You are on page 1of 11

Science of the Total Environment 748 (2020) 141396

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

Low-cost sensors for measuring airborne particulate matter: Field


evaluation and calibration at a South-Eastern European site
G. Kosmopoulos a, V. Salamalikis a, S.N. Pandis b,c,d, P. Yannopoulos e, A.A. Bloutsos e, A. Kazantzidis a,⁎
a
Laboratory of Atmospheric Physics, Department of Physics, University of Patras, Patras GR 26500, Greece
b
Department of Chemical Engineering, University of Patras, Patras GR 26500, Greece
c
Institute of Chemical Engineering Sciences, Foundation for Research and Technology Hellas (FORTH/ICE-HT), Patras, Greece
d
Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA
e
Department of Civil Engineering, University of Patras, Patras GR 26500, Greece

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Low-cost PM sensors were calibrated in


ambient conditions.
• The original sensor values were reason-
ably well correlated.
• The sensor response was different dur-
ing periods with high dust.
• The relative humidity had a negligible
effect on fine PMs.

a r t i c l e i n f o a b s t r a c t

Article history: Low-cost sensors are useful tools for the collection of air quality data, augmenting the existing regulatory mon-
Received 12 May 2020 itoring networks and providing an unprecedented opportunity to increase their spatial coverage. This study pre-
Received in revised form 27 July 2020 sents a calibration process of a low-cost PM sensor (PurpleAir PA-II, PAir) in ambient conditions in the city of
Accepted 29 July 2020
Patras, Greece, during 18 months of 2017–2018.
Available online 31 July 2020
The hourly PM1 and PM2.5 measurements using the original sensor values were reasonably well correlated (R2 =
Editor: Pavlos Kassomenos 0.82 for PM1 and R2 = 0.56 for PM2.5) with the reference instrument, but with a high mean bias and root mean
square error. There was a small improvement of around 10% for the daily averages. For PM1–2.5 (particles with
Keywords: diameters between 1 and 2.5 μm), PM2.5–10 (diameters between 2.5 and 10 μm) and PM10, the performance of
Airborne particulate matter the low-cost sensors was poor in this area with R2 < 0.37 in all cases.
Fine particles The response of the PAir sensor for PM1 and PM2.5 changed significantly compared to the reference instrument
Low-cost sensors during periods with high dust (or other coarse particle) concentrations. These periods were excluded and a sim-
Calibration ple linear calibration was then developed for the rest of the fine PM measurements. A method for the identifica-
South Europe
tion of these high dust periods based on regional model predictions is proposed. This calibration reduces the
relative mean error for hourly PM1 to 19% (1.1 μg m−3) and for PM2.5 to 18% (1.1 μg m−3). The corresponding
root mean square errors are 25% (1.4 μg m−3) for hourly PM1 and 25% (1.6 μg m−3) for PM2.5. The biases of
the corrected values are, as expected, practically zero. Surprisingly, the relative humidity had a negligible effect
on fine PM measurements of the PAir in this location and for the conditions of the study.
© 2020 Elsevier B.V. All rights reserved.

⁎ Corresponding author.
E-mail address: akaza@upatras.gr (A. Kazantzidis).

https://doi.org/10.1016/j.scitotenv.2020.141396
0048-9697/© 2020 Elsevier B.V. All rights reserved.
2 G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396

1. Introduction among two collocated sensors. Linear regression of hourly averaged


measurements of PMS sensor with GRIMM indicated slopes between
Air pollution affects both human health and climate. Airborne partic- 0.51 and 0.83 for all PM fractions and intercepts between 1.3 μg m−3
ulate matter (PM), is a significant pollutant in both urban and rural (for PM1) and 8 μg m−3 (for PM10). Sayahi et al. (2019) evaluated the
areas widely investigated during the last decades (Aguilera et al., performance of PMS sensors for PM2.5 and PM10 measurements during
2015; Cesari et al., 2018; Crippa et al., 2013; Kassomenos et al., 2014; different seasons. The comparison with reference instruments revealed
Putaud et al., 2010). Airborne particles are generated by various anthro- a good correlation, but seasonal differences for PM2.5. The performance
pogenic and natural sources (biomass burning, fuel combustion, trans- of the sensors for PM10 was much poorer.
portation, sea salt, volcanoes, dust events, etc.) affecting visibility and Despite the above progress, it is still not clear whether the perfor-
the Earths' energy balance by scattering and absorbing light or indi- mance of the PAir sensors is the same in all environments and condi-
rectly affecting clouds (IPCC, 2013). PM is classified as carcinogenic tions. This study focuses on the evaluation and the calibration of PAir
(IARC, 2013) while it has been also associated with aggravated asthma, sensors installed in the city of Patras, Greece, during an 18-month pe-
cardiovascular diseases, and increasing mortality in various epidemio- riod. The contribution of meteorology and especially humidity and tem-
logical studies (Analitis et al., 2006; Beelen et al., 2014; Penttinen perature conditions on the sensor's performance is also assessed. The
et al., 2001; Pope and Dockery, 2006, Steinle et al., 2015, Hackmann results of this calibration apply strictly to the conditions (aerosol size
and Sjöberg, 2017). distribution and composition, aerosol levels, etc.) and meteorology of
Particles with an aerodynamic diameter smaller than 2.5 μm (PM2.5) the specific area, but still provides important insights about the proper
are considered more dangerous due to their deeper penetration into the use of the same or similar sensors in other areas.
respiratory system. Coarse particles, particles with an aerodynamic di-
ameter between 2.5 and 10 μm, are also associated with respiratory 2. Methods
symptoms, cardiovascular diseases, and premature death (Dominici
et al., 2006; Hoek et al., 2013). The World Health Organization and 2.1. Study area and data description
most countries have established health-based guidelines and limits
aiming to reduce PM concentrations and their negative impacts. Patras is a typical medium-sized city (around 200,000 inhabitants) in
Traditionally, high quality, certified instruments are used to provide the Eastern Mediterranean. The PM originates from a wide variety of an-
measurements of PM concentrations for both regulatory and research thropogenic sources (traffic, industry, biomass burning, long-range trans-
applications. These monitors are expensive, relatively large, heavy, port) but also includes both local and Sahara dust (Gerasopoulos et al.,
and require skilled personnel to ensure the quality of their measure- 2011). Noteworthy are the city Port and a small industrial zone (~15 km
ments. Consequently, stations of air quality networks are sparsely dis- to the west of the city center). Higher PM concentrations occur during
tributed and are unable to assess the PM spatial variability across an the late fall, winter, and early spring due to domestic heating emissions
urban area or away from major cities (Bell et al., 2011; Kelly et al., and agricultural waste burning (Manousakas et al., 2018; Pikridas et al.,
2017). During the last years, there has been rapid development of 2013). The annual average PM10 concentration during the measurement
low-cost air pollutant sensors for both gases and PM (Kumar et al., period (2017–2018) was 32 μg m−3, within EU limits, as defined by the
2015; Lewis and Edwards, 2016). These low-cost devices are promising Directive 2008/50/EC (EU, 2008), but there were several days with
tools for the collection of air quality data with high spatial and temporal PM10 values above 50 μg m−3. Chemical characterization studies showed
coverage. Low-cost sensors are small and compact and have low energy that the fine particles in the area consist mainly of sulfates and organics
demands. Most of them are equipped with wireless communication (Florou et al., 2017; Pikridas et al., 2013).
modules providing real-time data transmission. Such low-cost net- The site, used for the evaluation of PAir sensors, is located at the Uni-
works are implemented in many countries to monitor air pollution versity of Patras campus in a suburban area, approximately 8 km north-
(Heimann et al., 2015; Jiao et al., 2016). east of the city center. The regular University activities are the primary
Despite their advantages, these sensors may be less reliable, a lot less local source on this site. Meteorological measurements were provided
accurate, or less precise when compared against the conventional in- by the weather station of the Laboratory of Atmospheric Physics. During
struments used for regulatory and research measurements (Williams the 18-month measurement period (May 2017–November 2018), the av-
et al., 2014; Shapiro et al., 2014). The main challenge for the successful erage hourly values of air temperature and relative humidity were 19 ±
use of low-cost sensors is their calibration for the environments and 6 °C (range 3–39 °C) and 65 ± 16% (range 9–97%), respectively. The ex-
ambient conditions under which they operate (Manikonda et al., perimental setup was a PAir sensor including two PMS sensors and a
2016; Papapostolou et al., 2017; Rai et al., 2017). For example, previous GRIMM EDM 180 dust monitor. The sensors operated reliably during
evaluations of low-cost PM sensors have revealed that the concurrent these 18 months and provided 97% of the possible hourly measurements.
meteorological conditions (temperature and relative humidity) may af-
fect their performance (Wang et al., 2015; Sousan et al., 2017; Jayaratne 2.2. Low-cost sensors
et al., 2018)
Plantower sensors for PM, PMS 1003 and 3003 (used by PAir) have Each PAir sensor includes two PMS 5003 sensors, named channels A
been calibrated in both laboratory-controlled (wind tunnel) and ambi- and B. The two PMS sensors are located in a water-proof box and are
ent conditions against reference instruments. For an urban residential connected to a Wi-Fi module in a custom-made package provided by
area in Salt Lake City, Kelly et al. (2017) reported that PM2.5 data from PurpleAir. Each PMS sensor measures PM concentrations continuously
PAir sensors correlated well with the higher-class instruments, with with 1-min temporal resolution. The operation of PAir sensors is based
R2 values higher than 0.88 when compared to daily measurements on light scattering principles. Α built-in fan draws air and particles
and R2 = 0.83–0.92 for hourly measurements. The South Coast Air Qual- into the measurement chamber. The particles pass through a laser
ity Management District also assessed the performance of PAir sensors beam, and a detector detects scattered light; the output signal is used
(SCAQMD, 2017). Their field evaluation against two reference instru- to calculate PM1, PM2.5, and PM10 mass concentrations by a proprietary
ments, a GRIMM EDM 180 and a Met One 1020 Beta Attenuation Mon- algorithm. The PurpleAir software provides uncorrected values (named
itor (BAM), showed a high correlation (R2 > 0.85) with hourly and daily correction factor CF = 1 in the corresponding output file) and corrected
PM1 and PM2.5 values between the sensor and the reference instrument values (named CF = atm in the file). The differences between the two
values. The correlation between the PM10 values was substantially values for PM2.5 < 30 μg m−3, the range of most of our measurements,
lower (R2 = 0.60–0.73 for hourly and 0.68–0.75 for daily averages). In are almost zero. The uncorrected values (CF = 1) are used in the rest
both studies, the PMS sensor intra-variability was low with R2 > 0.99 of this work.
G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396 3

