The optical particle sensor (OPC) under evaluation in the current work is the OPC-N2 manufactured commercially by Alphasense (http://www.alphasense.com) and is described in detail in Sousan et al. (2016). The OPC-N2 can be considered as a miniaturised OPC as it measures 75 mm × 60 mm × 65 mm and weighs under 105 g, and as such is significantly cheaper (approx. GBP 250) than the comparable reference instruments (see next section). The OPC-N2 samples via small fan aspirator and measures particle number concentration over a reported size range of 0.38 to 17 µm across 16 size bins, and maximum particle count of 10 000 s-1. The minimum time resolution is 10 s. The measured particle number concentration is converted via on-board factory calibration to particle mass concentrations for PM1, PM2.5 and PM10 size fraction according to European Standard EN481 (OPC-N2 manual). According the OPC-N2 manual, the standard definition for PM10 in EN481 extends beyond the particle size measured by the OPC-N2 and may consequently underestimate PM10 value by up to 10 %. Further discussion on calculations for conversion from particle number to mass concentrations is given in Sect. 2.3. All OPC-N2s in this study used firmware version 18.

The OPC-N2 is designed to log data via a laptop using software supplied by Alphasense; however, this may not be practical when using multiple OPC-N2s at once or for personal monitoring. Therefore, we developed a custom built system for logging the OPC-N2 during the inter-comparison, utilising either a Raspberry Pi 3 or Arduino system. The Python code to log the outputs from OPC-N2 on a Raspberry Pi 3 is made available in the Supplement. The Python code makes use of the py-opc Python library for operating the OPC-N2 written by Hagan (2017).

The first reference instrument was a TSI 3330 optical particle spectrophotometer (OPS), which measures particles number concentrations between 0.3 and 10 µm across 16 size bins, with a maximum particle count of 3000 particles cm-3. A GRIMM portable aerosol spectrometer (PAS-1.108, hereafter referred to as the GRIMM) was also utilised, which records particle number concentrations in 15 bins from 0.3 to 20 µm. The TSI 3330 and GRIMM were both recently calibrated and serviced. All measurements of airborne particle concentrations are inherently operationally defined and as a result the TSI 3330 and the GRIMM were chosen as reference instruments as they measure particle size in similar size bins by a similar photometric technique to the Alphasense OPC-N2.

For the sake of this inter-comparison, we have taken the TSI 3330 and GRIMM data as an accurate measure of particle mass concentrations. The reference instrument used for the factory calibration of the OPC-N2 by Alphasense is the TSI 3330 (Sousan et al., 2016) and hence included for comparison.

The instruments were housed within the Elms Road Observatory Station (EROS), located on the University of Birmingham campus. The site is classed as urban background, with emissions from nearby road and a construction site the major sources of particles. Fourteen OPC-N2s were deployed at EROS, enabling the precision of the OPC-N2 to be assessed along with the accuracy relative to the reference instruments, the TSI 3330 and GRIMM. An intensive inter-comparison ran for just over 5 weeks, from 26 August until 3 October 2016, during which all 14 OPC-N2s, the TSI 3330 and the GRIMM sampled ambient air. Minimal lengths (12 cm) of stainless steel tubing (OPC-N2) and conductive black tubing (TSI 3330 and GRIMM) were used to sample outside air, with each OPC having its own inlet at a height of 1.5 m. The vertical inlet for the TSI 3330 necessitated a bend in the tubing; however the calculated sampling efficiency (using von der Weiden et al., 2009) was 92 % for particles with a diameter of 10 µm. Therefore, while the inlet arrangement of the TSI 3330 may have affected the inter-comparison, particularly when considering the accuracy of the OPC-N2, we were limited to what was practical. Sampling intervals for the OPC-N2, TSI 3330 and GRIMM were 10, 60 and 6 s, respectively. In addition, RH measurements from the nearby Elms Road Meteorological station were also obtained which is located approximately 100 m away from EROS.

At the conclusion of the intensive inter-comparison, a subset of the OPC-N2 (5) continued to sample at EROS along with the GRIMM, to test the robustness and suitability of the OPC-N2 for longer-term monitoring. The long-term monitoring concluded on 1 February 2017, meaning that these OPC-N2s sampled ambient air for up to 5 months.

