Analytical performance and calibration strategies of low-cost particulate matter sensors for indoor and workplace monitoring—a review

Abstract

This review summarizes current evidence on low-cost particulate matter (PM) sensors for indoor and occupational environments and proposes a framework that links performance evaluation, calibration, and uncertainty to decision-making. Results from laboratory and field co-locations are synthesized to define reporting standards—accuracy, precision, dynamic range, detection capability, and temporal response—and to compare calibration strategies. Optical sensors consistently capture temporal dynamics of indoor sources but show mass bias that depends on concentration range, aerosol composition, and humidity. Context-specific reporting, with conditioning on environmental state and source regime, is therefore essential. Calibration practices range from simple linear corrections, often adequate at low to moderate concentrations, to multivariate or nonlinear models incorporating humidity, temperature, or volatile organic compounds, which reduce residual bias under high or mixed-source loadings. A staged quality assurance and quality control workflow—including procurement screening, bench checks, co-location with blocked validation, external validation for transportability, and rotating “gold-node” drift checks—ensures reproducibility and decision-relevant uncertainty. Deployment studies demonstrate that event-aware sensor networks can support targeted ventilation and filtration, reducing exceedance time and cumulative exposure without unnecessary energy use. Standardized reporting tables, model versioning, applicability limits, and anomaly-handling rules further enhance reliability and governance. Overall, low-cost PM sensors can provide decision-relevant data when embedded in calibrated, uncertainty-aware pipelines with explicit scope statements. While reference-grade methods remain necessary for compliance, calibrated networks are well suited to hotspot detection, intervention design, and operational optimization in buildings and workplaces.

---

1. Introduction

Particulate matter (PM), especially fine particles with aerodynamic diameters less than 10 µm (PM₁₀), 5 µm (PM₅), 2.5 µm (PM_2.5) and 1 µm (PM₁), poses significant risks to human health and the environment. Upon inhalation, these particles can penetrate deep into the respiratory tract,¹ reaching the alveolar region,^2–4 where they trigger oxidative stress, inflammation,^5,6 and systemic effects.^7,8 Long-term exposure has been linked to a range of adverse outcomes, including cardiovascular and respiratory diseases, impaired cognitive development, and increased mortality.⁹ In occupational and industrial environments, exposure to coarse-mode particles such as PM₅ from sources like coal dust, mineral processing, and construction materials is also of growing concern due to its cumulative impact on lung function and elevated risk of pneumoconiosis.^10,11 Consequently, accurate and high-resolution monitoring of PM concentrations remains a public health imperative and regulatory priority worldwide.

Particulate matter (PM) in buildings arises from both infiltration and indoor sources, with short, high-intensity events superimposed on background levels.^12,13 Multiple monitoring approaches are used to characterize these dynamics. Reference-grade methods—gravimetric samplers with size-selective inlets (PM_2.5/PM₁₀),¹⁴ beta attenuation monitors,¹⁵ and tapered element oscillating microbalances¹⁶—provide traceable mass measurements and robust long-term comparability, but they are expensive, require maintenance and filter handling, and are typically limited to sparse, fixed locations with minute-to-hour resolution. Optical instruments—research-grade nephelometers and optical particle counters—offer second-to-minute resolution and good temporal fidelity but convert scattering to mass using assumptions about particle size distribution, shape, and refractive index. Their accuracy degrades with composition changes and high humidity. Electrically based instruments such as diffusion charging and photoelectric sensors, as well as black carbon and elemental carbon analyzers, provide improved specificity for certain particle classes but introduce additional cost and power demands that restrict their use in dense monitoring networks. Similarly, wearable and personal samplers that employ gravimetric or optical measurement principles can capture fine-scale exposure differences within buildings and across microenvironments. However, these devices also present challenges related to airflow aspiration during movement, participant burden, and potential gaps in data completeness.¹⁷ Low-cost optical sensors integrated into IoT nodes enable dense spatial coverage and 1–60 s logging at far lower cost, making them well suited for event detection and networked building applications. However, they exhibit range- and composition-dependent bias, sensitivity to relative humidity and temperature, device-to-device variability, drift, and occasional data loss.¹⁸ In practice, building managers and researchers often combine methods: sparse reference instrumentation for traceability and calibration, complemented by dense networks for coverage and responsiveness.

Despite rapid adoption, several gaps limit the decision relevance of indoor PM monitoring. First, reporting remains heterogeneous: many studies emphasize correlations rather than absolute errors and omit state conditioning, dynamic range, or detection capability. Second, calibration is often site-specific and validated only internally. External validation across seasons, buildings, and source regimes is uncommon, leaving transportability uncertain. Third, uncertainty is seldom carried forward from calibration to the decision-related indicators such as exceedance duration, exposure dose, or ranking consistency, and the boundaries of applicability—including the types of particles and the range of environmental conditions—are rarely stated explicitly. In addition, quality assurance and quality control across the entire instrument lifecycle, including processes for screening, drift detection, recalibration management, anomaly resolution, and version tracking, are often insufficiently reported, even though these factors determine long-term reliability and data defensibility. Finally, translation to action lags behind measurement: relatively few studies couple event-resolved sensing to ventilation/filtration control and evaluate interventions with causal designs, and issues of data governance, privacy, and integration with building management systems are inconsistently addressed.

