the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 May 2020
Laboratory evaluation of particle-size selectivity of optical low-cost particulate matter sensors
Low-cost particulate matter (PM) sensors have been underinvestigation as it has been hypothesized that the use of low-cost andeasy-to-use sensors could allow cost-efficient extension of the currentlysparse measurement coverage. While the majority of the existing literaturehighlights that low-cost sensors can indeed be a valuable addition to thelist of commonly used measurement tools, it often reiterates that the riskof sensor misuse is still high and that the data obtained from the sensorsare only representative of the specific site and its ambient conditions. Thisimplies that there are underlying reasons forinaccuracies in sensor measurements that have yet to be characterized. The objective of this study is toinvestigate the particle-size selectivity of low-cost sensors. Evaluatedsensors were Plantower PMS5003, Nova SDS011, Sensirion SPS30, SharpGP2Y1010AU0F, Shinyei PPD42NS, and Omron B5W-LD0101. The investigation ofsize selectivity was carried out in the laboratory using a novel referenceaerosol generation system capable of steadily producing monodisperseparticles of different sizes (from∼0.55 to 8.4 µm)on-line. The results of the study show that none of the low-cost sensorsadhered to the detection ranges declared by the manufacturers; moreover,cursory comparison to a mid-cost aerosol size spectrometer (Grimm 1.108, 2020)indicates that the sensors can only achieve independent responses for one ortwo size bins, whereas the spectrometer can sufficiently characterizeparticles with 15 different size bins. These observations provide insight intoand evidence of the notion that particle-size selectivity has an essentialrole in the analysis of the sources of errors in sensors.
- Article
(1143 KB) - Full-text XML
- Supplement
(885 KB) - BibTeX
- EndNote
The recent emergence of low-cost sensors has enabled new possibilities intraditional air quality monitoring (Kumar et al., 2015; Morawska et al.,2018; Snyder et al., 2013). As a result of low unit costs and compact size,sensors can be deployed to the field in much higher quantities than before,thus enabling higher-resolution spatiotemporal data. Few studies havedemonstrated applications of sensor networks (Caubel et al., 2019; Feinberget al., 2019; Gao et al., 2015; Jiao et al., 2016; Popoola et al., 2018;Yuval et al., 2019). Distributed sensing of air quality can be seen as animportant progression towards a more comprehensive understanding ofcity-scale air quality dynamics as air pollution, and particulate matter(PM) in particular, may have highly localized concentration “hot spots” inurban areas. Practical limitations, such as expensiveness and bulkiness,constrain the use of conventional instrumentation in monitoring networks;therefore, low-cost sensors could have an essential role in the spatialextension of measurement coverage.
Numerous field studies have been conducted previously; the majority haveunderlined the potential usefulness of optical particulate matter sensors(Karagulian et al., 2019; Rai et al., 2017). However, the literature hasalso emphasized that the risk of sensor misuse is still high and that someexternal factors, such as relative humidity, may produce significantmeasurement artifacts in the data (Jayaratne et al., 2018; Kuula et al.,2018; Liu et al., 2019). In comparison to gas sensing, PM measurements arenotably more challenging when ambient particle sizes and their respectivedistributions vary significantly from source to source and from location tolocation. Along with size, particle physical properties such as shape andrefractive index also affect the sensor output. Several studies have pointedout that along with dynamic adjustment for meteorological parameters,on-site calibrations are required in order to achieve higher levels ofaccuracy and precision (Zheng et al., 2018). However, when consideringadvanced calibration techniques, Schneider et al. (2019) have raised a validpoint noting that it may be unclear whether the sensor data resulting fromcomplex correction and conversion processes (e.g., machine learning) are stilla legitimate and independent product of the sensor measurement and not acombination of secondary data and statistical model prediction. This is animportant remark when evaluating the usability of sensors as it highlightsthe need to identify the reasons behind inaccuracies in low-cost sensormeasurements.
While field evaluations are a natural step towards understanding anddeveloping sensors, they provide limited information about the detailedsensor response characteristics. In particular, less attention has been paidto the investigation of particle-size discrimination of sensors. Although afew studies have noted that the detectable particle-size ranges of sensorsmay be significantly different from the ones declared in their respectivetechnical specification sheets (Budde et al., 2018; Levy Zamora et al.,2019), this factor is not commonly considered when assessing sensoraccuracy. Thus, more research is needed. The objective of this study was toinvestigate and characterize the size selectiveness of some of the opticallow-cost sensors commonly appearing in the literature. The evaluated sensorswere Plantower PMS5003, Nova SDS011, Sensirion SPS30, Sharp GP2Y1010AU0F,Shinyei PPD42NS, and Omron B5W-LD0101. Along with these low-cost sensors, amid-cost optical aerosol size spectrometer (Grimm model 1.108, Grimm AerosolTechnik GmbH, Germany) was evaluated cursorily to highlight the differencesbetween the responses of low-cost and mid-cost devices. The investigation ofsize selectivity was carried out in the laboratory using a novel referenceaerosol generation system capable of steadily producing monodisperseparticles of different sizes. Sensor responses were compared to a referenceinstrument (APS, aerodynamic particle sizer 3321, TSI Inc., USA), anddetectable particle-size ranges of the sensors were obtained.
2.1 Evaluated sensors
The sensors evaluated in this study, and their main detection properties,are listed in Table 1. The optical detection configurations of these sensorswere arranged in either a 90 or 120∘ scattering angle, and either ared laser or an infrared (IR) light-emitting diode (LED) was used as a lightsource. Sensors utilizing an LED were equipped with additional lightfocusing lenses. The optical chamber itself was composed of aninjection-molded plastic body which was placed onto an electronic circuitboard. The PMS5003, SDS011, and SPS30 use fans to generate sample flow,whereas the PPD42 and B5W utilized natural convection resulting from aheating resistor. The sampling of the GP2Y1010AU0F was based on diffusion.The optical configurations and plastic body layouts are shown in Fig. S1 in the Supplement. Three units for each sensor model were evaluated inorder to assess their inter-unit variation.
