Long-Term Characterization of Indoor Air Quality at a Research Area Building: Comparing Reference Instruments and Low-Cost Sensors

,


Introduction
Indoor air pollution has a more detrimental effect on human well-being than outdoor pollution since people spend a major portion of their time in indoor environments.In their review, [1] estimated 1.8 million deaths associated with household air pollution in 2017.Despite this, in contrast to ambient air quality, there is no standard regulation worldwide for indoor air quality except for a few countries and regions (e.g., Canada, China, Portugal, and Taiwan) as reported in [2].In their work, authors list the existing national indoor air quality (IAQ) standards and discuss the World Health Organization (WHO) guidelines related to IAQ [3].
According to [4], people spend on average most of their time at home (58-69%), followed by work environments (28%).Among working places, offices constitute the most widespread professional environment, especially in industrialized countries, where about 67% of the workers are office employees [5].The main sources of indoor particles in homes and offices are outdoor particle penetration, smoking, including new generation devices (e-cigs or HNB products), cooking, and burning incense and candles [4,6].In recent years, several studies have established how each of these sources poses a risk for human health based on measurements of hazardous compounds such as polycyclic aromatic hydrocarbons (PAHs), volatile organic compounds (VOCs), formaldehyde, UFP, and black carbon (BC) [7][8][9][10].
In many cases, the metric used for the evaluation of indoor pollution impact is the mass of particles, often derived by converting number of particles to mass data from real-time optical counters.However, the choice of a density factor for the conversion is arbitrary when no simultaneous measurements with a reference instrument are available [11,12].Moreover, most indoor sources significantly increase particle number concentration (NC) with negligible impact on mass, especially when UFP are involved, and the knowledge of NC is necessary to predict the penetration and deposition dose in the human respiratory tract [13][14][15].
Generally, studies on indoor sources are conducted in test chambers and, only in a minor part, in real-life conditions [16][17][18][19][20] owing to the limitations associated with uncontrolled and unrepeatable experimental conditions.On the other hand, measurements carried out in real environments permit to evaluate the potential impact of aerosol transformation processes on indoor air quality [21,22].An example could be the possibility to better characterize second and third hand smoking exposure as described in [17].
Studies on indoor air quality are usually based on shortterm datasets, thus not receiving epidemiologists' requests of long-term studies [23][24][25].Conducting exposure studies to quantify indoor air quality would also require high spatial resolution, which is often unfeasible indoors due to the high cost, dimensions, noise, nuisance, and low portability of traditional instrumentation.In recent years, the use of PM LCS for indoor air quality monitoring has grown rapidly, helping to address these issues, and increasing the feasibility of a more widespread monitoring [26][27][28][29].
LCS implies, however, the collection of data with limited accuracy compared to reference instruments [30,31,32,33,34].Although many efforts have been made to assess the state of the art about LCS, there is currently a lack of uniformity in the methods to evaluate their performance and to calibrate them [35][36][37].The US EPA has developed a report to provide a reliable methodology for sensor performance evaluation, including testing protocols, metrics, and target values [38][39][40].For Europe, the technical specification "CEN/TC 264/WG 42 -Air quality sensors" is developing a technical guide for data quality objectives to be achieved by PM LCS and is expected to be released in the near future [41][42][43].Both documents focus mainly on outdoor applications.As regards indoor use of PM LCS, some reviews were produced within the scientific community describing various measurement campaigns carried out using PM LCS or trying to assess their performance [44][45][46][47].Giordano et al. [45] suggested performing multiple tests to assess sensor linearity in their response to different aerosol sources.Ródenas García et al. [47] highlighted the potentiality and opportunity of LCS used indoor, indicating as a forthcoming requirement, the development of LCS for UFP and nanoparticles.
The aim of the present work is the long-term characterization of indoor aerosols in a laboratory, inside a research building.Short-duration high-emission events caused by four commonly used indoor sources (incense, traditional cigarette, e-cigs, and HNB) are also included in the analysis.In view of the possible future set-up of a low-cost network for indoor air quality control, two low-cost sensors were used together with the scientific instruments and their performances were evaluated along the measurement period.The aerosol characterization was carried out measuring particle number size distributions in the range 0.3-1 μm and eBC for a ten-month period, using a TSI Optical Particle Sizer (OPS) model 3330 and an aethalometer Magee AE33, respectively.Furthermore, to assess the emissions of BrC from smoldering combustion processes, the AAE was estimated.Finally, two collocated PM LCS, Sensirion SPS30, were used together with the reference instrument for both the long-term analysis and transient events.SPS30 sensors were chosen for their superior performance compared to others [24,48,49], their low cost, and compact size which make them easily adaptable for use in indoor environments with limited spaces.To our knowledge, this is the first longterm study to examine the precision, accuracy, and durability of SPS30 sensors for indoor use.

