Aerodynamic particle size spectrometers are a well-established method to measure number size distributions of coarse mode particles in the atmosphere. Quality assurance is essential for atmospheric observational aerosol networks to obtain comparable results with known uncertainties. In a laboratory study within the framework of ACTRIS (Aerosols, Clouds, and Trace gases Research Infrastructure Network), 15 aerodynamic particle size spectrometers (APS model 3321, TSI Inc., St. Paul, MN, USA) were compared with a focus on flow rates, particle sizing, and the unit-to-unit variability of the particle number size distribution.
Flow rate deviations were relatively small (within a few percent), while the
sizing accuracy was found to be within 10 % compared to polystyrene latex
(PSL) reference particles. The unit-to-unit variability in terms of the
particle number size distribution during this study was within 10 % to
20 % for particles in the range of 0.9 up to 3
In order to perform a quantitative quality assurance, a traceable reference
method for the particle number concentration in the size range 0.5–3
Coarse aerosol particles in the atmosphere can have a significant influence on the optical properties of the atmospheric aerosol as well as on the total particle mass concentration. Generally, aerodynamic and optical particle size spectrometers are employed in atmospheric observational aerosol networks to directly measure the number size distribution of the coarse mode particles.
The Aerodynamic Particle Sizer spectrometer (APS model 3321, TSI Inc., St. Paul, MN, USA) is based on the acceleration of aerosol particles immersed in an air flow through a nozzle (Agarwal et al., 1979; Chen et al., 1985). The time of flight (TOF) of individual particles after acceleration is determined between two laser beams. Due to their longer relaxation time, the TOF of larger particles is longer than for smaller particles. The conversion of TOF to aerodynamic particle size classes is achieved by a calibration with polystyrene latex (PSL) spheres. Compared to optical particle size spectrometers with coherent light sources, the measuring principle of an APS is not influenced by ambiguities in the relation of the detected signal to a particle size, meaning that the calibration curve has a monotonic response over its full size range. Due to the measuring principle of the APS model 3321 made by TSI, it is however possible to measure the aerodynamic (TOF) and optical properties (scattered light) of individual particles at the same time in the so-called “correlated mode”.
Nevertheless, the measurements of the aerodynamic particle size spectrometer can be influenced by a variety of errors, depending on the version or type. In general, the sizing accuracy is known and has been published by Peters and Leith (2003). The issue of coincidence of older versions of the APS was solved with the production of the model 3320. However, for this model, Armendariz and Leith (2002) showed a discrepancy between the results of the summed aerodynamic mode and the correlated measuring mode, which was resolved in the latest APS model 3321. However, Peters and Leith (2003) showed that this model had a lower counting efficiency than its predecessor.
Only a few of the published performance studies deal with results of more than one device of the same type; e.g. Volckens and Peters (2005) reported on a study with three units APS model 3321. In general, a better knowledge of the unit-to-unit variability is essential in terms of the particle number size distribution. In particular, this aspect becomes important for analysis and interpretation of the results from observational atmospheric aerosol networks. Wiedensohler et al. (2012) have emphasized, that due to the growing number of measurement sites, quality controls are important to achieve comparability due to well-known uncertainties of the particle number size distribution.
Overview of compared TSI 3321 devices of the specific institute (Institute of Chemical Process Fundamentals ICPF, Institute for Atmospheric Sciences and Climate ISAC, Joint Research Center JRC, Navarino Environmental Observatory NEO, Leibniz Institute for Tropospheric Research TROPOS, Umweltbundesamt UBA, University of Helsinki UHEL) and sorted/indexed by age.
In the framework of ACTRIS, an intercomparison workshop for aerodynamic
particle size spectrometers was carried out at the facility of the World
Calibration Center for Aerosol Physics (WCCAP). This study dealt with the
comparability of 15 aerodynamic particle size spectrometers in terms of
their sizing accuracy and the unit-to-unit-variability of the particle size
distribution in the size range from 0.6 to 5
The core element in the measurement setup is a cubic mixing chamber with a
volume of approximately 0.5 m
In this intercomparison study, 15 units APS model 3321 (TSI Inc.) have been analyzed. An overview of all devices is given in Table 1. For the majority of devices, the last official calibration from manufacturer is not older than 3 years. Because of the limited number of chamber outlets, the devices were divided into two groups using one device (TROPOS F) as a relative reference in both runs.
