Performance assessment of low-cost environmental monitors and single sensors under variable indoor air quality and thermal conditions

13 Recent technological advancements have enabled the development and deployment of low -cost 14 consumer grade monitor s for ubiquitous and time - resolved indoor air quality monitoring. With their 15 reliable performance, this technology could be instrumental in enhancing automatic controls and human 16 decision making. We conducted a comprehensive performance evaluation of eight consumer grade multi- 17 parameter monitors and eight single - parameter sensors in detecting particulate matter, carbon dioxide, 18 total volatile organic compounds, dry - bulb air temperature, and relative humidity. In the controlled 19 chamber, we generated eight air pollution sources, each at two thermodynamic conditions — cool and 20 dry (20±1°C, 30±5%) , and warm and humid (26±1°C, 70±5%). The majority of tested devices under - 21 reported reference particle measurements by up to 50%, provided acceptable responses for carb on 22 dioxide within 15% and diverging results with poor quantitative agreement for total volatile organic 23 compounds. Despite the reported disparities in quantitative agreements, most of the low - cost devices 24 could detect source events and were strongly correlated with the reference data, suggesting that these 25 units could be suitable for measurement - based indoor air quality management. Most of the tested devices 26 have also proven to competently measure air temperature (within +/ - 0.6°C) and relative humidity (wit hin 27 +/ - 5% RH) and maintained a stable measurement accuracy over the two thermodynamic conditions. show fine part icles. the Clarity and Kaiterra were the closest to the reference concentration. They under - reported the reference on average by 52% and 53% respectively, followed by Foobot with 57%, and AirVisual and Awair with 67% and 73%, respectively. The uHoo showed by far the worst results with an MRE of 90% and PCC of -0.30. For mosquito coil burning that produced the PM 2.5 concentrations up to 492 μg/m³, we observed that all monitors exhibited very strong correlation to the miniWRAS data (PCC > 0.9) and different quantitative response: Kaiterra was the closest to the reference with an MRE of 11%, followed by Clarity 12%), AirVisual 31%), Awair (MRE 37%) and Foobot (MRE = 44%). Even in uHoo and under - 90%, cannot used for reliable measurements of fine particulate In the case of popcorn cooking, the consumer grade monitors were strongly correlated ( PCC > 0.97) with MRE within 70% for all monitors. Foobot performed the best with the MRE of 19% compared to the reference. Interestingly, in the case of popcorn cooking, uHoo performed much better than during other activities and showed similar results as the rest of consumer grade monitors, suggesting a lower sensitivity of the PM to sub - micron particles. This paper presents a comprehensive performance evaluation of low -cost consumer grade monitors and 608 single- parameter sensors in detecting five indoor environmental parameters – particulate matter, carbon 609 dioxide, total volatile organic compounds, dry -bulb a ir temperature and relative humidity. Eight 610 experiments were chosen to simulate indoor air pollutant sources that were carried out at two distinct 611 climatic conditions – cool & dry, and warm & humid.


Introduction
38 Increasingly strict energy efficiency requirements for buildings have led to tightening of building 39 envelopes to reduce uncontrolled outdoor air infiltration. As a result, unless adequate ventilation is 40 provided, air pollutants emitted inside buildings could be present at higher concentrations due to less 41 dilution [1]. This has exacerbated concerns about health effects from indoor exposures to air pollutants. 42 Some indoor air pollutants can be recognized by their immediate impacts on our body, such as throat 43 irritation or watery eyes [2]. Others, which often bypass the human olfactory radar, are not necessarily 44 benign. According to the US Environmental Protection Agency, some health impacts like respiratory 45 diseases, heart disease, and cancer can show up years after exposure [3]. This highlights the importance 46 of proper indoor air quality (IAQ) management including monitoring of air pollutants. 47 According to ASHRAE Standard 62.1-2019 [4], acceptable indoor air quality has "air in which there are no 48 known contaminants at harmful concentrations, as determined by cognizant authorities, and with which 49 a substantial majority (80% or more) of the people exposed do not express dissatisfaction". Multiple field 50 studies, however, showed that buildings often do not meet even the minimum standard requirements 51 [5]. Even when average concentrations in a building meet requirements, air pollutants are often non-52 homogeneously distributed which may result in elevated exposures at some locations [6][7][8]. 