PARTICLE BY PARTICLE CHARACTERISATION USING SEM-EDXA

6.1 Introduction

Over 12,000 particles that deposited to both dry and total Frisbee deposition collectors were characterised for size, shape and associated element composition for material collected during runs 1-4 at the Met. Station. Particle characterisations were carried out by manual and automated individual particle analysis (AIPA) modes using a scanning electron microscope coupled to an energy dispersive X-ray analyser (SEM-EDXA).These analyses were carried out at the Imperial College, Royal School of Mines building, London and I would like to thank John Watt who oversaw this work.

6.1.2 Experimental

Particle analysis was carried out on a JEOL 733 Superprobe SEM which was connected to a Link AN1000 X-ray analysis system. The Link EDXA system collects and stores emitted X-ray spectra with associated size/shape data for individual particles. Data acquisition was controlled by a ä Digiscan software package which assigns characteristic x-ray energies of user-specified elements into energy bands termed regions of interest (ROIs). This software allows up to 25 ROIs to be determined and four ROIs were used for background subtraction. A suite of user-defined lithophilic elements; Na, Al, Si, P, S, Cl, Ca, Cs and Ti which are characteristic of soils from Chilton Berkshire England were determined using energy-dispersive X-ray analysis (Cawse 1974 from Rahn 1976).

A second particle class of the anthropogenically-enhanced elements Fe, Co, Pb, U, Sr and Pu were also determined by EDXA. Elements such as Fe, Pb and U are naturally present in soils and it was assumed that relative percentage composition did not exceed 10% for Fe and Pb (Strahler & Strahler 1973) and 6% relative percentage composition for U (Rahn 1976). Relative percentage compositions in excess of these values may suggest an anthropogenic source although there are a number of analytical problems in identifying man-made from naturally-enhanced concentrations of some minerals such as galena and uranite (Rahn 1976 ; Hamilton & Adie 1982 ;Hunt et al 1992).

Elemental composition data are presented as relative percentage composition and are therefore qualitative. This was due to the unavailability of suitable standards for nuclides such as plutonium, uranium and americium. Also, the SEM laboratory was not equipped nor authorised to deal with radioactive substances. Full details of energy calibration and other procedures can be found in Appendix 8.

6.1.3 SEM manual mode

At the end of each air and deposition sampling period the carbon disks which were placed on the 1 m dry deposition Frisbee inserts were removed and placed in a container to avoid contamination. The circular discs of carbon (Agar Scientific Stansted Essex) have a diameter of 2.5 cm. The collecting surface was coated with an adhesive substrate which ensured that impacted particles were retained and this also helped to alleviate problems of particle bounce and re-entrainment. Prior to SEM analyses the carbon discs were coated with a thin layer of carbon a few Angstroms thick to avoid sample ‘charging’ or electron build-up on top of the sample.

Qualitative information on particle loading and morphological features such as shape were carried out using secondary electron imaging mode. Photomicrographs figures 6.1a and 6.1b show material deposited to one of the carbon discs which was located at Petersburgh Farm. Figure 6.1a clearly shows an assortment of particle shapes and sizes. Energy-dispersive X-ray analyses of these samples indicated that the majority of particles were predominantly aluminosilicates suggesting they were predominantly soil-derived.

Figure 6.1a

Figure 6.1b

Figure 6.1b illustrates a number of large particles which are angular in shape and contained a high relative percentage composition of aluminiosilicates. It was observed that the farmer occasionally used the field in which sampling took place as temporary storage areas for both livestock and agricultural equipment. Based on the SEM projected diameter of these particles which are in excess of 100 um, it is unlikely that they remain airborne for long periods. This suggests that these particles are locally resuspended, possibly re-entrained from vehicular tyres.

6.1.4 Scanning electron microscopy, automated mode

Material deposited to carbon stubs were analysed in the manual scan mode to determine particle loading and agglomeration; areas of high loading and particle agglomeration were avoided. Automated particle size data were obtained by rastering the electron beam between specified grid points using an automated scan mode. With a 1024 by 1024 grid superimposed on an image with a field width of 221.7 um the actual spacing between each point on the grid can be calculated (see figure 6.1c) by

Point space = 221.7

1024 = 0.216 (um)

Figure 6.1c

Grid spacing and Feret measurements

With reference to figure 6.1c particle size data was calculated from a number of Feret measurements. The top particle diameters are two projections taken at right angles to each other where (a) = minimum Feret and (b) = maximum Feret. The particle at the bottom of the SEM grid has an increased number of projections at right angles to each other which helps to improve the accuracy of the mean Feret diameter where mean Feret = projections (a+b+c+d) / 4

Further Feret measurements improves the accuracy of the projected area and 60 Feret measurements were used in this work to determine projected particle size. Data for particle size distribution were defined relative to the projected (mean) mid Feret measurement. For some runs up to 4000 particles were analysed per carbon disks and the large number of particles analysed reduced any inherent bias in choosing this particular Feret measurement over others (see below).