For regulatory purposes, an accepted method for measuring particle mass concentrations is a tapered element oscillating microbalance (TEOM), and therefore we also compared the OPC-N2 to this technique despite the difference in particle measurement approaches. An urban background air monitoring station part of the UK Automatic and Rural Urban Network (AURN) nearby EROS (Tyburn Rd) was chosen for this inter-comparison. At the Tyburn Rd AURN station, the TEOM monitor was fitted with a filter dynamic measurement system (FDMS) (Grover et al., 2006), to correct for semi-volatile particle loss. A subset of OPC-N2s (4) and the GRIMM PAS 1.108 that were deployed at EROS sampled at Tyburn Rd station for 2 weeks during February 2017. The OPC-N2 was housed individually within waterproof boxes on the roof of the cabin near to the TEOM inlet in order to keep the inlet length the same as used at EROS. The GRIMM sampled from a nearby separate inlet.

The two light-scattering optical particle counters used as reference instruments in this study were found to be well correlated (r2 > 0.9), with the GRIMM recording between 20 and 30 % higher concentrations for all three particle mass fractions (Fig. S1, Supplement). The GRIMM is known to overestimate number concentration (Sousan et al., 2016, and references therein) and this difference may reflect differing efficiencies in particle detection between the two instruments.

The performance of the custom built logging systems varied between 44 and 94 % successful data capture, with the Arduino and Raspberry Pi systems giving 44–65 and > 92 %, respectively. The Raspberry Pi data logger system was used for the long-term measurements and for the inter-comparison with the AURN site due to its better performance. The data losses were due to hardware issues and not related to performance of the OPC-N2. Due to the missing data, only a subset of measured PM2.5 concentrations when all 14 OPC-N2s were logging are shown in Fig. 1, along with measured concentrations by the reference instruments. From Fig. 1, while there are times when there appears to be excellent agreement between the OPC-N2s and the reference instruments, there are times when the OPC-N2s record a significant positive artefact, and during these times the spread in measured concentrations increases. For example, on the morning of 18 September, the range of measured concentrations by the individual OPC-N2 was from approximately 30 to 150 µg m-3, whereas the reference instruments reported ∼ 10 µg m-3. The cause of the positive artefact is investigated in later sections, but it points to the individual OPC-N2 responding differently to this artefact. Similar trends were also observed for PM1 and PM10; see Fig. S2 in the Supplement.

As there is a considerable spread in response for the OPC-N2 relative to the reference instruments, we then quantified whether it was always the same OPC-N2 reading low and high. Due to the aforementioned data capture issues, this analysis was only applied to days when all 14 OPC-N2s were running, 21–24 September (Fig. 1). The results are shown as a rank order plot, where the OPC-N2 observations are ordered from the highest reported value to the lowest over this period, normalised to the median concentration at the start of the analysis (t = 0), shown for PM2.5 mass concentration in Fig. 2. The ranking of the OPC-N2s showed some variability over time within periods of 1–6 h, which was particularly noticeable during periods when the OPC-N2 signals underwent large changes in concentrations. This demonstrates that the highest and lowest reporting OPC was not consistently reporting the highest and lowest the lowest PM2.5 concentrations, respectively, over the whole 3-day period. The same trend was also observed for PM1 and PM10 mass concentrations, as shown in Fig. S3 (Supplement).

For the 3-day time period (21–24 September) we applied the rank order analysis. Two subsets of concentrations measured by the OPC-N2 were evident in the time series (Fig. 1), one a period of highly variable mass concentrations (00:00 BST, 21 September, to 12:00 BST, 22 September 2016) of September) followed by more stable mass concentrations (12:00 BST, 22 September 2016 onward). This was reflected in the corresponding rank order plots where relatively consistent OPC rank orders were observed throughout the variable and comparatively stable PM concentrations periods. However, there is a noticeable transition between the two periods in the rank order plot, observed at approximately 12:00 BST on 22 September. This transition in rank orders would reflect the difference in OPC PM sensitivities, random noise and offset values between each OPC. Over the 3-day period the OPCs appeared to hold their response characteristics and hence rank orders well, suggesting that over this timescale quantitative concentrations could be directly compared. Due to the changing response and the incomplete data coverage, for the rest of the analysis in this paper, when comparing to the reference instruments, the median and inter-quartiles concentrations of all 14 OPC-N2s were used.