2.3. GRIMM Dust Monitor/reference instrumentation PM2.5 mass concentrations of suspended particulate matter in ambient
air by sampling the particulate matter in ambient air. In this study, the
A collocated GRIMM EDM 180 (GRIMM Environmental Dust Monitor measurements were averaged in hourly resolution.
180) particle analyzer, operated according to the standards of the man-
ufacturer, was used to evaluate the PAir sensor measurements. EDM 3. Results
180 uses light-scattering to determine PM1, PM2.5, and PM10. Particles
scatter light produced from a diode laser (660 nm), which is captured 3.1. Intra-variability and sensors precision
by a detector. The signal is assigned to 31 size channels (0.25–3 μm)
and is converted to mass concentration with 0.1 μg m−3 resolution. The hourly measurements of the two channels (A and B) of the PAir
The sample flow rate is 1.2 L min−1 and up to 3 × 106 particles L−1 were highly correlated (R2 = 0.99) for PM1, PM1–2.5 and PM10, with the
are detected with reproducibility higher than 97%. EDM 180 is also R2 decreasing slightly to 0.95 for hourly PM1–2.5 and to 0.88 for PM2.5–10
equipped with an integrated Nafion dryer to remove water. The (Fig. 1). The R2 of the measurements of the two sensors even at 10-min
GRIMM EDM 180 follows the European standard EN 12341 that de- intervals was quite high. These results are summarized in Table 1 and
scribes a standard gravimetric method for determining the PM10 or are consistent with previous studies examining the PAir intra-channel

Fig. 1. Scatterplots of 12,954 hourly (left panels a–c) PMS1, PMS1–2.5, PMS2.5, PMS2.5–10 and PMS10 mass concentrations (μg/m3) for channels A and B during the period 25/05/2017–30/11/
2018. The dashed black line corresponds to the “1:1” relationship.
4 G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396

