Verification and Usability of Indoor Air Quality Monitoring Tools in the Framework of Health-Related Studies

Abstract

Indoor air quality (IAQ) significantly impacts human health, particularly in enclosed spaces where people spend most of their time. This study evaluates the performance of low-cost IAQ sensors, focusing on their ability to measure carbon dioxide (CO₂) and particulate matter (PM) under real-world conditions. Measurements provided by these sensors were verified against calibrated reference equipment. The study utilized two commercial devices from inBiot and Kaiterra, comparing their outputs to a reference sensor across a range of CO₂ concentrations (500–1200 ppm) and environmental conditions (21–25 °C, 27–92% RH). Data were analyzed for relative error, temporal stability, and reproducibility. Results indicate strong correlation between low-cost sensors (LCSs) and the reference sensor at lower CO₂ concentrations, with minor deviations at higher levels. Environmental conditions had minimal impact on sensor performance, highlighting robustness to temperature and humidity within the tested ranges. For PM measurements, low-cost sensors effectively tracked trends, but inaccuracies increased with particle concentration. Overall, these findings support the feasibility of using low-cost sensors for non-critical IAQ monitoring, offering an affordable alternative for tracking CO₂ and PM trends. Additionally, LCSs can assess long-term exposure to contaminants, providing insights into potential health risks and useful information for non-expert users.

1. Introduction

Indoor air quality (IAQ) has gained significant importance in recent years due to its impact on human health, particularly considering that, on average, a person spends between 80% and 90% of their time indoors. Poor IAQ is associated with a range of adverse health effects, including cardiovascular diseases, respiratory illnesses, and mental health disorders [1,2,3], contributing to approximately 3.2 million deaths in 2020 [4,5,6]. This prolonged exposure to enclosed environments highlights the need to monitor and maintain good air quality. Despite the growing awareness of outdoor air pollution, indoor environments can be up to five times more polluted [1,2], making indoor monitoring in places such as homes, schools, hospitals, and senior homes a crucial area of research, especially in enclosed spaces where ventilation would be very limited. It is essential not only to react quickly when air quality levels deteriorate but also to develop proactive solutions that help prevent exposure to harmful pollutants. Monitoring IAQ is thus essential to mitigate health risks and promote well-being. Despite the availability of high-cost reference-grade sensors, their widespread use is limited by financial and logistical barriers, particularly in non-critical environments such as residential and educational settings. Consequently, low-cost sensors (LCSs) have gained traction due to their affordability, compact size, and potential for widespread deployment.

Among the diverse pollutants present in indoor air, particulate matter (PM) has been extensively studied for its health impacts, with recent findings emphasizing its role as a carrier for viruses and other harmful pollutants, potentially enhancing the transmission of infectious diseases in enclosed environments. PM_2.5, in particular, poses significant risks due to its ability to penetrate deeply into the respiratory system, contributing to cardiovascular, respiratory, and other systemic diseases [7,8]. Sources of indoor PM include activities such as cooking, heating, smoking, and the use of cleaning products, as well as external pollution from traffic and industrial emissions entering through ventilation systems [8]. Moreover, the synergistic effects of PM with other pollutants, such as volatile organic compounds (VOCs) and nitrogen dioxide (NO₂), amplify its impact on human health, further complicating the issue [7].

Volatile organic compounds (VOCs) are chemical compounds that volatilize easily at room temperature and are found in products such as disinfectants, paints, and furniture. Among these, formaldehyde is of particular concern due to its toxicity and potential to irritate the respiratory and ocular pathways, as well as its suggested carcinogenic risk following prolonged exposure [9]. Recent studies have demonstrated that long-term exposure to these chemicals, even at low concentrations, has been linked to both short-term health effects like sick building syndrome, a condition that causes symptoms such as eye, nose, and throat irritation, fatigue, and headaches [10], and long-term risks, including cancer [2]. The wide variability in the types and concentrations of VOCs present in indoor environments significantly complicates their evaluation, making it a challenge to assess their impacts comprehensively with current monitoring technologies. This underscores the need for more advanced and adaptable methods to accurately characterize and mitigate their potential health risks.

Carbon dioxide (CO₂), meanwhile, is primarily produced as a by-product of human respiration. While not toxic at low concentrations, elevated indoor levels may indicate poor ventilation. Prolonged exposure to high concentrations can lead to headaches and fatigue [11]. Beyond these health effects, research has also shown that certain elements of mental health and well-being, such as anxiety, reduced productivity in academic and work environments, and severe mental disorders, are affected by IAQ [12,13]. Additionally, CO₂ is increasingly recognized not only as an indicator of ventilation efficiency but also as a potential marker of airborne disease transmission risk, as higher concentrations indicate insufficient ventilation, which can lead to an accumulation of infectious aerosols [14]. Therefore, maintaining low CO₂ levels in indoor environments is therefore crucial to reducing the transmission risk of respiratory pathogens.

Traditionally, IAQ monitoring has relied on expensive and high-maintenance reference instruments that accurately measure pollutants such as PM, gases like CO₂, VOCs, and other harmful gases. However, these systems are often inaccessible to the general public due to their higher costs [4,5]. In response, low-cost sensors (LCSs) have emerged as an affordable solution for real-time pollutant monitoring [5], offering the potential for widespread deployment in homes and public buildings [15]. Their lower cost enables broader installation compared to traditional devices, promoting use in homes, offices, schools, and hospitals. Additionally, their compact size and portability facilitate the creation of distributed sensor networks, providing data with greater spatial and temporal coverage, thereby enabling more effective IAQ control. These LCSs are also compatible with IoT technologies, allowing real-time data transmission and visualization through mobile apps and online platforms. This not only facilitates continuous monitoring and rapid response to pollution spikes but also enables users to identify patterns and make informed decisions to improve IAQ [2]. Additionally, the integration of LCSs into smart building systems is a promising avenue for real-time IAQ management [2,16].

