Current Issue

Table 1 presents locations and instrumentation used for this study. Simultaneous observations were made using the DC1700 device and standard monitor of PM_2.5 at the Nagoya City Institute for Environmental Sciences (NCIES) located in Nagoya city, central Japan, during March 2017-March 2018. Fig. 1 portrays a location map of the observation sites in Nagoya, a major city with about 2.3 million residents. Yamagami et al. (2019) reported details of NCIES and PM_2.5 measurements at the site. The air mass at the site is often affected by long-range transport from continental eastern Asia in addition to domestic emissions. Particle diameters detected using the DC1700 show separation at >0.5 μm and >2.5 μm. The DC1700 was put in a weather shield box fitted with air ducts passing outside air easily. The weather shield box was installed outdoors under eaves to prevent sunlight and rain exposure. Hourly PM_2.5 mass concentrations were monitored using a β attenuation mass monitor (PM712; Kimoto Electric Co. Ltd.) installed in an outdoor case on the flat roof of the NCIES building. The PM712 is a PM_2.5 automatic measuring instrument that uses the standard method accepted in Japan. The temperature and relative humidity of the sample air and ambient air were measured respectively downstream of a PTFE tape filter and outside of the outdoor case. The PM712 also measured the mass concentration of optical black carbon (OBC, M_OBC) based on attenuation of near-IR scattering on a sampling spot of PM_2.5. Daily samples of PM_2.5 were collected using a pair of FRM-2000 samplers (Rupprecht and Patashnick Co. Inc., Albany, NY, USA) with PTFE filters (TK15-G3M; Pall Corp., Port Washington, NY, USA) for ion components and quartz fiber filters (2500QATUP; Pall Corp.) for carbon at NCIES. These samples were collected from 10:00 a.m. through 9:30 a.m. the next day. Details of FRM samplers at NCIES were reported by Ueda et al. (2016) and by Yamagami et al. (2019). Although original data obtained using the DC1700 were recorded every minute, their hourly averages were used for comparison with PM_2.5 mass concentrations.

Fig. 1. 
Map showing locations of observation sites (Nagoya City Institute for Environmental Sciences (NCIES), Nagoya University (NU), National air monitoring station in Nagoya at Kanokoden (KNK), and the local meteorological observatory (Met. Obs.)) in Nagoya, Japan.

The number-size distribution of aerosol particles, which was measured continually at Nagoya University (NU), was also used to evaluate data at NCIES. The distance between NCIES and NU is about 8 km (Fig. 1). The number-size distributions were measured using an optical particle counter (KC01D; Rion Co. Ltd.) with a laser diode. The KC01D measured the number concentrations for particles larger than 0.3, 0.5, 1.0, 2.0, and 5.0 μm diameter. The irradiated volume v of the laser was 5.0×10^-4 cm³. Sample air was introduced into the KC01D located in the observation room of the seventh floor after drying in a diffusion dryer to less than 20% relative humidity (RH). The length of tubing from the inlet to KC01D was about 3 m. The sample air flow rate was 0.5 L min^-1. A diffusion dryer used to keep the sample RH below 20% consisted of a 20-cm-long, 7-cm-diameter acrylic pipe and two desiccants mainly consisting of magnesium chloride (Nisso Dry-M; Nisso Fine Co., Ltd.). The KC01D was replaced by another instrument of the same type for maintenance during 12 May 2017 and 8 December 2017 because of air flow clogging. The use of KC01D data was limited to evaluation of the relative change of particle size distribution. Data of the optical particle counter needed to be corrected for the loss for coincidence counting error, depending on v and the particle number concentration. For comparison to data at the NCIES, measurements using the DC1700 in a weather shield box were also conducted at NU. In addition, PM_2.5 mass concentrations were monitored at the national air monitoring station in Nagoya at Kanokoden (KNK), which is near NU (within 2 km). Those data were used to evaluate regional differences of PM_2.5 mass concentrations. From March 2017, the PM_2.5 mass concentrations were measured hourly (ACSA-14; Kimoto Electric Co. Ltd.) using the same method as that used for PM712.