Table 1 Table 3
Comparison of the two PMS sensors A and B of the PAir units for different temporal reso- Evaluation metrics of the PAir hourly and daily average measurements compared to the
lutions during the 18-month study. Nobs defines the number of observations. reference instrument. 11,681 hourly measurements and 512 daily measurements were
used.
Nobs R2 Slope Intercept (μg/m3) Precision (μg/m3)
MBE rMBE MAE rMAE RMSE rRMSE R2
Abs mean Abs range
(μg m−3) (μg m−3) (μg m−3)
PM1
PM1 Hourly 5.79 1.03 5.86 1.04 7.31 1.29 0.82
10-Min 74,690 0.99 0.92 −0.39 1.37 0–9.5
Daily 5.71 1.02 5.73 1.02 6.79 1.21 0.85
Hourly 12,954 0.99 0.92 −0.40 0–7.47
PM1–2.5 Hourly 1.85 1.59 2.31 1.99 3.21 2.77 0.1
Daily 547 0.99 0.92 −0.44 0–4.59
Daily 1.92 1.70 2.28 2.03 2.95 2.62 0.1
PM1–2.5 PM2.5 Hourly 7.83 1.15 8.20 1.21 10.36 1.53 0.56
10-Min 74,690 0.95 1.12 0.08 0.40 0–17.56 Daily 7.73 1.15 7.97 1.18 9.54 1.42 0.56
Hourly 12,954 0.95 1.06 0.21 0–3.21 PM2.5–10 Hourly −0.45 −0.34 1.64 0.97 2.92 1.73 0.02
Daily 547 0.96 1.06 0.21 0–1.98 Daily −0.39 −0.22 1.59 0.91 2.67 1.52 0.16
PM10 Hourly 7.42 0.86 8.51 0.99 11.23 1.31 0.36
PM2.5 Daily 7.33 0.86 8.15 0.95 10.09 1.18 0.37
10-Min 74,690 0.99 0.96 −0.33 0.96 0–9.76
Hourly 12,954 0.99 0.96 −0.32 0–7.77
Daily 547 0.99 1.06 −0.29 0–3.03
fine PM (PM1 and PM2.5) measurements were also characterized by
PM2.5–10
10-Min 74,690 0.87 1.14 0.02 0.19 0–8.57 high bias, exceeding 100%. The bias for the PM10 was 86%. The statistics
Hourly 12,954 0.88 1.10 0.06 0–3.66 for the corrected values reported by the PAir (not shown here) are quite
Daily 547 0.92 1.14 0.01 0–2.7 similar for the modest concentrations during the study. This indicates
PM10 rather poor performance of the PAir sensors if original sensor values
10-Min 74,690 0.99 0.96 −0.19 0.76 0–14.45 are used.
Hourly 12,954 0.99 0.96 −0.19 0–8.94 Despite the discrepancies in the reported absolute concentrations,
Daily 547 0.99 0.96 −0.13 0–5.5 the PAir measurements for PM1 are relatively well-correlated with the
reference instrument (Fig. 2). The R2 of the hourly PM1 values is 0.82.
However, it is clear that the corresponding measurements fall into two
variability (Kelly et al., 2017; SCAQMD, 2016; Zheng et al., 2018). Addi-
groups: most of the measurements in which the PAir1 is much higher
tionally, the discrepancies between the two channels were lower than
than the reference PM1 and a few measurements for which they are ap-
1.4 μg m−3 for all fractions for the whole study.
proximately equal. We will return to this important issue in the next
In order to ensure the high quality of the retrieved data, only the
section. However, the correlation between the measurements of the
measurements with a difference between the two channels of less
two instruments decreases dramatically for supermicrometer particles.
than 20% of their average or less than 2 μg m−3 were considered valid
The R2 for the hourly PM1–2.5 is only 0.1 and of the hourly PM2.5–10
and were included in the subsequent analysis. Τhe second constraint
0.02. Given that most of the PM2.5 and a lot of the PM10 is the sub-
was needed to avoid the exclusion of the low concentration values
micrometer range there is modest R2 of 0.56 for the PM2.5 and of 0.37
from the dataset. The limit of 2 μg m−3 was estimated as the upper
for the PM10. However this is driven mainly by the PM1. These results
fence of the measured concentration differences of the two channels.
suggest that the PAir may have significant difficulties in providing useful
Based on this algorithm more than 98% of the hourly PM1, PM2.5 and
information for the super-micrometer particles at least for the concen-
values collected during the 18-month study were considered as valid.
tration levels and conditions of this study.
3.2. Evaluation of the accuracy of PAir PM measurements
3.3. Effect of high coarse PM levels on PAir measurements
The PAir sensor reported an average PM1 concentration (denoted as
Careful analysis of the conditions under which the behavior of the
PAir1) that was approximately two times higher than the average PM1
PAir sensor changes compared to the majority of the cases (Fig. 2), sug-
of the reference instrument (Table 2) for the relatively low concentra-
gested that some of these points correspond to periods with significant
tions (5.6 μg m−3 of the site). The discrepancy increased to a factor of
concentrations of Sahara dust. Our hypothesis is that high levels of
2.7 for the PM1–2.5 which had even lower levels (1.1 μg m−3). As a result,
coarse particles (as determined by the GRIMM measurements) lead to
the average PAir2.5 (PM2.5 reported by the PAir) was 2.1 times more
different behavior of the PAir sensor for PM1, PM1–2.5, PM2.5, and PM10.
than the PM2.5 measured by the GRIMM. The low-cost sensor showed
These points appear to follow a line with a much smaller slope com-
the opposite behavior for the coarse PM (PM2.5–10), reporting an aver-
pared to the rest of the measurements and they should be corrected dif-
age value that was lower by 22% compared to the reference instrument.
ferently during the calibration process.
This resulted in a PAir10 that was approximately 1.9 times higher than
In an effort to test our hypothesis the ratio, fc, of the coarse PM to the
the reference PM10.
fine PM:
Examining the full dataset, the relative root mean square error
(rRMSE) for daily average concentrations of PM1 was 121%, of PM2.5 f c ¼ PM2:5–10 =PM2:5 ð1Þ
142%, and of PM10 it was equal to 118% (Table 3). The corresponding er-
rors for the hourly averages were approximately 10% higher. In absolute was calculated based on the GRIMM measurements. This ratio fc varied
terms, the RMSE was around 7 μg m−3 for the PM1, around 3 μg m−3 for in our data set (hourly measurements) in the range from 0.01 to ap-
the PM1–2.5 and around 2.7 μg m−3 for the PM2.5–10. The low-cost sensor proximately 3. We then excluded from the data set the measurements
during periods for which fc exceeded a threshold value fc* and calculated
Table 2 the R2 for the hourly PMx values and the fraction of the remaining points
Study period averaged PM values for the two instruments, GRIMM and PAir. (data completeness). We repeated this exercise for all values of fc from 3
PM1 (μg m−3) PM1–2.5 PM2.5 (μg m−3) PM2.5–10 PM10 (μg m−3) to 0.01 with a step of 0.01 and the results are shown in Fig. 3. As the fc*
(μg m−3) (μg m−3) decreases the R2 increases and at the same time the fraction of the data
GRIMM PAir GRIMM PAir GRIMM PAir GRIMM PAir GRIMM PAir
remaining in the data set decreases. However, this change in R2 is mar-
ginal for fc* above 1 partially because only a few measurements fall in
5.60 11.30 1.12 3.04 6.73 14.47 1.76 1.37 8.54 15.86
this range. When fc* = 3 all measurements are included and
G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396 5

Fig. 2. Comparison of the hourly average concentrations reported by the GRIMM and the original PAir values for: (a) PM1; (b) PM1–2.5; (c) PM2.5; (d) PM2.5–10 and (e) PM10. The blue dots
denote measurements during Sahara dust episodes. The 1:1 line is also shown.

R2 = 0.56, while for fc* = 1, 3.6% of the measurements are excluded and support our hypothesis that PAir measurements during periods for which
R2 = 0.58. As the fc* decreases further, the R2 increases to 0.67 for f⁎c = 0.51 the coarse PM is significant compared to the fine PM (more specifically
and then rapidly to 0.85 for f⁎c = 0.47 (with 17% of the measurements ex- when PM2.5–10 > 0.47 PM2.5) have a different behavior and require differ-
cluded). Further reductions of the f⁎c lead to marginal increases of the R2 ent correction compared to the rest. Given that these measurements are a
but significant reductions of the remaining data points. Removing half small fraction of the overall dataset (17%) in this study, we will first focus
of the data corresponds to an increase of the R2 by only 0.01. Based on this on the other 83% in an effort to characterize first the PAir response during
analysis, the threshold of f⁎c = 0.47 is selected for the analysis. This allows periods of low coarse PM levels.
us to keep 83% of the PAir measurements during the study period and This result could be attributed to the losses of coarse particles inside
eliminates most of the outlying points as shown in Fig. 3. These results the low-cost sensor and the low sensitivity of its detector to such
6 G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396

Fig. 3. R2 (red symbols) between the hourly PMx measurements of the GRIMM and the PAir sensor as a function of fc = PM2.5–10/PM2.5. Also shown the data completeness (%) as a function
of fc (blue symbols) and selected threshold fc* = 0.47 (black diamond).