Description of the Site
The measurement site is the Potenza research area located in South Italy (Tito Scalo, 40.60 °N, 15.72 °E, 750 m a.s.l.) in a small industrial zone, placed in a large rural area surrounded by Apennine Mountains and near a few small plants, a busy main road, and a shopping mall.The research area hosts a complex building (Figure 1) with three institutes of the National Research Council including laboratories, offices, a library, a data center, a conference room, and a guesthouse.About 100 workers are expected to carry out their activities daily in the area.

Indoor Air
The complex building consists of three two-level edifices (edifices 1-3 in Figure 1) hosting laboratories (ground floor) and offices (first floor), a two-level edifice (edifice 4 in Figure 1) hosting offices and a conference room (ground floor) and offices and a library (first floor), and a three-level edifice hosting common facilities (ground floor), offices (first floor), and a guesthouse (third floor).The five edifices are interconnected, whereas edifice 6, devoted to plant engineering, is an independent building.Measurements were conducted in a laboratory/office located on the ground floor of edifice 1 (Figure 1).During working hours (8:30 a.m.-17:00 p.m. local time (LT), approximately from Monday to Friday), one or two employees are present in the room.
Apart from the infiltration (building envelop characteristics) and penetration (opening and closing the door at 8-9 a.m. and 4-5 p.m. local time (LT)) of outdoor particles, possible indoor sources, related to the activities in the lab/office room, include particle resuspension due to people moving and operation of  An exhaustive characterization of the outdoor sources of aerosols at the site, as suggested by previous works [50][51][52][53], classified the site as semirural, often affected by Saharan dust intrusions each year, and with the influence of vehicular traffic emissions from the close main road.

Materials and Methods
3.1.Instruments.The indoor aerosol characterization was obtained by collecting multi-instrumental data on a longterm basis.The instruments used were a TSI OPS 3330 to obtain number particle size distributions in the size range 0.3-10 μm (16 channels) and an aethalometer Magee AE33 to measure the eBC concentrations and estimate AAE.Further details on the instrumentation can be found in [54].In addition, four experiments were performed to characterize aerosol transient emissions from some common indoor sources at the research area such as incense, electric and traditional cigarettes, and HNB products.During these short-term experiments, a TSI versatile water-based condensation optical particle (V-WCPC) model 3789 was used to determine the number concentrations (NC) of UFP and fine particles down to 2.5 nm in diameter with 1 min time resolution [55].
The testing of two PM LCS Sensirion model SPS30 (units 5F65C and BB687) was carried out both on a long-term basis and for the four transient events using the OPS as a reference instrument.The SPS30 sensor utilizes a measurement technology based on laser scattering and provides particle size distribution in four bins in the range 0.3-10 μm, as well as PM 1 , PM 2.5 , PM 4 , and PM 10 mass concentrations.The sample airflow is generated by means of a fan, and the sampling time is 1 s.The SPS30s were calibrated by the manufacturer for PM 2.5 number size distribution using a TSI OPS model 3330.Additionally, SPS30 has a contaminationresistance technology that prevents the sensor's optical components from contamination because of particle accumulation [56].The Sensirion SPS30 is equipped with an automatic fan cleaning procedure which ensures the sensor's accuracy even during monitoring in a high polluted environment [57].
It also has an associated viewer software for data logging, visualization of real-time data, setting of the averaging time of sensor outputs, and selection of the auto cleaning interval if auto cleaning is enabled.However, due to the lack of internal storage, the SPS30 must be connected to a computer to gather data.
Owing to the very low number count density in PM 4 and PM 10 fractions, attention should be paid when considering coarse particle detection as suggested by the manufacturer [56].The limited reliability of SPS30 in measuring large particle concentrations was verified in [24,48,49].In the same works, a good performance of SPS30 in revealing fine particle was assessed.
Some tests were carried out to choose the suitable averaging time of sensor outputs, which was set to 100 s (see Figure S1 of the Supplementary material).