Hourly precipitation and temperature data were obtained from a local meteorological observatory at Nagoya. The distance from NCIES to KNK and the meteorological observatory is about 9 km (Fig. 1). The meteorological observatory is also within 2 km of NU.

Mass concentrations of PM_2.5 used for routine air monitoring such as the Federal Reference Method (FRM), which is widely regarded as reliable, were measured under the condition of 35±5% RH around the sample. PM712 provides the mass concentration of PM_2.5 equivalent to the value of FRM after correcting the water contents of ambient particles (Kimoto Electric Co. Ltd., 2007).

Therein, M_{total, PM2.5}, M_{dry, PM2.5}, and M_{w, PM2.5} respectively represent the mass concentrations of PM_2.5 at the relative humidity of the sample air (RH_sample), the mass concentration of dried PM_2.5, and the mass concentration of water contained in PM_2.5 at RH_sample. In addition, a and b (respectively 0.010 and 6.000) are coefficients ascertained from correlations of daily data between the FRM and the PM712 obtained in Japan. A correction curve based on equation (1) for RH<90% is similar to a theoretical equation, assuming hygroscopicity of ammonium sulfate and ammonium nitrate (Iwamoto et al., 2018; Snider et al., 2016).

For equivalence checking of PM_2.5 mass concentration in this study, the PM_2.5 mass concentration found using PM712 at NCIES was compared with that estimated from daily sample of PM_2.5 by FRM. Fig. 2 presents scatter plots of daily PM_2.5 mass concentrations obtained using FRM and PM712 during the observation period. Daily PM_2.5 mass concentrations of PM712 were averaged for 24 hr of data obtained during 10:00 a.m. to 10:00 a.m of the next day, using hourly data after correction by equation (1). Although plots for PM_2.5 mass concentration before humidity correction (M_{total, PM2.5}) show dispersion, the relation between PM_2.5 mass concentration by FRM and M_{dry, PM2.5} showed good correlation of R²=0.95 and slope=1.01. The greatest outlier plot (arrowed plot in Fig. 2) was that of data from a Kosa (dust) event of 7 May 2017. However, such cases were rarely observed. Most of the PM712 data showed a nearly 1:1 relation with those of FRM, suggesting that the daily average of mass concentration of PM_2.5 by PM712 after humidity correction using equation (1) was usually reliable to that by FRM. For this study, the hourly PM_2.5 mass concentration by PM712 was referenced to compare data of DC1700. The manufacturer states that averaged and standard deviations of hourly blanks of PM712 are within ±3.5 μg m^-3 for more than 15 data. A recent report has described that PM_2.5 mass concentration by some instruments using β attenuation can have different bias between night and daytime (Hasegewa et al., 2018). However, their test for 12-hr samples using FRM showed good correspondence with the PM_2.5 mass concentration of PM712 in many cases, except at some times during daytime in winter.

Fig. 2. 
Scatter plots of mass concentrations of PM_2.5 by PM712 and FRM at NCIES for the year of Mar. 2017 through Mar. 2018.

Based on the difference between particle number concentrations of >0.5 μm and >2.5 μm diameter obtained using DC1700, the number concentrations (N_0.5-2.5) for 0.5 μm to 2.5 μm particles are calculable. The mass concentration (M_0.5-2.5) for sizes of 0.5 μm and 2.5 μm was estimated from N_0.5-2.5, given as

where ρ and D respectively stand for the particle density and particle diameter. The D value was calculated as the geometric average between 0.5 μm and 2.5 μm. The ρ value was assumed for density (1.8 g/cm³) of ammonium sulfate as the most abundant compound in PM_2.5 at Nagoya (Yamagami et al., 2019; Ueda et al., 2016; Ikemori et al., 2015). According to their reports, sulfate, organic carbon, ammonium, elemental carbon, and nitrate were the major components, comprising more than 65% of the PM_2.5 mass concentration for all seasons. Chloride, sodium, potassium, magnesium, and calcium were less than 5% in PM_2.5. Among the other densities of materials considered from major components, the respective densities of sulfuric acid and ammonium nitrate are 1.84 and 1.73 g/cm³, which are similar values to those found for ammonium sulfate. The density of carbonaceous particles (approx. 1.5 g/cm³; Slowik et al., 2004) is slightly lower than that of ammonium sulfate. Therefore, the value of M_{dry, 0.5-2.5} using the density of ammonium sulfate might be overestimated by as much as 17% under the condition that all particles are carbonaceous particles.

