Low Cost, Multi-Pollutant Sensing System Using Raspberry Pi for Indoor Air Quality Monitoring

: Deteriorating levels of indoor air quality is a prominent environmental issue that results in long-lasting harmful effects on human health and wellbeing. A concurrent multi-parameter monitoring approach accounting for most crucial indoor pollutants is critical and essential. The challenges faced by existing conventional equipment in measuring multiple real-time pollutant concentrations include high cost, limited deployability, and detectability of only select pollutants. The aim of this paper is to present a comprehensive indoor air quality monitoring system using a low-cost Raspberry Pi-based air quality sensor module. The custom-built system measures 10 indoor environmental conditions including pollutants: temperature, relative humidity, Particulate Matter (PM) 2.5 , PM 10 , Nitrogen dioxide (NO 2 ), Sulfur dioxide (SO 2 ), Carbon monoxide (CO), Ozone (O 3 ), Carbon dioxide (CO 2 ), and Total Volatile Organic Compounds (TVOCs). A residential unit and an educational ofﬁce building was selected and monitored over a span of seven days. The recorded mean PM 2.5 , and PM 10 concentrations were signiﬁcantly higher in the residential unit compared to the ofﬁce building. The mean NO 2 , SO 2 , and TVOC concentrations were comparatively similar for both locations. Spearman rank-order analysis displayed a strong correlation between particulate matter and SO 2 for both residential unit and the ofﬁce building while the latter depicted strong temperature and humidity correlation with O 3 , SO 2 , PM 2.5 , and PM 10 when compared to the former.


Introduction
In the past decade, declining air quality has emerged as one of the largest global environmental issues due to its harmful effects on human health. The American Lung Association (ALA) "State of the Air ® 2020" estimated that nearly 49% of the United States population lived in harmful environmental air conditions [1]. An average person spends upward of 90% of time indoors and owing to the COVID-19 pandemic, this percentage has further increased [2]. A total of 1.6 million premature deaths worldwide are attributed to indoor air pollution [3]. Harm inducing health conditions such as Sick Building Syndrome (SBS) and Building Related Illness (BRI) are proven to be the effects of unhealthy levels of indoor air quality [4][5][6][7]. According to the US Environmental Protection Agency (EPA), the major constituents of indoor air pollution include Particulate Matter (PM), Nitrogen dioxide (NO 2 ), Sulfur dioxide (SO 2 ), Carbon monoxide (CO), Ozone (O 3 ), Carbon dioxide (CO 2 ), and Total Volatile Organic Compounds (TVOCs) [8]. Particulate Matter (PM) refers to a mixture of airborne particles, which are generally categorized by diameter as PM 10 (2.5 µm to 10 µm) and PM 2.5 (less than 2.5 µm) [6,8,9]. Fine particles of aerosol in the form of PM serve as means for viruses to propagate, causing various respiratory diseases. NO 2 is a reddish-brown gas that is mainly generated by vehicular traffic in the outdoor environment, while indoors, it is caused by tobacco smoke and fuel-burning attaching multiple gas sensors to a Raspberry Pi board [34][35][36][37][38]. The measured data can be visualized in real-time and also extracted remotely using a local network by the end-user. The Open Source Building Science Sensors (OSBSS) platform developed by Akram et al. provides a low-cost Arduino-based system for monitoring and recording long-term indoor parameters such as light and CO 2 [39]. Thomas, Alex, and Richard (2019) [40] analyzed the calibration data of 100 air quality monitoring devices from 24 office building to measure the accuracy of monitors using the Monte Carlo simulation. The existing LCAQS studies primarily focus on pollutants such as PM2.5, CO 2, and PM10 more frequently than CO, Ozone, and NO 2, while SO 2 and TVOC are rarely encountered [8,41,42].
The aim of this paper is to present a comprehensive indoor air quality monitoring system using a low-cost Raspberry Pi-based air quality sensor module. The proposed system measures environmental parameters such as temperature, relative humidity, PM 2.5 , PM 10 , NO 2 , SO 2 , CO 2 , CO, O 3 , and TVOC in real-time to determine air quality. The remainder of this paper is structured as follows. Section 2 (materials and methods) describes the LCAQS system with its constituent sensors. The monitoring methods and different data analysis techniques employed are also detailed. Section 3 (results and discussions) presents a case study where the monitored data and results are discussed. The conclusion highlights the paper's contributions and possible future work in Section 4.