Experimental
Set-Up and Dataset Description.All the measurements were carried out in a laboratory/office with a surface area of approximately 40 m 2 placed on the ground floor of edifice 1 (Figure 1).It has natural ventilation (doors and windows) and a mechanical ventilation system which is used in wintertime for heating.No filtration system is present in the mechanical ventilation system.The instruments were placed on a table (90 cm high) near the door, opposite the desk where two employees usually work (Figure 1).
During the short-term experiments, a window and the door of the laboratory/office were left open.All sources were placed at approximately 1.5 m from the instrumentation.
The long-term dataset, lasting from October 2021 to July 2022, included hourly and daily averaged number size distributions from OPS data and the two Sensirion SPS30 units, respectively, as well as hourly and daily averaged eBC mass concentration and AAE values from aethalometer measurements.

Indoor Air
As regards the four transient pollution events, data from all instruments were collected on March 1, 2, 4, and 8, 2022. 1 min CPC and aethalometer data were smoothed in a 5 min moving average and compared to 5 min OPS measures.These days were excluded from the long-term dataset analysis.

Results and Discussion
4.1.Long-Term Dataset 4.1.1.OPS Size Distribution.OPS mean size distributions were calculated separately for working and nonworking days and for the cold (October-April) and hot (May-July) seasons, as shown in Figure 2. Working and nonworking days' size distributions show a similar shape over the entire dataset, with a mode at 1.56 μm which can be associated to particle resuspension due to the traffic source, as previously assessed in [54].Working days are characterized by a slightly enhanced contribution of coarse particles, becoming more pronounced during the hot season.
To compare the fine and coarse contribution to number concentrations, OPS data were grouped into two fractions, NC 1 including particles with diameter from 0.3 to 1 μm and NC 1-10 for particles with diameter from 1 to 10 μm.Looking at the diel cycle of the NC 1 fraction shown in Figure 3(a), for the whole period, an increased concentration is found during nighttime.This pattern is partly driven by the cold season behavior owing to the biomass burning from domestic heating source, whereas during the hot season, a flatter pattern is obtained.The boundary-layer dynamics could have contributed to these trends [58].
Despite the high standard deviation of NC 1-10 diel cycles, especially for hot season data, a clear increase of concentration during this season can be inferred.This can probably be due to the semirural nature of the site, the reduction of rain events, and to the higher frequency of long-range dust advection episodes registered during the hot season, as detailed in Section 4.1.3.Moreover, a peak during the morning is well distinguishable when working days are considered, which is likely associated with the resuspension of coarse particles from the traffic sources.For both NC 1 and NC 1-10 fractions, a higher variability characterizes the hot season data.
4.1.2.eBC and AAE.Looking at Figure 4, a clear pattern for eBC diel cycles is obtained, indicating the traffic from outdoor as the main source of indoor eBC.In [53], outdoor eBC measurements in this area highlighted the two-peak behavior of carbonaceous particle concentration, typical of traffic sources.In their work, the authors concluded that the outdoor eBC at the research building area was strongly influenced by the presence of a main busy arterial (about 20,000 vehicles per day [59]) passing at about 1 km from the site.The two peaks' behavior shows seasonal features in terms of morning/evening peak intensity ratio.In fact, during the cold season, the morning and evening peaks show similar intensities though the evening peak is broader.Conversely, during the hot season, the morning peak prevails over the evening peak.This can be partly attributed to a different dynamic of the boundary layer between the two periods, as found in [58] for a similar site.
As expected, a nighttime increase in AAE is visible during the cold season when higher values are typically found compared to the warm season.This is mainly caused by the biomass burning for domestic heating [53].In general, AAE values are higher during weekends compared to working days, likely due to agricultural activities [54].During the cold season, AAE values are slightly higher during weekends than on working days, as biomass burning for domestic heating is likely used more frequently than on weekdays due to people spending more of their time at home.