In this context, LCSs are becoming indispensable for monitoring these pollutants, but their accuracy, reliability, and long-term performance in real-world environments remain unresolved [2,16]. While initial verification in controlled environments is essential, it is equally important to evaluate their performance after some time of operation to ensure that they provide useful and reliable data during their lifetime. So, verification against high-cost reference equipment is crucial, as LCSs are prone to inaccuracies caused by environmental factors such as temperature (T) and humidity (RH) [3,16]. The calibration of LCSs has been widely discussed in recent literature, with various methods proposed to improve their performance. Colocation calibration with reference instruments, where LCSs are placed next to reference instruments to adjust their readings, has been shown to be effective in both indoor and outdoor environments [16,17]; however, this approach does not account for sensor drift over time or environmental variability [16,18].

Many low-cost PM sensors, including those used in this study, rely on optical particle counting (OPC) technology, which estimates particle mass concentration based on laser scattering. While this approach is widely used due to its cost-effectiveness and practicality, the accuracy of the readings is heavily dependent on the algorithms employed to convert particle counts into mass concentrations [19]. However, the lack of standardization in their proprietary algorithms often leads to significant variability in reported concentrations across different devices. This variability becomes particularly evident when comparing measurements from different sensors under varying aerosol compositions, as each algorithm processes particle size distributions and densities differently, introducing biases specific to certain particle types. As highlighted by Wallace [20], such algorithms often rely on simplified models that may overlook key factors like the proportion of particle sizes or aerosol heterogeneity, resulting in systematic over- or underestimations. Addressing these discrepancies is critical for enhancing the reliability of low-cost sensors and ensuring their effective integration into indoor air quality monitoring networks. As a result, while these sensors offer valuable insights into IAQ, it is essential to understand their limitations, especially when evaluating health-related risks linked to PM exposure.

Many CO₂ sensors, including those used in this study, employ advanced algorithms to maintain calibration and ensure accuracy over time. These algorithms ensure that the sensors remain accurate across a range of concentrations, though they may introduce variability in performance, particularly when internal recalibrations are triggered or when the environmental conditions deviate from the ideal assumptions of the algorithm.

Moreover, the role of LCSs in citizen science initiatives has also been explored, where non-experts use these sensors to monitor indoor air quality in their local environments. This democratization of air quality data has increased public engagement with environmental issues and provided large datasets that can complement traditional air monitoring networks [1,2]. However, this approach presents challenges, as data from LCSs often require significant processing before they are suitable for quantitative analysis, as environmental factors like humidity can introduce inaccuracies [16,18].

The importance of effective communication of indoor air quality data to non-expert users is essential, especially with the rise of citizen science projects and the growing use of LCSs by the general public. Studies have highlighted the need for clear communication strategies that translate technical air quality data into actionable information for the general public [17]. For example, color-coded systems and simplified air quality indices can help non-experts understand the risks associated with poor IAQ without overwhelming them with technical details [18]. Mobile applications and web platforms can also provide real-time updates and personalized recommendations to improve indoor air quality [3]. This involves not only ensuring the quality and accuracy of the data but also presenting them in a way that is accessible and meaningful for non-experts, allowing them to make informed decisions regarding their exposure to indoor air pollutants [16,21].

This report is focused on the verification of several IAQ sensors in real environments. The main objective is to assess their performance under everyday conditions and to determine their accuracy and consistency in detecting key pollutants. In this study, two low-cost IAQ monitoring devices (inBiot and Kaiterra) are verified by comparing them with reference-grade equipment under a variety of indoor conditions. Additionally, the study seeks to explore how these devices can be employed, acknowledging their limitations, for the ultimate goal of evaluating the impact of IAQ on human health, following several procedures [22]. This study advances the ongoing discussion on the role of LCSs in IAQ monitoring and provides insights into how these devices can be effectively used in real-world settings.

Moreover, the findings have implications for improving the utility of LCSs in longitudinal health-related studies and predictive modeling. For instance, insights derived from stable sensor performance over extended periods can inform the design of distributed IAQ monitoring networks in homes, schools, and hospitals, providing actionable data for both end-users and researchers. This study also identifies key limitations and avenues for future work to optimize the accuracy and durability of LCSs, contributing to their broader adoption in IAQ management strategies.

This research contributes to the objectives of the K-HEALTHinAIR project (https://k-healthinair.eu/ accessed on 22 November 2024), which aims to enhance our understanding of IAQ determinants and develop practical solutions to reduce exposure to harmful pollutants in indoor environments across Europe. By assessing the stability and accuracy of sensors after 6000 h of operation, this research provides critical insights into their durability and practical value for indoor air quality monitoring. These findings not only validate the feasibility of deploying such devices for extended periods but also highlight areas for improvement, contributing to the ongoing development of accessible and scalable air quality monitoring solutions.

2. Materials and Methods

2.1. Reference Equipment

For the accurate characterization of events and sensor verification, the following reference equipment was used:

• A Testo 440 multifunctional Air Quality Meter (Testo Industrial Services GmbH, Kirchzarten, Germany) was used for CO₂ measurement as the reference instrument. It is equipped with a CO₂ probe (Model 0632 1552) and was employed as the reference instrument to evaluate the performance of the low-cost sensors under study. This device integrates temperature and humidity sensors, providing high accuracy and reliability for IAQ measurements. Calibration was certified by ENAC (October 2024), ensuring traceability to international metrological standards. The calibration process included testing under controlled laboratory conditions (20 ± 5 °C and <80% RH) across multiple reference points to verify its precision. The expanded uncertainty of the CO₂ measurements was determined according to EA-4/02 M:2022 guidelines (European Accreditation EA, 2022 [23]), with a 95% confidence level. This robust certification guarantees the suitability of the Testo 440 as a reliable reference for verifying the accuracy and performance of other air quality monitoring devices.
• A PCE-PQC 34 Air Quality Monitor (PCE Deutschland GmbH, Meschede, Germany) was employed for PM_2.5 measurement as the reference instrument to evaluate the performance of the low-cost PM sensors included in this study. This device features advanced particle counting technology, capable of measuring concentrations of particulate matter with a diameter < 10 μm (PM10), particulate matter with a diameter < 5 μm (PM5), particulate matter with a diameter < 2.5 μm (PM_2.5), and particulate matter with a diameter < 1 μm (PM1), with a flow rate of 2.83 L/min and a detection range from 0.3 to 25 μm. The system utilizes a long-life laser diode as the light source, ensuring high accuracy and repeatability in particle detection. Calibration of the instrument is traceable to standards, with particle size bins pre-configured for factory calibration, enabling compliance with ISO 21501-4 [24] (December 2022). The PCE-PQC 34’s high sensitivity and built-in mass concentration mode allow for precise particle mass calculations in μg/m³, making it a reliable reference for validating the accuracy of low-cost PM sensors under real-world conditions.