Low Cost Air Quality System
The custom built LCAQS monitoring system used for this project was developed on a Raspberry Pi 3B Plus module. This module is a low-power, compact, easily configurable SBC which is powered by an ARM Cortex-A53 processor [43]. The Raspberry Pi module encompasses features such as a Micro SD port for external storage access, Bluetooth, wireless LAN, USB ports, and GPIO pins for external communication. A nine-port expansion board was integrated into the Raspberry Pi module. A Wi-Fi USB adapter was connected to the Gigabit Ethernet port to establish access to the internet. In order to dissipate excess operational heat, a dual fan and heat sink module was attached to the Raspberry Pi system. Figure 1 shows the custom-built LCAQS monitoring system, and Figure 2 depicts a detailed block diagram with all electrical connections. The sensors used in this study were selected from the author's previous publication [8] accounting for repeatability, tolerance, and response time.  The DHT22 sensor is utilized to measure humidity and temperature levels. The sensor is connected to the Raspberry Pi module through a 5k pull-up resistor using I/O pins. The measuring capacity of the sensor ranges between −40 • F and 176 • F with ±0.5 • F accuracy for temperature and from 0% to 100% with ±2-5% accuracy or humidity [44]. DHT22 outputs humidity and temperature values as serial data. The SDS 011 sensor measures particulate matter between 0.3 and 10 µm in diameter using a laser scattering technique. The sensor with a built-in fan is interfaced with the Raspberry Pi module using an UART (universal asynchronous receiver-transmitter) connection. SDS 011 with a    The SPEC-DGS-NO2 968-043 is a low-power IoT integrated gas sensor used to measure NO2 levels. The sensor with a circuit voltage of 3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-NO2-968-043, with a response time of fewer than 30 s and repeatability less than ±3% of the reading, can measure NO2 concentrations The SPEC-DGS-NO2 968-043 is a low-power IoT integrated gas sensor used to measure NO 2 levels. The sensor with a circuit voltage of 3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-NO2-968-043, with a response time of fewer than 30 s and repeatability less than ±3% of the reading, can measure NO 2 concentrations between 0 and 5 ppm [46]. The SPEC-DGS-SO2-968-038 is a high-performance IoT integrated gas sensor used to measure SO 2 levels. The sensor with a circuit voltage of 3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-SO2 968-038 with a response time of less than 30 s and repeatability less than ±3% of the reading can measure SO 2 concentrations between 0 and 20 ppm [47]. The Senseair K30 is an OEM (original equipment manufacturer) module which is customized as a sensor platform to measure CO 2 gas. The sensor uses the NDIR (non-dispersive infrared) technique to detect CO 2 levels in a region. The module is interfaced with Raspberry Pi using an UART connection. SenseAir K30, with a quick response time of 20 s, can measure CO 2 concentrations between 0 and 5000 ppm with accuracy of ±3% of the reading [48]. The SPEC-DGS-CO 968-034 is a lightweight IoT integrated gas sensor used to measure CO levels. The sensor with a circuit voltage of 3.3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-CO-968-034 with a typical response time of 15 s and repeatability less than ±3% of the reading (or 2 ppm, whichever is greater) can measure CO concentrations between 0 and 1000 ppm [49]. The SPEC-DGS-O3-968-042 is an ultra-low power consumption IoT integrated gas sensor used to measure O 3 levels. The sensor with a circuit voltage of 3.3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-O3-968-042 with a response time of less than 30 s and repeatability less than ±3% of the reading can measure O 3 concentrations between 0 and 5 ppm [50]. The uThing::VOC™ is an open access USB module which utilizes the Bosch BME680 air quality sensor to detect VOCs (volatile organic compounds) [51]. The BME680 sensor mainly detects compounds such as ethane, ethanol, acetone, and isoprene with an accuracy of 5%. The module is integrated with Raspberry Pi using a simple USB connection, and the VCP (virtual communication port) interface is utilized to export data as a CSV (comma-separated values) file. The sensor with a response time of less than one second outputs an IAQ index between 0 and 500 based on EPA guidelines, which corresponds to the VOC concentrations [51]. The specifications and the market price of this LCAQS are shown in Table 1. Based on the official datasheets and specifications, the sensor modules chosen for this LCAQS are pre-calibrated in the factory with FRM/FEM equipment and therefore required no end-user calibration.