4.1.3.
Influence of Outdoor Sources.From the results obtained, the influence of traffic and domestic heating as outdoor sources of indoor fine fraction abundance can be inferred.A confirmation of the influence of the traffic can be derived by the scatterplot between NC 1 and eBC over the long-term dataset, as shown in Figure 5.The correlation coefficient of 0.48 indicates that eBC has an influence on NC 1 , although it is not the only influencing factor.
The measurement site has semirural nature, with soil erosion by wind, and there is dust resuspension; thus, the outdoor particle penetration is expected to have an influence on the indoor coarse fraction.This was verified by comparing the time series of daily indoor NC 1-10 and PM 10 measured at a monitoring station close to the measurement site and managed by the Regional Agency for the Environment of Basilicata (ARPAB) resulting in a R 2 of 0.63 as shown in Figure 6.The comparison also indicates the long-range transport of desert dust and sea salt as an additional outdoor source of coarse particles as shown in Figure 6.In Figure S2 of the Supplementary Material, NAAPS maps (https://www.nrlmry.navy.mil/aerosol/index_frame.html)and Hysplit back-trajectories (https://www.ready.noaa.gov/HYSPLIT.php),supporting this finding, can be found.6

Emission
Indoor Air used at the research area.In Table 1, details of the experiments are listed, whereas the temporal variations of the parameters are plotted in the graphs of Figure 7. UFP concentration shows a broad peak starting at the ignition of the incense cone and decaying at a few minutes after the cone's extinguishing, as reported in Figure 7.The peak concentrations ranged from about 5000 #/cm 3 to a maximum of 66000 #/cm 3 , in accordance with the UFP incense peak concentrations reported in [60].The temporal trend of total number particle concentration peak is similar to those described in [16,61] though they reported maximum concentration 20-to 30-fold higher than the background, whereas in our case, it was found 10-fold higher.This discrepancy could be attributed to the different burning times of incense, varying room conditions (door closed), and experimental settings.The NC 1 fraction shows a double peak: the first occurring a few minutes after the ignition, when the UFP peak starts growing, and the second during the cone burning, highlighting the emission of fine particles.The eBC concentration follows NC 1 , rather than UFP number concentrations.For comparison, some studies such as those of [62,63] have investigated brown carbon content in relation to smoldering biomass combustion using incense as a surrogate for stronger absorption at shorter wavelengths.The authors calculated the AAE parameter as the negative slope of the absorption coefficient versus wavelengths (355, 405, 532, 870, and 1047 nm) in a log-log plot obtaining a value of 4.53.High values of AAE (>4) were also found in [62], where the couples of wavelengths 405-532 nm and 532-781 nm were considered for AAE calculation.In this work, incense-related AAE was 2.72, the highest among all other sources tested.

E-Cigarette.
The test with the e-cig consisted of three vaping sessions performed by the same volunteer.The ecigarette was filled with e-liquid containing 2.2 mg mL -1 of nicotine.It was lit for ten minutes from 2:30 p.m. on March 2, then lit again at 3:36 p.m., and turned off at 3:42 p.m., and finally, it was vaped from 4:34 p.m. to 4:38 p.m.As regards the UFP particles, comparable number concentrations were found in [17], who used similar experimental set-up for ecig testing.However, it is worth noting that the types of kit and e-liquid used may cause substantial variability in e-cig emissions [19,64].Moreover, even in the absence of combustion processes during vaping, the eBC peaks are distinguishable in the graphs of Figure 7, with a maximum after the third vaping session.This could be due to the burning of residual incrustation on the coil promoted by the high temperatures reached (>1000 °C) when no saturation with e-liquid occurs in the tank [65].