particles which scatter less light per unit mass compared to smaller ones needed. Such a method applicable to areas affected by Sahara dust is
(Schmidt-Ott and Ristovski, 2003). Particles in the PAir sensor are forced discussed in the next section.
to take two 90o turns before passing the laser beam (Sayahi et al., 2019).
Similar problems during high coarse PM periods have been reported by 4. PAir sensor calibration
Mukherjee et al. (2017). Finally, the PMS manufacturer does not provide
the algorithm for the conversion of the scattering signal to mass concen- A lot of the high coarse PM periods in the site are associated with
tration. Dust particles, due to composition and size but also issues re- transport of Sahara dust to the area. The proposed correction method
lated to their flow inside the low-cost sensors, may need a different identifies these based on available dust forecasts. These forecasts are
calibration algorithm to reduce the higher biases observed. based on the model SKIRON (Nickovic et al., 2001; Spyrou et al., 2010)
The evaluation of the PAir measurements against those of the refer- and on the Barcelona Dust Forecast Center (Haustein et al., 2012;
ence instrument were repeated for the subset of the measurements Pérez et al., 2011).
with relatively low coarse particle concentrations. Given that only 17%
of the data points were excluded, the improvement in the error and 4.1. Sahara dust events identification
bias metrics was modest with improvements of a few percent (Table 4).
For example, the RMSE of the hourly PM1 measurements decreased During March–April 2018 Sahara dust was transported to Greece in a
from 7.31 to 7.22 μg m−3 and the mean bias from 1.03 to 1 μg m−3. Clearly few periods with the hourly PM10 mass concentration in Patras reaching
the most significant improvement is the improvement of the R2 which for up to 140 μg m−3. Dust forecasts by SKIRON (0.24° × 0.24° resolution)
the same hourly PM2.5 values increased from 0.56 to 0.85. The corre- revealed increased dust loads across all over Greece while Barcelona
sponding change for PM1 was from 0.82 to 0.87. This will be important Dust Forecast Center (on 0.1° × 0.1° resolution) dust prediction models
in the next step for the calibration of the PAir sensor for the periods reported dust clouds moving and covering Southern Europe region.
with relatively low coarse PM levels. This effect can be seen in the This approach identified 10% of the hourly measurements as periods
slope of the linear regression of the hourly PAIr PM2.5 values versus during which Sahara dust was evident (PM10 > 40 μg m−3) in the area.
the reference PM2.5 that increases from 1.51 to 2.38 when the high These data points are shown in Fig. 2 (with blue color). This subset is in-
coarse PM periods are excluded. The effect on the slope of the PM1 cluded in the set of measurements identified as Sahara dust cases by the
line is a lot smaller as it changes only from 1.94 to 1.91. We conclude GRIMM measurements (16% of the data), shown in Fig. 4, given that
that the effect of the coarse PM on the PAir behavior is more impor- there are periods during which the high coarse PM in the site could be
tant for PM2.5 measurements than for PM1. This is expected given due to other sources (local dust, sea salt, etc.) or may have been Sahara
that a significant fraction of the lower tail of the dust size distribution dust that was not predicted by the corresponding models.
is in the 1–2.5 μm size range. The evaluation of the PAir measurements excluding these high Sa-
While the preceding analysis shows that the PAir behavior changes hara dust period is summarized in Table 4. Excluding the predicted
during periods of high coarse PM levels, the detection of these data high Sahara dust periods from the data set has a small effect on the
points has relied on the measurements of a reference instrument. PM1 error and bias metrics compared to the full data set and leads to
These reference measurements will not be available in general, there- modest improvements of the PM2.5 metrics. For example, the RMSE
fore a detection method based on readily available information is for PM2.5 changes by approximately 1%.

Table 4
Evaluation of the PAir performance hourly performance excluding high coarse PM and high Sahara dust periods.

Nobs ΜΒE (μg/m3) rMBE MAE (μg/m3) rMAE RMSE (μg/m3) rRMSE Slope Intercept (μg/m3) R2

Excluding periods of high coarse PM (based on GRIMM measurements)


PM1 9710 5.76 1.00 5.81 1.01 7.22 1.26 1.94 0.36 0.87
PM2.5 8.29 1.29 8.36 1.30 10.50 1.63 2.38 −0.62 0.85

Excluding periods of predicted high Sahara dust levels (based on transport models)
PM1 11,644 5.81 1.04 5.87 1.05 7.32 1.31 1.98 0.30 0.84
PM2.5 7.94 1.20 8.14 1.23 10.26 1.55 2.16 0.26 0.74
G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396 7

Fig. 4. Hourly PAir (a) PM1 and (b) PM2.5 measurements versus the corresponding GRIMM measurements. The data points with fc* > 0.47, based on the GRIMM measurements are in blue,
while the rest are in red.

This model-based approach also leads to significant improvements corrected values. Eqs. (2) and (3) ignore any potential dependence of
of the R2 between the PAir and GRIMM measurements. For hourly the sensor response to temperature and RH. The lack of correlation be-
PM2.5 values R2 increased from 0.56 to 0.74 (it was 0.85 when the tween the PAir measurements and the reference values for the coarse
GRIMM measurements were used to identify high coarse PM periods). PM makes the use of the sensor for PM10 measurements problematic.
The corresponding change for PM1 was from 0.82 to 0.84 (it was 0.87 The correlation between the calibrated PAir measurements and the ref-
when the reference instrument was used). These results suggest that erence instrument is shown in Fig. 5.
the use of the readily available transport model predictions is an excel- The use of this relatively simple correction reduces the relative mean
lent alternative for the identification of the high coarse PM periods if error for hourly PM1 to 18.6% (1.07 μg m−3) and for PM2.5 to 17.7%
one is interested in PM1 and a good alternative for PM2.5 analysis in (1.14 μg m−3). The corresponding root mean square errors are 24.5%
areas affected by Saharan dust. (1.42 μg m−3) for hourly PM1 and 24.6% (1.56 μg m−3) for PM2.5. The
biases of the corrected values are, as expected, practically zero. The dra-
4.2. Correction of the PAir measurements for fine PM matic improvement of the accuracy of the measurements with these
simple corrections is encouraging for the ability of these sensors to pro-
The first correction approach that we investigated relied simply on vide reasonably accurate concentration measurements of fine PM.
the linear regressions shown in Table 4. The dataset excluding measure- To quantify the performance of the proposed calibration technique,
ments with high coarse to fine PM ratios, as identified by the GRIMM the entire data set was randomly split into two subsets: the training
measurements, was used. The equations for the correction of the PAir subset with 75% of the data and the testing with the other 25%. The
PM1 and PM2.5 values in this case are: training set was used to calculate the calibration equation (slope and in-
 tercept) which was then implemented on the testing subset for the
PAir1;cor ¼ 0:52 PAir1 –0:18 μg m−3 ð2Þ evaluation. This procedure was conducted for both PM1 and PM2.5 frac-
 tions. In order to increase the robustness of the analysis, 1000 such inde-
PAir2:5;cor ¼ 0:42 PAir2:5 þ 0:26 μg m−3 ð3Þ pendent tests were conducted.
For PM1, the slopes and intercepts of the training data sets ranged
where PAir1 and PAir2.5 are the PM1 and PM2.5 values reported by the from 0.5 to 0.52 and from 0.06 to 0.24 μg m−3 respectively. The calcu-
PAir sensor (for CF = 1) and PAir1,cor, PAir2.5,cor are the corresponding lated averages are 0.51 and 0.18 μgm−3 and they are similar to the
8 G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396