This review responds to these gaps through a practice-oriented synthesis of indoor and workplace particulate matter (PM) monitoring using low-cost sensors and complementary methods. It defines key performance metrics and reporting standards that emphasize decision-relevant accuracy, detection capability, and state conditioning, while comparing co-location and calibration strategies ranging from simple linear approaches to multivariate and nonlinear models. These comparisons incorporate diagnostics that prevent information leakage and include external validation to ensure model transportability. The review further examines the main factors affecting measurement quality—such as humidity, temperature, aerosol composition and size, sensor placement, and operational artifacts—and establishes a comprehensive QA/QC and uncertainty framework that incorporates versioned recalibration, anomaly-handling rules, and auditable data provenance. Building on this methodological foundation, the paper translates the synthesized evidence into deployment guidance for residential, institutional, and occupational environments, addressing siting, logging, and data analytics considerations. Finally, it outlines intervention pathways that connect event detection with targeted ventilation and filtration control, evaluating their effectiveness through indicators such as exceedance time, exposure dose, and ranking stability, and concludes with a forward-looking research agenda on calibration transferability, composition and size sensitivity, and reliable performance under low-concentration conditions, accompanied by governance practices that ensure long-term, transparent, and responsible sensor use in buildings and workplaces. This review extends beyond previous summaries of low-cost PM sensing by explicitly connecting analytical performance, calibration methodology, and uncertainty quantification to decision-making processes in both indoor and occupational environments. It introduces a comparative perspective that integrates laboratory and field evidence, highlights range- and context-dependent calibration, and clarifies how performance needs differ between residential and workplace settings, thereby linking sensor evaluation with practical deployment and exposure management.

2. Literature search and methodology

This review was developed through a structured literature search across major scientific databases, including Web of Science Core Collection, Scopus, IEEE Xplore, and Google Scholar for English-language publications from 1 Jan 2000 to 30 Sep 2025 on low-cost optical PM sensing in buildings and workplaces. Keywords such as “low-cost particulate matter sensors”, “indoor air quality”, “workplace monitoring”, “calibration”, and “uncertainty” were used in various combinations. Studies were included if they reported on the application, evaluation, or calibration of low-cost PM sensors in indoor, residential, institutional, or occupational environments. Publications focusing solely on outdoor air monitoring, purely theoretical modeling without experimental validation, or non-PM pollutants were excluded.

To ensure methodological consistency, we prioritized peer-reviewed articles and review papers published in English, with a focus on studies that provided explicit performance metrics, calibration strategies, or uncertainty assessments. Additional references were identified by examining the bibliographies of relevant papers. This approach ensures that the review captures both seminal and recent contributions while maintaining relevance to decision-making in building and workplace contexts.

3. Performance evaluation and calibration framework

This section articulates a comprehensive framework for the evaluation and calibration of low-cost particulate matter (PM) sensors deployed in indoor and occupational environments, with the dual aims of traceability to reference methods and demonstrable fitness-for-purpose.^19,20 The framework is organized into four interlocking components. First, Performance metrics and reporting standards specify the quantitative constructs—accuracy and precision, linearity and dynamic range, pre-/post-calibration error statistics, temporal response, stability, cross-sensitivities, detection capability, and data completeness—together with reporting conventions that enable comparability across instruments, microenvironments, and concentration regimes.²¹ Second, Co-location and calibration methods define laboratory and field protocols for establishing equivalence to reference measurements and for deriving correction models ranging from parsimonious linear adjustments to multivariate and machine-learning approaches, with explicit criteria for model selection, internal and external validation, and transportability. Third, Factors affecting measurement quality synthesizes the physical and operational determinants of performance—hygroscopic growth²² and temperature effects,²³ aerosol size²⁴ and refractive index,²⁵ source mixtures and ventilation states,²⁶ aspiration and optical fouling²⁷—that give rise to range- and context-dependent biases.²⁰ Finally, QA/QC pipeline and uncertainty reporting codifies pre-deployment screening, burn-in, zero/span and rolling co-locations, anomaly detection, versioned recalibration, and range-conditioned uncertainty statements, ensuring reproducibility, defensible inference, and clear applicability limits. Together these elements provide a rigorous, transparent basis for designing studies, benchmarking instruments, and communicating limitations in real-world sensing.

3.1 Performance metrics and reporting standards

A rigorous evaluation of low-cost particulate matter (PM) sensors for indoor and occupational deployments must begin with clear performance constructs and unambiguous reporting conventions. At a minimum, reports should explicitly describe the main performance indicators, including accuracy and bias, which refer to the level of agreement with a traceable reference, and precision, which reflects repeatability within and between instruments and is often represented through a coefficient of variation or a pooled standard deviation. They should also document linearity and dynamic range by reporting slope, intercept, and goodness of fit across the relevant concentration intervals, as well as error statistics both before and after calibration, such as the root-mean-square error, mean absolute error, and mean absolute percentage error. Temporal performance should be discussed in terms of response time and the capability to track transient events, together with an assessment of long-term stability and potential drift. Furthermore, cross-sensitivities to environmental variables like relative humidity and temperature, and to interfering substances such as volatile organic compounds, should be examined. Finally, data completeness and detection capability, including the limit of detection and the effective functional range, must be clearly reported to ensure transparency and comparability among studies.^20,21 Across indoor evaluations that span multiple pollutants and hygro-thermal states, consumer-grade monitors typically exhibit strong temporal correlation with reference instruments yet retain substantial mass bias—often approaching a factor of two—so both correlation and absolute error metrics should be reported explicitly, together with the operating ranges over which they were computed.^28,29 Because indoor measurement quality is shaped by the microenvironment, performance statistics must be contextualized by source regimes and hygro-thermal states. In controlled two-season experiments, most devices produced RH within ±5% yet tended to over-report RH in cool–dry and under-report in warm–humid conditions.^28,30 PM readings tracked events well but displayed quantitative bias, again supporting standardized reporting of both time-correlation and mass accuracy stratified by environmental state. For occupational microenvironments and high-concentration domains, reference-traceable accuracy is particularly sensitive to range effects. Coal-dust experiments show that linearity degrades at elevated concentrations, while absolute errors (RMSE) increase substantially, demonstrating the imperative to report performance by range and to provide range-specific calibration or uncertainty models. Normalized RMSE remaining below ∼20% despite absolute error increases may be informative in risk management contexts, but should not substitute for absolute accuracy reporting.^31,32 In source-rich indoor environments, event detection capability and lag structure relative to occupant location and ventilation states should be treated as reportable performance facets. Personal/indoor/outdoor concurrent deployments show that indoor sources dominate short-term exposure peaks. Stepwise models indicate indoor PM explains the majority of variance in personal exposures during at-home periods, whereas relying on outdoor concentrations alone performs poorly.^33,34 Consequently, sensor evaluations should report I/O and P/I ratios, event increments, and model fit (R²) for interpretable, use-case-relevant performance. Reviews of indoor low-cost sensor applications and development trajectories consistently endorse this multi-metric, context-aware reporting practice.