Conventional Cigarettes.
The test with the traditional cigarette consisted of two smoking sessions, performed by the same smoker, lasting seven and five minutes, respectively,  UFP concentration showed two peaks, one at the end of the first smoking session and the other, broader and more intense than the first, a few minutes after the end of the second smoking session, as shown in Figure 7.The eBC concentration increased simultaneously with the second UFP peak.A similar behavior was also seen for AAE and NC 1 although there was a slight growth visible a few minutes after the first cigarette as well.Together with incense, traditional cigarettes were found to be the strongest emitters of BrC according to AAE values in Figure 7, but different dynamics can be identified.In the case of cigarettes, a narrow peak of AAE values is visible, similar to those of eBC, NC, and NC 1 , whereas for incense, AAE values decay more slowly compared to eBC and NC 1 .This difference 8 Indoor Air can be attributed to the presence of primary absorbing particles emitted during cigarette burning compared to incense, when a secondary origin for BrC can be hypothesized.The delay of NC peak with respect to NC 1 peak in the case of incense further supports this hypothesis.
4.2.4.Heat-Not-Burn (HNB) Product.The HNB devices heat the tobacco stick without burning it.For the experiment with HBN, the same device was used during three sessions, with one tobacco stick smoked, for each session, by the same volunteer.The HNB was lit at 1:15 p.m. on March 8 with a vaping time of two minutes, then it was vaped for four minutes from 1:55 p.m., and finally it was lit at 2:57 p.m. and turned off at 3:01 p.m. UFP concentration shows three peaks, each a few minutes after the lighting of the HNB.Three broader peaks were found for NC 1 but shifted compared with CPC ones as illustrated in Figure 7.The temporal trends of eBC and AAE seem to be unaffected by HNB emissions, suggesting no production of absorbing particles from this source as found in [19].In their experiment, carried out inside a traveling car, traditional cigarette, e-cig, and HNB BC emissions were compared, resulting in HNB as the only smoking device with no spike in BC concentration.Looking at Figure 7, significantly lower concentrations of both UFP and fine particles were found for HNB compared to e-cig and traditional cigarettes in accordance with the recent literature on the matter [66][67][68][69][70][71].However, comparing, for example, the similar experiments carried out by [67,68], where emissions from two HNB devices with different sticks were analyzed, an elevated variability in particle emission is evident.For this reason, the results obtained for the HNB device used in the present work may not be directly comparable to other brands or of different configurations of the same product.

Source Comparison.
A comparison among different source emissions was carried out as summarized in Table 2.
As regards UFP, the most intense source is e-cig, followed by incense and traditional cigarettes, demonstrating a similar impact, and lastly HNB.This confirms the findings of [26] about the relevance of e-cig source in the production of UFP particles.Figure 7, shows how e-cig and HNB emissions exhibit the quickest decay in agreement with the results reported in [19].In terms of NC 1 behavior, the most intense source was e-cig, followed by traditional cigarette and incense then HNB.Among the four sources, e-cig showed the most intense peaks in fraction NC 1-10 in agreement with the findings of [72], which used an aerodynamic particle sizer (APS) in conjunction with a Scanning Mobility Particle Sizer (SMPS) to detect a mode at approximately 1000 nm.Similarly, the recent research by [73] also made note of considerable emissions of 1-2 μm aerosols.It is possible that the release of metal and silicates from the coil and tank, as mentioned in previous studies [65,[74][75][76], contributes to this phenomenon.According to [65], metal emission can be favored by the degradation of the coil due to the high temperatures reached (>1000 °C) when the wick is not saturated with e-liquid.The degradation of the coil, visible for the e-cig used in the present experiment, corroborates this hypothesis.As far as combustion-related emissions are concerned, the most intense eBC emitter is the traditional cigarette, followed by incense and e-cig (similar) and HNB, whereas incense and traditional cigarettes show higher AAE values, followed by e-cig and HNB.In Figure 8, the temporal trends of particle size distribution for the four experiments are shown.The plots highlight the emission behavior of the short-duration sources with an increase of particles in the range 0.3-1 μm for all sources; in some cases, super micron particles were also emitted.
These fractions were poorly characterized in the literature [6,16,72,73], since generally FMPS or SMPS were used for similar experiments.However, in real-life conditions, the coagulation and condensation processes can lead to the formation of coarser particles, which can increase the amount of aerosol deposited in the upper respiratory tract, compared to UFP.