Fig. 5. Comparison of hourly corrected PAir measurements against the corresponding GRIMM measurements for PM1 (upper panel) and PM2.5 (lower panel). 9710 data points are shown.
The periods with high coarse PM concentrations have been excluded. The 1:1 line is also shown.

ones calculated for the entire dataset (Eq. (2)). The RMSE (and rRMSE) low coarse PM concentrations periods were used. This allows us to sepa-
values for the testing data set are 1.3–1.54 μg m−3 (0.23–0.27) with a rate to a certain extent the two issues: response of the sensor to RH and
mean value 1.41 μg m−3 (0.25). The difference of these results with response to coarse particles.
the use of the full data set is only 0.01 μg m−3 (0.05). The MAE and The error, fEx, is defined as
rMAE mean values using the testing datasets were practically identical
with those derived from the full set. Similar agreement was found for fEx ¼ PAircor;x −PMx ð4Þ
the PM2.5 measurements. For the training datasets, the slope ranged
where x is the PM size range, PMx is the reference measurement, and
from 0.41 to 0.42 with a mean value of 0.42 that is identical to the one
PAircor,x is the corresponding PAir measurement corrected using
derived from the entire dataset (Eq. (2)) while the mean value of inter-
Eqs. (2) and (3). These previous corrections neglect the effect of the RH.
cept differs by only 0.01 μg m−3. The mean RMSE (rRMSE) and MAE
The error for PM1 and PM2.5 is shown in Fig. 6. Positive error suggests
(rMAE) values differed by less than 0.04 μg m−3 (0.01). These results
that the corrected sensor value overestimates the PM concentration (as
suggest that our analysis with the full dataset is quite robust.
given by the reference instrument) and vice versa. There is a positive
error for RH less 45% suggesting that our previous correction results in
4.3. Influence of RH on the performance of PAir sensors
overpredictions for low RH periods. This is as much as 1.5 μg m−3 for
25% RH. This suggests that a RH-independent calibration could provide
Atmospheric particles absorb water as the atmospheric relative hu-
reasonable results as long as it is applied to periods during which the RH
midity (RH) increases. This leads to increases of both their size and
is not extreme (high or low).
mass (Hänel, 1976, Hegg et al., 1993; Brock et al., 2016, He et al.,
To quantify the improvement in performance if a RH-dependent cor-
2019). Light scattering by the particles also increases (Zhao et al.,
rection was used, the errors were fitted with third-degree polynomials
2019). The reported PM concentrations should not include this water.
(Fig. 6):
High-quality instruments are equipped with dryers or heaters to re-
move particle water. Low-cost sensors, such as PAir, do not include PAircor;x −PMx ¼ f x ðRHÞ ¼ ax þ bx RH þ cx RH2 þ dx RH3 ð5Þ
dryers or heaters. Their response of PAir is based on light scattering by
the particles, therefore it is expected to be sensitive to RH. In this sec-
tion, we try to quantify this effect for the conditions of our campaign. and Eq. (5) (after solving for PMx) was used to correct the data. Overall,
The effect of RH on the PM concentrations reported by the PAir is the RH effect on statistical metrics was almost negligible (Table 5). For
assessed by examining the relationship between the PAir error and the example the RMSE for PM1 was decreased from 1.42 to 1.37 μg m−3.
RH. The error is calculated as the difference between the corrected hourly This very low sensitivity of PAir measurements to RH is unexpected
PAir measurements (using Eqs. (2) and (3)) and the reference measure- and not consistent with results for other sensors (Jayaratne et al.,
ments. In this first stage of the analysis, only the data corresponding to 2018; Zheng et al., 2018; Wang et al., 2015), however it should be
G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396 9

Fig. 6. Differences between the corrected PAir and GRIMM measurements as a function of relative humidity (RH) for PM1 (upper panel) and PM2.5 (lower panel). The black dots and the
solid red lines correspond to the mean and median differences for each RH class. The bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The number of
the observation points for each RH bin is shown on top of each boxplot. The solid blue lines are the third-degree polynomials (Eq. (5)) between the PM differences and RH.

viewed as a positive attribute of the sensor. Following the examination 2017 to November 2018 (11,681 hourly measurements) and covered
of RH effect on PAir sensors' performance, the effect of ambient temper- a PM2.5 concentration range of 0.1–95 μg m−3 (hourly average 6.7 ±
ature was also examined and was found to be negligible. 3.9 μg m−3), a coarse PM range of 0.01–41 μg m−3 (hourly average
1.8 ± 2.9 μg m−3), ambient temperature from 3 to 39 °C (hourly average
5. Conclusions 20 ± 6 °C), and RH from 9 to 97% (hourly average 62 ± 16%).
The measurements of PMS sensors were reproducible with an
This study investigates and evaluates the performance of the com- R2 > 0.88 and discrepancies lower than 1.37 μg m−3 between the two
mercially available low-cost PM sensor, PurpleAir PA-II (PAir), against channels, for all fractions. The uncorrected hourly PAir PM1 measure-
a GRIMM instrument in Patras, Greece. These PAir units use two ments (using a unity correction factor in the Purple Air data processing)
Plantower PMS 5003 sensors. Our study lasted 18 months from May were well-correlated (R2 = 0.82) with the reference instrument. The R2
for the hourly PM1–2.5 was only 0.1 and of the hourly PM2.5–10 0.02. This
suggests that the PAir is best suited for PM1 measurements. Given that
Table 5
most of the PM2.5 during the study was in the sub-micrometer range
Evaluation of the RH-independent and RH-dependent corrected hourly fine PM PAir
measurements. the PM2.5 correlation was also reasonable (R2 = 0.56).
The PAir appears to have difficulties in the measurement of coarse
MAE (μg/m3) rMAE RMSE (μg/m3) rRMSE R2
particles. Its reported values for PM10 are mainly influenced by the
RH-independent correction sub-micrometer fraction. If the PM10 includes a significant coarse frac-
PM1 1.07 0.186 1.42 0.245 0.87 tion, the reported PM10 concentration by this low cost sensor appears
PM2.5 1.14 0.177 1.56 0.246 0.85
to be questionable.
RH-dependent correction The uncorrected (original, indicated as CF = 1) PAir values for PM1
PM1 1.03 0.179 1.37 0.239 0.88 and PM2.5 were approximately two times higher than the values of the
PM2.5 1.09 0.170 1.54 0.241 0.86
reference instrument. The relative root mean square error (rRMSE) for
10 G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396