3.2 Co-location and calibration methods

Co-location establishes traceability and enables model-based corrections that convert raw optical counts or nephelometric responses into mass metrics comparable to reference instruments.^35–37 For indoor/workplace studies, both laboratory (chambers with reference aerosols) and field co-locations are warranted. Laboratory co-locations facilitate controlled perturbation of RH/T and aerosol size/composition, including zero/span operations and step changes to probe response dynamics. Field co-locations enable contextual calibration under true source mixes and ventilation states.²⁹ Calibration models span from simple linear regressions³⁸ to nonlinear machine learning (ML) approaches,³⁹ often augmented with RH and temperature.^23,40 These models can leverage large datasets to learn flexible mappings between sensor outputs and reference instruments, thereby improving calibration accuracy under complex and dynamic environmental conditions. Building upon this foundation, recent studies have explored artificial intelligence techniques, including deep learning architectures,⁴¹ ensemble frameworks,⁴² and hybrid physical–statistical models, to further enhance adaptability and predictive performance. Growth and refractive-index effects are captured by explicit humidity and temperature terms, by functions that approximate hygroscopic growth and phase behavior, and by interaction terms that allow coefficients to change across environmental states. Event dynamics associated with condensation, evaporation, and nucleation are encoded via short-window temporal derivatives and phase indicators that distinguish rising and decay segments, while deposition and ventilation effects enter through empirically estimated decay constants derived from log-linear segments after emission cessation. Coagulation and composition changes are addressed indirectly by context variables that proxy source class and by multi-task or hierarchical models that share information across microenvironments yet permit regime-specific parameters. In machine-learning and AI frameworks, these physics-aware features and constraints improve identifiability and reduce overfitting, whereas uncertainty communication relies on range- and state-conditioned error summaries and prediction intervals evaluated under temporally blocked validation and truly external tests. In practice, this combination—mechanism-aware covariates, dynamic structure, and rigorous validation—yields calibrations that are both more accurate and more transportable across seasons, buildings, and task regimes. Such AI-based systems are capable of modeling high-dimensional relationships and contextual dependencies,⁴³ allowing them to dynamically adjust to temporal and spatial variations, automate recalibration, and enable transfer learning across different monitoring environments. Despite their promising accuracy and robustness, these data-driven methods also pose challenges related to model interpretability, computational demand, and the need for extensive, high-quality training data. Addressing these issues is critical for ensuring the transparency, reproducibility, and long-term applicability of AI-assisted calibration in both indoor and occupational sensing contexts. Evidence from high-dust occupational settings shows that linear models can be adequate in lower ranges but may over- or under-estimate systematically at higher concentrations, while ML or ensemble models improve fit across the full range and reduce range-dependent bias.³⁹ A two-layer ensemble that combines best-performing base learners has been shown to yield R² ≈ 0.97–0.98 across low–high concentration regimes with substantial RMSE reductions. Yet authors caution that model choice may be sensor- and condition-specific, underscoring the need to report model selection rationale, training data coverage, and external validation.³² Recent evaluations indicate that non-linear learners can materially improve calibration under rapidly varying meteorology and traffic-related source mixes when models are trained and assessed at event-scale temporal resolution; their advantage diminishes with coarse temporal aggregation or out-of-domain validation.⁴⁴ Under high-humidity and high-concentration regimes, strong correlations with reference monitors can coexist with inflated errors at the upper range, underscoring the need for range- and state-conditioned error summaries and for communicating both confidence and prediction intervals.⁴⁵ Taken together, these findings motivate preserving fine temporal resolution where feasible, incorporating physics-aware covariates that reflect humidity, temperature, and flow conditions, and enforcing temporally blocked internal validation alongside truly external tests to assess transferability and avoid information leakage.⁴⁶ In source-rich workplaces, multi-variable calibration that explicitly includes VOC metrics can be critical. In a diesel generator workshop, SVM calibration with RH, temperature, and TVOC covariates raised post-calibration R² by >12%, reduced RMSE to <5 µg m⁻³, and brought relative error below 15%.⁴⁷ The analysis further showed that the TVOC correction coefficient substantially influenced model performance, a finding consistent with expected optical interferences and condensation effects in mixed combustion plumes. Consequently, inclusion of a VOC covariate is recommended where oily mists, solvent vapors, or exhaust plumes co-occur with PM.⁴⁸ Calibration design should follow sound statistical principles. It is important to avoid information leakage by maintaining a strict separation between training, validation, and test datasets, and by applying blocked resampling when temporal autocorrelation is present. Adequate coverage should be ensured so that the calibration includes low, median, and high concentration ranges as well as representative environmental covariates. The limits of extrapolation need to be clearly documented, and external validation should be conducted to evaluate the model's transportability. Reviews and methodological syntheses from the indoor sensor literature provide additional guidance and should be cited together with empirical co-location studies. For networked deployments, calibration and co-location can be organized in sequential stages. A centralized co-location phase, where all nodes are compared simultaneously against a reference instrument, allows the estimation of inter-unit variance and the derivation of either unit-specific or pooled calibration models. This can be followed by an in situ rotation phase, in which a reference or “gold node” is periodically moved among locations to capture spatial heterogeneity and assess model drift. Reporting should include pre- and post-calibration scatterplots and residual diagnostics to demonstrate model performance and limitations.⁴⁹ Where feasible, pairwise inter-node regressions can quantify network consistency and underpin transfer learning strategies across microenvironments. To facilitate a quick overview of the calibration strategies discussed in this section, Table 1 summarizes the main approaches, their defining features, typical performance indicators, and common application contexts.