53 The European Respiratory Society (ERS) has identified particulate matter (PM 2.5 and PM 10 ), volatile organic 54 compounds (VOCs) and carbon dioxide (CO 2 ) as key air pollutants [9]. Most of these indoor pollutants 55 derive from indoor or outdoor anthropogenic sources [9] and their control can be achieved either by 56 limiting or eliminating the emitting source(s) or through adequate ventilation and filtration. To assure 57 adequate control, IAQ monitoring is an important aspect that can trigger the right chain of actions, via 58 real-time feedback to encourage human actions or through direct activation of automated control 59 devices. While there is no universal air pollutant metric established that benchmarks indoor air quality 60 [10], indoor CO 2 concentrations have been used as an indicator of human bio-effluents in occupied 61 buildings and as a control metric for rooms equipped with demand-controlled ventilation [11]. However, 62 in buildings with low or no occupancy, or where other air pollutant sources which emit VOCs or particles 63 are problematic, ventilation control based on CO 2 concentration only may not be sufficient [12]. This 64 highlights the importance of monitoring multiple relevant air pollutants. 65 66 Historically, indoor air quality monitoring has been performed by professionals with certified reference 67 instruments [13]. The high capital cost and large size makes such devices unsuitable for ubiquitous and 68 continuous IAQ monitoring in buildings [14]. Recently, technological advances in metal oxide for the real-time monitoring of air temperature and relative humidity, along with several IAQ parameters, 73 commonly including PM 2.5 , PM 10 , CO 2 and total VOCs (TVOCs) [18]. Some of the consumer grade monitors 74 include sensors for other gases, such as carbon monoxide, nitrogen dioxide, ozone, or other parameters 75 such as air pressure and sound level. The commonly available consumer grade monitors typically store 76 data on IoT servers, and the measurements can be visualized through the web or mobile applications. The 77 increased availability on the market of such consumer grade monitors and single low-cost sensors (devices 78 that measure individual IAQ parameters and send data to a logger) has drawn the attention of many 79 researchers. 80 81 To date, several studies examined the performance of low-cost sensors and monitors in detecting the PM 82 indoors [19][20][21][22][23][24][25] and outdoors [26][27][28][29]. Singer et al. [20] tested the performance of low-cost air quality 83 monitors in detecting fine particles from residential sources. They found a quantitative agreement within 84 a factor of two for most of the sources but very little response for particles with an optical diameter below 85 0.3 µm. These results were recently confirmed by Wang et al. [19]. Other studies found that the 86 performance of the integrated PM sensors into consumer grade monitors can be influenced by the air 87 temperature and relative humidity [30,31]. The accuracy of CO 2 measurement with low-cost NDIR 88 sensors, frequently deployed within consumer grade monitors, was also found to be dependent on the 89 air temperature and relative humidity [32]. Beyond direct measurements, some devices estimate CO 2 90 concentration from TVOC measurements, resulting in substantial errors [33]. To bridge the knowledge gap, we evaluated the performance of various IAQ monitors and sensors under 107 a controlled range of indoor air pollution and thermal conditions. In an environmental chamber, we tested 108 the response of eight consumer grade multi-parameter monitors in measuring PM, CO 2, and TVOC emitted 109 from eight common indoor sources. We also tested their response to the two main thermo-hygrometric 110 parameters, namely air temperature and relative humidity.  Switzerland. The chamber is equipped with a dedicated heating, ventilation, and air conditioning (HVAC) 121 system that enables control of air temperature, relative humidity, ventilation rate, and airflow 122 distribution. The conditioned air was supplied through a 2-stage media filter to eliminate nearly all 123 exogenous airborne particle contributions from outdoors to the chamber. The air was supplied through 124 six floor-mounted diffusers and exhausted via six diffusers on the ceiling. 