The value of M_0.5-2.5 from DC1700 must be corrected for the hygroscopic increase of particle mass because measurements are conducted under ambient conditions without humidity control. For this study, the dried mass concentration (M_{dry, 0.5-2.5}) obtained using the DC1700 was estimated, given as

where RH_ambient represents the relative humidity of ambient air. This correction uses the same method for PM712 as that shown in equation (1). Instead of RH_sample of equation (1), RH_ambient was applied to the correction for DC1700 because the length to the optical sensor in DC 1700 is short.

Fig. 3 portrays temporal variations in March 2017 as an example of well-synchronized variation of M_{dry, PM2.5} and M_{dry, 0.5-2.5}: (a) the number concentration of aerosol particles measured using DC1700 at NCIES and NU; (b) M_{dry, 0.5-2.5}, M_{dry, PM2.5}, and M_PM2.5 by FRM and M_{dry, 0.5-2.5}/M_{dry, PM2.5} at NCIES; (c) M_OBC at NCIES; (d) the size-segregated volume concentration (0.3-5.0 μm) by KC01D at NU; (e) the relative humidity of ambient air measured at outside of PM712 (RH_ambient) and downstream of the sample filter of PM712 (RH_sample); and (f) the temperature and precipitation amounts recorded at the local meteorological observatory. The volume concentrations were calculated from the number concentrations for the respective size ranges of the KC01D, given as shown below.

Fig. 3. 
Temporal variations are shown: (a) number concentrations found using DC1700 at NCIES and NU; (b) mass concentrations of 0.5-2.5 μm diameter particles found using DC1700 (M_{dry, 0.5-2.5}), mass concentration of PM_2.5 by PM712 (M_{dry, PM2.5}), and by FRM (M_PM2.5 FRM), and the ratio of M_{dry, 0.5-2.5} and M_{dry, PM2.5} at NCIES; (c) mass concentration of OBC (M_OBC) found using PM712; (d) volume concentrations of aerosols with 0.3-5 μm diameter by OPC at NU; (e) relative humidity of ambient air measured at outside of PM712 (RH_ambient) and downstream of the sample filter of PM712 (RH_sample) at NCIES; and (f) temperature and hourly precipitation at a local meteorological observatory for March 2017.

Therein, D_i and N_i respectively stand for the minimum particle diameter and number concentration of channel i counting for particles larger than D_i of KC01D.

As portrayed in Fig. 3a, the temporal variations of the number concentrations obtained using DC1700 at NU and NCIES were almost equal. Such synchronous variation was observed throughout the observation period. Fig. 4 presents a scatter plot of the number concentrations (>0.5 μm and >2.5 μm) obtained by DC1700 at NCIES and NU and scatter plots of PM_2.5 mass concentrations at NCIES and KNK for the observation periods. Their respective coefficients of determination (R²) and slope values of linear regression were 0.78-0.87 and 0.91-1.13. The good degree of correlation suggests that the air masses of these locations can be regarded as almost equivalent. As portrayed in Fig. 3b, the value of M_{dry, PM2.5} was almost equal to that of M_PM2.5 by FRM. The temporal variation of M_{dry, 0.5-2.5} correlated well with that of M_{dry, PM2.5} (Fig. 3b) and the volume concentration (0.3-5.0 μm) obtained by KC01D (Fig. 3d). However, the values of M_{dry, 0.5-2.5} were always less than that of M_{dry, PM2.5}: about 50% of M_{dry, PM2.5} in March 2017. Also, RH_ambient was usually less than 80%, as depicted in Fig. 3e. The period of RH_ambient higher than 80% in the period was recorded during precipitation events (Fig. 3f). During such events, M_{dry, PM2.5} tended to be lower. Higher M_{dry, PM2.5} (>20 μg m^-3) events were observed mostly under conditions of less than 70% RH_ambient. Although ambient and sample air humidity changed diurnally, M_{dry, 0.5-2.5}/M_{dry, PM2.5} (Fig. 3b) is less dependent on humidity. The values of M_OBC (Fig. 3c) were usually less than 10% of M_{dry, PM2.5}.