Table 1 Summary of key calibration approaches, performance metrics, and application contexts

Calibration approach	Key features	Application contexts	Strengths	Limitations
Linear regression	Simple linear fit between sensor and reference values	Laboratory and controlled field tests	Transparent and easy to interpret	Limited under non-linear or high humidity conditions
Multivariate regression	Includes RH, temperature, and aerosol characteristics as covariates	Indoor and occupational environments	Accounts for environmental variability	Requires larger datasets and careful covariate control
Machine learning	Random forest, support vector regression, neural networks	Dynamic or mixed-source environments	Captures non-linear dependencies, adaptive to conditions	Risk of overfitting, limited interpretability
Hybrid physical–statistical	Combines physical models with statistical learning	Long-term and cross-site deployments	Mechanistic grounding and improved generalization	Computationally demanding
Adaptive or online calibration	Periodic update using feedback or transfer learning	Continuous monitoring networks	Maintains accuracy over time	Needs reference link and additional computation

---

3.3 Factors affecting measurement quality

Three families of mechanisms drive variability in low-cost optical PM sensors: environmental, aerosol-physical, and operational.⁵⁰ First, relative humidity (RH) and temperature alter particle hygroscopic growth and scattering, perturbing mass conversions even when the instrument's optics are stable.⁵¹ Comparative indoor evaluations show systematic RH-dependent bias—over-reporting in cool–dry and under-reporting in warm–humid conditions—and motivate RH-stratified calibration or inclusion of RH/T as covariates.⁵² Error reporting should therefore be state-conditioned rather than pooled. Second, aerosol size, morphology, and refractive index reshape the response curve.^53,54 In workplace settings, combustion aerosols and process dusts present distinct optical signatures. Data demonstrate range-dependent linearity degradation in high coal-dust regimes, with precision deteriorating as concentration increases—indicative of counting statistics saturation, coincidence losses, or changes in size distributions.³² Reporting by source class and range is thus more informative than a single global slope. Physicochemical transformation processes further shape measurement bias and variability by continuously altering particle size distributions, composition, and residence times near the sensing volume. Coagulation reduces number concentrations at small diameters while shifting mass toward larger sizes, which can depress optical particle counter counts in fine bins and change the scattering-to-mass conversion within the instrument's internal algorithm. Condensation and subsequent evaporation modify both diameter and refractive index, producing humidity- and composition-dependent changes in scattering efficiency and, in semi-volatile plumes, hysteresis between rising and falling phases of an event. Nucleation generates ultrafine particles initially below the lower cut of many low-cost optics, contributing little to mass but seeding growth that enters the detectable range with a delay, thereby introducing apparent lags between reference and sensor responses. Deposition removes particles at size-dependent rates through gravitational settling, diffusion, and surface interactions along inlets and chambers, which leads to under-capture of coarse particles and enhanced loss of ultra fines, especially under low flow or complex inlet geometries. Together, these transformations explain range-dependent linearity, context-dependent bias under humid or compositionally rich conditions, and mismatches between number- and mass-centric signals that must be acknowledged when interpreting indoor and workplace measurements. Third, co-pollutant interference affects optical paths and inference models. In diesel generator facilities, TVOC levels materially impacted SVM-based calibration performance. Incorporating TVOC into multivariate models improved agreement with reference data and reduced residual bias. In indoor exposure studies, incense and cooking generate episodic fine particles with strong refractive and absorptive features, creating pronounced short-term increments that test both response time and conversion algorithms.⁵⁵ Evaluations that quantify event increments, I/O ratios, and partial R² for source terms provide a richer picture of sensor capability in real homes and mixed microenvironments. Ventilation and occupancy modulate both baselines and peaks. Concurrent personal–indoor–outdoor campaigns reveal that indoor PM often explains most variability in personal exposure during at-home periods, with ventilation state systematically shifting indoor levels.⁵⁶ Consequently, assessments that ignore ventilation metadata risk attributing variance to sensor error rather than to airflow and microenvironment transitions. Reporting should therefore include ventilation states and, where possible, lag analyses between rooms or between indoor and outdoor monitors. Beyond physical and environmental effects, data integrity is another determinant of measurement quality. In practice, raw data streams from low-cost sensors may contain gaps, outliers, or implausible values arising from power interruptions, transmission errors, or short-term sensor drift. To ensure continuity and reliability of the time series used for calibration and analysis, missing or faulty observations were corrected through imputation prior to statistical modeling. Depending on the temporal structure and variability of the dataset, linear interpolation, nearest-neighbor substitution, or regression-based imputation was applied to restore plausible values while preserving underlying temporal trends. These imputation steps were conducted conservatively to avoid artificial smoothing or bias amplification, thereby maintaining data representativeness during co-location, calibration, and QA/QC assessments. Explicitly addressing imputation ensures that subsequent calibration models operate on coherent datasets, reducing propagation of error and improving the reproducibility of reported performance metrics.

Finally, device architecture and sampling hardware—including the design of the air inlet, the stability of the airflow, and the cleanliness of the optical window—introduce operational biases over time. Network and device papers that describe the electronics, flow paths, and telemetry—such as systems based on LoRaWAN microcontroller architectures and defined sampling intervals—are valuable to cite because these design choices influence noise, packet loss, and power-related sampling gaps, which in turn appear as data completeness issues and temporal aliasing in performance statistics.^57–59 Where portable or personal devices are used, body-worn placement and movement alter aspiration efficiency and exposure representativeness, so studies should differentiate fixed-site vs. personal performance and avoid conflating the two.