Table 1Basic features of the evaluated sensors declared by themanufacturers.
a Standard particle (CF =1) output was used.b Manually adjusted threshold voltage was set to 0.5 V asrecommended by the manufacturer.
The mid-cost Grimm 1.108 spectrometer, which was used here for demonstration purposes, isan optical aerosol size spectrometer with 15 size bins (from 0.23 to 20 µm). Previous evaluations of the Grimm 1.108 spectrometer have shown its responseto be similar to that of the APS (Peters et al., 2006); furthermore, its accuracy(mass of C-factor-adjusted total suspended particles) is comparable to that of massmeasurement methods such as the filter weighing method (Burkart et al.,2010).
2.2 Reference aerosol
2.2.1 Vibrating orifice aerosol generator and gradient elution pump
The aerosol sampled by the low-cost sensors was generated using a vibratingorifice aerosol generator 3450 (VOAG, TSI Inc., USA). The operatingprinciple of the VOAG is based on the instability and breakup of acylindrical liquid jet. Mechanical disturbances of a resonance frequencyvibration disintegrate the cylindrical jet into uniform droplets, which aredispersed into an aerosol flow system with appropriate dilution air.Dispersed droplets evaporate before significant coagulation occurs and formparticles from the non-volatile solute dissolved in the volatile liquid. Ifthe droplet liquid is non-volatile, the particle diameter and dropletdiameter are equal. Otherwise, the produced particle size is calculable fromthe volumetric fraction of the non-volatile solute, as shown in Eqs. (1)–(2):
whereDd is the generated droplet diameter,Q is the solution feedrate, andf is the disturbance frequency.
whereDp is the diameter of the formed particle,C is the volumetricconcentration of the non-volatile solute in the volatile liquid (typically2-propanol or purified water), andI is the volumetric fraction ofimpurity in the volatile liquid.
According to Berglund and Liu (1973), the output aerosol numberconcentration of the VOAG has a relative standard deviation of less than 3 %, and the formed particle-size distribution is monodisperse having ageometric standard deviation (GSD) less than 1.014. These, and particularlythe capability to produce highly monodisperse size distribution, areimportant features regarding sensor size selectivity evaluation; whilepolydisperse aerosol can be used, for instance, to estimate responsestability and linearity to varying concentration levels (Hapidin et al.,2019; Papapostolou et al., 2017; Sayahi et al., 2019a), the presence ofmultiple different-sized particles prevents the distinction between sensorresponse and specific particle size. The most significant deficiency of theVOAG (and the main limitation of this study) is that its smallest producibleparticle size is in practice limited by the impurity within the carrierliquid to approximately 0.55 µm.
The novelty of the aerosol generation method used in this research is basedon the observation that the particle size of the monodisperse and constantnumber concentration reference aerosol can be controlled by feedingsolutions with different non-volatile concentrations to the VOAG, one afterthe other. Such an aerosol generation technique was first utilized by Kuula etal. (2017), who accomplished the solution blending with a supplementarysyringe pump and a manually operated three-way valve. In this study,however, the solution feeding was done with a gradient elution pumptypically used in ion chromatography (GP50, Dionex Inc., USA). The GP50gradient pump has four different eluent channels and is capable ofdispensing liquids with high pressure (max. 5000 psi) and accurate volumeflow rate (0.04–10.0 mL min−1 in increments of 0.01 mL min−1). The foureluent channels can be mixed with a resolution of 0.1 % (combined outputof the four channels is always 100 %); furthermore, the GP50 has auser interface that enables the operator to generate parameterized eluent-dispensing programs. In essence, the utilization of the GP50 allows the userto freely choose and produce monodisperse aerosols of desired particle sizeswithout the tuning of VOAG running parameters or manual alternation of theliquid concentrations. Additionally, the preconfigured dispensing programsare fully automated, making the comparison of consecutive test runs morereliable.
2.2.2 Sampling configuration
A schematic figure of the used test setup is shown in Fig. 1. Referenceaerosol was generated using the VOAG–GP50 system as described in theprevious section. Dioctyl sebacate (DOS, density of 0.914 g cm−3) was usedas a non-volatile solute in a 2-propanol solvent (>99.999 %,Sigma-Aldrich), and the formed particles were transparent oil droplets.Although the reference instrument APS is known for having decreased countingefficiency for liquid droplets over∼5 µm in size(Volckens and Peters, 2005), no additional corrections were used. Runningparameters of the VOAG and GP50 are shown in Table S1. Thethree different DOS concentrations (A–C) refer to the four different eluentchannels of the GP50 (the use of three channels was sufficient for thisstudy).
Figure 1Schematic of the sensor evaluation setup. The Grimm 1.108 spectrometer drew itssample from where the sensor enclosure is now shown.
The GP50 used an automated program for dispensing the liquids. A programinvolves a number of consecutive time steps in which the blending ratios ofeluent channels, step durations, and volumetric flow rate of the liquid canbe defined separately. Executing the program means that the GP50 dispensesthe liquids according to the settings determined in each step. The programused in this evaluation consisted of 10 steps in which the produced particlesizes were logarithmically spaced from 0.45 to 9.78 µm. Thecalculated blending ratios and the respective particle sizes are shown inSupplemental Table S2. Step duration of 5 min was used; a single testrun thus lasted approximately 60 min. Dead volumes in the GP50 and VOAGslightly extend the theoretical run time duration. A complete test run canbe performed as quickly as in 15 min, which results in fewermeasurement points and weaker statistical power though. An example of the producedreference aerosol number size distribution measured with the APS is shown inFig. 2. It is worth underlining that the number of steps used in the GP50dispensing program does not dictate the number of different particle sizesproduced. The number of steps and the parameters assigned to them simplydefine the minimum (blending ratio of the first step) and maximum (blendingratio of the last step) particle size and the rate (step duration) at whichthe particle-size gradient evolves from the minimum size to maximum size.The word “gradient” is used to note that a step from 2 to 3 µm, forinstance, does not lead to a discontinuous and sudden step from one particlesize to another.