2.2. Low-Cost Sensors

This study evaluates two low-cost indoor air quality (IAQ) monitoring devices: inBiot (MICA) and Kaiterra (Sensedge Mini). These sensors were selected due to their wide availability in the European market and their inclusion of critical features for indoor air quality (IAQ) monitoring. Both devices are certified under standards such as RESET and WELL, ensuring their compliance with recognized building and health standards. Additionally, these sensors represent two distinct manufacturers, enabling a comparative evaluation of performance across brands. The sensors were tested against reference-grade equipment to assess their accuracy, consistency, and applicability for monitoring IAQ in non-critical environments.

Table 1. Comparative Specifications of IAQ Monitoring Devices.

Device	MICA (inBiot) [25]	Sensedge Mini (Kaiterra) [26]
Measurement Parameters	CO2, PM1, PM2.5, PM4.0, PM10, Formaldehyde, TVOC, Temp, RH	CO2, PM2.5, PM10, TVOC, Temp, RH
CO2 Accuracy	±50 ppm + ±3% (0–5000 ppm)	±40 ppm + ±3% (400–10,000 ppm)
CO2 technology	NDIR sensor	NDIR sensor
PM Accuracy	±5 μg/m3 + ±5% (0–100 μg/m3); ±10% (100–1000 μg/m3)	±3 μg/m3 (0–30 μg/m3); ±10% (30–1000 μg/m3)
PM technology	laser-based optical sensor	Laser-based optical sensor
Certifications	RESET, WELL	RESET, WELL
Connectivity	My inBiot platform	Mobile app and web platform
Estimated Cost (€)	~500	~750

below provides a detailed comparison of the specifications and features of the two devices, highlighting their measurement parameters, accuracies, certifications, and connectivity options.

The MICA and Sensedge Mini sensors utilize laser scattering technology for particulate matter (PM) measurement. This method estimates particle mass concentrations by analyzing the scattering intensity of a laser beam when it interacts with airborne particles. For PM_2.5, some sensors may rely on proprietary algorithms that estimate concentrations based on the detected size distribution of PM₁₀ and other parameters. These algorithms, while effective for trend monitoring, can introduce variability depending on the aerosol composition and density.

All IAQ monitors used had a minimum of 6000 h of previous operation. These devices were selected to evaluate how they behaved after a significant period of use. The sensors performed measurements every 5 s, and the data were averaged over one-minute intervals. These averaged values were collected from their respective platforms, providing a high-resolution overview of the monitored IAQ parameters.

Two sensors of each model were selected for the study, hereafter named INB01, INB02 for inBiot sensors and KAI01, KAI02 for Kaiterra sensors. The measured variables were the concentrations of PM₁₀, PM_2.5, PM₁, CO₂, T, and RH, which were collected every minute. These data formed a dataset resulting from the measurements by the four sensors involved in the study and the fixed reference (hereafter named REF) from September to November 2024.

2.3. Chamber for Performance Tests

The experiments were conducted in a hermetically sealed 1 m³ test chamber equipped with a lateral door for sample introduction and retrieval, a ceiling-mounted agitation system, and dedicated inlets for injecting contaminants. To simulate realistic indoor air quality scenarios, CO₂ levels were increased by researchers exhaling within the chamber, achieving approximately the concentrations required for the tests. PM was generated using separately lit candles. The controlled environment of the chamber allowed for the precise monitoring of IAQ parameters using reference-grade instruments. A photograph of the test chamber setup is provided in Figure 1 to illustrate the experimental configuration.

Standard gases with known concentrations and monodisperse particulate matter (PM) were not used in this study for several reasons. First, the use of reference-grade equipment, calibrated to international standards, and a controlled test chamber ensured the homogeneity of air quality parameters across all sensors during measurements. Second, the study employed realistic sources of particulate matter emissions, such as candles, which are representative of conditions typically encountered in indoor environments. This approach enhances the applicability of the findings to real-world scenarios. Finally, the use of standard gases and monodisperse PM introduces significant logistical and financial costs, which may limit their feasibility in studies focused on evaluating low-cost sensors for practical applications. By relying on a combination of reference equipment and controlled emission sources, this study maintains a balance between rigor and real-world relevance.

2.4. CO₂ Sensor Verification

This study aimed to verify four CO₂ sensors used for IAQ monitoring by comparing them to a reference-grade device. The reference sensor used in this study was the TESTO 400, which was professionally calibrated according to ENAC standards. The verification process involved exposing the test sensors to stable indoor conditions, measuring CO₂ concentrations between 500 ppm and 1200 pm, as well as other environmental parameters such as T and RH. To simulate realistic IAQ scenarios, CO₂ levels were increased by researchers exhaling within the chamber, achieving approximately the concentrations required for the tests. Once CO₂ was introduced into the chamber, the chamber was then sealed for a sufficient time to allow the CO₂ levels to stabilize. Adequate time, around 1 h, was given to collect sufficient and representative data.

The tests were conducted over stable periods of CO₂ concentration, allowing for the analysis of average deviation (the mean difference between the reading of each sensor and the reference sensor in ppm), percentage error (the relative error in percentage of each sensor compared to the reference sensor), and temporal stability of the CO₂ sensor readings (the variability in sensor readings over time, measured as the standard deviation of the CO₂ concentrations.). The data were compared to those of the reference sensor to evaluate the precision of the sensors.

2.5. PM Sensor Verification

For this study, a candle was used as the primary source for generating particles due to its affordability and the relative ease of controlling the particle profiles by using burn time as the main parameter to adjust the particle levels for testing. However, additional tests were conducted with particles of varying characteristics in isolated experiments to evaluate differences in the calculation algorithms employed by each manufacturer.