3.4 QA/QC pipeline and uncertainty reporting

Converting “low-cost sensors” into decision-grade data requires a documented QA/QC pipeline and explicit uncertainty quantification. In practice, the workflow begins with procurement-level quality assurance—screening units from the same lot to characterize inter-unit variability—followed by bench checks that verify firmware versions, confirm fan/laser functionality, and perform zero/span tests for gas channels alongside temperature and RH sanity checks.^19,60,61 A burn-in period is then used to stabilize early drift before field activities. Co-location with a traceable reference is conducted for sufficient duration and environmental coverage to support robust calibration. Model development proceeds with strict separation of training and testing time blocks to prevent leakage, complemented by internal validation. Transportability is demonstrated through external validation across distinct microenvironments or seasons. During routine operation, systems are monitored for health flags, data completeness, and drift, using periodic re-co-location or rotating “gold” units as sentinels.⁶² Anomaly handling procedures address spikes, negative or saturated values, and other implausible observations through predefined rules. Finally, recalibration is managed under version control with detailed change logs, preserving a complete audit trail from raw signal to reported metric. Routine QA/QC operations should include scheduled zero checks, span checks, and field rotations. Data QA rules should specify validity ranges per channel, RH/T plausibility checks, flagging logic for rapid changes inconsistent with physics, and procedures for gap filling and down-sampling to standard averaging times for comparability. Where networks operate at high temporal resolution, telemetry QA—including packet loss tracking and clock drift correction—must be reported, as these factors directly influence data completeness and alignment with reference time bases. Engineering-style network descriptions, such as those detailing modular node design, MCU/LoRaWAN architecture, and the choice of indoor versus outdoor sampling intervals, provide templates for documenting how sampling strategies can bias QA indicators. For instance, differences in sampling frequency between external nodes and internal nodes can introduce systematic gaps in data resolution and comparability.⁶³

When reporting uncertainty, it is essential to present both point estimates and their associated confidence or prediction intervals and explicitly communicate performance across concentration and environmental gradients, including relative humidity and temperature.⁶⁴ Evaluations should detail pre- and post-calibration metrics—such as RMSE, MAE, and R²—across defined concentration ranges and RH categories, and include residual diagnostics against RH, temperature, and, where relevant, VOC levels to reveal systematic deviations. For machine learning models, calibration workflows must incorporate cross-validation protocols that block temporal dependencies and further assess generalizability through evaluations in independent microenvironments or seasons. Given the dominance of microenvironmental drivers in indoor exposure, we recommend that QA/QC documentation and uncertainty statements incorporate use-case-specific endpoints: for example, if the deployment aims to rank shift-average exposures across stations, then rank-based statistics and mis-ranking risk are more relevant. If the deployment aims to trigger ventilation or process controls, then event detection sensitivity/specificity, response time, and false-alarm/false-miss rates must be included. Studies of personal versus fixed-site monitoring demonstrate that indoor PM explains most variance in personal exposure during at-home periods. Consequently, ventilation state metadata and I/O, P/I ratios should be folded into reporting as auxiliary QA channels that help interpret whether deviations are sensor errors or contextual shifts. Finally, all QA/QC actions should be explicitly versioned—covering sensor firmware, calibration model identifiers, parameter sets, and periods of validity—and documented in a transparent and reproducible manner. A structured checklist, presented either within the manuscript or as an appendix, should summarize pre-deployment baselines, the design and coverage of co-location exercises, calibration model architecture and included covariates, results from both internal and external validation, strategies for anomaly and data-loss handling, and maintenance procedures such as optical window cleaning schedules. In addition, clear criteria for re-calibration and an explicit statement of applicability limits—specifying particle types, environmental ranges, and maximum reliable concentrations before saturation or coincidence losses occur—should be provided. Recent reviews of indoor air quality monitoring and sensor development highlight the importance of such standardized documentation, which is critical for advancing low-cost sensor networks from exploratory research tools to reliable systems that can support practical decision-making in health and occupational contexts.

4. Evidence and deployment in indoor and workplace environments

Below, this section synthesizes deployment evidence and practical guidance for low-cost particulate matter (PM) sensors across indoor and occupational contexts. It first consolidates what is known for homes and apartments, then distills lessons for institutional and workplace environments, and finally translates these findings into actionable deployment and analytics practices. Emphasis is placed on linking microenvironmental source regimes and hygro-thermal states to observed performance and on documenting calibration and QA/QC steps so that data are “decision-grade.”

4.1 Residential settings

Residential environments are characterized by short-duration, high-intensity source events superimposed on an infiltration background, with large intra- and inter-room gradients driven by occupant behavior and ventilation states, as shown in controlled residential evaluations of indoor air quality monitors.⁵² Laboratory and field investigations across multiple source types demonstrate that consumer-grade light-scattering monitors capture the temporal dynamics of events but consistently display substantial mass bias, underscoring the importance of context-rich reporting and careful calibration.^55,65 In an evaluation involving 21 indoor sources, six home monitors yielded only semi-quantitative responses for PM_2.5, ranging from about half to twice the reference values, with negligible sensitivity to ultrafine aerosols and variable responses for coarse particles, highlighting the need to report both correlation and absolute error stratified by source class. Concurrent personal–indoor–outdoor campaigns using corrected low-cost devices show that indoor events dominate short-term personal exposures, while ventilation conditions, such as open windows or air conditioning, modulate infiltration and decay. Regression models attribute the majority of exposure variance to indoor activities such as cooking and incense burning.⁵⁶ Within homes, cooking remains the canonical driver of transient PM peaks, with incense and environmental tobacco smoke also producing large increments and high indoor/outdoor ratios, particularly in poorly ventilated conditions. Multi-room deployments with one outdoor and several indoor nodes indicate that kitchens show the most pronounced spikes during meal preparation, whereas bedrooms accumulate resuspended dust and may present higher daily averages of PM_2.5 and PM₁₀ when occupied. PM₁ concentrations tend to reflect outdoor infiltration in highly ventilated rooms. Source apportionment studies combining low-cost sensors with positive matrix factorization reveal that up to 95% of PM₁ in indoor spaces originates outdoors, while the outdoor contribution declines progressively with larger size fractions, illustrating the complex interplay between indoor generation and infiltration.⁶⁶ Infiltration factors and indoor/outdoor ratios depend strongly on building leakage, HVAC (Heating, Ventilation, and Air Conditioning) operation, and ambient levels. Case studies show that indoor/outdoor ratios are typically less than one when outdoor concentrations are elevated in naturally ventilated homes, but greater than one under strong indoor source regimes, highlighting the need to report ratios across states. Deployment implications follow directly: placement should be source-aware, with one node in the kitchen positioned at breathing height but away from steam plumes, one in the principal sleeping area, and one outdoor node shielded from rain and sun to provide infiltration context.^59,67 Averaging times of one to five minutes are recommended to capture event dynamics without excessive smoothing, complemented by finer resolution sampling during interventions. Calibration should be performed by co-locating with a reference device across different thermal and humidity conditions, while ensuring temporal blocking between training and testing sets to avoid information leakage.⁴⁷ Event-level performance should be reported in terms of slopes, intercepts, R², and error metrics stratified by source class and hygro-thermal state, as sensors tend to over- or under-estimate mass when relative humidity is high due to hygroscopic growth and optical effects. Networked residential deployments further demonstrate that combining thresholding with change-point detection improves the ability to flag events while minimizing false positives caused by humidity or steam transients.^68,69 Large-scale household sensor networks have been used to quantify the benefits of filtration interventions and the spatial heterogeneity of exposure across rooms, leading to actionable recommendations such as range hood use and filter upgrades.^70–72 Finally, context metadata such as ventilation state, occupancy, and activity patterns significantly improve the interpretability of sensor data and the robustness of cross-home models, as shown in personal exposure studies where the inclusion of such covariates explained a substantial share of the observed variance.