Figure 2An example of the produced reference aerosol. Decreasing numberconcentrations below 1 and above 5 µm result fromapproaching the lower detection limit (0.5 µm) of the APS andincreasing inertial deposition losses in the sampling lines, respectively(concentration range 30–90 cm−3 of particle number concentration). This had, however, no effect onthe evaluation results as the sensor response was normalized against theconcentration measured by the APS. Along with the lower detection limit ofthe APS, another limiting factor of the study was the smallest producibleparticle size, which was approximately 0.55 µm. The GSD of the sizedistribution remained below 1.2.
Formed particles were neutralized in the dispersion outlet of the VOAG andfurther fed into a flow splitting section where the reference aerosol wassymmetrically directed to both the reference instrument (aerodynamicparticle sizer 3321, TSI Inc., USA) and sensor. The sensors wereencapsulated in 3D-printed airtight enclosures with an external pumpconnected to them in order to ensure appropriate sample flow through thesensor. The sample flow rate was set to be 1 L min−1 – the aerosol flowrate of the APS (sheath flow of the APS taken from the laboratory air).Although there is no clear theoretical basis as to why a different flow ratewould affect the way the sensor discriminates different particle sizes(apart from different particle-size-specific sampling losses), additionaltests were conducted with flow rates of 0.5 and 2 L min−1 to ensure thatthis was indeed the case (see Fig. S2). For the PMS5003 andSPS30 sensors, an exhaust deflector was used to prevent unwanted samplemixing resulting from the fan outlet, which for these sensors, was situatedright next to the sensor inlet. An illustration of the PMS5003 samplingarrangement is shown in Fig. 3. A schematic figure of all the samplingarrangements is shown in Fig. S3.
Figure 3A cross-section view of the sampling arrangement of the PMS5003sensor. The flow deflector is shown in light grey, the inlet pipe in red,and the exhaust connector of the housing in yellow.
All sensor units were in the original condition except for the PPD42 and B5Wsensors which had their air heating resistors removed. The evaluationplatform used in this study did not require independent means of sampleflow. Furthermore, holes were drilled into the plastic body of the PPD42 toensure that the sample aerosol could reach to the optical detection volume.The inlet of the PPD42 was originally designed to be on top of the plasticbody (facing towards the electronic circuit board); therefore, when theelectronic circuit board of the sensor was oriented in parallel with thesample stream, the majority of the particles would have bypassed the sensor.In general, along with the PPD42, the plastic body layouts of the PMS5003and SPS30 are susceptible to inertial deposition losses due to their90∘ elbows in particle stream pathways. However, the more stablesample flow system (i.e., fan instead of convection) might help compensatefor the sub-optimal layouts of these sensors.
2.3 Data processing
The output signal of the evaluated sensor and APS was measured synchronouslyusing a 10 s time resolution and moving average. Any raw measurementpoint which had GSD (calculated from the APS data) exceeding 1.2 wasdisregarded (∼2.1 % of the data), but typically the GSDvalues ranged between 1.04 and 1.08. The sensor bias was set to zero bysampling clean air for 10 min (60 data points) and then subtracting theclean air response from the test aerosol response. The bias correction wasonly relevant for the GP2Y1010AU0F and B5W sensors. In order to preventarbitrary unit comparisons, the sensor response was normalized using Eq. (3):
wherei is theith measurement point, “sensor” is the sensor signal, andAPS is the APS total mass concentration. The maximum sensor ∕ APS ratiorefers to the maximum ratio measured during a single test run.
The normalized 10 s resolution data were divided into 30 logarithmicallyspaced size bins (from 0.45 to 9.73 µm) according to the count mediandiameters (CMDs, aerodynamic) measured by the APS. An average sensor responseas a function of average CMD was then calculated for each size bin. Thedecision to divide the data into 30 bins was based on the clarity of theproduced figure and statistically sufficient number of measurement pointsbelonging to each bin. This process was completed for three different sensorunits, and a combined (average and standard deviation) sensor response wascalculated. Valid detection ranges, which were defined as the upper half ofthe detection efficiency curve, of the sensors were linearly interpolatedfrom the average response functions. A detailed example of how the data wereprocessed and how the valid detection ranges were calculated is shown in theSupplement. The cursory evaluation of the Grimm instrument was conductedusing the same data processing method. The size bins of PMS5003, SPS30,SDS011, and B5W were discretized so that no overlapping signals wereobtained. For example, the outputs of the SDS011 were used as PM2.5 andPM10−2.5 (PM10−2.5 calculated as PM10–PM2.5) instead of PM2.5 and PM10.
The PMS5003, SDS011, and SPS30 sensors have digital outputs whereas theothers are analog-based. Along with the PM mass fractions listed in Table 1,the PMS5003 and SPS30 sensors also output particle number concentrations,but these signals were not used because the response comparison to thereference instrument was carried out using only mass concentration values.This decision was based on the observation that low-cost sensors have beenpredominantly used to measure mass concentration and not numberconcentration.
3.1 Grimm model 1.108
The normalized detection efficiencies of the 15-bin Grimm 1.108 spectrometer are shown inFig. 3. The normalized detection efficiency of 70 %–90 % results fromthe average efficiency from multiple data points and, in this case, does notimply that the Grimm spectrometer would systematically underestimate particle massconcentrations. The same applies to the respective sensor response figures(next section).