Fig. 4. 
Scatter plots of number concentrations of >0.5 μm and 2.5 μm measured using DC1700 at NCIES and NU, and of the mass concentrations of PM_2.5 at NCIES and KNK.

Scatter plots of M_{dry, PM2.5} and M_{dry, 0.5-2.5} at NCIES for two months (a, March-April 2017; g, March 2018) and for all data in one plot (h) are presented in Fig. 5. The M_{dry, 0.5-2.5} values are shown below the 1 : 1 line to M_{dry, PM2.5} values. The coefficient of determination (R²) and slope values of linear regression for all data in this study were, respectively, 0.68 and 0.40. Although the observation values for the whole period (Fig. 5h) showed large dispersion, the relations within two months (as shown in Figs. 5a, b, e and g) were found to have stronger correlation. For the scatter plot showing data of January and February (Fig. 5f ), greater variation was apparent at higher concentrations during particular events in February (red dots in Figs. 5f and h), as discussed in section 3.4. The respective maximum and minimum values of R² were 0.90 found for March-April 2017 (Fig. 5a) and 0.62 found for July-August (Fig. 5c) 2017. The respective maximum and minimum values of slopes were also 0.55 found for March-April 2017 and 0.24 found for July-August 2017. The slope values of linear regression differed by season. Fig. 6 portrays monthly box plots for ratios of M_{dry, 0.5-2.5}/M_{dry, PM2.5} at NCIES. For March-April 2017 and December 2017-March 2018, the median values of M_{dry, 0.5-2.5}/M_{dry, PM2.5} were 0.4-0.5. However, the median values were less than 0.4 for May-November 2017. They were especially low (approx. 0.25) in summer ( July-August). The ratio of M_{dry, 0.5-2.5}/M_{dry, PM2.5} decreased once in summer but recovered in winter and the following spring.

Fig. 5. 
Scatter plots of mass concentrations of 0.5-2.5 μm in diameter by DC1700 versus mass concentration of PM_2.5 by PM712 for two months (a-g) and for a year from Mar. 2017 through Mar. 2018. Periods of events B (18:00 19-8:00 20 February 2018), and C (21:00 254-9:00 26 February 2018) and D (16:00-20:00 28 February 2018) are denoted by red dots.

Fig. 6. 
Monthly box plots for ratio of mass concentration of 0.5-2.5 μm by DC1700 (M_{dry, 0.5-2.5}) to mass concentration of PM_2.5 by PM712. The lower boundary of the box shows the 25th percentile, the line within the box represents the median, and the upper boundary of the box shows the 75th percentile. Whiskers above and below the box respectively show the 90th and 10th percentiles. Numbers shown above boxes are the numbers of observed data.

Han et al. (2017) also measured ambient aerosols using DC1700 in Houston, Texas. Correlation between the number concentration of 0.5-2.5 μm and PM_2.5 mass concentration found using standard instruments for 12 days was a similar level (R²=0.778). Seasonal changes in detection for another low-cost sensor to measure atmospheric PM_2.5 also reported from results of several studies. Sayahi et al. (2019) measured mass concentrations of particulate matter using PMS 1003 and 5003 at Salt Lake City, Utah, in the USA. Correlation to 24-hr averaged PM_2.5 mass concentration by FRM was good (R²>0.88) in winter, but poorer (R² of 0.18-0.32) in spring. Nakayama et al. (2017) evaluated a new palm-sized optical PM_2.5 sensor (PM_2.5 sensor; Panasonic Corp.) at four urban and suburban sites in Japan. Their slope to standard PM_2.5 mass concentration tended to be higher in winter than in summer, which was inferred as attributable to small particles by photochemical formation. They also reported that the sensor tended to overestimate the PM_2.5 mass concentration by hygroscopic growth of particles.