In both runs, eight devices were mounted vertically underneath the
individual outlets (see Fig. 1). This arrangement basically ensures no
particle losses due to impaction or sedimentation from the mixing chamber to
the individual devices. For all devices, a special attachment for the inlet
was used, which decouples the aerosol flow (1 L min
For analyzing the sizing accuracy, PSL spheres have been re-suspended, using
a nebulizer in combination with a silica-gel aerosol diffusion dryer. To
optimize the experimental design, the sampling matrix of the PSL size
calibrations has been done with two mixtures of three different PSL particle
sizes (0.7, 1.0 and 2.0
To obtain the unit-to-unit variability of the aerodynamic particle number size distribution over a wide particle size range, two procedures were
carried out: (a) overnight measurements of the ambient aerosol and (b) by
using a custom-made coarse-mode-nebulizer to produce coarse mode ammonium
sulfate particles up to 5
The quality of an APS in terms of sizing accuracy or particle number
concentration (distribution) strongly depends on its aerosol and sheath flow
rates. The manufacturer specifies the aerosol flow with
Photo of the measuring setup for the intercomparison of eight units APS 3321.
Measured aerosol and sheath flow rates of the initial state.
The mean particle diameters were determined by fitting a multi-modal logarithmic function to the measured particle number distributions of the re-suspended PSL mixtures. These results were compared to aerodynamic diameters calculated from the manufacturer's data, considering the Cunningham slip correction, but no ultra-Stokes effects (Wang and John, 1987). The relative deviation between both values is shown in Fig. 3.
For the majority of devices, the deviations in terms of sizing are less than
10 %, with a few exceptions. ICPF A shows significantly higher values over
a wide range. This may be a result of its flow re-adjustment, while the TOF
calibration was untouched. Also for NEO, the internal TOF calibration
parameters seem unsuitable and incorrect for the re-adjusted flow rates. On
average, an optimum for 1.6
Relative deviation of the measured aerodynamic diameter of six PSL sphere sizes.
The sizing first had to be corrected to merge the results of the runs of the
different sets of instruments and to make them comparable. This was done to
decouple the variability in sizing from the concentration measurements.
Because of the diverse influences for smaller and larger particles, the
sizing for the entire particle size range was corrected using only the
results from 1.6
The results are shown in Fig. 4. The particle number size distributions
for the 15 devices strongly deviate, especially in the sub-micron size
range. For the lowest size channels, the deviation is up to a factor of 10.
The mean relative deviation (95 % confidence interval) decreases steadily
from approximately 60 % for the smallest size channels and reaches a
minimum with values of 10–20 % in the size range from 0.9 up
to 3
Four devices stood out from the general behaviour and showed a poorer
performance, especially in the size range with the smallest variability
among the units overall (0.9–3.0
Merged results of both runs for ammonium sulfate (left column) and ambient aerosol (right column), particle number size distribution (upper row) and relative deviation from average (lower row). The grey shaded range is the mean deviation (95 % confidence interval) of the selected values.
Relative deviation of the calculated total number, surface and
volume concentration for the measured distribution relative to the averaged
distribution: ammonium sulfate (left column) and ambient aerosol (right
column), full size range (upper row) and for particles larger 0.9
The relatively large unit-to-unit variability up to 60 % between the particle number size distributions did not meet the expectations. The measured flow rates lay within the specified range or were re-adjusted to the reference values. On average, the size accuracy was within 10 %. Furthermore, although no TOF-recalibration has been performed, the deviations in sizing were corrected roughly in a post-processing step.