Site-Description
The two sites ( Figure 3) chosen for this experiment were a university lab building and the living room of a two-bedroom residential apartment unit, both located in the city of Gainesville, Florida, United States. Gainesville is an inland city located in Alachua County of Central Florida, which experiences a humid subtropical climate all year-round. The lab building is an educational office space with a plywood floored area of 321 sq.ft situated at a distance of 245 ft from the nearest major road. A 1.5-ton ducted split-type indoor air conditioning unit (Mitsubishi PEA-A18AA) provides centralized cooling for the room while the air handling unit maintains a constant airflow rate of 235 CFM [52]. This HVAC unit is coupled with a washable MERV-8 air filter with a PM 2.5 removal efficiency of 20-35% [53]. The mechanically ventilated space has three operable windows, which are closed at all times. Due to the COVID pandemic, the occupancy capacity is limited to a maximum of three people at a given time. The living room is part of a standard twobedroom residential unit. The unit is located on the second floor of a three-story building with a carpet floor area of 421 sq.ft. A 4-ton centralized air conditioning unit (GOODMAN GSX13048-AWUF24051BA) provides cooling, and the air handling unit maintains a constant airflow rate of 835 CFM [54]. This HVAC unit is coupled with a MERV-8 air filter with a PM 2.5 removal efficiency of 20-35% [53]. The mechanically ventilated space has two operable windows, which are closed at all times. The occupancy capacity is limited to a maximum of three people throughout the year. The distance between the two experimental sites is 2272.30 ft. The air conditioning units at both sites continuously function throughout the year. The experimental sites were chosen to study variations in air quality among two similar spaces with different functional purposes.

Measuring and Sampling Methods
Two distinct time periods spanning seven days were selected to conduct the experiments at the two respective locations. For the residential unit, data were collected between 8 September 2020 and 14 September 2020 while for the educational building, air quality was monitored from 17 September 2020 to 23 September 2020. In total, the 10 air quality parameters measured were relative humidity, temperature, PM 2.5 , PM 10 , NO 2 , SO 2 , CO 2 , CO, O 3 , and TVOC. In order to structure the monitoring process to limit measurement uncertainties, the standardized EPA monitoring protocol for indoor air quality was implemented [10]. The ASHRAE 62.1-2019 standard was utilized for analyzing each air pollutant concentration [55]. In both monitoring locations, the LCAQS system was placed 3.7 feet above floor level, 1.5 feet away from a wall, and at least a 4.8 feet distance from any corner. Each of the two LCAQS systems deployed at both locations collects 24-h continuous data in intervals of 10-min for all of the pollutants mentioned [10,27,56,57].  The SPEC-DGS-NO2 968-043 is a low-power IoT integrated gas sensor used to measure NO2 levels. The sensor with a circuit voltage of 3 volts is interfaced with the Raspberry Pi module using an UART connection. DGS-NO2-968-043, with a response time of fewer than 30 s and repeatability less than ±3% of the reading, can measure NO2 concentrations

Data Analysis
The data measured by the LCQAS systems are extracted and subjected to statistical analyses to understand the effects of indoor pollutant concentrations on air quality. Python (version 3.6.12) language was used to conduct descriptive analysis to determine statistical parameters such as mean, standard deviation, and variance. A quantile-quantile (Q-Q) plot was produced in Python to study the distribution pattern of the data set. Correlation studies were carried out in RStudio software (version 1.3.1093) to study the inter-dependency and influence of one pollutant on another. The monotonic relationship between pollutants was determined using Spearman rank-order correlation. The Spearman correlation coefficient ranging between −1 and 1 was utilized to establish a degree of correlation within pollutants, while a minimum sensitivity of <0.05 was considered significant [58][59][60].