daily average concentrations of PM1 was 121% and of PM2.5 142%. The Programme “Competitiveness, Entrepreneurship and Innovation” (NSRF
low-cost sensor fine PM (PM1 and PM2.5) measurements were also char- 2014–2020) and co-financed by Greece and the European Union
acterized by high bias, exceeding 100%. This indicates rather poor perfor- (European Regional Development Fund).
mance of the PAir sensors for fine PM, if the uncorrected values are used.
The PM1 and PM2.5 PAir measurements during periods for which the References
coarse PM was significant compared to the PM2.5 have a different behav-
Aguilera, I., Eeftens, M., Meier, R., Ducret-Stich, R.E., Schindler, C., Ineichen, A., Phuleria,
ior compared to the rest of the measurements. This is probably due to the H.C., Probst-Hensch, N., Tsai, M.Y., Künzli, N., 2015. Land use regression models for
different aerosol size distribution during these periods. Excluding the crustal and traffic-related PM2.5 constituents in four areas of the SAPALDIA study. En-
measurements during periods for which PM2.5–10 > 0.47 PM2.5 (16% of viron. Res. 140, 377–384. https://doi.org/10.1016/j.envres.2015.04.011.
Analitis, A., Katsouyanni, K., Dimakopoulou, K., Samoli, E., Nikoloulopoulos, A.K., Petasakis,
the hourly data points) improved the R2 for the hourly PM1 from 0.82
Y., Touloumi, G., Schwartz, J., Anderson, H.R., Cambra, K., Forastiere, F., Zmirou, D.,
to 0.87. A method to identify these periods of high coarse PM due to Sa- Vonk, J.M., Clancy, L., Kriz, B., Bobvos, J., Pekkanen, J., 2006. Short-term effects of am-
hara dust based on atmospheric transport model predictions was evalu- bient particles on cardiovascular and respiratory mortality. Epidemiology 17,
ated. This approach identified 8% of the hourly measurements as high 230–233. https://doi.org/10.1097/01.ede.0000199439.57655.6b.
Beelen, R., Raaschou-Nielsen, O., Stafoggia, M., Andersen, Z.J., Weinmayr, G., Hoffmann, B.,
dust periods and increased the R2 for the hourly PM1 from 0.82 to 0.85. Wolf, K., Samoli, E., Fischer, P., Nieuwenhuijsen, M., Vineis, P., Xun, W.W.,
Use of this readily applicable method (the model results of SKIRON and Katsouyanni, K., Dimakopoulou, K., Oudin, A., Forsberg, B., Modig, L., Havulinna, A.S.,
the Barcelona Dust Forecast Center are publicly available) is recom- Lanki, T., Turunen, A., Oftedal, B., Nystad, W., Nafstad, P., De Faire, U., Pedersen, N.L.,
Östenson, C.G., Fratiglioni, L., Penell, J., Korek, M., Pershagen, G., Eriksen, K.T.,
mended for sites affected by Sahara dust episodes. The calibration of the Overvad, K., Ellermann, T., Eeftens, M., Peeters, P.H., Meliefste, K., Wang, M., Bueno-
sensor needs to be different for the high coarse PM periods. The rest of De-Mesquita, B., Sugiri, D., Krämer, U., Heinrich, J., De Hoogh, K., Key, T., Peters, A.,
the work excluded these periods from the analysis focusing on the re- Hampel, R., Concin, H., Nagel, G., Ineichen, A., Schaffner, E., Probst-Hensch, N.,
Künzli, N., Schindler, C., Schikowski, T., Adam, M., Phuleria, H., Vilier, A., Clavel-
maining 84% of the hourly data points. Chapelon, F., Declercq, C., Grioni, S., Krogh, V., Tsai, M.Y., Ricceri, F., Sacerdote, C.,
The use of a simple relative-humidity independent linear correction Galassi, C., Migliore, E., Ranzi, A., Cesaroni, G., Badaloni, C., Forastiere, F., Tamayo, I.,
reduced the relative root mean square error to 24.5% (1.41 μg m−3) for Amiano, P., Dorronsoro, M., Katsoulis, M., Trichopoulou, A., Brunekreef, B., Hoek, G.,
hourly PM1 and to 24.6% (1.58 μg m−3) for PM2.5. This is a significant re- 2014. Effects of long-term exposure to air pollution on natural-cause mortality: an
analysis of 22 European cohorts within the multicentre ESCAPE project. Lancet 383,
duction by a factor of 4–5 compared to the use of the uncorrected 785–795. https://doi.org/10.1016/S0140-6736(13)62158-3.
measurements. Bell, M.L., Ebisu, K., Peng, R.D., 2011. Community-level spatial heterogeneity of chemical
The influence of RH on the performance of PAir sensors was also ex- constituent levels of fine particulates and implications for epidemiological research.
J. Expo. Sci. Environ. Epidemiol. 21, 372–384. https://doi.org/10.1038/jes.2010.24.
amined and was found to be surprisingly small. The application of a RH- Brock, C.A., Wagner, N.L., Anderson, B.E., Attwood, A.R., Beyersdorf, A., Campuzano-Jost, P.,
dependent correction relying on a third-degree polynomial reduced the Carlton, A.G., Day, D.A., Diskin, G.S., Gordon, T.D., Jimenez, J.L., Lack, D.A., Liao, J.,
RMSE by only 1%. Markovic, M.Z., Middlebrook, A.M., Ng, N.L., Perring, A.E., Richardson, M.S., Schwarz,
J.P., Washenfelder, R.A., Welti, A., Xu, L., Ziemba, L.D., Murphy, D.M., 2016. Aerosol op-
The good agreement between the corrected PM1 and PM2.5 values of tical properties in the southeastern United States in summer - part 1: hygroscopic
the PAir sensor and the reference instrument suggests that the PAir can growth. Atmos. Chem. Phys. 16, 4987–5007. https://doi.org/10.5194/acp-16-4987-
be used for ambient measurements under ambient conditions similar 2016.