The M_{dry, 0.5-2.5} found in this study was corrected for hygroscopic growth. Therefore, differences between M_{dry, 0.5-2.5} and M_{dry, PM2.5} reflect the contribution of other factors. The M_{dry, 0.5-2.5} values were always less than those of M_{dry, PM2.5}. Sousan et al. (2016) reported tests of the detection efficiency of DC1700 using monodispersed fine salt (0.1, 0.2 and 0.3 μm) and large oleic acid (1.3, 2, 3 and 5 μm) particles. The actual detection efficiency for the DC1700 in the size range of 0.5-2.5 μm was 52% for 1.3 μm particles. The difference of actual detection efficiency from manufacturer specifications can affect the underestimation of M_{dry, 0.5-2.5} to M_{dry, PM2.5}. Moreover, the difference of M_{dry, 0.5-2.5} to M_{dry, PM2.5} in this study tended to vary seasonally. This finding suggests that the difference can change along with the seasonal change of atmospheric aerosols. As factors affecting the seasonal change of M_{dry, 0.5-2.5}/M_{dry, PM2.5}, we can infer the existence of light-absorbing particles and smaller fine particles below the detection limit, and infer a difference of size distribution in the estimation equations to that of actual atmospheric aerosols. Effects on estimation of the mass concentration of PM_2.5 using DC1700 and the seasonal change by each factor are explained below.

3. 3. 1 Existence of Light-absorbing Particles

Light-absorbing particles are detected inappropriately as being of smaller size because of the lower intensity of light scattering for the same physical size of non-light-absorbing particles. According to Sousan et al. (2016), who described laboratory tests for DC 1700, the detected diesel exhaust particles were one or more orders fewer than those detected using non-light absorbing salts having the same mass concentration. In the present study, the average fraction of OBC mass concentrations to PM_2.5 was 7.8±9.5% (average±standard deviation for this observation period). The light absorption of aerosols might partly affect underestimation at such a level. Fig. 7a, b and c present monthly box plots for M_{dry, PM2.5}, M_OBC and ratios of M_OBC/M_{dry, PM2.5} at NCIES. Median values of M_OBC/M_{dry, PM2.5} were less than 0.2 for all data (Fig. 7c). The ratios were slightly high in October with slight variation. In that year, M_{dry, PM2.5} of October was lowest (Fig. 7a) because of the arrival of multiple typhoons. M_OBC of October was of a comparable level to the other months (Fig. 7b). Therefore, the mass fraction of black carbon was apparently high. However, comparison to M_{dry, 0.5-2.5}/M_{dry, PM2.5} in Fig. 6 shows that the seasonal variation of M_OBC/M_{dry, PM2.5} was less and was poorly matched.

Fig. 7. 
Monthly box plots for mass concentrations of PM_2.5 (a) and OBC (b) and the ratio of mass concentration of OBC to PM_2.5 (c) by PM712 at NCIES and the ratio of volume concentrations of 0.5-2 μm to 0.3-2 μm (d) measured by OPC at NU. The lower boundary of the box shows the 25th percentile, the line within the box represents the median, and the upper boundary of the box shows the 75th percentile. Whiskers above and below the box respectively show the 90th and 10th percentiles. The numbers shown above boxes are the numbers of observed data.

3. 3. 2 Existence of Smaller Fine Particles below the Detection Limit

The minimum size detectable by DC1700 is 0.5 μm as the optical diameter. Earlier studies have also demonstrated the significant contribution of particles with diameters smaller than the detectable size limit of the DC 1700 sensor (Zikkova et al., 2017; Manikoda et al., 2016; Sousan et al., 2016). Fig. 7d presents monthly box plots for the volume ratios of 0.5-2.0 μm to 0.3-2.0 μm (V_{0.5-2.0 μm}/V_{0.3-2.0 μm}) by KC01D at NU. The median values were maximum (0.5) in December and minimum (0.3) in July. The seasonal variation of high values in winter and low values in summer were close to those of M_{dry, 0.5-2.5}/M_{dry, PM2.5} in Fig. 6. The volume fraction of particles smaller than 0.5 μm in PM_2.5 might affect the seasonal difference between M_{dry, 0.5-2.5} and M_{dry, PM2.5}. Fig. 8 portrays the monthly averaged number-size distribution measured using the KC01D device at NU and their volume-size distribution for December and July 2017. For the volume-size distribution, the area of each bar for a size range corresponds to the volume concentration for the size range. Aerosol measurements were taken using the DC1700 device under ambient relative humidity. Therefore, some hygroscopic particles having less than 0.5 μm dry diameter can be counted if the particles become larger than 0.5 μm under ambient conditions. The dry size (D_{dry, 0.5}) of the minimum detectable size (0.5 μm) by the DC1700 in ambient conditions was estimated using the hygroscopicity of ammonium sulfate (Snider et al., 2016). Actually, D_{dry, min} was 0.39±0.06 μm, on average, during the observation periods. Similarly, the dry size (D_{dry, 2.5}) of the upper detection size (2.5 μm) obtained using the DC1700 device was estimated as 1.9±0.3 μm. The D_{dry, 0.5} and D_{dry, 2.5} are shown as dashed and solid red lines in Fig. 8b.