Results and Discussion
The custom-built LCAQS system was deployed in two locations, namely: site 1 (a living room within a residential apartment unit) and site 2 (an educational office space). The system successfully gathered concurrent data from all of the individual sensor modules without hindrance at both respective locations. The values of monitored pollutants were extracted from the LCAQS system and transferred to an external computer without data attenuation. Table 2 shows the monitored average environmental parameters along with its standard deviation, minimum, and maximum levels. Site 1 has a mean temperature and humidity of 80.5 • F and 45.5%, while site 2 has a relatively lower mean temperature of 75.5 • F and a higher mean humidity of 70.4%. The PM 2.5 and PM 10 concentrations for site one range between 0.20 and 50.5 µg/m 3 and 0.20 and 80.9 µg/m 3 while recording mean values of 8.53 and 10.2 µg/m 3 , respectively. The average PM 2.5 and PM 10 concentrations for site 2 are comparatively lower at 2.33 and 2.43 µg/m 3 . The mean NO 2 concentration for site 2 at 60.3 ppb is relatively higher than that of site 1 at 41.8 ppb, and the corresponding standard deviation follows a similar trend at 2.02 for site 1 and 6.94 for site 2. Site 1 has a mean SO 2 value of 39.9 ppb, while site 2 has a relatively lower value of 29.6 ppb. The minimum CO 2 concentration of 761 ppm recorded at site 1 is greater than the maximum value of 522 ppm registered at site 2. Similarly, the mean concentration of 2195 ppm at site 1 is significantly higher than the mean value of 432 ppm at site 2. Since the collected CO data for site 2 was mostly negligible in value, it was deemed as statistically insignificant for this study. For site 1, the average CO concentration is 1.05 ppb, and the maximum value recorded is 2.12 ppb. The mean ozone level for site 1 at 12.1 ppb is over five times greater than the mean concentration of 2.37 ppb registered at site 2. The recorded TVOC values were measured against the IAQ index scale ranging from 0 to 500. Site 1 has an average TVOC level of 139.2, and site 2 has a value of 121 while the maximum value for both sites is 250. The PM 2.5 and PM 10 concentrations follow a similar pattern for both site 1 and site 2 as plotted in Figure 4c,d. For the first three days, distinctive peaks varying in amplitudes are visible around midday for the particulate matter at Sites 1 and 2. The crests with higher values at site 1 can be attributed to scheduled cooking activities when compared to relatively shorter peaks at site 2, which may be due to increased human activity during lunch breaks and increased nearby vehicular traffic. These peaks can be compared against the ASHRAE 62.1-2019 standard for the particulate matter, which is 35 µg/m 3 (24 h) for PM 2.5 and 150 µg/m 3 (24 h) for PM 10 . Some of the data points for PM 2.5 exceed the ASHRAE 62.1-2019 [55] standard during mid-day, while PM 10 exclusively lies below the standard. On the latter half of the fifth day, a spike is detected starting in the late evening. This anomaly might be due to an organized house event containing three people, subsequently leading to increased house activity. This event can also be viewed in terms of increased pollutant concentrations simultaneously across PM 10 , CO 2 , SO 2 , and TVOC. From Figure 4e, the NO 2 concentrations for site 1 show a significantly stable pattern across the seven-day monitoring period. The potential reason for this trend could be the lack of NO 2 emitting sources. The recurring smooth crests for site 2 can be attributed to a gradual increase in traffic-related pollution over the course of the day peaking at midday. The ASHRAE 62.1-2019 standard for NO 2 is 100 ppb (1 h), and at both sites the NO 2 levels recorded are below this standard. The SO 2 concentrations follow a relatively similar trend for both site 1 and site 2, as plotted in Figure 4f. The ASHRAE 62.1-2019 standard for SO 2 is 75 ppb (1 h), and at both sites, some data points recorded oscillate above this standard. The CO 2 concentrations for site 2 show a significantly low value exhibiting a stable pattern across the span of seven-days, as viewed in Figure 4g. The potential reason for this trend could be because of increased sedentary human behavior within the office space. The ASHRAE 62.1-2019 standard for CO 2 is 1000 ppm (24 h), and at site 2, the CO 2 levels registered are below this standard. Most of the data points are above the standard for Site 1, which might be due to factors such as CO 2 exhalation from increased people count, cooking, and emissions from household cleaning products. The CO concentrations detected in site 1 portray a relatively low value when compared to the ASHRAE 62.1-2019 standard for CO, which is 9 ppm (8 h). Figure 4i shows O 3 concentrations where site 1 displays a significantly stable pattern across the entirety of the study period. The possible reason for this trend could be a lack of nearby vehicular-emissions, which is a primary source of ground-level O 3 . The repeating low-amplitude smooth peaks for site 2 can be attributed to locational proximity to traffic-related pollution over the course of the day. Both sites recorded O 3 levels below the ASHRAE 62.1-2019 standard of 70 ppb (8 h). The measured TVOC data is mapped onto the IAQ index scale, which classifies air quality on a scale of 0 to 500 as: good (0-50), average (51-100), little bad (101-150), bad (151-200), worse (201-300), and very bad (301-500) [51,61]. As depicted in Figure 4j, the data oscillates, leaving only a few observation points below the acceptable good standard, and some even exceed the worse standard line. Both sites display TVOC concentrations that peak at varying amplitudes throughout the study period around midday. The only exception is day five at site 2 where the decline in TVOC levels can be attributed to the absence of occupants. This decline in concentration is also simultaneously seen in PM 2.5 , PM 10, and O 3 .