One measure of the precision of a group of instruments is the coefficient of variance (CV) and this was calculated for the measured ambient mass concentrations of all 14 OPC-N2s to assess the variability between 14 instruments. The average CV was 0.32 ± 0.16, 0.25 ± 0.14 and 0.22 ± 0.13 for PM1, PM2.5 and PM10 mass concentrations, respectively. This is higher than the value of 0.1 considered acceptable for duplicate instruments by the US EPA (see Sousan et al., 2016, and references therein) but perhaps not unreasonable for low-cost sensors. This may in part be due the OPC-N2 all sampling from separate but identical inlets but suggests the precision of the OPC-N2 would need to be considered when comparing multiple units. To analyse whether the CV for the OPC-N2 varied over the month, the median concentration was plotted along with the CV (shown for PM2.5 in Fig. 3). Throughout the measurement period, the CV was fairly consistent (mean of 0.22 ± 0.13), with spikes in CV values evident during periods of high PM2.5 concentrations, in agreement with trends observed in Fig. 1. We observed a similar trend of consistent CV values for both PM1 and PM10 concentrations, suggesting reasonably stable agreement between all OPC-N2s over a 5-week period.

The median and inter-quartiles of the measured PM concentrations from the 14 OPC-N2s were used to compare the measured particle mass concentrations to the reference instruments (Fig. 4). From Fig. 4, the notably similar distributions across all three particle size fractions for the first and third quartiles indicate good agreement between the 14 OPC-N2s, further highlighting the reasonable degree of precision between the OPC-N2s as shown in the previous section. At typical ambient PM2.5 and PM10 mass concentrations for the UK, similar distributions were observed for the OPC-N2s and reference instruments (Fig. 1), suggesting reasonable agreement between the devices. In contrast, different distributions were observed for the PM1 fraction, with the OPC-N2 and GRIMM in agreement but appearing to overestimating the PM1 mass concentrations with respect to the TSI 3330. The OPC-N2 has a higher particle size cut-off (0.38 µm) compared to the TSI (0.3 µm), and this may explain the observed difference in frequency distribution for PM1 (Fig. 1). While the TSI and GRIMM have the same particle size cut-off (0.3 µm), these instruments have been shown to disagree (Fig. S1), possibly due to different particle collection efficiencies.

When the median and inter-quartile OPC-N2 concentrations were plotted against the TSI and GRIMM concentrations, the slope was greater than unity for all three size fractions (Table 1), indicating that the OPC-N2s were overestimating the ambient particle mass concentrations (approx. 2 to 5 times, Table 1). Overall, the OPC-N2s and GRIMM were in better agreement compared to the TSI for all size fractions (Table 1). The GRIMM was found to record PM concentrations 20–30 % higher compared to the TSI (Fig. S1), and this could in part account for the observed lower slopes between the GRIMM and the OPC-N2s.

The time series of the median OPC-N2 PM2.5 concentrations along with the two reference instruments are shown in Fig. 5, and for a large portion of the inter-comparison all instruments appear to be in agreement. However, there were a number of times when the OPC-N2 readings were up to an order of magnitude higher relative to the reference (e.g. 15 September), pointing to a significant instrument artefact. On 15 September, the GRIMM and TSI also move out of agreement and may point to the same artefact affecting the GRIMM. Similar trends were also observed for the PM1 and PM10 mass fractions (Fig. S4, Supplement) with the OPC-N2 overestimating the PM10 concentration by several orders of magnitude on 15 September (peak mass concentrations in the order of 15 000 µg m-3). Note that as EROS is an urban background site, it was unlikely to be affected by plumes from sources such as vehicles and as a result these high concentrations spikes may not be real.

The factors contributing to this apparent artefact shown by the OPC-N2 were investigated. In Fig. 6, the agreement between the OPC-N2 and the TSI instrument appears to vary as a function of ambient RH, with better agreement observed between the two instruments during periods of relatively low ambient RH. However, during times when the RH was high (> 90 %), the OPC-N2 recorded concentrations markedly higher than that measured by the TSI 3330 (Fig. 6). Similar trends were also observed for PM1 and PM10 mass concentrations (Fig. S5, Supplement). Thus, it points to ambient RH as a significant contributing factor affecting the particle mass concentrations measured by the OPC-N2, and this is tested further in later sections. There are distinct differences in design in OPC-N2 compared to the reference instruments (GRIMM and TSI 3330) as both the TSI 3330 and GRIMM utilise a sheath flow unlike the OPC-N2. The sheath flow in both devices will be warmed to temperatures higher than the ambient air due to proximity to the instrument pumps and electronics. This would mean that they measure at a lower RH than ambient and could explain why no RH dependence was observed on measured particle concentrations by the GRIMM and TSI 3330.