The response characteristics of the Grimm spectrometer are in line with its technicalspecifications showing that each size bin only corresponds to its specificdetection range. A flat response curve would indicate that the strength ofthe output signal remains unchanged regardless of the particle size, whichwould show that the size bin is unable to make a distinction betweendifferent particle sizes. Some mismatch between the particle sizing of theAPS and the Grimm spectrometer can be observed as a result of different particle sizingtechniques (time of flight and optical), but this is trivial, consideringthe objective of this study. The purpose of this figure is to highlight howan aerosol measurement device with several particle sizing bins shouldrespond to the evaluation method used in this study.
3.2 Low-cost sensors
Response functions of the evaluated sensors are shown in Fig. 4a–f.
Figure 4Normalized detection efficiency of the 15 particle-size bins as afunction of the count median diameter of the reference aerosol.Consecutively increasing and decreasing response curves indicate that theparticle sizing of the instrument is functioning correctly. For the sake ofclarity, degrees of measurement variation have been excluded from thefigure. Bins 14 and 15, which correspond to 10–15 and 15–20 µm,respectively, are not shown as they did not produce any response (asexpected).
Figure 5Normalized detection efficiency of discretized PM mass fractionsreported by the low-cost sensors as a function of the count median diameterof the reference aerosol. The colored circles represent the calculatedaverage responses of the three sensor units, and the shaded background areasrepresent the respective standard deviations. Standard deviations of theaverage CMDs were negligible due to the reliable and reproducible testmethod. Figure legends correspond to the bin size ranges stated by thecorresponding manufacturer.
Plantower PMS5003
According to Fig. 4a, it is apparent that the PMS5003 does not accuratelydistinguish between PM1, PM2.5, and PM10 size fractions. The first and thesecond bin (supposedly corresponding to 0.3–1.0 and 1.0–2.5 µm) aresimilar, with valid detection ranges of approximately<0.7 and<0.8 µm, respectively (valid detection ranges weredefined as the upper half of the detection range; see the section “Dataprocessing”). The lower cut points of these bins may reach close to0.3 µm, as stated by the manufacturer; however, this could not beconfirmed using the VOAG–GP50 system. As the larger standard deviationsindicate, the third bin is noisier and significantly off of its stateddetection range (2.5–10 µm).
Based on the test, the PMS5003 cannot be used to measure coarse-modeparticles (2.5–10 µm); furthermore, its ability to measure PM2.5depends on the stability of the ambient air size distribution: for example,if the proportions of mass in<0.8 and>0.8 µmfractions change significantly, the PMS5003 is susceptible to inaccuraciesbecause its valid detection range cannot account for changes occurring inparts of the size distribution that it can hardly observe. However, if theambient size distribution is stable, the PMS5003 can be adjusted to measurePM2.5 with reasonable accuracy (Bulot et al., 2019; Feenstra et al., 2019;Magi et al., 2019; Malings et al., 2019). Similarly, the validity of PM10measurements can only be ensured when the proportion of mass in>0.7 or>0.8 µm size fractions is either constant ornegligible with respect to the total PM10 mass. In reality, this is rarelythe case, which poses a high risk of sensor misuse. This observation is inline with the findings from previous studies (Laquai, 2017b; Li et al.,2019; Sayahi et al., 2019b) which show, for instance, that the PMS5003 couldnot detect a substantial dust storm episode while deployed in the field. Themost accurate and reliable results are most likely achieved for the PM1 sizefraction by using either bin 1 or bin 2 signals.
Nova SDS011
The response function of the SDS011 is shown in Fig. 4b. Contrary to thePMS5003, the SDS011 exhibits two clearly different detection ranges: thefirst bin (0.3–2.5 µm) corresponds approximately to<0.8 µm, and the second bin (2.5–10 µm) corresponds approximatelyto 0.7–1.7 µm. Similarly to the PMS5003, the SDS011 is not suitablefor the measurement of coarse-mode particles, and the measurements of PM10can be grossly inaccurate, as also noted by Budde et al. (2018) and Laquai (2017a). However, due to the clearer difference between bin 1 and bin 2detection ranges, the SDS011 has the potential to measure PM2.5 moreaccurately than the PMS5003. For example, by calculating the ratio of bins 1and 2, it is possible to approximate the distribution of mass in the0.3–2.5 µm size range, thus using an additional correction factor toobtain more accurate results. Previous studies have shown that the SDS011can be reasonably accurate in the measurements of PM2.5 (Badura et al.,2018; Liu et al., 2019).
Sensirion SPS30
The response function of the SPS30 is shown in Fig. 4c. The validdetection range of the first bin (0.3–1.0 µm) is approximately<0.9 µm. The second, third, and fourth bins (supposedlycorresponding to 1.0–2.5, 2.5–4.0, and 4.0–10 µm) are nearlyidentical, with valid detection ranges of approximately 0.7–1.3 µm.The identical detection ranges indicate that these bins may have beenfactory calibrated using the same test aerosol. The SPS30 is a relativelynew sensor (introduced to the markets in late 2018), and neither Web ofScience nor Scopus showed any existing studies as of September 2019.However, the South Coast Air Quality Management District (SCAQMD, USA) hasconducted a preliminary field test where three SPS30 units were compared tothree different federal equivalent method (FEM) monitors (SCAQMD, 2019). Theresults of this test showed that the SPS30 sensors had very low cross-unitvariability (∼1 %, 1.3 %, and 2.4 % for PM1, PM2.5, and PM10,respectively), and, more importantly, the coefficient of determinations forthe measurement of PM1, PM2.5, and PM10 decreased fromR2∼0.91 to 0.83 and further down to 0.12, respectively. These observationsstrongly align with the results of this study; furthermore, they illustratehow a sensor with limited operational range may exhibit a near-regulatory-grade performance if the measured size fraction is in alignment with thevalid detection range of the sensor (<0.9 µm and PM1). Onthe other hand, the severity of data misinterpretation is apparent when thesensor measurement is extended to cover particle sizes that it cannotobserve.