The tests varied as follows. A lit candle was placed inside the chamber, where all sensors, including the reference sensor, were located. The candle remained lit for a controlled period to generate PM. After the designated time, the candle was removed and extinguished outside the chamber. This is because extinguishing the flame in an uncontrolled manner indoors would generate larger particles that could interfere with the measurements of smaller particles. The chamber was then sealed and left unopened for 2 h to complete the test. PM data were extracted from the reference sensor, processed, and analyzed using Microsoft Excel.

As previously mentioned, particle density and size distribution are critical factors in the quantification of airborne particles. The density of particles emitted by a candle has been reported as 1.6 g/cm³ [27]. Using the same test chamber in future works, other alternative methods will be employed to generate aerosols (water, 1 g/cm³, and Titanium Dioxide P25, 4.3 g/cm³) in air, although these methods produce fewer stable aerosols and are more challenging to standardize for experiments.

2.6. Data Analysis Methodology

The data analysis in this study focused on assessing the accuracy, precision, and variability of the low-cost sensors by comparing their measurements to those obtained from calibrated reference equipment under controlled conditions. The key steps in the methodology included the following:

• Error Calculation: Absolute error was computed as the difference between the low-cost sensor reading and the reference equipment measurement (i.e., |500 − 400| = 100 ppm). Relative error was then calculated as the percentage of the absolute error relative to the reference value (i.e., (100/500) × 100 = 20%).
• Temporal Stability Assessment: For each parameter (e.g., CO₂, PM), measurements were collected over extended periods to evaluate the temporal stability of the sensors. Variability was quantified using standard deviation and interquartile range (IQR) metrics.
• Comparison with Intrinsic Accuracy: In the error assessment process, the intrinsic accuracy of both the reference device (inherent uncertainty stated in its calibration certificate and interpolated with the verified values) and the low-cost sensors was considered. For each measurement, the specified accuracy ranges of the devices were used to establish acceptable discrepancies. To interpret such discrepancies, the combined accuracy of both devices was considered as a potential source of variability. For example, for the INB01 with a measurement of 473 ppm (trial #1) CO₂, the expected range would be 50 + 0.03 × 473 + 13 (from the reference device) = 77 ppm. If the reference value for this trial is 493 ppm, the measurement is OK. This approach allowed accounting for the inherent uncertainties in the measurements and assessing whether the observed errors exceeded the expected ranges based on device specifications.
• Graphical Representation: Box plots and scatterplots were used to visualize the error distributions, highlighting median values, interquartile ranges, and outliers for each sensor.

3. Results

3.1. CO₂

Several trials were conducted with different values of CO₂ from 500 to 1200 ppm approximately. The averaged readings for each trial from the calibrated reference sensor (REF) and the test sensors (INB01, INB02, KAI01, and KAI02) are in

Table 2. Summary of the results of reference tool and studied sensors for CO₂.

			REF	INB01		INB02		KAI01		KAI02
Trial	T	RH	CO2 (ppm)	CO2 (ppm)	Errrel (%)	CO2 (ppm)	Errrel (%)	CO2 (ppm)	Errrel (%)	CO2 (ppm)	Errrel (%)
#1	21	44	493	473	4.13	433	12.23	463	6.23	462	6.39
#2	23	45	516	486	5.88	494	4.33	497	3.71	489	5.29
#3	25	72	579	570	1.49	537	7.25	546	5.67	554	4.23
#4	22	41	582	557	4.32	509	12.53	546	6.24	543	6.69
#5	24	37	584	546	6.38	543	6.88	549	5.91	551	5.60
#6	25	81	677	658	2.79	617	8.89	636	6.03	642	5.10
#7	25	27	744	699	6.01	659	11.43	683	8.15	692	6.91
#8	24	59	766	734	4.18	722	5.75	702	8.36	724	5.56
#9	25	64	794	765	3.66	725	8.75	739	6.93	747	5.98
#10	22	50	730	710	2.70	678	7.13	683	6.41	671	8.08
#11	24	78	982	946	3.68	919	6.46	907	7.69	921	6.26
#12	24	92	1035	1003	3.06	977	5.66	1044	0.89	974	5.92
#13	22	43	1190	1125	5.46	1081	9.17	1106	7.06	1090	8.43

.

To evaluate the precision of the sensors and to verify them, averaged data during the whole trial were compared to those from the reference sensor. The difference between the reference sensor and the sensor under study was determined by the uncertainties of both devices. If this difference was within this range, the sensor was deemed to comply; otherwise, it was considered non-compliant. As is shown in Table 1, the low-cost sensors showed a general alignment with the reference sensor, particularly at lower concentrations of CO₂. Deviations from the reference sensor become more pronounced as the CO₂ concentration increases. In spite of these differences, all sensors appeared to be within a similar range, and all of them complied with the verification specifications. All experiments described in Table 1 were conducted under stable CO₂ concentrations within the chamber, except for experiment #10, where intentional fluctuations were introduced to evaluate sensor performance under varying conditions. Additionally, Figure 2 shows the graphical evolution of CO₂ concentration during the whole trial. As examples, trials #6 (a) and #10 (b) have been chosen.

The readings from all sensors were plotted against those of the reference sensor to visualize their alignment in Figure 3, which shows that all sensors had a strong positive correlation with the reference sensor (R² = 0.999 for INB01, R² = 0.996 for INB02, R² = 0.994 for KAI01, and R² = 0.999 for KAI02). The low-cost sensors studied can reliably track CO₂ concentration trends across the measured range, exhibiting higher accuracy at lower CO₂ concentrations (400–600 ppm), with measurements closely aligning with the ideal line. At higher CO₂ levels (above 800 ppm), some sensors demonstrated increasing deviation from the ideal line. These deviations may be attributable to manufacturing tolerances or minor calibration differences but do not appear significant, as all the sensors complied with the verification compared with the reference sensor. So, these results suggest that low-cost sensors can serve as viable alternatives for monitoring CO₂ concentrations in non-critical environments. Further calibration may enhance their performance in applications requiring higher CO₂ measurements. Also, sensors from the same brand (inBiot and Kaiterra) appear to produce fairly consistent readings among themselves (INB01 vs. INB02, KAI01 vs. KAI02), indicating strong reproducibility within each manufacturer’s sensor model line.