to our study taking into account the corresponding uncertainty of its Cesari, D., De Benedetto, G.E., Bonasoni, P., Busetto, M., Dinoi, A., Merico, E., Chirizzi, D.,
Cristofanelli, P., Donateo, A., Grasso, F.M., Marinoni, A., Pennetta, A., Contini, D.,
measurements. Moreover, a number of important insights have emerged. 2018. Seasonal variability of PM2.5 and PM10 composition and sources in an urban
The response of the sensors for fine PM is affected by the coarse PM. The background site in Southern Italy. Sci. Total Environ. 612, 202–213. https://doi.org/
RH appears to have little effect on the response and a RH-independent 10.1016/j.scitotenv.2017.08.230.
Crippa, M., Decarlo, P.F., Slowik, J.G., Mohr, C., Heringa, M.F., Chirico, R., Poulain, L., Freutel,
calibration performs practically as well as a RH-dependent one. This is a F., Sciare, J., Cozic, J., Di Marco, C.F., Elsasser, M., Nicolas, J.B., Marchand, N., Abidi, E.,
positive but also surprising result and additional studies in other areas Wiedensohler, A., Drewnick, F., Schneider, J., Borrmann, S., Nemitz, E., Zimmermann,
are needed to assess its generality and also to explain it. The use of the R., Jaffrezo, J.L., Prévôt, A.S.H., Baltensperger, U., 2013. Wintertime aerosol chemical
composition and source apportionment of the organic fraction in the metropolitan
uncorrected values reported by the PAir is recommended as the basis
area of Paris. Atmos. Chem. Phys. 13, 961–981. https://doi.org/10.5194/acp-13-961-
for the subsequent calibration and correction. Based on the results of 2013.
this study, additional analysis of the response of the sensor to coarse par- Dominici, F., Peng, R.D., Bell, M.L., Pham, L., McDermott, A., Zeger, S.L., Samet, J.M., 2006. Fine
ticles and lack of response to RH is recommended. particulate air pollution and hospital admission for cardiovascular and respiratory dis-
eases. J. Am. Med. Assoc. 295, 1127–1134. https://doi.org/10.1001/jama.295.10.1127.
EU, 2008. Directive 2008/50/EC of the European Parliament and the Council of 21 May
CRediT authorship contribution statement 2008 on Ambient Air Quality and Cleaner Air for Europe.
Florou, K., Papanastasiou, D.K., Pikridas, M., Kaltsonoudis, C., Louvaris, E., Gkatzelis, G.I.,
Patoulias, D., Mihalopoulos, N., Pandis, S.N., 2017. The contribution of wood burning
G. Kosmopoulos: Methodology, Validation, Formal analysis, Data and other pollution sources to wintertime organic aerosol levels in two Greek cities.
curation, Writing - original draft. V. Salamalikis: Methodology, Valida- Atmos. Chem. Phys. 17, 3145–3163. https://doi.org/10.5194/acp-17-3145-2017.
tion, Formal analysis, Writing - review & editing. S.N. Pandis: Conceptu- Gerasopoulos, E., Amiridis, V., Kazadzis, S., Kokkalis, P., Eleftheratos, K., Andreae, M.O.,
alization, Methodology, Writing - review & editing. P. Yannopoulos: Andreae, T.W., El-Askary, H., Zerefos, C.S., 2011. Three-year ground based measure-
ments of aerosol optical depth over the Eastern Mediterranean: the urban environ-
Data curation, Writing - review & editing. A.A. Bloutsos: Data curation, ment of Athens. Atmos. Chem. Phys. 11, 2145–2159. https://doi.org/10.5194/acp-
Writing - review & editing. A. Kazantzidis: Conceptualization, Method- 11-2145-2011.
ology, Writing - review & editing. Hackmann, D., Sjöberg, E., 2017. Ambient air pollution and pregnancy outcomes: a study
of functional form and policy implications. Air Qual. Atmos. Heal. 10, 129e137.
Hänel, G., 1976. The properties of atmospheric aerosol particles as functions of the relative
Declaration of competing interest humidity at thermodynamic equilibrium with the surrounding moist air. Adv.
Geophys. 19, 73–188.
The authors declare that they have no known competing financial Haustein, K., Pérez, C., Baldasano, J.M., Jorba, O., Basart, S., Miller, R.L., Janjic, Z., Black, T.,
Nickovic, S., Todd, M.C., Washington, R., Müller, D., Tesche, M., Weinzierl, B.,
interests or personal relationships that could have appeared to influ- Esselborn, M., Schladitz, A., 2012. Atmospheric dust modeling from meso to global
ence the work reported in this paper. scales with the online NMMB/BSC-Dust model - part 2: experimental campaigns in
Northern Africa. Atmos. Chem. Phys. 12, 2933–2958. https://doi.org/10.5194/acp-
12-2933-2012.
Acknowledgments He, Y., Gu, Z., Lu, W., Zhang, L., Okuda, T., Fujioka, K., Luo, H., Yu, C.W., 2019. Atmospheric
humidity and particle charging state on agglomeration of aerosol particles. Atmos.
We acknowledge the support of this work by the project “PANhellenic Environ. 197, 141–149.
Hegg, D., Larson, T., Yuen, P.F., 1993. A theoretical study of the effect of relative humidity on
infrastructure for Atmospheric Composition and climatE change” (MIS
light scattering by tropospheric aerosols. J. Geophys. Res.-Atmos. 98, 18435–18439.
5021516) which is implemented under the Action “Reinforcement of Heimann, I., Bright, V.B., McLeod, M.W., Mead, M.I., Popoola, O.A.M., Stewart, G.B., Jones,
the Research and Innovation Infrastructure”, funded by the Operational R.L., 2015. Source attribution of air pollution by spatial scale separation using high
G. Kosmopoulos et al. / Science of the Total Environment 748 (2020) 141396 11