Sharp GP2Y1010AU0F
The response function of the GP2Y1010AU0F is shown in Fig. 4d, and itsvalid detection range appears to be approximately<0.8 µm.Like the previously discussed sensors, the GP2Y1010AU0F can be used tomeasure small particles (e.g., PM1) but not coarse-mode particles. Severallaboratory evaluations have been previously conducted for the GP2Y1010AU0F,but none of these have assessed its detection range using monodisperse testaerosols (Li and Biswas, 2017; Manikonda et al., 2016; Sousan et al., 2016).Wang et al. (2015) used atomized polystyrene latex (PSL) particles toevaluate the effect of particle size on the GP2Y1010AU0F response, but noconcluding remarks can be obtained from these results. The study methodutilized only three different sized PSLs; moreover, it was not designed toinvestigate the complete detection range of the GP2Y1010AU0F. However,according to the authors, the results implied that the sensor was moresensitive to 300 nm particles than to 600 and 900 nm particles, which is inslight disagreement with the results of this study whereby the normalizeddetection efficiency curve shows the highest sensitivity peak for 0.6 µm sized particles as well as a decreasing trend for particlessmaller than this. There is no obvious explanation for this discrepancy, butit is worth re-emphasizing the differences in the used evaluationapproaches.
Shinyei PPD42
Response functions of the three PPD42 sensor units are shown in Fig. 4e.Contrary to the other sensors, a combined response function was notcalculated as the three units exhibited significantly different responsecharacteristics. The circles and shaded background areas represent averageresponses and respective standard deviations of the individual sensor units(calculated from the∼300 raw data points). The validdetection range of the first unit is 1.0–2.1 µm, and it is likely tobe best suited for PM2.5 measurements. However, the low detection efficiencyof<1.0 µm sized particles may considerably hinder itsaccuracy. Valid detection ranges of the second and third units are>5.9 and 1.5–4.9 µm, indicating preferable applicabilityto coarse-mode particle measurements. Previous laboratory evaluations havenoted that the PPD42 output is a function of particle size but could notprovide a more detailed analysis of the complete detection range (Austin etal., 2015; Wang et al., 2015). A study of Kuula et al. (2017) reported avalid detection range of approximately 2.5–4.0 µm, which is in thesame range as the third unit of this study.
Due to the apparent inter-unit inconsistency in valid detection ranges, itis evident that the response characteristics of the PPD42 have to bequantified case by case before reliable measurements can be achieved.Accordingly, the inconsistent response characteristics may also contributeto the fact that previous field evaluation studies have achieved varyingresults regarding the performance of PPD42; Bai et al. (2019) and Holstiuset al. (2014) reportedR2 values of 0.75 and 0.55–0.60, respectively, forthe measurement of PM2.5, whereas N. E. Johnson et al. (2018) and K. K. Johnson et al. (2018) reported more modest values of 0.36–0.51 and 0–0.28, respectively (Bai et al., 2019; Holstius et al., 2014; N. E. Johnson et al., 2018, K. K. Johnson et al., 2018). On the other hand, Kuula et al. (2017, 2018) showed thathigher levels of accuracy can be achieved if the measured size fraction istargeted to correspond to the characteristic response function of the PPD42(R2=0.96 andR2=0.87, respectively).
Omron B5W
The response function of the B5W is shown in Fig. 4f. The two size binsexhibit two different detection ranges (0.6–1.0 and>3.2 µm) that are reasonably close to the ones declared bythe manufacturer (0.5–2.5 and>2.5 µm).In fact, out of all sensors, the B5W appears to be the most promising sensorfor the ambient monitoring of PM2.5 and PM10–2.5 size fractions. Incomparison to SDS011 and SPS30, for instance, the usability of the B5W maybe hindered by its temperature-gradient-based sampling method, which is notas reliable as the respective fan-based method. Nonetheless, it is the onlysensor capable of measuring both fine- and coarse-fraction particles. NeitherWeb of Science nor Scopus showed existing studies for the Omron B5W.
According to the results obtained in this study, low-cost optical sensorsexhibit widely varying response characteristics regarding theirsize selectivity (from<0.7 to>5.9 µm, Table 2). However, none of the sensors have precisely the same responsecharacteristics stated by their manufacturers, which provides evidence ofthe fact that particle-size selectivity may play an essential role in theanalysis of the sources of errors in sensors and underlines that scientists,as well as manufacturers, need to acknowledge the limitations related tothis: attempts to artificially extend the operational range of sensorsbeyond their practical capabilities using complex statistical models may beunreasonable and lead to misleading conclusions. Empirical corrections forknown artifacts, such as humidity, can be justifiable; however, sensor dataand advanced modeling techniques should be merged cautiously in order toretain both the validity and representativeness of the data.
Table 2Valid detection ranges of the evaluated sensors. Symbols of“greater than” or “smaller than” refer to cases where the other end ofthe size cut point was outside of the particle-size range producible by theVOAG–GP50 system (0.45–9.73 µm). Units are in micrometers.
* Valid detection ranges of the individual sensors, not bins.
A cursory comparison to a mid-cost aerosol size spectrometer (Grimm 1.108)shows that low-cost sensor development is still considerably behind its moreexpensive alternative: while the Grimm 1.108 spectrometer could sufficiently characterizeparticle sizes with up to 15 different size bins, the low-cost sensors couldonly achieve independent responses for one or two bins, which is asignificant weakness, considering that the ability to measure particle sizecorrectly is at the foundation of accurate mass measurement (massαdp3). The development of low-cost sensors should focus on increasing thenumber of size bins, and more importantly, making sure that each size bin iscalibrated correctly. Improperly configured bin sizing poses a significantrisk of data misinterpretation and will inevitably lead to inaccuratemeasurements. A low number of size bins limits the valid operational rangeof sensors; however, it is unclear how the number of advanced measurementfeatures and low unit cost should be reconciled.