Figure 4 highlights the precision and spread of errors for each sensor, with some showing larger variability. The sensors showed varying levels of relative error, with median errors generally falling between 4% and 8%, which can be observed also in Table 1. This indicates that while the sensors are reasonably accurate, there is room for improvement in precision. The variability in relative errors, as shown by the interquartile ranges (IQRs), differs among sensors. The box plot represents the lower quartile (25%) and upper quartile (75%), defining the interquartile range (IQR), with the line inside the box indicating the median. The ‘whiskers’ extend to the minimum and maximum values within 1.5 times the IQR from the quartiles, and points beyond this range are considered outliers. Some sensors exhibit narrower error distributions, reflecting more consistent performance, while others show wider spreads, indicating greater variability in their measurements. For example, INB01 displays the lowest median relative error (4%), with a small IQR, and INB02 exhibits the largest errors, with a median close to 8% and a wider IQR. Both KAI sensors have similar error distributions, with medians around 6%. While all sensors are suitable for basic monitoring purposes where small deviations from the reference are tolerable, those with smaller errors and tighter distributions are better candidates for applications requiring higher accuracy and reproducibility.

It can be seen in Figure 5 that the relative error for all sensors remains relatively stable across the temperature range (21–25 °C), suggesting that temperature has a negligible effect on sensor performance within this range, as no clear trends were observed, indicating that temperature variability in this study was not a significant factor influencing accuracy. The negligible correlation with temperature (0.088) implies the same; the sensors were not sensitive to temperature variations within the tested range. Likewise, the relative error does not show a clear trend with changes in relative humidity (27–92%), indicating that humidity had no measurable influence on sensor accuracy within the tested range. The errors appear evenly distributed regardless of humidity levels, which reinforces the conclusion that the sensors were not sensitive to variations in relative humidity under these conditions. Moreover, the low positive correlation (0.38) with relative humidity shows the same. So, the sensors demonstrated good robustness to temperature and humidity variations within the tested range.

3.2. PM

3.2.1. PM₁ Results

Several trials were conducted to evaluate the performance of the low-cost sensors (INB01 and INB02) compared to the reference sensor (REF) under varying conditions.

Table 3. Summary of the results obtained with the reference tool and the studied sensors for PM₁.

			REF	INB01		INB02
Trial	T	RH	PM1 (μg/m3)	PM1 (μg/m3)	Errrel (%)	PM1 (μg/m3)	Errrel (%)
#1	27	46	10.12	8.67	14.29	7.82	22.72
#2	27	42	12.98	10.32	20.49	10.15	21.77
#3	27	51	13.07	13.21	1.08	12.10	7.45
#4	26	52	14.62	12.59	13.86	12.39	15.20
#5	26	45	18.22	14.10	22.59	14.42	20.85
#6	28	40	27.35	26.20	4.21	25.20	7.87
#7	27	50	55.38	56.98	2.89	55.20	0.32
#8	26	43	63.73	59.83	6.12	59.69	6.35

summarizes the averaged PM₁ concentrations measured during each trial, along with the relative errors for both INB01 and INB02. Some deviations are observed, particularly at lower concentrations, where INB01 and INB02 exhibit larger relative errors, reaching up to ~22.7%. Despite these deviations, all PM₁ measurements obtained from the low-cost sensors were within the accuracy ranges specified by their manufacturers, demonstrating compliance across all tested conditions.

Figure 6 shows the graphical evolution of PM₁ concentration during the whole trial (example trial #3 has been chosen).

Figure 7 illustrates the alignment between the average data from the sensor measurements and the ideal line to assess sensor precision and verify their accuracy. As can be seen, both sensors INB01 and INB02 track PM₁ concentrations effectively across the measured range, demonstrating a high correlation with the reference sensor (R² = 0.996 for INB01, and R² = 0.998 for INB02), indicating good consistency in their readings. Moreover, both sensors comply with the manufacturer’s specifications for PM₁, which state an accuracy of 5 μg/m³ ± 5% of the value (0–100 μg/m³) and ±10% of the value (100–1000 μg/m³), further verifying their performance within the tested conditions.

Figure 8 illustrates the distribution of relative errors for each sensor. The black line with a cross indicates the mean relative error for each sensor. Additionally, the mean value obtained from the reference equipment is included as a baseline, without any associated error, serving as a benchmark for comparison. This measure complements the median values shown within the box plots, offering a clearer picture of overall accuracy. The distribution of relative errors highlights that INB01 generally performs with lower median errors (~6%), while INB02 exhibits a wider variability in its error distribution, with medians closer to ~8%.

3.2.2. PM_2.5 Results

The data for PM_2.5 were collected during the same trials previously described for PM₁, using the four low-cost sensors (INB01, INB02, KAI01, and KAI02) alongside the reference sensor (REF) under controlled indoor conditions.

Table 4. Summary of the results obtained with he reference tool and the studied sensors for PM_2.5.

			REF	INB01		INB02		KAI01		KAI02
Trial	T	RH	PM2.5 (μg/m3)	PM2.5 (μg/m3)	Errrel (%)	PM2.5 (μg/m3)	Errrel (%)	PM2.5 (μg/m3)	Errrel (%)	PM2.5 (μg/m3)	Errrel (%)
#1	27	46	26.43	9.36	64.59	8.51	67.81	16.89	36.12	15.67	40.71
#2	27	42	31.22	11.17	64.23	11.00	64.77	23.72	24.03	22.80	26.97
#3	27	51	22.32	13.84	38.00	12.72	42.99	30.67	37.45	27.20	21.87
#4	26	52	31.74	13.38	57.86	13.21	58.38	34.18	7.67	32.79	3.28
#5	26	45	41.81	15.28	63.45	15.52	62.89	30.18	27.81	32.35	22.63
#6	28	40	53.26	27.57	48.24	26.39	50.45	82.77	55.39	80.05	50.30
#7	27	50	109.75	60.14	45.20	58.39	46.80	137.00	24.83	135.98	23.90
#8	26	43	200.35	71.06	64.53	70.59	64.77	133.65	33.29	142.28	28.99

summarizes the PM_2.5 concentrations recorded during each trial, along with the relative errors for each sensor. As PM_2.5 levels increase, these sensors tend to underestimate the concentrations, with relative errors ranging between 38% and 65% in most trials for INB01 and INB02, respectively. Similarly, KAI01 and KAI02 show deviations at higher concentrations, but their error ranges are narrower, particularly that for KAI02, which displayed the lowest median error of all sensors (~27.3%). It is important to note that none of the sensors met the manufacturer’s specifications when compared to the reference sensor in any of the trials conducted.