4.2 Institutional and workplace settings

Institutional buildings exhibit load patterns dominated by occupancy, cleaning, printing, and intermittent maintenance, while workplaces span task-specific aerosol regimes. This diversity produces concentration ranges and particle properties that can challenge the linearity and specificity of low-cost optics, particularly at high mass loadings or under elevated VOCs, and thus requires environment-specific calibration, placement, and QA/QC.^73,74 In offices and classrooms, occupant activity and cleaning trigger re-suspension and short spikes, whereas printing and renovation generate distinct particle size distributions. Neonatal units and clinical spaces show sensitive microenvironments where even routine activities elevate PM.^75,76 Low-cost nodes have been successfully used to map such episodic behaviors and to inform operational controls provided the calibration is maintained across seasonal HVAC modes.^77,78 A practical strategy is to combine a few fixed “sentinel” nodes at high-traffic or process-proximal locations with roving mappers during suspected emission periods, rotating devices to quantify inter-unit spread and drift. In production spaces, multiple studies demonstrate that low-cost sensors can achieve high correlations with reference instruments but may show systematic bias that grows with concentration or with interfering vapors, motivating multivariate or nonlinear calibration.^18,79 In a machine shop, a LoRa-based oil-mist sensor network achieved R² ≈ 0.96 against a reference, operated stably over months, and documented that peak concentrations during production exceeded NIOSH's 0.5 mg m⁻³ recommended value in most areas. Lowering RH improved sampling accuracy, and lowering temperature reduced oil-mist levels.⁸⁰ From a worker-protection perspective, network optimization and hazard mapping are essential. Studies have emphasized using prior task/hazard maps to guide node placement, then leveraging incoming data to iteratively refine coverage of “hot zones,” an approach long advocated in occupational hygiene and now facilitated by inexpensive sensors. For large plants, hybrid networks mixing fixed and wearable nodes, synchronized with time–motion logs, enable attribution of exposure peaks to specific tasks and locations. In addition, low-cost, distributed environmental monitors have been piloted for factory health applications, demonstrating feasibility of long-term, multi-pollutant monitoring with unit costs below traditional instrument budgets.⁵⁹

4.3 Deployment and analytics

Translating the above evidence into operational guidance requires attention to network design, data pipelines, event analytics, and integration with building/industrial control systems. Begin with a conceptual model of sources and air pathways, including emission points such as cooking surfaces, printers, and weld bays, along with the relevant mixing volumes and removal processes such as filtration, deposition, and exfiltration. Use a small number of fixed sentinel nodes to characterize background and infiltration, with one positioned outdoors or in the supply air and another in the main occupied zone, and then place additional nodes near the hypothesized hotspots. In residences, the triad of kitchen–bedroom–outdoor is repeatedly justified by event patterns and infiltration dynamics.⁷¹ In workplaces, add task-proximal nodes and at least one downwind node per process line. Wearable samplers can be used during specific tasks to align personal exposure with fixed-site fields.^17,81

For source-rich settings, use 1 min logging to capture transients. 5 min rolling averages provide interpretable event metrics without overwhelming storage. Duty-cycling can extend battery life for wearables between task windows, while fixed nodes can be mains-powered with UPS backup for resilience. Industrial deployments have leveraged low-power wide-area networks to coordinate dozens of nodes with multi-month uptime. Follow good statistical practice: ensure separation of training, validation, and test time blocks. Cover low, median, and high deciles of concentration and environmental covariates. Clearly document the limits of extrapolation, such as very high relative humidity or atmospheres with abundant volatile organic compounds, and conduct true external validation at a different site, season, or task to quantify transportability.^82,83 Field studies indicate that machine-learning models (RF, SVM, ANN) outperform simple linear corrections when cross-sensitivities are substantial or when high-load, process aerosols are present.⁸⁴ Still, document prediction intervals and range-specific performance so downstream users can adjudicate fitness for use. Event detection and inference can be robustly implemented by integrating simple thresholding with change-point or gradient detection methods, thereby enabling accurate identification of emission onsets and offsets while minimizing artefacts induced by humidity.⁸⁵ Event type classification can be achieved using supervised models trained on labeled periods, with total VOC included as a predictor in VOC-rich environments to enhance calibration.⁸⁶ Sensor networks can integrate with building management or smart control systems to activate ventilation or filtration when particulate matter exceeds thresholds, and with industrial process controls to trigger alarms. Network optimization frameworks from occupational hygiene—“hazard mapping” and iterative placement—have been successfully combined with low-cost arrays to minimize blind spots and improve detection of hot zones.⁸⁷ Quality assurance, quality control, and governance can be ensured through a documented pipeline that includes procurement screening to quantify inter-unit variability, bench checks of firmware and zero/span responses as well as RH/T plausibility, sensor burn-in, co-location for calibration with sufficient duration and coverage, internal validation using blocked resampling, external validation, deployment monitoring with health flags, anomaly handling for negative or saturated values, and version-controlled recalibration with documented change logs. Uncertainty should be communicated transparently, including calibration errors, concentration-specific performance, environmental influences, inter-unit consistency, temporal stability, cross-sensitivities and their mitigation, data completeness, and applicability limits. Studies in homes and shops demonstrate how such reporting enables end users to interpret ranks vs. magnitudes correctly and to decide when additional reference measurements are needed.⁸⁸ Collectively, the deployment and analytics literature converges on a simple message: place sensors according to sources and airflow, calibrate with context, validate externally, and automate analytics that account for infiltration and events. Done well, low-cost arrays deliver the spatial and temporal resolution necessary to diagnose problems and reduce exposures across homes, institutions, and workplaces at sustainable cost.