The VOAG–GP50 aerosol generation system described in this study introduced anovel approach to the quick and efficient evaluation of aerosol measurementdevices. The use of a GP50 gradient pump eliminates much of the manual laborthat previously was an inseparable part of the VOAG operation, thus making thegeneration of reference aerosols more consistent and reliable. Its automateddispensing programs allow for highly repeatable testing; furthermore, thefour different eluent channels enable the operator to pick and choose thedesired particle size to be produced freely. Along with saving manual laborand time, this is also a cost-saving feature as traditionally usedpolystyrene latex (PSL) particles are not needed. Considering these matters,the VOAG–GP50 system can potentially be scaled to an industrial-leveloperation, which is an intriguing feature when considering the massdeployment of sensors and their respective quality assurance and control.
The data can be openly accessed upon request.
The supplement related to this article is available online at: https://doi.org/10.5194/amt-13-2413-2020-supplement.
JK and TM designed the experimental setup, and JK carried out the tests. KThad an important role in refurbishing the gradient elution pump. SM and OGprovided some of the sensors. JK was responsible for the data analysis,although all co-authors provided valuable feedback, particularly TM. JKwrote the manuscript with the help of all co-authors.
The authors declare that they have no conflict of interest.
This research has been supported by the European Regional Development Fund (Urban innovative actions initiative project HOPE; UIA03-240 grant) and Horizon 2020 (iSCAPE grant no. (689954)).
This paper was edited by Murray Hamilton and reviewed by two anonymous referees.
Austin, E., Novosselov, I., Seto, E., and Yost, M. G.: Laboratory evaluationof the Shinyei PPD42NS low-cost particulate matter sensor, PLoS One, 10,1–17,https://doi.org/10.1371/journal.pone.0137789, 2015.
Badura, M., Batog, P., Drzeniecka-Osiadacz, A., and Modzel, P.: Evaluation oflow-cost sensors for ambient PM2.5 monitoring, J. Sensors, 2018, 5096540,https://doi.org/10.1155/2018/5096540, 2018.
Bai, L., Huang, L., Wang, Z., Ying, Q., Zheng, J., Shi, X., and Hu, J.:Long-term field Evaluation of Low-cost Particulate Matter Sensors inNanjing, Aerosol Air Qual. Res., 20, 1–12,https://doi.org/10.4209/aaqr.2018.11.0424, 2019.
Berglund, R. N. and Liu, B. Y. H.: Generation of Monodisperse AerosolStandards, Environ. Sci. Technol., 7, 147–153,https://doi.org/10.1021/es60074a001,1973.
Budde, M., D. Schwarz, A., Müller, T., Laquai, B., Streibl, N.,Schindler, G., Köpke, M., Riedel, T., Dittler, A., and Beigl, M.:Potential and Limitations of the Low-Cost SDS011 Particle Sensor forMonitoring Urban Air Quality, 3rd International Conference onAtmospheric Dust (DUST2018), ProScience, 5, 6–12,https://doi.org/10.14644/dust.2018.002, 2018.
Bulot, F. M. J. J., Johnston, S. J., Basford, P. J., Easton, N. H. C. C.,Apetroaie-Cristea, M., Foster, G. L., Morris, A. K. R. R., Cox, S. J., andLoxham, M.: Long-term field comparison of multiple low-cost particulatematter sensors in an outdoor urban environment, Sci. Rep., 9, 7497,https://doi.org/10.1038/s41598-019-43716-3, 2019.
Burkart, J., Steiner, G., Reischl, G., Moshammer, H., Neuberger, M., andHitzenberger, R.: Characterizing the performance of two optical particlecounters (Grimm OPC1.108 and OPC1.109) under urban aerosol conditions, J.Aerosol Sci., 41, 953–962,https://doi.org/10.1016/j.jaerosci.2010.07.007, 2010.
Caubel, J. J., Cados, T. E., Preble, C. V., and Kirchstetter, T. W.: ADistributed Network of 100 Black Carbon Sensors for 100 Days of Air QualityMonitoring in West Oakland, California, Environ. Sci. Technol., 53,7564–7573,https://doi.org/10.1021/acs.est.9b00282, 2019.
Feenstra, B., Papapostolou, V., Hasheminassab, S., Zhang, H., Boghossian, B.Der, Cocker, D., and Polidori, A.: Performance evaluation of twelve low-costPM2.5 sensors at an ambient air monitoring site, Atmos. Environ.,216, 116946,https://doi.org/10.1016/j.atmosenv.2019.116946, 2019.
Feinberg, S. N., Williams, R., Hagler, G., Low, J., Smith, L., Brown, R.,Garver, D., Davis, M., Morton, M., Schaefer, J., and Campbell, J.: Examiningspatiotemporal variability of urban particulate matter and application ofhigh-time resolution data from a network of low-cost air pollution sensors,Atmos. Environ., 213, 579–584,https://doi.org/10.1016/j.atmosenv.2019.06.026, 2019.
Gao, M., Cao, J., and Seto, E.: A distributed network of low-cost continuousreading sensors to measure spatiotemporal variations of PM2.5 in Xi'an,China, Environ. Pollut., 199, 56–65,https://doi.org/10.1016/j.envpol.2015.01.013,2015.
Grimm 1.108: Aerosol spectrometer user manual, availableat:https://wmo-gaw-wcc-aerosol-physics.org/files/opc-grimm-model--1.108-and-1.109.pdf (last access: 11 February 2020), GRIMM Aerosol Technik GmbH, Germany, 2020.
Hapidin, D. A., Saputra, C., Maulana, D. S., Munir, M. M., and Khairurrijal,K.: Aerosol chamber characterization for commercial particulate matter (PM)sensor evaluation, Aerosol Air Qual. Res., 19, 181–194,https://doi.org/10.4209/aaqr.2017.12.0611, 2019.