Figure 9 shows the graphical evolution of PM_2.5 concentration during the whole trial (example trial #3 has been chosen). The remaining trials exhibited evolution graphs with a morphology similar to the one shown, but with varying intensities.

To assess the accuracy and precision of the sensors, the average values from each trial were compared to the reference sensor, accounting for their respective uncertainties (Figure 10) and showing that all the sensors exhibit a reasonable linear alignment with the reference sensor at lower concentrations. As PM_2.5 levels increase, these sensors tend to underestimate the concentrations. However, the correlation coefficients for the data from the low-cost sensors against the reference sensor were as follows: INB01 (R² = 0.954), INB02 (R² = 0.959), KAI01 (R² = 0.875), and KAI02 (R² = 0.903), indicating a strong linear relationship between the sensor readings and the reference measurements.

Figure 11 highlights the distribution of relative errors among the sensors. The box plot represents the lower quartile (25%) and upper quartile (75%), defining the interquartile range (IQR), with the line inside the box indicating the median. The ‘whiskers’ extend to the minimum and maximum values within 1.5 times the IQR from the quartiles, and points beyond this range are considered outliers. INB01 and INB02 exhibit higher relative errors (median errors ~50–65%), while KAI01 and KAI02 demonstrate improved performance, with medians around 25–35%. This suggests that the Kaiterra sensors are more precise for PM_2.5 monitoring compared to the inBiot models, particularly under higher particle concentrations. Overall, while all sensors show limitations in accuracy for PM_2.5, particularly at higher concentrations (>100 µg/m³), their strong correlation with the reference sensor and relative consistency make them suitable for non-critical IAQ monitoring. Further calibration and refinement are recommended for applications requiring higher precision, especially where the nature of the particles can be known.

3.2.3. PM₁₀ Results

Several trials were conducted to evaluate the performance of the four low-cost sensors compared to the reference sensor under controlled indoor conditions.

Table 5. Summary of the results obtained with the reference tool and the studied sensors for PM₁₀.

			REF	INB01		INB02		KAI01		KAI02
Trial	T	RH	PM10 (μg/m3)	PM10 (μg/m3)	Errrel (%)	PM10 (μg/m3)	Errrel (%)	PM10 (μg/m3)	Errrel (%)	PM10 (μg/m3)	Errrel (%)
#1	27	46	33.49	9.72	70.97	8.92	73.37	20.79	37.93	18.43	44.98
#2	27	42	69.89	27.57	60.55	26.45	62.16	89.86	28.57	86.02	23.08
#3	27	51	29.39	13.87	52.81	12.77	56.54	33.95	15.53	29.30	0.31
#4	26	52	40.90	13.57	66.81	13.34	67.38	38.87	4.97	36.38	11.06
#5	26	45	53.10	15.58	70.65	15.73	70.37	35.70	32.77	36.28	31.67
#6	28	40	69.89	27.57	60.55	26.45	62.16	89.86	28.57	86.02	23.08
#7	27	50	130.99	60.61	53.73	58.71	55.18	152.27	16.24	148.73	13.55
#8	26	43	289.33	81.50	71.83	80.61	72.14	167.19	42.22	171.91	40.58

summarizes the PM₁₀ concentrations recorded during each trial, along with the relative errors for each sensor. INB01 and INB02 tend to underestimate PM₁₀ concentrations as values increase, with relative errors ranging from ~52% to ~73%. On the other hand, KAI01 and KAI02 show better performance, particularly at moderate concentrations, but still demonstrate significant variability. KAI02 consistently exhibited the smallest relative errors, from ~0.3 to ~45%. It is important to note that, while most of the sensors did not meet the manufacturer’s specifications when compared to the reference sensor, the Kaiterra devices achieved compliance in some isolated trials. The remaining trials displayed evolution graphs with morphology similar to those shown, albeit with varying intensities.

Figure 12 shows the graphical evolution of PM₁₀ concentration during the whole trial (example trial #3 has been chosen). The remaining trials exhibited evolution graphs with a morphology similar to the one shown, but with varying intensities.

To assess the sensors’ precision and verify their performance, the average data from each trial were compared to those from the reference sensor in Figure 13, showing that all sensors exhibit increasing deviations at higher PM₁₀ concentrations. The correlation coefficients for the data from the low-cost sensors against the reference sensor are as follows: INB01 (R² = 0.953), INB02 (R² = 0.958), KAI01 (R² = 0.866), and KAI02 (R² = 0.888), indicating strong but slightly weaker linear relationships compared to PM_2.5.

Figure 14 highlights the distribution of relative errors across the sensors. The box plot represents the lower quartile (25%) and upper quartile (75%), defining the interquartile range (IQR), with the line inside the box indicating the median. The ‘whiskers’ extend to the minimum and maximum values within 1.5 times the IQR from the quartiles, and points beyond this range are considered outliers. INB01 and INB02 display higher variability and larger relative errors, whereas KAI01 and KAI02 demonstrate narrower error ranges, reflecting better consistency and precision.

Overall, the results suggest that while all sensors provide useful data for general monitoring purposes, their accuracy for PM₁₀ concentrations requires improvement, particularly at higher levels (>150 µg/m³). KAI02 stands out as the most reliable option among the tested sensors for PM₁₀ monitoring. Further calibration and refinement are recommended for applications requiring higher precision or stricter thresholds.