We deployed a subset of the OPC-N2 devices (4) and the GRIMM at an urban background AURN station to enable comparison of the measured ambient particle mass concentrations to a TEOM-FDMS. The time series of the measured concentrations of PM10 and PM2.5 for all instruments is shown in Fig. 7. The two reference instruments were found to be well correlated (r2 > 0.91, Fig. S6, Supplement), but the GRIMM reading was about 20 % lower than the TEOM, in agreement with previous work (Grover et al., 2006). From Fig. 6, periods of agreement between the four OPC-N2s and the reference instruments (GRIMM and TEOM) were apparent, along with times when the four OPC-N2s measured concentrations that were notably higher than the reference instruments. Overall, when compared to the TEOM, the OPC-N2 measurements were 2.5–3.9 times higher for both the PM10 and PM2.5, with considerable scatter observed (Table 2).

Closer inspection of Fig. 7 indicated that the times when the four OPC-N2s overestimated the particle mass concentrations were during times of high RH (e.g. 12–14 February), as observed in the previous section. However, there were periods of high RH when the four OPC-N2s and the TEOM were in better agreement (e.g. 20 February onwards), indicating that the large positive artefact observed in the OPC-N2 was not just related to RH. Rather, it appears that positive artefact was observed during times when the volatile fraction measured by the TEOM was relatively high, as well as higher RH, as was observed on 12–14 February (Fig. 7). Thus, it suggests that the ambient aerosol composition also contributed to the significant positive artefact in the OPC-N2. A recent laboratory study found that the particle mass concentrations measured by OPC-N2 for all three size fractions were highly linear with respect to gravimetrically corrected reference instruments but that the slope was dependent on the aerosol type (Sousan et al., 2016). Sousan et al. (2016) observed in the PM10 fraction slopes greater than unity for Arizona road dust but less than unity for salt and therefore suggest that changes in aerosol composition may also account for the differences observed between the reference instruments and OPC-N2 (Fig. 7). This result highlights a limitation when comparing optical methods to gravimetric – as differences may be due to changes in particle mass, size distribution or composition: as all can affect the ability of a particle to scatter light (Holstius et al., 2014).

From Fig. 6, the times when there was a large positive artefact in the OPC-N2 occurred when the RH was above 85 %. If we exclude these times when the RH was over this threshold, better agreement between the four OPC-N2s and the TEOM was observed, with slopes between 1.1 and 1.7 for both size fractions (Table 2). One of the OPC-N2s recorded notably higher mass concentrations compared to the reference instruments (OPC11), compared to the other three OPC-N2s (Table 2), and this highlights the need to calibrate each OPC individually before use in field measurements.

After the conclusion of the intensive measurements at EROS (Sect. 3.1), five of the OPC-N2s continued monitoring for a further 4 months to examine whether there was any evidence of instrument drift over time, along with the GRIMM as reference. One of the OPC-N2s failed in December, and so was excluded from this analysis. The remaining four OPC-N2s were compared to GRIMM and in January after running for 4 months (Fig. 9a), and while three of the OPC-N2s had a similar distribution to the GRIMM (OPC12, 13 and 14), OPC9 appeared to show evidence of instrument drift as the mode has shifted relative to the GRIMM. However, the increased frequency of higher mass concentrations not observed by the GRIMM but by all four OPC-N2s (Fig. 9a) suggests that ambient RH is also a factor, as the average RH in January (91 %) was higher than September (84 %). Therefore, we calculated the correction for RH as described in the previous section (Eq. 5), as changes in aerosol composition would affect the particle hygroscopicity. In addition, the κ was only fitted for the data with RH < 95 % since the hygroscopicity of aerosol is highly sensitive to any error in the RH measurement above this value. Application of the RH correction factor resulted in better agreement between each of the OPC-N2s, with similar corrected distributions observed (Fig. 9b). Furthermore, the corrected OPC-N2 concentrations also had better agreement with the GRIMM during January (Fig. 9b) compared to uncorrected concentrations (Fig. 9a), suggesting that changes in the particle water content were the cause. Thus, at least over a 4-month measurement period, there appears to be no evidence of instrument drift in the OPC-N2, once appropriate correction factors were applied.