standard. Most of the data points are above the standard for Site 1, which might be due to factors such as CO2 exhalation from increased people count, cooking, and emissions from household cleaning products. The CO concentrations detected in site 1 portray a relatively low value when compared to the ASHRAE 62.1-2019 standard for CO, which is 9 ppm (8 h). Figure 4i shows O3 concentrations where site 1 displays a significantly stable pattern across the entirety of the study period. The possible reason for this trend could be a lack of nearby vehicular-emissions, which is a primary source of ground-level O3. The repeating low-amplitude smooth peaks for site 2 can be attributed to locational proximity to traffic-related pollution over the course of the day. Both sites recorded O3 levels below the ASHRAE 62.1-2019 standard of 70 ppb (8 hr). The measured TVOC data is mapped onto the IAQ index scale, which classifies air quality on a scale of 0 to 500 as: good (0-50), average (51-100), little bad (101-150), bad (151-200), worse (201-300), and very bad (301-500) [51,61]. As depicted in Figure 4j, the data oscillates, leaving only a few observation points below the acceptable good standard, and some even exceed the worse standard line. Both sites display TVOC concentrations that peak at varying amplitudes throughout the study period around midday. The only exception is day five at site 2 where the decline in TVOC levels can be attributed to the absence of occupants. This decline in concentration is also simultaneously seen in PM2.5, PM10, and O3.  (e) (f) Quantile-Quantile (Q-Q) plots [62,63] are used to characterize the distribution of measured values at Site 1 and 2. Based on the spread of observational points, the type of distribution can be determined by visually characterizing the data sets with respect to the Quantile-Quantile (Q-Q) plots [62,63] are used to characterize the distribution of measured values at Site 1 and 2. Based on the spread of observational points, the type of distribution can be determined by visually characterizing the data sets with respect to the identity line. If most of the measured points are concentrated along the identity line sharing a similar slope, then the data are normally distributed while a deviation above (high-bias) or below (low-bias) the identity line indicates a skewed distribution. From Figure 5, the PM 2.5 and PM 10 data are normally distributed with limited deviation from the identity line for low to medium concentrations and exhibit a low-bias towards higher concentrations. For NO 2 , the observational points are concentrated towards the low and medium spectrum of the data range, while a few points exist away from the distribution for higher concentrations. The slope of TVOC distribution is the most identical to that of the identity line, thereby displaying a high degree of normal distribution while indicating a high closeness in trend between Site 1 and 2. Pollutants such as SO 2 and CO 2 depict a substantial positive and negative deviation respectively across medium and higher concentrations resulting in a skewed distribution. From lower to medium pollutant levels, O 3 has a relatively normal distribution. At higher concentrations, the plot shows a greater positive deviation from the identity line.
