 0.62 0.20 Sheep/goats 26 0.22 0.14 0.27 0.15 0.10 0.15 0.64 0.71 0.28 Poultry 31 0.39 0.33 0.26 0.02 0.01 0.03 0.60 0.64 0.25 Swine 25 0.54 0.51 0.15 0.07 0.05 0.07 0.40 0.40 0.14 Handling of liquid manure 18 0.16 0.07 0.23 0.15 0.11 0.16 0.70 0.80 0.23 aSample size.bArithmetic mean. cSample standard deviation.J. Environ. Monit., 1999, 1, 379–382 381particles were observed thereafter. Analysis of the test materials Large diVerences between the proportions of the particle types during diVerent tasks were found. Except for tending of showed that potato starch particles could clearly be classified on the basis of the carbon count. Silicon counts were primarily swine and poultry, the proportion of organic particles was low.Although many work operations in farming involve used for chemical classification of particles. Inorganic particles showed substantial and variable carbon counts due to the handling of organic materials, the generated dust is dominated by inorganic particles. A possible explanation is that the carbon coating which complicated the classification.The carbon count was therefore subtracted from the total X-ray surface of farm materials like grain and hay is contaminated with particles from the soil which can settle from air or deposit count after which quartz could be clearly distinguished from the silicates wollastonite and feldspar (Fig. 2). As the total by rain-splash. During handling of farm materials particles on the surface are more likely to be dispersed into the air than X-ray count excluding carbon could be very small for organic particles, the relative silicon count could be high even with a the organic substrate itself.The soil as a source of inorganic contamination is also plausible since the mineral soils which low silicon count. Therefore carbon counts >80% of the total count were used as a primary criterion for classification of dominate in the region of the study have an organic content which is often less than 5% by weight (Singh et al.7).The organic particles. About 35% of the particles that were not classified as silica or inorganic were finally classified as organic silica content of the dust varied from 2 to 17% which also may point at soil as an important source of dust as soil is as they were relatively high in carbon and low in silicon.It is possible, however, that some of these particles were not organic probably the only likely source of silica dust in the farm environment. or were aggregates of both organic and inorganic particles. The proportion of organic particles may therefore have been overestimated. However, this error is counteracted by under- Conclusions estimation of size and number of some organic particles.Accuracy was further studied by the reproducibility of the An automated method for sizing, counting and chemical characterisation of particles collected in the farm environment estimated proportions of the particle types. The overall standard deviation was 0.078 when on average 200 particles were by scanning electron microscopy-energy dispersive spectroscopy was developed.It was possible to classify particles as analysed per analysis. The precision was suYcient for the comparison of field samples from the farm environment where organic, silicon-rich (silica) and other inorganic particles. Large diVerences in composition of the airborne dust were highly significant diVerences between tasks were found.The standard deviation was smaller in samples where a larger observed during diVerent types of farm work. Most of the dust was inorganic and probably due to contamination from number of particles had been analysed as should be expected. However, a much larger number of particles need to be the soil. It therefore seems likely that the results can be used in etiologic studies, but further validation is needed for quanti- classified to improve the precision substantially.Hanna et al.2 mention ranges of 5–11% and 26–37% for the proportions of tative purposes. two chemical particle types in samples collected in outdoor air. These data suggest standard deviations of 0.015 to 0.03 References which were found when 1000 or 5000 particles where classified per analysis (personal communication with the author).Low 1 E. Melbostad, W. Eduard and P. Magnus, Scand. J. Work Environ. Health, 1997, 23, 271. (<0.25) and high (0.75) proportions were estimated with 2 R. B. Hanna, K. J. Karcich and D. L. Johnson, Scanning Electron better precision which is similar to the binomial distribution. Microsc, 1980, 1, 323. The results probably have a bias due to many assumptions 3 B. Ho� flich, Diploma Thesis, Technical University of Darmstadt, that have been made in the quantification. In addition soluble 1997, (in German). particles as salts and sugars were lost as samples needed to be 4 S.Gunst, Diploma Thesis, Technical University of Darmstadt, 1997, (in German). resuspended to obtain an optimal particle density on the SEM 5 G. S. Fritz, J. J. McCarthy and R. J. Lee, in Proceedings of the 16th specimen. If the bias does not depend on the type of farm Annual Conference of the Microbeam Analysis Society, ed. work, i.e., particle types are similar, the results may still be R. H. Geiss, San Francisco Press, CA, 1981, pp. 57–60. useful in etiologic research. As the farmer study aims at 6 D. L. Johnson, Scanning Electron Microsc, 1983, 3, 1211. identifying the most important components in airborne dust 7 B. R. Singh, T. Børresen, G. Uhlen and E. Ekeberg, in Management for respiratory eVects among farmers, further validation of of Carbon Sequestration in Soil, CRC Press, Boca Raton, New York, 1988, pp. 195–208. the method, e.g., by comparison to standard methods, was postponed until more was known about the importance of organic dust. Paper 9/02941I 382 J. Environ. Monit., 1999, 1, 379–3