4. Discussion

The findings from this study underscore the potential of low-cost sensors (LCSs) for IAQ monitoring in non-critical environments. A key contribution of this research is the evaluation of sensor stability and accuracy after one year of continuous use in indoor office and residential settings, addressing a gap in the literature where most studies focus on newly manufactured sensors or short-term deployments. This long-term perspective is particularly relevant for real-world applications where devices are subject to environmental variability and wear over time.

An important observation from this study is that, even after over 6000 h of operation, the performance of the low-cost sensors did not show noticeable signs of deterioration. This highlights their potential for long-term use in monitoring indoor air quality. However, continued evaluation in future years will be necessary to confirm the stability of their performance over longer time periods.

This study demonstrates that low-cost sensors can maintain reliable performance after extended use, reinforcing their potential for long-term IAQ monitoring. The results provide valuable benchmarks for sensor deployment in residential and office settings and inform strategies for integrating these devices into distributed monitoring networks. By contributing to the characterization of IAQ in diverse environments, this research supports the development of cost-effective and scalable solutions for monitoring air quality at a granular level. Moreover, the findings highlight the importance of continued evaluation and calibration to ensure sustained reliability, particularly in high-pollution settings or when data are used for health-related studies.

The results obtained for CO₂ concentrations demonstrate that the tested low-cost sensors generally align well with the reference sensor, particularly at lower concentrations (400–600 ppm), where measurements closely follow the ideal line. As CO₂ levels increase beyond 800 ppm, some deviations from the reference are observed. These discrepancies, although minor, can likely be attributed to manufacturing tolerances or calibration differences. Importantly, all sensors complied with verification criteria, indicating their reliability for non-critical applications.

Notable differences in performance between sensor models were observed. For example, INB01 exhibited the lowest median relative error (~4%) with a narrow IQR, indicating high consistency. Conversely, INB02 showed a wider spread of errors, with median values near 8%, suggesting variability in manufacturing quality or calibration. The Kaiterra sensors displayed similar performance, with median relative errors around 6%, reflecting good reproducibility within this sensor model. These findings suggest that while all sensors are suitable for basic monitoring, those with tighter error distributions are better candidates for research scenarios requiring higher precision.

Environmental conditions, including temperature (21–25 °C) and relative humidity (27–92%), had negligible effects on sensor accuracy, as shown by the lack of clear trends and low correlation coefficients (0.088 and 0.38, respectively). While these results highlight the robustness of the sensors in typical indoor environments with moderate fluctuations, the testing conditions were intentionally limited to this range. Future studies should aim to expand testing to include more extreme conditions, such as higher temperatures and humidity levels, to gain a more comprehensive understanding of the sensors’ performance under diverse environmental scenarios. This would ensure broader applicability for various settings.

The results obtained for PM concentrations highlight both the potential and limitations of the tested low-cost sensors for IAQ monitoring. Across all trials, the sensors showed strong linear relationships with the reference sensor, with correlation coefficients ranging from 0.87 (PM₁₀) to 0.99 (for PM₁ only; inBiot). However, the performance varied significantly between models and at different concentration ranges.

At lower PM concentrations (especially for PM₁), all sensors provided relatively consistent readings compared to the reference sensor. As concentrations increased, INB01 and INB02 demonstrated a tendency to underestimate PM levels (PM_2.5 and PM₁₀), with relative errors exceeding 50% in several trials. In contrast, the Kaiterra sensors exhibited better alignment with the reference sensor, particularly KAI02, which consistently achieved the lowest relative errors. This suggests that while both sensor types can track general trends in PM concentrations, the Kaiterra models are more reliable for higher concentrations.

The analysis of relative error distributions further underscores the variability in sensor performance. INB01 and INB02 showed wider error ranges, indicating greater inconsistencies, while the Kaiterra sensors demonstrated narrower distributions, reflecting better reproducibility. This variability could be attributed to differences in the algorithms used by each manufacturer to process particle counts and calculate mass concentrations, as well as the calibration standards applied during sensor production. On the other hand, further analyses using different types of particles and with slightly broader environmental conditions are necessary to complete the study.

However, the study also acknowledges limitations, particularly regarding the accumulation of dirt on PM sensors in high-pollution environments, which could impact long-term accuracy. In this study, the sensors were primarily used in indoor office and residential environments, where exposure to high levels of pollutants was occasional. This context may have contributed to their durability, and results might differ in more extreme conditions, such as in industrial or urban areas with consistently high particulate matter concentrations. Future research should focus on testing LCSs under diverse environmental conditions and exploring hybrid approaches that combine LCSs with high-precision instruments for enhanced reliability.

By addressing these challenges and leveraging the strengths of LCSs, this study contributes to the broader objective of making IAQ monitoring accessible, scalable, and actionable for diverse stakeholders.

Relevance to Air Quality and Health Research

This study is being conducted as part of a broader project aimed at characterizing indoor air quality in diverse environments using these sensors. As the project progresses, ongoing use of these devices in real-world scenarios will provide further data on their durability and potential performance drift, offering deeper insights into their long-term reliability and applicability for extended deployments.

The study titled “Protocol for the Enhanced Management of Multimorbid Patients with Chronic Pulmonary Diseases: Role of Indoor Air Quality” [22] outlines a clinical trial aimed at improving the management of patients with chronic pulmonary conditions by integrating IAQ monitoring into their care. The protocol emphasizes the importance of continuous IAQ monitoring, including parameters such as PM and VOCs, to identify potential triggers for respiratory exacerbations.

In this context, the usability of low-cost CO₂ sensors becomes particularly relevant. While CO₂ itself is not a direct pollutant, its concentration levels serve as proxies for ventilation efficiency and occupancy rates, which can be indirectly related to the accumulation of indoor pollutants like PM and VOCs. By deploying these sensors in patients’ homes, healthcare providers can gain real-time insights into ventilation patterns and potential exposure to harmful pollutants.