SPS30 Performance Evaluation.
To test the performance of SPS30 sensors, the number concentrations were calculated by aggregating the data into two size intervals NC 1 and NC 1-10 , as previously done for OPS data.However, following [49], the analysis of the results was carried out for the NC 1 fraction only, owing to the low reliability of SPS30 estimates of number concentration in the range 1-10 μm.
SPS30s were evaluated both for precision, by intramodel comparison, and accuracy, by testing them against the OPS on a 1 h and 24 h time basis.Moreover, the long-term stability was checked.Coefficient of determination (R 2 ) and slope from a linear fit, Spearman's coefficient (ρ), root mean square error (RMSE), and coefficient of variation (CV) were used as evaluation metrics [37,40,[77][78][79].The assessment of CV was considered for precision evaluation only, as a metric to compare the two units of the SPS30 sensor model.3 for NC 1 fraction.As far as drift is concerned, the temporal trend of the difference between sensor and reference data was analyzed over the entire dataset.
The values of both R 2 and slope fall in the range indicated in [40] as satisfactory to infer a good sensor performance.This translates in the capability of SPS30 to follow the variations of number concentrations as measured by OPS in the range 0.3-1 μm, as confirmed by the time series of daily values plotted in Figure 9.
The high linearity (0 87 < R 2 < 0 90) found between the number concentrations measured by SPS30s and OPS (Figure 9) is consistent with [80].Sousan et al. [80] found a high correlation between the SPS30 sensors and the reference instrument for mass concentration of three types of laboratory-generated aerosol in both environmental and occupational setting conditions.
Considering the slope of the regression line between the two SPS30 units and OPS, values higher than one indicate an overestimation of the number concentration.Similar results were obtained in the South Coast Air Quality Management District (SCAQMD, USA) field test, where mass concentrations from three SPS30 units were compared to those obtained by three reference monitors in outdoor conditions [81].Additionally, a response comparable to this was registered in [57] when comparing SPS30 and aerodynamic particle sizer (APS) number concentrations for laboratory-generated aerosols.
To investigate the SPS30 sensor response to distinct levels of particle loading, the Pearson coefficient (r) between SPS30 and OPS NC 1 hourly data was calculated separately for each quartile of OPS number concentrations.Results are illustrated in Figure 10 showing a better performance of both SPS30 sensors for increasing aerosol number concentration, in agreement with the findings of [40,79] in the case of mass concentrations.
The low values of CV in Table 2 highlight a high precision for collocated SPS30 sensors in accordance with [57], where SPS30 was more precise than Plantower PMS5003 in measuring laboratory-generated ammonium sulfate, nebulized Arizona road dust, NIST Urban PM, or oil mist.
To evaluate the possible drift in sensor outputs over time, the temporal pattern of the absolute value of the difference between the sensor and the reference daily number concentration was calculated, as reported in Figure 11.A slightly increasing trend is visible indicating the effect of prolonged recording time on sensor measures [25].

Transient Events.
Measurements of number concentrations by means of the two SPS30 sensors were carried out during the tests with the selected sources described in Section 4.2, allowing examination of their efficacy in detecting real short-lasting pollution events in an indoor environment.SPS30 data were time paired with the 5 min OPS measures for comparison, and the corresponding time series are shown in Figure 12.R 2 coefficients for both sensor units were calculated and reported in the upper corner of the graph.As it is possible to see, SPS30s succeeded in the detection of all events with varying accuracy: a good agreement (R 2 ≥ 0 6) was obtained in the case of incense, traditional cigarette, and HNB, whereas a worse result was found for    Indoor Air e-cig.This could be related to the duration of the emission peaks of the different sources.In fact, among the sources, e-cig peaks show shorter temporal extents with faster changes in number concentration.

Limitations and Further Research Directions
In the present work, only a single model of each smoking device (e-cig, traditional cigarette, and HNB) was used as short-duration high-emission source in the experiments.Given the large variety of models and technologies available, especially in the new-generation smoking devices, the results obtained here may not account for the variability of emissions of such sources.The experiments were conducted in real-life conditions, and an extensive analysis of the influence of parameters such as temperature or humidity on LCS performances has not been carried out yet.The development of a low-cost network for indoor air quality control based on LCS represents a further research direction.To this aim, the performance evaluation of LCS under different ambient parameters will be assessed in indoor environments.Moreover, the potentiality of new-generation LCS for UFP and nanoparticles detection will be studied.13 Indoor Air