concentrations. The slope of TVOC distribution is the most identical to that of the identity line, thereby displaying a high degree of normal distribution while indicating a high closeness in trend between Site 1 and 2. Pollutants such as SO2 and CO2 depict a substantial positive and negative deviation respectively across medium and higher concentrations resulting in a skewed distribution. From lower to medium pollutant levels, O3 has a relatively normal distribution. At higher concentrations, the plot shows a greater positive deviation from the identity line. The Spearman rank-order analysis was used to estimate the strength of the correlation between two measured parameters by assigning a ranked value (R) between −1 and 1 [58,59,64]. According to Figure 6, site 2 exhibits a relatively stronger correlation among The Spearman rank-order analysis was used to estimate the strength of the correlation between two measured parameters by assigning a ranked value (R) between −1 and 1 [58,59,64]. According to Figure 6, site 2 exhibits a relatively stronger correlation among environmental parameters compared to site 1. Site 1 shows that SO 2 has a strong correlation with PM 2.5 (R = 0.9) and PM 10 (R = 0.83). This trend can also be found at site 2 among PM 2.5 (R = 0.92) and PM 10 (R = 0.9). This can be attributed to cooking activities in site 1 and vehicular pollution at site 2, which both contribute to increasing levels of SO 2 , PM 2.5 , and PM 10 [17,65,66]. Both sites 1 and 2 have the same degree of negative correlation between O 3 and humidity (R = −0.78). The higher O 3 concentrations can be due to increased solar irradiation corresponding to low relative humidity-high temperature surroundings. The temperature values recorded at site 2 shows strong correlations with O 3 (R = 0.78), SO 2 (R = 0.8), PM 2.5 (R = 0.88), and PM 10 (R = 0.9). An increase in temperature results in dry air, which might aggregate higher concentrations of PM 2.5 , and PM 10 in a city environment. The relationship between temperature and O 3 , SO 2 , and PM 2.5 is similar with the results observed by Mahmoud while exhibiting stronger degrees of correlation [67]. O 3 has a strong correlation with PM 2.5 (R = 0.8) and PM 10 (R = 0.82) at site 2. There is a moderate correlation between temperature and TVOC (R = −0.65) at site 1. Similarly, temperature and TVOC share a moderate correlation (R = −0.72) at site 2. The similar negative correlation between TVOC and temperature within an office space was observed by Otolorin et al. [68]. Site 1 displays a moderate correlation between humidity and SO 2 (R= −0.62), while site 2 exhibits a comparatively stronger correlation (R= −0.8). SO 2 has the tendency to react with water vapor present in building walls and furniture, which may lead to decreasing concentrations of SO 2 as the humidity increases [12,13,20]. There is a moderate correlation (R = 0.58) between humidity and CO 2 at site 1 and a strong correlation (R = 0.78) at site 2. Site 1 has a low correlation between NO 2 and CO 2 (R = 0.47), while site 2 exhibits a relatively stronger correlation (R = 0.77). This might be due to human activities in site 1 and traffic-related pollution at site 2. This correlation between NO 2 and CO 2 is consistent with the study performed by S.H. Hwang [69]. There is a strong correlation at site 2 between humidity and NO 2 (R = 0.98) and a relatively weaker correlation at site 1 (R = 0.45). At site 1 there is a moderate correlation between CO 2 and O 3 (R = −0.5) while site 2 displays a stronger correlation (R = −0.9) and this may be because of insufficient ventilation [6,15,70]. Site 2 registered a moderate correlation between humidity and TVOC (R = 0.72). The NO 2 values of site 2 show a moderate correlation with TVOC (R = 0.73) and have a relatively stronger correlation with humidity (R = 0.98). This may be attributed to the presence of high-emitting building materials and fuel-burning household appliances at the residential unit [17,20,66].
temperature values recorded at site 2 shows strong correlations with O3 (R = 0.78), SO2 (R = 0.8), PM2.5 (R = 0.88), and PM10 (R = 0.9). An increase in temperature results in dry air, which might aggregate higher concentrations of PM2.5, and PM10 in a city environment. The relationship between temperature and O3, SO2, and PM2.5 is similar with the results observed by Mahmoud while exhibiting stronger degrees of correlation [67]. O3 has a strong correlation with PM2.5 (R = 0.8) and PM10 (R = 0.82) at site 2. There is a moderate correlation between temperature and TVOC (R = −0.65) at site 1. Similarly, temperature and TVOC share a moderate correlation (R = −0.72) at site 2. The similar negative correlation between TVOC and temperature within an office space was observed by Otolorin et al. [68]. Site 1 displays a moderate correlation between humidity and SO2 (R= −0.62), while site 2 exhibits a comparatively stronger correlation (R= −0.8). SO2 has the tendency to react with water vapor present in building walls and furniture, which may lead to decreasing concentrations of SO2 as the humidity increases [12,13,20]. There is a moderate correlation (R = 0.58) between humidity and CO2 at site 1 and a strong correlation (R = 0.78) at site 2. Site 1 has a low correlation between NO2 and CO2 (R = 0.47), while site 2 exhibits a relatively stronger correlation (R = 0.77). This might be due to human activities in site 1 and traffic-related pollution at site 2. This correlation between NO2 and CO2 is consistent with the study performed by S.H. Hwang [69]. There is a strong correlation at site 2 between humidity and NO2 (R = 0.98) and a relatively weaker correlation at site 1 (R = 0.45). At site 1 there is a moderate correlation between CO2 and O3 (R = −0.5) while site 2 displays a stronger correlation (R = −0.9) and this may be because of insufficient ventilation [6,15,70]. Site 2 registered a moderate correlation between humidity and TVOC (R = 0.72). The NO2 values of site 2 show a moderate correlation with TVOC (R = 0.73) and have a relatively stronger correlation with humidity (R = 0.98). This may be attributed to the presence of high-emitting building materials and fuel-burning household appliances at the residential unit [17,20,66].