5. From monitoring to action

This section develops a translational pathway that turns calibrated, quality-assured signals from low-cost PM sensors into operational decisions in buildings and workplaces. The argument proceeds from evidence on event-resolved dynamics and air-pathways to concrete control choices, then evaluates those choices through exceedance time, dose, and distributional metrics while disclosing range-conditioned uncertainty. It then frames cost and governance so deployments remain economical, compliant, and ethically defensible at scale, and concludes with a forward agenda that improves transferability, long-term robustness, and equity. The synthesis emphasizes scene-specific determinants—source regimes, hygro-thermal states, ventilation strategies—and insists on model provenance so that any claimed effect size is reproducible and suitable for regulatory dialogue.

5.1 Interventions and control strategies

Across residential, institutional, and occupational settings, the most policy-relevant structure in indoor PM is generated by short-duration, high-intensity events superimposed on an infiltration background governed by air-exchange and filtration. Residential campaigns consistently show minute-scale increments during cooking, incense use, and cleaning, with calibrated low-cost instruments capturing the timing and relative magnitude of peaks despite source-dependent mass bias. Institutional buildings display analogous spikes synchronized with occupancy, printing, and custodial activity, while workshops exhibit task-bound surges linked to diesel operation, oil-mist machining, welding, woodworking, and mineral dust handling. When co-located and calibrated in context, multi-node networks based on optical particle counters or nephelometers provide sufficient temporal fidelity to trigger local capture, filtration boosts, and ventilation phasing that compress exceedance time and reduce dose without blanket over-ventilation.

In homes and apartments, siting one node in the kitchen, one in a primary sleeping area, and one outdoors (or in supply air) reveals the coupling between emission onsets, indoor decay constants, and infiltration. Event-aware control follows directly: range-hood activation at the onset of frying or broiling, short post-event boosts, and outdoor-air flushing only when ambient is cleaner than indoors. Because optical response and decay rates shift with RH and temperature, both calibration and reporting must be stratified by hygro-thermal state. Otherwise, over- or under-control is likely when hygroscopic growth elevates optical signals or when high RH induces artefactual peaks.⁷¹ Personal–indoor–outdoor concurrent deployments reinforce the primacy of indoor source control for exposure reduction during at-home periods: regressions attribute large fractions of short-term personal dose to kitchen events and incense, and ventilation state modulates the magnitude and persistence of peaks, so timing controls to event onsets yields disproportionate benefits relative to uniform air-exchange increases.⁷²

Institutional buildings—offices, schools, clinics—benefit from the same state-contingent logic, but with system-level coordination. Filtration boosts and outdoor-air fractions are aligned to occupancy waves and cleaning schedules, not held constant. When outdoor episodes raise the risk of importing pollution (I/O > 1), the strategy shifts toward internal recirculation with enhanced filtration. When ambient is clean, make-up air is a cost-effective diluent. Because occupant resuspension and equipment use produce distinct size distributions and decay profiles, intervention evaluation emphasizes event increments, upper-tail suppression, and decay-constant shortening, all reported alongside calibration scope and seasonal coverage so performance claims remain transportable across HVAC modes.⁸⁹

Workshops and industrial spaces require a task-aligned hierarchy of controls—substitution where feasible, local exhaust ventilation (LEV), isolation, and enclosure—guided by calibrated network maps. Diesel generator facilities illustrate the approach: Plantower-based arrays, co-located to research-grade references and calibrated with humidity, temperature, and TVOC covariates, recover systematic bias and raise R² by more than ten percentage points while reducing RMSE to <5 µg m⁻³. Calibrated heat-maps localize zones peaking around 120–220 µg m⁻³ during operation and inform enclosure tightening, LEV capture positioning, and task scheduling.⁴⁷ In oil-mist machining shops, LoRa-connected networks achieve R² ≈ 0.96 against reference photometers over month-scale campaigns, with a minority of drifting units detected and corrected through rotation. Lowering RH improves sampling accuracy and reducing process temperature curbs mist generation—two setpoints that directly translate sensor evidence into controls.⁸⁰

Impact must be demonstrated with causal designs rather than descriptive plots. Pre–post and case–control frameworks quantify changes in exceedance time, time-integrated dose (area under the curve), event increments, and decay constants under comparable outdoor conditions and HVAC states, where interventions target ranking, robust rank measures and mis-ranking risk are more decision-salient than mean absolute error, whereas compliance-adjacent screening requires prediction intervals and residual diagnostics stratified by concentration and RH/T bins. In every case, action logic is bound to model provenance—co-location duration, microenvironment coverage, range specificity, and external validation—so that operational decisions remain auditable.