Fig. 8. 
Monthly averaged number-size distribution (a) and volume-size distributions (b) of aerosol particles by KC01D for July and December 2017. Red dash-dotted and solid lines of (b) respectively represent the yearly averaged dried size of 0.5 μm and 2.5 μm particles estimated by assuming ammonium sulfate.

PM_2.5 particles by PM712 were classified as less than 2.5 μm under ambient conditions by a virtual impactor. Therefore, the size range of dried PM_2.5 was also estimated as less than D_{dry, 2.5} (ca. 2 μm). Although particles less than D_{dry, 0.5} can affect underestimation of M_{dry, 0.5-2.5} to M_{dry, PM2.5}, the estimated volume concentration for sizes of 0.3-2 μm in Fig. 8b was a maximum at 0.3-0.5 μm. The D_{dry, 0.5} was in 0.3-0.5 μm. According to parameters for atmospheric volume-size distributions derived from averages of measurements reported by Whitby (1978), the respective mode peaks of the accumulation and coarse mode particles were 0.32 μm and 5.7 μm for urban aerosols. The volume fraction of particles smaller than 0.5 μm to particles smaller than 2.5 μm can be estimated as 62% based on Whitby’s parameters. Considering size changes attributable to hygroscopic growth, the volume fraction of particles smaller than 0.39 μm (D_{dry, min}) among particles smaller than 2 μm was 54%. The result suggests that the contribution of particles smaller than the detection limit especially accounts for the underestimation of M_{dry, 0.5-2.5} to M_{dry, PM2.5}.

Although the volume concentration of particles smaller than 0.3 μm was unknown for this study, the volume ratio of 0.3-0.5 μm to 0.5-2 μm of July 2017 was minimum (Fig. 7d). The volume fraction of particles smaller than the detection limit (D_{dry, 0.5}) was expected to be high in July 2017; consequently, the values of M_{dry, 0.5-2.5}/M_{dry, PM2.5} were small. Given a volume concentration of less than 0.3 μm as V_{<0.3 μm}, the ratio of M_{dry, 0.5-2.5} to M_{dry, PM2.5} was almost equivalent to the ratio of V_{0.5-2.0 μm} to (V_{<0.3 μm}+V_{0.3-2.0 μm}). The ratio of M_{dry, 0.5-2.5} to M_{dry, PM2.5} is expected to be somewhat larger because the DC1700 counted hygroscopic particles of less than 0.5 μm. In consideration of general and observed volume-size distribution, we assumed simple values of V_{<0.3 μm}: the values of V_{<0.3 μm} are V_{0.3-2.0 μm} and 0.5·V_{0.3-2.0 μm}, respectively, when the accumulation mode diameters are small (as in July) and large (as in December). Then, the ratios of V_{0.5-2.0 μm} to (V_{<0.3 μm}+V_{0.3-2.0 μm}) were estimated respectively as 0.15 and 0.33. Consequently, the size change can become a factor of difference of 0.18 by this assumption, which was comparable to a difference of M_{dry, 0.5-2.5}/M_{dry, PM2.5} between July and December (ca. 0.2) in Fig. 6. As implied by this estimate and by Whitby’s size distribution, the seasonal variation of V_{0.5-2.0 μm}/V_{0.3-2.0 μm} suggests that a particle volume smaller than the detection limit of DC1700 varies with the season. The contributions of smaller particles might be greater in summer because the formation and growth of secondary particles is more active in summer (Okada, 1985). These results suggest that simple conversion of number data from Dylos to PM_2.5, which is independent of the size distribution, would engender large bias, with variation derived from seasonal changes in the size distribution.