6.1.5 Derivation of projected particle diameter and number volume density

Digiscan software calculates mid, minimum, maximum and mean projected diameters on the basis of Feret measurements. As figure 6.1c show these comprise pairs of tangents which enclose the particle. There are a number of limitations in using this approach to size particles in that two particles may have similar Feret projections as measured by this method but may have totally different shapes. A further limitation to particle sizing using SEM-EDXA is the inclusion of elements within a particle area which are below the user-defined threshold. Threshold selection is determined by what elements you want to measure. Particles containing elements below the threshold will present a ‘hole’ to the electron beam and no information is recorded for that area of the particle (Watt personal communication 1993).

From the projected particle diameter data it is possible to work out a number frequency of particles which fell into size classes based on the mean mid Feret measurements. Assuming spherical shape and a density of 2 g cm³ for these particles it was possible to calculate a mass number volume density NVD (g) for each particle size class. Finally, the NVD (g) value for each particle size class was converted into a size-selective particle mass flux (g m^-2 s^-1). Size selective NVD mass fluxes were referenced against the Pm₁₀ and rod 3 air sample concentrations to derive size-selective NVD dry particle deposition velocities (Mamane & Knoll 1985 ; Schultz 1992).

The size class distribution of particles were determined based on their projected mid Feret diameter (um) for element z > Na. The number frequency of particles within a particular size class N are defined below;

N = ((S - 0.5)/ r) / 1E+06 um

where

S = size class interval of 1 um

(S - 0.5) = mid-point of size class interval

r = radius of size class interval (m)

1E+06 = correction factor used to convert from units of um

The number of particles which fell into a particular size class were tallied to produce a number frequency size distribution. The mean projected mid Feret diameter size class can then be converted to a particle volume mean diameter V by:

V = (4/3) IIr³

An assumed density D of 2 g cm³ was based on the work of ( Rahn 1976)

The number volume density NVD is calculated by;

N*V*D

where:

N = number frequency of particles within each size class

V = volume for each size class mid point (g)

D = assumed density at 2.0 g cm^-3

A mass NVD flux (g m^-2 s^-1) was calculated by dividing NVD (g) by the SEM scanned area (m^-2) / by the sampling period (s).

The mass NVD flux was calculated by:

Mass NVD (g) / Scanned area m^-2

sample period s

where;

the scanned area is a product of the SEM field of view (6.2E-08 m^-2 ) times the number of fields of view scanned for each sample.

The number of fields of view for each run were 228, 117, 45 and 222 for runs 1 - 4 respectively. The sampling period times were 3.2E+6, 1.28E+6, 7.96E+5 and 8.14E+5 seconds for runs 1 - 4 respectively.

6.1.6 Results for projected particle diameter and NVD

Figures 6.1d and 6.1g illustrate the total number frequency and calculated NVD (g) for Frisbee material at the Met. Station for runs 1 - 4. The SEM particle size class data show that for all runs the number frequency of deposited material were dominated by particles less than 10 um projected diameter. The distribution was bi-modal with maxims at 3 um and 6 um.

Particle NVD mass (g) however, was overwhelmingly associated with large particle size in excess of 20 um projected diameter. This was in agreement with the observations by a number of workers that deposition fluxes around industrial plants are dominated by coarse particle deposition (Sehmel 1980 ; Garland & Nicholson 1991 ; Vallack and Chadwick 1989, 1992). For all runs the greater than 11 um size fractions contained in excess of 86% of deposited mass with runs 1 and 3 having in excess of 93% of deposited material.

Table 6.1 illustrates the difference in NVD mass flux and actual mass flux (g m^-2 s^-1 ) deposited to each of the Frisbee dry deposition collectors at 1 m above ground level. Sample masses to the Frisbees were measured using a five decimal place precision balance with a measurement uncertainty of less than 5%.

In general there was a factor of four difference between the Frisbee gravimetric mass flux and the NVD mass flux for all runs except run 3 which shows good agreement between the two. The discrepancies between NVD and actual mass fluxes may be due to (a) Non-representative particle deposition between the Frisbee and carbon stub (b) The SEM scanned field of view was not representative of actual deposition to the Frisbee (c) The assumptions of particle shape and density may be too simplistic under the wide range of environmental conditions experienced during the runs and (d) Errors in gravimetric measurement of actual mass flux where it is possible that the inclusion of Fe-rich particles from rusting of the Frisbee mast were not present on the carbon disk.