Taking into account the specified range for the aerosol flow rate, a
variation of 10 % for the concentration seems to be acceptable. Because of
insufficient sizing accuracy in combination with a moderately sloping
particle number size distribution, the expected variability is slightly
larger (approximately 20 %). With increasing slope this aspect becomes
more important, e.g. for ammonium sulfate larger than 3
The large unit-to-unit variability of the number size distribution in the
sub-micron range certainly results from individual differences in unit
counting efficiencies. This issue was analyzed in several studies for
previous TSI APS models as well as for the latest model 3321 (Karg et al.,
1991; Armendariz and Leith, 2002; Peters and Leith, 2003; Volckens and
Peters,
2005). In general, the counting efficiency of a TSI APS model 3321 is
influenced by aspiration losses, transmission losses and detector errors
(Volcken and Peters, 2005). The detector error is associated with low pulse
height of the optical signals used for the TOF measurement. The effect can
be divided into two types.
Just one of the two signals is lower than a certain threshold. Such events
are rejected for the particle number size distribution. However, it is
marked and counted by the device as “Event Type I”. Neither of the two signals reaches the threshold, because the particle
misses the laser beam or scatters just too little light for other reasons.
Such particles are completely undetected by the device.
The unit-to-unit-variability in the sub-micron range should be primarily based on these two types of detector errors. Either the general quality of the optics (cleanliness of the optical components, detector sensitivity, laser beam focusing, etc.) or the precision of the alignment of the aerosol flow and the laser beam could be a reason for this variability.
Karg et al. (1991) already showed that the counting efficiency may depend also on the sample. This aspect is reasonable in the context that the counting efficiency in the sub-micron range is based on the detector error. Therefore, the counting efficiency is also a function of the optical properties (primarily the complex refractive index) of the sample, which means it is rather a function of the optical diameter than the aerodynamic diameter. Extended analysis of pulse pair type or the scattering signal in the correlated mode are necessary. Although a slight deviation between the results of the two samples is noticeable, an independent measuring principle is necessary to investigate this effect. The significant deviation between laboratory generated ammonium sulfate and ambient aerosol of some devices in the coarse mode size range cannot be explained by this argumentation.
During the intercomparison, no traceable reference method for a particle
number concentration was available for aerosol particles between 0.5 and
3
The resulting deviation for the calculated integral values of total number,
surface and volume concentration is shown in Fig. 5. The mean variability
of ammonium sulfate is much smaller than for ambient aerosol, due to the
higher concentration in the super-micron range. This is also the reason
for the lower variability of the total particle number concentration
compared to the total particle volume, due to the stronger weighting of
larger particles. Compared to the whole particle size range, the variability
for particles larger than 0.9
Quality controls are essential to get comparable and accurate results for atmospheric measurement networks. In the framework of ACTRIS, 15 aerodynamic particle size spectrometers were intercompared with a focus on the basic parameters: flow rates, size accuracy, and concentration.
For the majority of devices, the measured flow rates were in the specified
tolerance range of 0.9–1.1 and 3.9–4.1 L min
The most significant differences and variability can be found for the
concentration measurements. The size range up to 0.9
Naturally, the significant unit-to-unit variability propagates for the
derived integrated values (total number, surface and volume concentration).
Only for the size range larger than 0.9
Some devices have shown an extraordinarily poor quality based on technical
defects or insufficient calibration. These instruments have not been
considered in the final analysis. In conclusion, a few points should be
emphasized for the future, considering long-term measurements.
Quality checks for flow rates and size accuracy should be a standard
procedure in the field. After a readjustment of the flow rates, a TOF
re-calibration might be needed. Measured particle number size distributions are influenced by counting
efficiency effects. Individual correction functions are needed as a standard
data processing step to get comparable results. For quality controls of concentration measurements and to derive such
counting efficiency functions a traceable reference method is needed for
number concentrations in the particle size range from 0.5 to 3
This work was accomplished by the European research infrastructure project ACTRIS (Aerosols, Clouds, and Trace gases Research InfraStructure Network). The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007–2013) project No. 262254.
We would like to thank TSI for their support during the workshop. This laboratory study was independently performed and was not co-funded by TSI. Edited by: S. Malinowski