These findings indicate that while low-cost sensors provide an accessible means for continuous PM₁₀ monitoring, their accuracy at higher concentrations may not be sufficient for applications requiring stringent precision, such as health impact studies or regulatory compliance. However, their affordability and strong correlation with reference sensors make them valuable for identifying trends, detecting peaks, and supporting baseline IAQ assessments in non-critical environments.

The integration of CO₂ sensors into the clinical study offers several advantages:

• Early Detection of Poor Ventilation: Elevated CO₂ levels can indicate inadequate ventilation, prompting timely interventions to improve air exchange and reduce pollutant build-up.
• Personalized Patient Education: Providing patients with feedback on their indoor CO₂ and PM levels can empower them to adopt behaviors that enhance IAQ, such as opening windows or using exhaust fans.
• Data-Driven Clinical Decisions: Continuous IAQ monitoring data can inform clinicians about environmental factors contributing to respiratory symptoms, allowing for tailored management plans.
• Resource Allocation: Identifying homes with consistently poor IAQ can help prioritize resources and interventions for patients at higher risk of exacerbations.

The results of this study support the use of low-cost CO₂ sensors as viable tools for monitoring ventilation efficiency and human occupancy in non-critical environments. Their ease of use, affordability, and reasonable accuracy make them particularly suitable for large-scale deployments in health-related IAQ studies. Future work should explore methods to enhance their performance at higher CO₂ concentrations, as well as their integration with other IAQ parameters to provide a comprehensive picture of indoor environments.

The integration of low-cost sensors into health-related studies offers a cost-effective approach to continuous IAQ monitoring. These sensors, with costs ranging from EUR 500 to EUR 1000 per measurement point, provide continuous temporal data that are invaluable for understanding IAQ dynamics over time. Although their precision may not match that of high-cost reference equipment, the ability to gather consistent, real-time data across multiple points makes them an attractive alternative for large-scale deployments. In contrast, monitoring using reference-grade instruments or sampling methods, while highly accurate, often entails significantly higher costs. Moreover, these methods frequently lack the temporal resolution needed for detailed IAQ studies, as they typically rely on periodic sampling rather than continuous measurement. This disparity highlights a critical trade-off: precision versus cost and temporal resolution.

Additionally, it is worth emphasizing that the sensors evaluated in this study had been in use for over a year prior to testing. This provides valuable insights into their long-term performance and highlights their durability under real-world operating conditions, making the results more representative of practical applications.

The Project aims to bridge this gap by developing strategies to effectively leverage low-cost sensors in IAQ monitoring for health studies. By understanding their limitations and optimizing their deployment, it is exploring ways to create robust frameworks that maximize the utility of these sensors while minimizing the impact of their inherent inaccuracies. For example, low-cost sensors can complement reference methods by providing baseline data and identifying periods of interest for more-detailed analysis with high-precision equipment.

Ultimately, this approach allows for more accessible and scalable IAQ monitoring, which is essential for identifying and mitigating the health impacts of poor indoor air quality. By integrating these sensors into research methodologies, a balance can be achieved between cost-efficiency, data richness, and actionable insights, advancing our understanding of the role of IAQ in health outcomes.

5. Conclusions

This study evaluates the performance and usability of low-cost sensors (LCSs) for monitoring IAQ, with a focus on CO₂ and particulate matter (PM₁, PM_2.5, and PM₁₀). The findings demonstrate the potential of these sensors for continuous IAQ monitoring in non-critical environments, while also highlighting areas for improvement to enhance their reliability in health-related studies. It is important to note that the sensors evaluated in this study were not newly manufactured but had been in use for over a year, providing insights into their real-world performance and durability under extended operation.

In conclusion, this study offers a comprehensive evaluation of low-cost sensors, demonstrating their stability and accuracy after 6000 h of operation. The findings provide a foundation for their integration into long-term IAQ monitoring networks, contributing to the democratization of air quality data collection and analysis. Future research should focus on extending these evaluations to a broader range of environmental conditions and exploring methods for real-time data adjustment to enhance sensor reliability further.

• Sensor Performance: Low-cost sensors showed a strong correlation with reference equipment across all tested parameters, particularly for CO₂ measurements at lower concentrations. However, deviations were observed at higher CO₂ concentrations and for PM measurements, particularly for PM_2.5 and PM₁₀, where underestimations were common.
• Environmental Robustness: Temperature and relative humidity had minimal impact on sensor performance under common environmental conditions, underscoring their robustness for diverse indoor environments. This suggests that low-cost sensors can reliably operate in real-world conditions, provided their limitations are accounted for.
• Usability in Health Studies: The integration of low-cost sensors into health-related studies offers a cost-effective approach to IAQ monitoring, enabling continuous data collection that supports the identification of trends and critical exposure periods. While the accuracy of these sensors may not match that of reference-grade equipment, their affordability and accessibility make them valuable tools for large-scale deployment in studies assessing the health impacts of IAQ.
• Regarding future directions, to enhance the usability of low-cost sensors in health-related research, several avenues for future work have been identified:
• Evaluation of Additional Sensors: Beyond the current focus on CO₂ and particulate matter, further evaluations will be conducted on other sensors integrated into IAQ monitoring devices, including those measuring formaldehyde and TVOCs. These additional parameters are critical for a more comprehensive assessment of indoor air quality and its potential health impacts.
• Development of Alarm Thresholds: Strategies will be developed to establish concentration thresholds provided by the sensors that can serve as early warning signals for short-term exposures. These thresholds will help identify critical moments where immediate action may be required to reduce exposure risks.
• Assessment of Long-Term Exposure: The data collected from these sensors will also be used to evaluate medium- and long-term exposure levels for patients in their homes. This analysis will provide insights into the chronic effects of IAQ on health and support personalized interventions based on individual exposure profiles.
• Sensor Integration and Calibration: Efforts will focus on refining calibration methods and integrating sensor data with reference-grade instruments to enhance data reliability and usability. This will include exploring hybrid approaches that leverage both low-cost and high-precision monitoring tools.

By pursuing these directions, this research will further advance the application of IAQ monitoring technologies in health studies, enabling more effective strategies for managing indoor environments and mitigating health risks associated with poor indoor air quality.