Table 6.1

Comparison of dry deposition flux g m^-2 s^-1 to carbon stubs and Frisbees at the Met. Station

The NVD mass flux, however, is based on a projected area diameter which is different to aerodynamic diameter. Schultz (1993) argued that particles are rarely spherical because the surface area comprises voids and craters. Based on the work of Hinds (1982) Schultz used a factor of 0.7 to normalise projected area diameter d_ap to aerodynamic diameter d_ae for material collected by a cascade impactor and deposition plates coated with adhesive. Table 6.1.2 gives normalised data on size selective NVD which measured dry deposition Frisbee material at the Met. Station. These data are normalised by a factor of 0.7 to convert d_ap to d_ae aerodynamic diameter.

Table 6.1.2

NVD mass flux for < 10 um and > 11 um normalised to aerodynamic diameter

The normalised d_ae was referenced against the Pm₁₀ less than 10 um air concentrations and rod 3 air greater than 11 um air concentrations (g m^-3) to give a Vg based on mass (table 6.1.3). With reference to Table 6.1.3 all runs for the < 10 um size fraction deposition velocities were associated with small and sub-micron particle diameters of approximately 1 um. The mass Vg values for the > 11 um size fractions (table 6.1.3) were in agreement with coarse particle size where the lower particle cut-off particle diameters for the impaction rod was 11 um.

Table 6.1.3

Size selective Vg based on mass flux at the Met. Station

With reference to table 6.1.4, air concentration data was available from runs one and two to derive a deposition velocity for the > 11 um size fraction referenced against a combined aerosol (< 10 um + > 11 um) fraction from the Pm₁₀ and rod 3 air samplers. Table 6.1.4 gives data on the <10 um and >11 um mass fractions (g m^-2 s^-1) referenced against the mass loading (g m^-3) collected by the Pm₁₀ and rod 3 samplers for runs 1 and 2.

Table 6.1.4

Vg mass flux based on < 10 um + > 11 um size fractions

The NVD (< 10 um + > 11 um) Vg values can be compared with the Pm₁₀ and rod 3 referenced dry deposition velocities for Cs and Pu in Table 6.1.5.

Table 6.1.5

Comparison of mass based Vg and Vg (AMAD) for Cs and Pu

ND = not determined

With reference to table 6.1.5 there was good agreement between mass flux Vg and ¹³⁷Cs dry deposition velocity (AMAD) for run 1. Plutonium dry deposition velocities (AMAD) for run 1 were higher than mass Vg by a factor of 2.7 and 4 for ^239+240Pu and ²³⁸Pu respectively. The observed relationship between mass Vg and ¹³⁷Cs Vg (AMAD) for run 2 was significantly different by a factor of 37 between the two.

6.1.6 Conclusions to particle size data

The NVD mass flux particle size distribution data show that coarse particles dominated the deposition flux during all sampling periods at the Met. Station. Size-matched NVD mass fluxes with air sampler type (see table 6.1.3) show close agreement to expected values with a mean Vg of 3.97E-03 m s^-1 for the less than 10 um particle diameter and 1.68E-02 m s^-1 for the greater than 11 um fractions.

If the <10 um and >11 um NVD mass fluxes are summed and referenced against the Pm₁₀ mass loading, Vg increases to 1.78E-02 m s^-1 and 1.58E-02 m s^-1 for runs 1 and 2 respectively. This suggests that a representative Vg is more realistic where the deposition flux is matched with its corresponding size-matched air concentration. Thus, the Pm₁₀ referenced dry deposition velocities for Pu and Cs values reported in this work were higher due to an underestimation of large particle deposition to the Pm₁₀ air sampler.

Table 6.1.4 shows good agreement between actual mass Vg and NVD mass Vg with just over a factor of three difference between the two. There was also good correlation ( table 6.1.5) between actual mass Vg, NVD mass Vg and dry deposition velocities (AMAD) for ¹³⁷Cs, ^239+240Pu and ²³⁸Pu.

The data was however, limited. For example, data collected for sampling period 2 yield very different NVD mass Vg and ¹³⁷Cs (AMAD) with a factor of 37 difference between the two. This observed result, however, does not alter the general conclusion that when deposition flux and air concentrations are matched, the derived dry particle deposition velocity for mass and AMAD are lower. It must be noted, however, that the NVD mass flux size-matched dataset was limited to two runs with as yet unquantified errors and uncertainties for measuring Vg using the SEM.

In conclusion, this work shows that the derivation of Vg is subject to inherent sampling artifacts of large particle deposition when using common air sampler types such as the Pm₁₀. The extent of the effect of large particle deposition within the derivation of Vg is difficult to quantify due to the limited dataset and the uncertainties associated with SEM analyses. However, even where the deposition flux and the air concentration are matched, the mass Vg and dry particle deposition velocity AMAD were always greater than the assumed value of 1E-03 m s^-1 which is commonly used to model deposition-related dose assessments from atmospheric dispersion models.