Holstius, D. M., Pillarisetti, A., Smith, K. R., and Seto, E.: Field calibrations of a low-cost aerosol sensor at a regulatory monitoring site in California, Atmos. Meas. Tech., 7, 1121–1131,https://doi.org/10.5194/amt-7-1121-2014, 2014.
Jayaratne, R., Liu, X., Thai, P., Dunbabin, M., and Morawska, L.: The influence of humidity on the performance of a low-cost air particle mass sensor and the effect of atmospheric fog, Atmos. Meas. Tech., 11, 4883–4890,https://doi.org/10.5194/amt-11-4883-2018, 2018.
Jiao, W., Hagler, G., Williams, R., Sharpe, R., Brown, R., Garver, D., Judge, R., Caudill, M., Rickard, J., Davis, M., Weinstock, L., Zimmer-Dauphinee, S., and Buckley, K.: Community Air Sensor Network (CAIRSENSE) project: evaluation of low-cost sensor performance in a suburban environment in the southeastern United States, Atmos. Meas. Tech., 9, 5281–5292,https://doi.org/10.5194/amt-9-5281-2016, 2016.
Johnson, K. K., Bergin, M. H., Russell, A. G., and Hagler, G. S. W.: Fieldtest of several low-cost particulate matter sensors in high and lowconcentration urban environments, Aerosol Air Qual. Res., 18, 565–578,https://doi.org/10.4209/aaqr.2017.10.0418, 2018.
Johnson, N. E., Bonczak, B., and Kontokosta, C. E.: Using a gradient boostingmodel to improve the performance of low-cost aerosol monitors in a dense,heterogeneous urban environment, Atmos. Environ., 184, 9–16,https://doi.org/10.1016/j.atmosenv.2018.04.019, 2018.
Karagulian, F., Barbiere, M., Kotsev, A., Spinelle, L., Gerboles, M.,Lagler, F., Redon, N., Crunaire, S., and Borowiak, A.: Review of thePerformance of Low-Cost Sensors for Air Quality Monitoring, Atmosphere, 10, 506,https://doi.org/10.3390/atmos10090506, 2019.
Kumar, P., Morawska, L., Martani, C., Biskos, G., Neophytou, M., DiSabatino, S., Bell, M., Norford, L., and Britter, R.: The rise of low-costsensing for managing air pollution in cities, Environ. Int., 75, 199–205,https://doi.org/10.1016/j.envint.2014.11.019, 2015.
Kuula, J., Mäkelä, T., Hillamo, R., and Timonen, H.: Responsecharacterization of an inexpensive aerosol sensor, Sensors, 17, 2915,https://doi.org/10.3390/s17122915, 2017.
Kuula, J., Kuuluvainen, H., Rönkkö, T., Niemi, J. V., Saukko, E.,Portin, H., Aurela, M., Saarikoski, S., Rostedt, A., Hillamo, R., andTimonen, H.: Applicability of Optical and Diffusion Charging-BasedParticulate Matter Sensors to Urban Air Quality Measurements, Aerosol AirQual. Res., 122, 1–16,https://doi.org/10.4209/aaqr.2018.04.0143, 2018.
Laquai, B.: Impact of Particle Mass Distribution on the Measurement Accuracyof Low-Cost PM-Sensors, October, available at:https://www.researchgate.net/publication/320537200_Impact_of_Particle_Mass_Distribution_on_the_Measurement_Accuracy_of_Low-Cost_PM-Sensors (last access: 1 April 2020), 2017a.
Laquai, B.: Particle Distribution Dependent Inaccuracy of the PlantowerPMS5003 low-cost PM-sensor Particle Distribution Dependent Inaccuracy ofthe Plantower PMS5003 low-cost PM-sensor, October, availableat:https://www.researchgate.net/publication/320555036_Particle Distribution_Dependent_Inaccuracy_of_the_Plantower_PMS5003_low-cost_PM-sensor (last access: 10 September 2019), 2017b.
Levy Zamora, M., Xiong, F., Gentner, D., Kerkez, B., Kohrman-Glaser, J., andKoehler, K.: Field and Laboratory Evaluations of the Low-Cost PlantowerParticulate Matter Sensor, Environ. Sci. Technol., 53, 838–849,https://doi.org/10.1021/acs.est.8b05174, 2019.
Li, J. and Biswas, P.: Optical characterization studies of a low-costparticle sensor, Aerosol Air Qual. Res., 17, 1691–1704,https://doi.org/10.4209/aaqr.2017.02.0085, 2017.
Li, J., Mattewal, S. K., Patel, S., and Biswas, P.: Evaluation of NineLow-cost-sensor-based Particulate Matter Monitors, Aerosol Air Qual. Res.,20, 1–17,https://doi.org/10.4209/aaqr.2018.12.0485, 2019.
Liu, H. Y., Schneider, P., Haugen, R., and Vogt, M.: Performance assessmentof a low-cost PM2.5 sensor for a near four-month period in Oslo, Norway,Atmosphere, 10, 41,https://doi.org/10.3390/atmos10020041, 2019.
Magi, B. I., Cupini, C., Francis, J., Green, M., and Hauser, C.: Evaluationof PM2.5 measured in an urban setting using a low-cost optical particlecounter and a Federal Equivalent Method Beta Attenuation Monitor, AerosolSci. Technol., 0, 1–13,https://doi.org/10.1080/02786826.2019.1619915, 2019.
Malings, C., Tanzer, R., Hauryliuk, A., Saha, P. K., Robinson, A. L.,Presto, A. A., and Subramanian, R.: Fine particle mass monitoring withlow-cost sensors: Corrections and long-term performance evaluation, AerosolSci. Technol., 0, 1–15,https://doi.org/10.1080/02786826.2019.1623863, 2019.