Fig. 9 presents temporal variations of M_{dry, 0.5-2.5}, M_{dry, PM2.5} at NCIES, M_{dry, PM2.5} at KNK, and other parameters by KC01D, along with meteorological data for February 2018. As described earlier, M_{dry, 0.5-2.5}/M_{dry, PM2.5} was usually about 0.4. However, M_{dry, 0.5-2.5}/M_{dry, PM2.5} were occasionally higher than 0.7, with high mass concentration events such as B (18:00 19-8:00 20 February 2018), C (21:00 24-9:00 26 February 2018), and D (16:00-20:00 28 February 2018). As depicted in Fig. 5f and h with data colored in red, these high ratios of M_{dry, 0.5-2.5}/M_{dry, PM2.5} were rarely observed during the remainder of the year. For the other high mass concentration event A (8 February 2018) in Fig. 9a, the ratio of M_{dry, 0.5-2.5}/M_{dry, PM2.5} was at a normal level (ca. 0.5). Values of M_PM2.5 by FRM and M_{dry, PM2.5} at KNK during events A-D also corresponded well to M_{dry, PM2.5} at NCIES (Fig. 9a). These increases of M_{dry, PM2.5} in events A-D occurred during the dry condition (RH_ambient<70%) without precipitation (Fig. 9a, e and f).

Fig. 9. 
Temporal variations are shown: (a) mass concentration of 0.5-2.5 μm in diameter by DC1700 (M_{dry, 0.5-2.5}) and mass concentrations of PM_2.5 by PM712 and FRM (M_{dry, PM2.5} at NCIES and M_PM2.5 FRM) at NCIES and mass concentrations of PM_2.5 by ASCA-14 at Kanokoden (M_{dry, PM2.5} at KNK); (b) mass concentration of OBC; (c) volume concentrations of aerosols with 0.3-5 μm diameter by OPC at NU; (d) volume fractions of 0.3-0.5 μm, 0.5-1 μm and 1-2 μm particles to 0.3-2 μm particles; (e) relative humidity of ambient air (RH_ambient) and downstream of sample filter of PM712 (RH_sample) at NCIES; and (f) temperature and hourly precipitation at local meteorological observatory for February 2018. Periods of high-concentration events A, B, C, and D are highlighted by green and yellow bars.

As estimated in section 3.3, particles smaller than the detection limit can account for the large fraction of underestimation of M_{dry, 0.5-2.5} to M_{dry, PM2.5} among some factors. Therefore, size distribution differences might affect the ratio of M_{dry, 0.5-2.5} to M_{dry, PM2.5}. In events C-D, the fraction of 0.5-1.0 μm particles (Fig. 9d) increased with higher M_{dry, PM2.5}. In event A (8 February 2018), an increase of smaller (0.3-0.5 μm) particles contributed to the increase of M_{dry, PM2.5}. For comparison, volume-size distributions of events A and C are presented in Fig. 10. As the figure shows, the volume concentration of 0.5-1.0 μm particles of event C was clearly greater than that of event A. Therefore, the volume fraction of particles larger than D_{dry, 0.5} was apparently higher for event C. The precipitous change observed for the difference between M_{dry, 0.5-2.5} and M_{dry, PM2.5} was regarded as mainly reflecting rapid changes in the volume-size distribution.

Fig. 10. 
Volume-size distributions of aerosol particles by KC01D for events A and C. Red dash-dotted and solid lines respectively show the dried sizes of 0.5 μm and 2.5 μm particles estimated by assuming ammonium sulfate.

As shown in Fig. 5h, drastic events of M_{dry, 0.5-2.5}/M_{dry, PM2.5} change, such as events B, C, and D, rarely occurred throughout the year. Nevertheless, these events demonstrated that a particle fraction that is undetectable by DC1700 can change drastically in a single day, not only on the time-scale of seasons. Therefore, to use DC1700 data as a proxy of atmospheric PM_2.5, attention must be devoted to temporal changes of the number-size distribution.