125 The HVAC was turned off two minutes before the start of a pollutant generation and monitor testing, so 126 that the air exchange was provided solely by infiltration ( For reference monitoring of time-and size-resolved particle levels we deployed a Grimm Model 1371, 165 Aerosol Technik (miniWRAS). The miniWRAS combines an optical light scattering sensor unit that counts 166 particles in 31 bins from 0.25 to 35 μm and an electrical mobility analyzer that resolves particles in 10 bins 167 from 10 to 193 nm. Measurements were taken at 1-minute intervals. The calibration of the miniWRAS 168 was verified using monodispersed 1.005 μm and 2.005 μm diameter polystyrene latex particles (PSL, 169 Thermal Scientific, 405 US), with error below 10%. The use of a particle counter to determine particle 170 mass concentrations requires the adoption of a particle density. It is known that depending on the 171 pollutant source, particle density could vary significantly [19]. However, during this study, no mass-based 172 measurement was performed and the particle mass concentration from miniWRAS was determined 173 assuming spherical particles having a density of 1.68 g/cm³ for all experiments. We also performed a 174 complementary set of analyses with adjusted source dependent densities to quantify the degree of bias 175 introduced owing to the constant density assumption. 176 6 The LI-COR 850 Biosciences gas analyzer (LI-COR) was used for the reference measurements of CO 2 and 177 relative humidity. The LI-COR has a CO 2 measurement range of 0-20'000 ppm and the manufacturer-178 specified accuracy within 1.5% of reading. The LI-COR directly measures water vapor in the air (accuracy 179 of 1.5%), which is used together with atmospheric pressure and dry bulb air temperature values to 180 compute the relative humidity. The calculated error of the instrument, including the atmospheric data is 181 ± 2%. The reference measurements for CO 2 concentrations and relative humidity were taken at 10 second 182 intervals and averaged over 1 minute. The instrument response was confirmed through exposure to 183 calibration gases at 0 and 1'500 ppm. 184 For TVOC measurements, no true reference was considered owing to the current technological limitations 185 for measuring time-resolved TVOCs concentrations were recorded with 10-second resolution for GW and 1-minute for the AerPID. The GW 195 data were averaged at 1-minute intervals. Apart from TVOCs, GW IQ-610 has sensors that detect CO, CO 2 , 196 relative humidity and dry bulb temperature. There sensors were not calibrated nor used in subsequent 197 analyses. reported in Table S1.
The retail price was recorded in March of 2020.  Measurements were additionally performed with eight single low-cost sensors that can capture levels of 225 particulate matter, carbon dioxide, air temperature, and relative humidity. The sensors were chosen 226 according to in-house availability, personal interest in specific technologies, and their widespread use in 227 consumer grade monitors. Also, sensors (except SHT31) were specifically chosen not to overlap with the 228 sensors already tested within the monitors. Table 3 summarizes the information about sensor type, 229 measurement range, particle size range, accuracy, and price. 230 231 Littelfuse 11492 (Lit92) Air temperature The performance of the tested devices in terms of quantitative agreement with the reference data was 284 additionally assessed through the comparison of the mean relative error (MRE) across devices. 285 286

287
All results for the IAQ parameters PM, CO 2, and TVOC are first reported for the warm and humid 288 conditions. Insights about the seasonal performance comparison are presented in the subsection 3.4. 289

Particulate matter 290
The 16 experimental runs summarized in Table 1 generated a broad range of pollutants, including 291 particulate matter of different sizes, as outlined in Figure 2. Combustion of candles and mosquito coils 292 generated a substantial number of fine and ultrafine particles with the diameter mode at 0.2 μm. Popcorn 293 cooking created the most widespread particle size distribution with the most considerable fraction of 294 emitted mass centered at 3.0 μm. Vacuuming produced the highest particle mass concentration with the 295 diameter mode at 13.1 μm. Room deodorant, candles, mosquito coil, and popcorn all contributed to the 296 generation of fine particles with widely varied fine particle emissions. Room deodorant generated the 297 lowest particle concentration with a peak PM 2.5 of 4 μg/m³, while mosquito coil burning resulted in the 298 highest PM 2.5 concentration of 515 μg/m³. As shown in Figure 2, most of the consumer grade monitors 299 responded to particle concentration changes with a strong correlation to the miniWRAS data. For each 300 monitor, quantitative agreement varied across the sources. 306 Figure 3 shows the dynamic variation of PM 2.5 between the miniWRAS and the consumer grade monitors 307 and single sensors for candle burning, mosquito coil burning, and popcorn cooking. The peak particle 308 concentration for candle burning encompasses the effect of lighting the match at the beginning and 309 extinguishing the candles at the end of the experiment, as each instantly elevated the particle 310 concentration.