(site 1) (site 2) Figure 6. Spearman correlation heat map for indoor environmental parameters at two sites. Figure 6. Spearman correlation heat map for indoor environmental parameters at two sites.

Conclusions
In this paper, a comprehensive environmental monitoring system is presented which detects most of the critical indoor air pollutants. Multiple independent sensors were selected accounting for repeatability, tolerance, and response time. These individual sensor modules were integrated onto a Python-powered Raspberry Pi platform to form a single LCAQS system to measure, analyze, and determine indoor air quality. The multisensor module recorded environmental parameters such as temperature, relative humidity, PM 2.5 , PM 10 , NO 2 , SO 2 , CO 2 , CO, O 3 , and TVOC in real-time at regular 10-min intervals. The logged data was extracted from the LCAQS system and transferred to a computer using an open-source network file transfer application, PuTTY.
It was observed that both PM 2.5 and PM 10 concentrations were significantly higher in the residential apartment unit (site 1) than at the office space (site 2) while the former recorded maximum values of 50.5 and 80.9µg/m 3 , respectively. Sites 1 and 2 registered to mean NO 2 levels of 41.8 and 60.3 ppb, respectively, while site 2 displayed a wider range of observed data between 43.8 and 132 ppb. The highest difference in concentration levels was observed with CO 2 , as site 1 (2195 ppm) exhibited CO 2 values that were more than five times greater than those of site 2 (432 ppm). Although both sites 1 and 2 displayed comparatively close levels of SO 2 at 39.9 and 29.6 ppb, respectively, the maximum concentration was observed at site 1 at 161 ppb. The mean O 3 concentration recorded at site 1 at 12.1 ppb was significantly higher compared to site 2 at 2.37 ppb. Both sites registered the same maximum TVOC value of 250, while site 1 displayed a slightly higher average value of 139.2 to that of site 2 at 121.
Spearman rank-order analysis indicates that PM 2.5 , and PM 10 has a strong correlation with SO 2 at both site 1 and site 2. The potential reason for this can be cooking activities in site 1 and vehicular pollution at site 2, which both contribute to increasing levels of PM 2.5 , PM 10 , and SO 2 . O 3 concentrations at site 2 exhibit a strong correlation with PM 2.5 and PM 10 . NO 2 and CO 2 show a relatively stronger correlation at site 1 when compared to site 2, and this might be due to human activities in site 1 and combustion-related pollution at site 2. Similarly, NO 2 and TVOC display moderate correlation at site 2. There is a moderate correlation between CO 2 and O 3 at site 1, while site 2 shows a relatively stronger correlation, and this might be due to insufficient ventilation. The possible effect of an increased number of human occupants on environmental pollutants can be witnessed at site 2, which displays strong temperature and humidity correlation with O 3 , SO 2 , PM 2.5 , and PM 10 when compared to site 1.
The constituent sensors of the custom-built LCAQS system utilized in this study rely on factory calibration. It is a challenge to individually calibrate all the sensors owing to limited availability and high cost of calibration equipment in the market for certain pollutants. Owing to the latter, limited efforts have been taken to validate the obtained results with a reference monitor. For future studies, sensor parameters such as such as precision, short and long term drift, and power supply effect can be tested by devising operational rigidity experiments on the sensors to ensure the accuracy of obtained results. Furthermore, the correlation values derived in this paper may need to be examined further to arrive at a conclusive inference. This can be achieved by monitoring air quality for longer periods across different locational settings accounting for varied surroundingrelated influences. Moreover, Principal Component Analysis (PCA) can be undertaken to reduce the multicollinearity of measured parameters. Also, mixed-effect linear regression models are suggested to better study the association between indoor air pollutants and building characteristics.