Manikonda, A., Zikova, N., Hopke, P. K., and Ferro, A. R.: Laboratoryassessment of low-cost PM monitors, J. Aerosol Sci., 102, 29–40, 2016.
Morawska, L., Thai, P. K., Liu, X., Asumadu-Sakyi, A., Ayoko, G., Bartonova,A., Bedini, A., Chai, F., Christensen, B., Dunbabin, M., Gao, J., Hagler,G., Jayaratne, R., Kumar, P., Lau, A. K. H., Louie, P. K. K., Mazaheri, M.,Ning, Z., Motta, N., Mullins, B., Rahman, M. M., Ristovski, Z., Shafiei, M.,Tjondronegoro, D., Westerdahl, D., and Williams, R.: Applications of low-costsensing technologies for air quality monitoring and exposure assessment: Howfar have they gone?, Environ. Int., 116, 286–299,https://doi.org/10.1016/j.envint.2018.04.018, 2018.
Papapostolou, V., Zhang, H., Feenstra, B. J., and Polidori, A.: Developmentof an environmental chamber for evaluating the performance of low-cost airquality sensors under controlled conditions, Atmos. Environ., 171, 82–90,https://doi.org/10.1016/j.atmosenv.2017.10.003, 2017.
Peters, T. M., Ott, D., and O'Shaughnessy, P. T.: Comparison of the Grimm1.108 and 1.109 portable aerosol spectrometer to the TSI 3321 aerodynamicparticle sizer for dry particles, Ann. Occup. Hyg., 50, 843–850,https://doi.org/10.1093/annhyg/mel067, 2006.
Popoola, O. A. M., Carruthers, D., Lad, C., Bright, V. B., Mead, M. I.,Stettler, M. E. J., Saffell, J. R., and Jones, R. L.: Use of networks of lowcost air quality sensors to quantify air quality in urban settings, Atmos.Environ., 194, 58–70,https://doi.org/10.1016/j.atmosenv.2018.09.030, 2018.
Rai, A. C., Kumar, P., Pilla, F., Skouloudis, A. N., Di Sabatino, S., Ratti,C., Yasar, A., and Rickerby, D.: End-user perspective of low-cost sensors foroutdoor air pollution monitoring, Sci. Total Environ., 607–608, 691–705,https://doi.org/10.1016/j.scitotenv.2017.06.266, 2017.
Sayahi, T., Kaufman, D., Becnel, T., Kaur, K., Butterfield, A., Collingwoodd,S., Zhang, Y., Gaillardon, P.-E., and Kelly, K. E.: Development of acalibration chamber to evaluate the performance of low-cost particulatematter sensors, Environ. Pollut., 255, 113131,https://doi.org/10.1016/j.envpol.2019.113131, 2019a.
Sayahi, T., Butterfield, A., and Kelly, K. E.: Long-term field evaluation ofthe Plantower PMS low-cost particulate matter sensors, Environ. Pollut.,245, 932–940,https://doi.org/10.1016/j.envpol.2018.11.065, 2019b.
Schneider, P., Bartonova, A., Castell, N., Dauge, F. R., Gerboles, M.,Hagler, G. S. W., Hüglin, C., Jones, R. L., Khan, S., Lewis, A. C.,Mijling, B., Müller, M., Penza, M., Spinelle, L., Stacey, B., Vogt, M.,Wesseling, J., and Williams, R. W.: Toward a Unified Terminology ofProcessing Levels for Low-Cost Air-Quality Sensors, Environ. Sci. Technol.,53, 8485–8487,https://doi.org/10.1021/acs.est.9b03950, 2019.
Snyder, E. G., Watkins, T. H., Solomon, P. A., Thoma, E. D., Williams, R.W., Hagler, G., Shelow, D., Hindin, D. A., Kilaru, V. J., and Preuss, P. W.:The changing paradigm of air pollution monitoring, Environ. Sci. Technol.,47, 11369–11377,https://doi.org/10.1021/es4022602, 2013.
Sousan, S., Koehler, K., Thomas, G., Park, J. H., Hillman, M., Halterman, A., and Peters, T. M.: Inter-comparison of low-cost sensors for measuring themass concentration of occupational aerosols, Aerosol Sci. Technol., 50,462–473,https://doi.org/10.1080/02786826.2016.1162901, 2016.
South Coast Air Quality Management District (SCAQMD): Field EvaluationSensirion SPS30 Evaluation Kit Background, available at:http://www.aqmd.gov/docs/default-source/aq-spec/field-evaluations/sensirion-sps30-evaluation-kit---field-evaluation.pdf, last access: 5 November 2019.
Volckens, J. and Peters, T. M.: Counting and particle transmission efficiencyof the aerodynamic particle sizer, J. Aerosol Sci., 36, 1400–1408,https://doi.org/10.1016/j.jaerosci.2005.03.009, 2005.
Wang, Y., Li, J., Jing, H., Zhang, Q., Jiang, J., and Biswas, P.: LaboratoryEvaluation and Calibration of Three Low-Cost Particle Sensors forParticulate Matter Measurement, Aerosol Sci. Technol., 49, 1063–1077,https://doi.org/10.1080/02786826.2015.1100710, 2015.
Yuval, Magen Molho, H., Zivan, O., Broday, D. M., and Raz, R.: Application ofa sensor network of low cost optical particle counters for assessing theimpact of quarry emissions on its vicinity, Atmos. Environ., 211, 29–37,https://doi.org/10.1016/j.atmosenv.2019.04.054, 2019.
Zheng, T., Bergin, M. H., Johnson, K. K., Tripathi, S. N., Shirodkar, S., Landis, M. S., Sutaria, R., and Carlson, D. E.: Field evaluation of low-cost particulate matter sensors in high- and low-concentration environments, Atmos. Meas. Tech., 11, 4823–4846,https://doi.org/10.5194/amt-11-4823-2018, 2018.