12
Most of the tested consumer grade monitors do not report particle concentrations in size range larger 341 than 2.5 μm. Clarity is the only monitor that has this ability. As shown in Figure 4,     The tested single sensors, namely SCD40 and K30, were very strongly correlated (PCC = 0.99 for both) with 371 the LI-COR data. The SCD40 was the most accurate as its reported peak concentration deviated from the 372 reference just by 3%, and the MRE was 6%, while the K30 under-reported the CO 2 concentration by 12% 373 on average and showed a short time delay. The CO 2 increase was also observed for candle burning and 374 essential oil heating activities, although not significant enough to merit further analysis when compared 375 to CO 2 injection. 376  reference was calculated and compared. As presented in Table 4, Awair proved to be the most stable 405 monitor overall while having sensors in all of the categories. An equally high performance for PM and CO 2 406 was shown by Kaiterra which did not measure TVOCs. AirVisual and uHoo showed 20% higher MRE in cool 407 and dry conditions for PM measurements while the opposite can be said for Foobot. When measuring 408 CO 2 , the most deviation was shown by Foobot with more than 80% of a difference and with significant 409 error in both conditions. uHoo had an offset at 17% with better performance in cool and dry conditions. 410 Overall, it can be observed that similar magnitudes of MRE compared to reference were observed for the 411 tested monitors during the different seasons. This finding is supported by a one-year long evaluation of 3 412 consumer grade monitors which determined minimal measurement dependence on temperature and 413 relative humidity and minimal drift [47]. Even so, it needs to be stated that the majority of monitors were 414 slightly closer to reference in cool and dry conditions for PM 2.5 and CO 2 and in warm and humid conditions 415 in case of TVOC. 416 417 The relative humidity variations inside the chamber during both simulated seasons are shown in Figure 7.

422
The LI-COR reported relative humidity values from 24% to 30% RH (mean = 26% RH) for cool and dry and 423 51% to 73% RH (mean = 64% RH) for warm and humid conditions. The tested devices followed the 424 reference values well and responded to changes in the relative humidity.   [20] which reported the limit 479 of particle detection at around 0.25 μm. According to specifications, the majority of consumer grade 480 monitors are supposed to register particles with optical diameter between 0.3 µm and 2.5 µm. Depending 481 on the pollutant source and associated particle size distribution, a closer agreement with the reference 482 was found in case of optical particle diameter ranging from 1 μm to 2.5 μm where the majority of tested 483 devices reported around 50% of reference concentration at the worst. The agreement diminished when 484 the sources were dominated by submicron particles (< 1 μm) and during activities that generate coarse 485 particles (e.g. vacuuming). Studies [20, 21] also report that optical monitors (consumer, professional and 486 research grade) may be under-reporting the mass concentration of larger particles generated from 487 vacuuming if they have higher density. However, owing to the polydisperse nature of particle sources 488 indoors, the response of most of the sensors was time correlated. Strong correlation with reference data 489 was found also by Li et al.
[24] for the tested consumer grade monitors. This means that the devices are 490 dynamically keeping track of concentration changes and can be used to detect an event despite poor 491 quantitative agreement. Analyzed data suggests no consistent bias for PM 2.5 sensors. End-users should be 492 made aware that the PM data from the current low-cost sensors needs to be understood as an indication 493 of a state change or a rough estimation rather than actual concentration in indoor environments. 494 According to their specifications, Clarity and single sensors SPS30, OPC-R1 and OPC-N3 have the ability to 495 detect PM 10 . Our results showed that Clarity's sensor Plantower PMS 6003 and SPS30 are in the sub $50 496 category and that they can barely detect any PM 10 concentration changes. Kaittera uses the Plantower 497 3003 which has the ability to detect PM 10 but the manufacturer chooses not to relate that data to the 498 end-user. OPC-R1 with the double, and the OPC-N3 six-time higher price both correlate well to the 499 reference, with OPC-R1 still in the price range to be considered for a low-cost consumer grade monitor 500 integration. At their current state, Clarity and SPS30 cannot be used for determining PM 10 concentrations. 501 Improvements in the algorithms used to determine PM mass concentrations from optical particle counting 502 are needed to improve measurement accuracy for coarse-mode particles. 