Conclusions
A long-term indoor campaign was carried out at a research area building located in a semirural site in South Italy to characterize aerosol number size distributions and carbonaceous particles.Short-term intense emission events common to the research area were simulated and included in the analysis.Moreover, the performances of two PM LCS, Sensirion SPS30, were evaluated comparing LCS data with the number size distributions obtained by the reference instrument.The main results are summarized in the following: (i) The analysis of the long-term dataset suggests that infiltration and penetration of outdoor particles strongly influence the number particle concentration in both fine and coarse fractions; traffic (BC and resuspended dust) and domestic heating (BrC) are the main sources of indoor particles from outdoor (ii) The short-term experiments with the four indoor sources showed that HNB has the smallest impact on particulate loading; e-cig emits particles with dimensions spanning from ultrafine to coarse, probably due to the release of metal and silicate by the heating coil degradation (iii) The transformation processes modify the particle size distribution after emission, as highlighted during the experiments; this could influence penetration and deposition rate in the human respiratory system (iv) Concerning the comparison between OPS and LCS, a good agreement for both long-term and shortterm experiments is observed, along with a slight degradation of the LCS long-term performance The results obtained suggest best practices to adopt for the health of workers, especially when high emission sources are present.Moreover, the comparison of SPS30 PM LCS and OPS TSI 3330 data clearly demonstrates their feasibility in supporting indoor air quality evaluation on both the longterm basis and for transient events.

Figure 1 :
Figure 1: (a) General plan of the complex building hosting the Potenza research area.(b) Plan view of edifice 1 with the laboratory/office where the measurements were carried out.The red letters "I" and "S" indicate where the instruments and the sources were placed, respectively.

Figure 2 :
Figure 2: (a) Mean number size distribution for working and nonworking days for the whole dataset.(b, c) Mean number size distribution for working and nonworking days and for hot and cold seasons.Vertical bars represent standard deviations.For color images, please refer to the online version of the article.

Figure 3 :
Figure 3: (a) Diel (working and nonworking) cycles of NC 1 fraction for the whole period, cold, and hot season.(b) The same for NC 1-10 fraction.

Figure 4 :
Figure 4: (a) Weekly and diel (working and nonworking) cycles of eBC for the cold and hot season.(b) The same for AAE.

32 R 2 Figure 5 :
Figure 5: Scatterplot of daily mean NC 1 vs. eBC data over the whole dataset.

Figure 6 :
Figure 6: (a) Temporal trend of daily mean PM 10 and NC 1-10 and (b) corresponding scatterplot.Black arrows indicate Saharan dust advection episodes.For color images, please refer to the online version of the article.

Figure 7 :
Figure 7: Temporal variations of NC 1 , NC, eBC, and AAE data during the five tests.Please note that HNB source was plotted with a different y-axis extension to enable cleaner visualization.For color images, please refer to the online version of the article.

4. 3 . 1 .
Long-Term Dataset.Precision and accuracy of SPS30s were first evaluated on a long-term basis.Results are shown in Table

Figure 8 :
Figure 8: OPS particle number size distributions in the range 0.3-10 μm for (a) incense, (b) e-cig, (c) trad cigarette, and (d) HNB.For color images, please refer to the online version of the article.

Figure 9 :Figure 10 :
Figure 9: (a) Temporal trend of daily mean NC 1 measured by the OPS and the two SPS30s and (b) corresponding scatterplots.For color images, please refer to the online version of the article.

Figure 11 :
Figure 11: Time variation of the absolute value of the difference between daily mean NC 1 measured by the OPS and the two SPS30s.For color images, please refer to the online version of the article.

Figure 12 :
Figure 12: Temporal trend of NC 1 measured by the OPS and the two SPS30s during the four transient pollution events.Please note that different y-axis extensions were used to enable cleaner visualization.For color images, please refer to the online version of the article.

Table 1 :
List and duration of the tests carried out for the four indoor sources: incense, electric and traditional cigarettes, and heat-not-burn (HNB) products.

Table 2 :
Contributions of different sources to NC 1 , NC, eBC, and AAE increase, measured during tests.

Table 3 :
Sensor performance for NC 1 fraction measured by the two SPS30 units.Both precision and accuracy evaluation metrics include R 2 , the slope, and the Spearman coefficient (ρ).RMSE was calculated only for SPS30 and OPS comparisons, while the CV was determined only for intramodel comparison.Statistical parameters were calculated both on a 1 h and 24 h basis.