spatial density networks of low cost air quality sensors. Atmos. Environ. 113, 10–19. &ndash; part 1: model description, annual simulations and evaluation. Atmos.
https://doi.org/10.1016/j.atmosenv.2015.04.057. Chem. Phys. 11, 13001–13027. https://doi.org/10.5194/acp-11-13001-2011.
Hoek, G., Krishnan, R.M., Beelen, R., Peters, A., Ostro, B., Brunekreef, B., Kaufman, J.D., 2013. Pikridas, M., Tasoglou, A., Florou, K., Pandis, S.N., 2013. Characterization of the origin of
Long-term air pollution exposure and cardio-respiratory mortality: a review. Environ. fine particulate matter in a medium size urban area in the Mediterranean. Atmos. En-
Heal. A Glob. Access Sci. Source 12. https://doi.org/10.1186/1476-069X-12-43. viron. 80, 264–274. https://doi.org/10.1016/j.atmosenv.2013.07.070.
IARC (International Agency for Research on Cancer), 2013. Air Pollution and Cancer. WHO Pope, C.A., Dockery, D.W., 2006. Health effects of fine particulate air pollution: lines
Press, World Health Organization, Geneva, Switzerland. that connect. J. Air Waste Manag. Assoc. 56, 709–742. https://doi.org/10.1080/
IPCC, 2013. Summary for policymakers. In: Stocker, T.F., Qin, D., Plattner, G.-K., Tignor, M., 10473289.2006.10464485.
Allen, S.K., Boschung, J., Nauels, A., Xia, Y., Bex, V., Midgley, P.M. (Eds.), Climate Putaud, J.P., Van Dingenen, R., Alastuey, A., Bauer, H., Birmili, W., Cyrys, J., Flentje, H., Fuzzi,
Change 2013: The Physical Science Basis. Contribution of Working Group I to the S., Gehrig, R., Hansson, H.C., Harrison, R.M., Herrmann, H., Hitzenberger, R., Hüglin, C.,
Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cam- Jones, A.M., Kasper-Giebl, A., Kiss, G., Kousa, A., Kuhlbusch, T.A.J., Löschau, G.,
bridge University Press, Cambridge, United Kingdom and New York, NY, USA. Maenhaut, W., Molnar, A., Moreno, T., Pekkanen, J., Perrino, C., Pitz, M., Puxbaum,
Jayaratne, R., Liu, X., Thai, P., Dunbabin, M., Morawska, L., 2018. The influence of humidity H., Querol, X., Rodriguez, S., Salma, I., Schwarz, J., Smolik, J., Schneider, J., Spindler,
on the performance of a low-cost air particle mass sensor and the effect of atmo- G., ten Brink, H., Tursic, J., Viana, M., Wiedensohler, A., Raes, F., 2010. A European
spheric fog. Atmos. Meas. Tech 11, 4883–4890. https://doi.org/10.5194/amt-11- aerosol phenomenology-3: physical and chemical characteristics of particulate mat-
4883-2018. ter from 60 rural, urban, and kerbside sites across Europe. Atmos. Environ. 44,
Jiao, W., Hagler, G., Williams, R., Sharpe, R., Brown, R., Garver, D., Judge, R., Caudill, M., 1308–1320. https://doi.org/10.1016/j.atmosenv.2009.12.011.
Rickard, J., Davis, M., Weinstock, L., Zimmer-Dauphinee, S., Buckley, K., 2016. Commu- Rai, A.C., Kumar, P., Pilla, F., Skouloudis, A.N., Di Sabatino, S., Ratti, C., Yasar, A., Rickerby, D.,
nity Air Sensor Network (CAIRSENSE) project: evaluation of low-cost sensor perfor- 2017. End-user perspective of low-cost sensors for outdoor air pollution monitoring.
mance in a suburban environment in the southeastern United States. Atmos. Meas. Sci. Total Environ. 607–608, 691–705. https://doi.org/10.1016/j.scitotenv.2017.06.266.
Tech 9, 5281–5292. https://doi.org/10.5194/amt-9-5281-2016. Sayahi, T., Butterfield, A., Kelly, K.E., 2019. Long-term field evaluation of the Plantower
Kassomenos, P.A., Vardoulakis, S., Chaloulakou, A., Paschalidou, A.K., Grivas, G., Borge, R., PMS low-cost particulate matter sensors. Environ. Pollut. 245, 932–940. https://doi.
Lumbreras, J., 2014. Study of PM10 and PM2.5 levels in three European cities: analysis org/10.1016/j.envpol.2018.11.065.
of intra and inter urban variations. Atmos. Environ. 87, 153–163. https://doi.org/ SCAQMD, 2016. Field Evaluation Purple Air PM Sensor Background.
10.1016/j.atmosenv.2014.01.004.
Schmidt-Ott, A., Ristovski, Z.D., 2003. Measurement of airborne particles. Indoor Environ-
Kelly, K.E., Whitaker, J., Petty, A., Widmer, C., Dybwad, A., Sleeth, D., Martin, R., Butterfield,
ment: Airborne Particles and Settled Dust. Wiley-VCH Verlag GmbH & Co. KGaA,
A., 2017. Ambient and laboratory evaluation of a low-cost particulate matter sensor.
pp. 56–81.
Environ. Pollut. 221, 491–500. https://doi.org/10.1016/j.envpol.2016.12.039.
Shapiro, M.A., Swann, P/.G., Hartsough, M., 2014. Regulatory considerations of lower cost
Kumar, P., Morawska, L., Martani, C., Biskos, G., Neophytou, M., Di Sabatino, S., Bell, M.,
air pollution sensor data performance. Handb. Ther. Antibodies. vol. 1, pp. 277–300.
Norford, L., Britter, R., 2015. The rise of low-cost sensing for managing air pollution
https://doi.org/10.1002/9783527619740.ch12.
in cities. Environ. Int. 75, 199–205. https://doi.org/10.1016/j.envint.2014.11.019.
Sousan, S., Koehler, K., Hallett, L., Peters, T.M., 2017. Evaluation of consumer monitors to
Lewis, A., Edwards, P., 2016. Validate personal air-pollution sensors. Nature 535, 29–31.
measure particulate matter. J. Aerosol Sci. 107, 123–133. https://doi.org/10.1016/j.
https://doi.org/10.1038/535029a.
jaerosci.2017.02.013.
Manikonda, A., Zikova, N., Hopke, P.K., Ferro, A.R., 2016. Laboratory assessment of low-cost
PM monitors. J. Aerosol Sci. 102, 29–40. https://doi.org/10.1016/j.jaerosci.2016.08.010. South Coast Air Quality Management District (SCAQMD), 2017. Field evaluation purple air
Manousakas, M., Diapouli, E., Papaefthymiou, H., Kantarelou, V., Zarkadas, C., Kalogridis, (PA-II) PM sensor. available at. http://www.aqmd.gov/docs/default-source/aq-spec/
A.C., Karydas, A.G., Eleftheriadis, K., 2018. XRF characterization and source apportion- fieldevaluations/purple-air-pa-ii-field-evaluation.pdf?sfvrsn=2.
ment of PM10 samples collected in a coastal city. X-Ray Spectrom. 47, 190–200. Spyrou, C., Mitsakou, C., Kallos, G., Louka, P., Vlastou, G., 2010. An improved limited area
https://doi.org/10.1002/xrs.2817. model for describing the dust cycle in the atmosphere. J. Geophys. Res. 115. https://
Mukherjee, A., Stanton, L.G., Graham, A.R., Roberts, P.T., 2017. Assessing the utility of low- doi.org/10.1029/2009JD013682 D17211.
cost particulate matter sensors over a 12-week period in the Cuyama valley of Cali- Steinle, S., Reis, S., Eric, C., 2015. Science of the total environment quantifying human ex-
fornia. Sensors (Switzerland) 17. https://doi.org/10.3390/s17081805. posure to air pollution — moving from static monitoring to spatio-temporally re-
Nickovic, S., Kallos, G., Papadopoulos, A., Kakaliagou, O., 2001. A model for prediction of solved personal exposure assessment. Sci. Total Environ. 443, 184–193. https://doi.
desert dust cycle in the atmosphere. J. Geophys. Res. Atmos. 106, 18113–18129. org/10.1016/j.scitotenv.2012.10.098.
https://doi.org/10.1029/2000JD900794. Wang, Y., Li, J., Jing, H., Zhang, Q., Jiang, J., Biswas, P., 2015. Laboratory evaluation and cal-
Papapostolou, V., Zhang, H., Feenstra, B.J., Polidori, A., 2017. Development of an environ- ibration of three low-cost particle sensors for particulate matter measurement. Aero-
mental chamber for evaluating the performance of low-cost air quality sensors sol Sci. Technol. 49, 1063–1077. https://doi.org/10.1080/02786826.2015.1100710.
under controlled conditions. Atmos. Environ. 171, 82–90. https://doi.org/10.1016/j. Williams, R., Kaufman, A., Hanley, T., Rice, J., Garvey, S., 2014. Evaluation of Field-deployed
atmosenv.2017.10.003. Low Cost PM Sensors doi:EPA/600/R-14/464 (NTIS PB 2015-102104).
Penttinen, P., Timonen, K.L., Tiittanen, P., Mirme, A., Ruuskanen, J., Pekkanen, J., 2001. Zhao, P., Ding, J., Du, X., Su, J., 2019. High time-resolution measurement of light scattering
Number concentration and size of particles in urban air: effects on spirometric lung hygroscopic growth factor in Beijing: a novel method for high relative humidity con-
function in adult asthmatic subjects. Environ. Health Perspect. 109, 319–323. ditions. Atmos. Environ. 215, 116912.
https://doi.org/10.1289/ehp.01109319. Zheng, T., Bergin, M.H., Johnson, K.K., Tripathi, S.N., Shirodkar, S., Landis, M.S., Sutaria, R.,
Pérez, C., Haustein, K., Janjic, Z., Jorba, O., Huneeus, N., Baldasano, J.M., Black, T., Basart, S., Carlson, D.E., 2018. Field evaluation of low-cost particulate matter sensors in high-
Nickovic, S., Miller, R.L., Perlwitz, J.P., Schulz, M., Thomson, M., 2011. Atmospheric and low-concentration environments. Atmos. Meas. Tech 11, 4823–4846. https://
dust modeling from meso to global scales with the online NMMB/BSC-Dust model doi.org/10.5194/amt-11-4823-2018.

You might also like