503 An additional analysis was carried out to evaluate the effect of adopting different source dependent 504 particle densities for the reference miniWRAS. To calculate the mass concentration of particles, the 505 default densities of 1.68 g/cm³ for miniWRAS was adjusted with experimental values from literature for 506 each pollutant source. For the majority of tested devices, the PM 2.5 concentration was closer to the 507 reference data with adjusted density in case of candle burning, popcorn cooking and mosquito coil 508 burning, regardless from the season, as reported in Table S3. 509 The consumer grade monitors and sensors evaluated in the experiments include non-dispersive infrared 510 (NDIR) technology to detect CO 2 concentrations in the indoor environment except the Foobot. Despite 511 the same price range, the Foobot has no dedicated sensor and estimates the CO 2 concentration from the 512 TVOC data with the use of an algorithm. As a result, all the sensors except Foobot were very strongly 513 correlated with the reference (PCC > 0.8). Foobot's very poor performance is a direct consequence of 514 manufacturers design choice and suggests that currently there is no alternative to a dedicated CO 2 sensor. 515 The uHoo had a consistent offset from the reference which is indicative of a systematic instrument error. 516 The manufacturer could possibly correct this error with the use of better calibration procedures and 517 algorithms in future software updates. Despite the very strong correlation, Netatmo_i had a poor dynamic 518 response as it took ~15 minutes in both conditions to approach the reference, thereby not capturing the 519 peak CO 2 event, which is not acceptable. Unlike other tested sensors, the CO 2 sensor inside Netatmo_i 520 was introduced to the market over seven years ago. The results from newer devices suggest that the low-521 cost sensing technology has matured and is becoming more accurate and reliable. It is important to note 522 that all of the tested CO 2 sensors, except Awair, include automatic baseline correction (ABC). At initial 523 device startup, ABC can take from a week up to two weeks which makes the whole procedure 524 cumbersome. Further, devices go into ABC mode once a week. This could result in erroneous readings in 525 environments that do not periodically reach global background outdoor CO 2 levels, particularly in 526 buildings that are occupied continuously or have low enough ventilation and short periods without 527 occupancy. The data output on CO 2 concentrations from the majority of the tested modern low-cost 528 sensors can be used with confidence in decision making if the ABC requirements are met. 529 In all tested monitors and single sensors, relative humidity and air temperature were measured by a single 538 sensor. This sensor integrates two components, a capacitive relative humidity sensor and the band gap 539 air temperature sensor. Interestingly, the majority of the units use the Sensirion SHT sensors from series 540 2 (Foobot, Netatmo_i, Netatmo_o) and 3 (AirVisual, Kaiterra, Awair), which suggests a trend on the 541 market. However, the best performance was shown by uHoo and its Bosch BME 280 sensor practically 542 being true to the reference. There is no logical clustering of measurements with regards to the SHT sensor 543 series. This indicates that consumer grade monitor manufacturers use different procedures for sensor 544 calibration and use custom signal conversion algorithms. Additional reasoning for the result disparities 545 may be caused by variable algorithms employed to compensate for internal heat gains inside the custom-546 built monitor shells that affect final readings and justify the result disparities. The air temperature was 547 reported accurately by most of the tested devices, with 3 out of 8 consumer grade monitors being within 548 +/-0.5°C from the reference air temperature in both seasons and all the monitors being within +/-0.6°C 549 from the reference regardless of the season, except for Clarity in both climatic conditions and for AirVisual 550 in warm and humid conditions. These results confirm the suitability of consumer grade monitors, apart 551 from Clarity, to monitor the air temperature inside buildings. The AirVisual and Kaiterra represent the 552 monitors with color displays with a higher heat output as shown in Figure 9. Our results suggest that the 553 air temperature measurements were well compensated for the local heat production, except for Clarity. 554 On the other hand, the Lit92 sensor was installed on a housing that accommodated multiple single sensors 555 and was in proximity of a microcontroller with a power converter. The heat output from the 556 microcontroller likely interfered with the air temperature field which led to overestimated temperature 557 values.

562
To better summarize the performance of consumer grade monitors in both thermodynamic conditions, 563 we developed an overall performance grading. First, the performance of the monitors was averaged 564 across all 16 experimental conditions. Then, according to the classification for MRE (PM 2.5 , PM 10 , CO 2 and 565 TVOC) or MAE (relative humidity and temperature) and PCC, each monitor was given a grade from 1 to 5. 566 This was done by dividing the range between the minimum and maximum MRE or MAE for each 567 parameter into 5 categories where the grade 1 was assigned to the worst and 5 to the best category. The 568 score was averaged across two test thermodynamic conditions. For the PCC, the 5 categories were based 569 on the rating introduced in chapter 2.7. Table 5 shows the summarized performance for each monitor. 570 The MRE, MAE and PCC data used for the monitor ranking are given in the supplement Table S2. 571 572 pollutants, air temperature, and relative humidity; and it also scored highly for measuring TVOC 577 concentrations, unlike many other monitors. The Kaiterra monitor scored just a bit lower but lacks the 578 ability to report more than one gaseous pollutant (in this case TVOC). A slightly lower performance was 579 shown by the uHoo and AirVisual monitors, followed by Netatmo_i, but the latter monitor lacked the 580 ability to report PM and TVOC. The Clarity came in second to last despite not monitoring CO 2 and TVOC. 581 However, these results need to be considered carefully. We determined that the device was connected 582 to the proprietary device hub used for calibration, but when data log was analyzed, we discerned that no 583 calibration from the network to the device was received which could account for the erroneous 584 measurements. Foobot showed the worst overall performance, especially in the IAQ category, and the 585 Netatmo_o exhibited a good overall performance for relative humidity and temperature but is not 586 monitoring any of the pollutants. Contrary to the expectation, monitors on the lower price spectrum had 587 the best performance in the tested categories. End-users should not regard the price of the low-cost 588 monitors as an indicator of their performance. 589 Seasonal comparison did not show a clear influence of indoor thermodynamic conditions on the accuracy 590 and stability of the measurements. Each device displayed comparable performance in both conditions. 591 The main differences could be observed between devices, when measuring individual parameters 592 regardless of the condition. 593 While interpreting the reported results, several limitations must be acknowledged. Only a single new 594 device of each model was tested and their durability and consistency over time was not considered. The 595 study did not evaluate the impact of automatic baseline correction on CO 2 sensor performance and did 596 not consider the effect of intermittent high to very low ambient RH changes. Further, the performance 597 assessment did not consider the quality and richness of the real-time data reporting interface, nor the 598 accessibility and availability of the measured data. For PM measurements, miniWRAS was not adjusted 599 with the true size of particles with gravimetric measurements, and the default density of 1.68 g/cm³ was 600 used. correlated for the majority of tested devicesmeaning that the low-cost units could be used to detect 614 concentration changes of particulate matter spanning from 0.3 to 2.5 μm. On average, the best 615 performing monitor deviated from the reference by a factor of two. Among the single sensors, OPC-R1 616 provided the best results for PM 2.5 , while the OPC-N3 proved to be the best for PM 10 monitoring. The 617 majority of the tested units performed well in detecting CO 2 concentrations up to 3'500 ppm resulting in 618 errors within 25% from the reference, with the best monitors performing within 3% from the reference. 619 Foobot and uHoo monitors failed to accurately report the CO 2 concentration, with the mean relative error 620 exceeding 30%. Low cost TVOC monitors Awair, Foobot and uHoo showed a strong correlation with the 621 professional grade monitors despite a poor quantitative agreement. For relative humidity, the majority of 622 tested devices gave time-correlated and acceptable results within 5% difference from the reference with 623 the tendency to over-report relative humidity in cool and dry conditions and under-report it in warm and 624 humid conditions. The uHoo, SCD40 and SHT31 showed the best performance with less than 0.6% RH 625 difference, while the Clarity was the worst in class resulting in a 12% difference from reference. The air 626 temperature was reported within +/-0.5°C from the reference temperature in both seasons by 3 out of 8 627 consumer grade monitors and within +/-0.6°C by the majority of tested devices. Seasonal comparison 628 revealed that the majority of consumer grade monitors displayed comparable performance in both 629 conditions, with the majority of consumer grade monitors being slightly closer to reference in cool and 630 dry conditions for PM and CO 2 and in warm and humid conditions for TVOC. 631 Recent technological advancements have opened up an opportunity for more effective indoor air quality 632 control and management. The present study suggests that the majority of the tested low-cost consumer 633 grade monitors have the potential to be used to secure adequate indoor environments by triggering the 634 right chain of actions. This could be accomplished either via a feedback loop to encourage human actions 635 or through integration in a building management system with automated controllers and devices. To 636 assure continuous improvement of low-cost environmental sensing technology, future work should focus 637 on the examination of the longitudinal performance of these units, development of quality control 638 algorithms that minimize errors and remove bias, and development of the standards and guidelines for 639 their testing. 640  [1] for 6 unscented candles and 3 scented candles respectively. In case of the mosquito coil burning, the density was reduced to 1.17 g/cm³ [1], while for the popcorn cooking activity, the adjusted density was set to 1.10 g/cm³. For this latter, an average value between the densities determined for 90 g of popcorn heated in a microwave (1.32 g/cm³) [1] and 15 g of oil brought to bubble in a steel wok (0.88 g/cm³) [1] was taken. Contrarily, the density for carpet vacuuming was increased to 2.89 g/cm³ [1] and the average particle density from the room deodorant was not adjusted because no experimental value was found in literature.