5.2 Cost–benefit, compliance, and ethics

The value proposition of low-cost sensing is realized when spatial and temporal resolution is translated into targeted control that reduces exposure while avoiding indiscriminate energy use. Costs arise from devices, co-location and calibration, communications and power infrastructure, as well as maintenance of optical components and data services, whereas benefits derive from avoided exposure through fewer minutes above thresholds, reduced spatial extent of events, and suppression of upper-tail concentrations, together with avoided energy use achieved by shorter or less frequent filtration and ventilation runs under event-aware control. In residential and dormitory contexts, sensor-triggered hoods and post-event flushes cut both exposure and runtime compared with fixed schedules. In workshops, task-targeted LEV and enclosure adjustments yield larger returns than uniform air-flow increases and help maintain comfort and process stability.

Scaling from pilots to programs hinges on modularization. Centralized co-location produces unit-specific or fleet-level calibrations, while rolling “gold-node” rotations provide drift sentinels. Firmware and calibration models are versioned, anomaly rules for spikes, negative or saturated values are standardized, data-completeness thresholds are defined, and scheduled re-checks help keep uncertainty within bounds. LPWAN backbones, edge buffering, and compressed payloads extend battery life and simplify installation for wearables and temporary nodes across large facilities. Industrial campaigns report months-long stability with a small fraction of drifters that are easily identified and corrected via rotation and re-co-location. The cost of such governance is modest relative to the cost of poor data quality and mis-targeted controls.

Compliance-adjacent use is legitimate when uncertainty is explicit. Although low-cost nodes are not FRM/FEM instruments, calibrated and range-bounded PM data support hotspot detection, prioritization of confirmatory sampling, documentation of intervention effects, and operational tuning in buildings. Alignment with regulator and occupational-hygiene expectations requires disclosing calibration scope, external validation context, and aerosol-type applicability. Performance should be reported with slopes, intercepts, R², and RMSE/MAE stratified by concentration and RH/T bins, supplemented by inter-unit statistics and completeness, so screening remains clearly distinct from determinations while retaining credibility. Coal-dust, diesel, and oil-mist case studies demonstrate how range-aware calibration and model documentation enable practical workflows near compliance without over-claiming precision.

Ethical governance is necessary to maintain trust. In workplaces, personal monitoring should be voluntary with informed consent, identifiers must be anonymized or pseudonymized, and access should be role-based with audit trails. Data retention should be limited and purpose-specific, while cybersecurity measures such as credential management, encrypted communication, signed firmware, and timely patching are required to safeguard both privacy and operational integrity, especially where sensors control fans, dampers, or alarms. In residential and semi-public settings, cross-streams such as CO₂ and occupancy proxies can reveal behavior, so collection adheres to the minimum-necessary principle with opt-out options for noncritical channels. Reports that bind calibration and QA/QC to intervention timing and show uncertainty bands are consistently more persuasive to facilities managers and regulators than correlation alone, and they sustain adoption by clarifying risk, benefit, and residual uncertainty.

6. Research gaps and future directions

Although low-cost PM sensors have demonstrated substantial promise for indoor and occupational exposure monitoring, several unresolved challenges hinder their broader adoption. In residential environments, most studies emphasize short-term event detection, but systematic assessments of infiltration dynamics, background accumulation, and cross-room variability remain limited. Few investigations integrate ventilation metadata or occupant behavior into sensor interpretation, which hampers the accurate attribution of exposure peaks to sources versus airflow conditions. Longitudinal household studies that link event-resolved PM dynamics with ventilation strategies, HVAC operation, and intervention outcomes are still scarce, leaving important gaps in evidence-based recommendations for exposure reduction.

In occupational settings, calibration and deployment practices are even less standardized. High-concentration dust, welding fumes, oil mists, and mixed combustion aerosols often push sensors beyond their validated ranges, producing range-dependent biases that are seldom quantified. Cross-sensitivities with temperature, humidity, and volatile organic compounds further complicate interpretation, yet only a few workplace studies explicitly incorporate these covariates in calibration. Moreover, task-specific monitoring with wearable devices remains underexplored, and there is limited guidance on aligning personal exposure estimates with fixed-site measurements. Worker privacy, data governance, and ethical concerns are rarely addressed in published deployments, despite their central importance for long-term adoption.

Future work should aim to develop calibration approaches that are transferable across seasons, microenvironments, and source regimes, while explicitly incorporating environmental covariates. Greater emphasis on longitudinal, intervention-oriented studies is needed to evaluate how sensor data can guide ventilation, filtration, and process modifications in practice. Finally, standardized reporting protocols that include uncertainty ranges, applicability limits, and QA/QC documentation specific to indoor and workplace contexts will be essential to move low-cost PM sensing from exploratory research into reliable, decision-support systems.

7. Conclusions

Low-cost particulate matter sensors provide new opportunities for monitoring air quality in indoor and occupational environments where traditional reference methods are often impractical due to cost and logistical constraints. Evidence across residential, institutional, and workplace studies demonstrates that such sensors are effective in capturing temporal dynamics and identifying high-intensity emission events, while also offering the spatial resolution needed to characterize exposure variability within and across microenvironments. Nevertheless, their quantitative accuracy is strongly influenced by humidity, temperature, aerosol composition, and concentration range, which underscores the necessity of rigorous calibration and quality assurance protocols.

To translate sensor data into decision-grade evidence, uncertainty must be explicitly reported and linked to concentration levels and environmental conditions, while deployment strategies should be tailored to source regimes and ventilation patterns. Although current applications remain limited by challenges of transferability, long-term stability, and standardization, emerging approaches in machine learning calibration, network-based QA/QC, and hybrid sensing show considerable potential. By addressing these gaps and adopting transparent, standardized reporting practices, low-cost PM sensors can move beyond exploratory research to become reliable tools for guiding ventilation and filtration strategies, informing occupational health interventions, and ultimately protecting public health in indoor spaces and workplaces.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

No new primary data were generated for this review. The data supporting the analysis of the methods discussed in this article are sourced from previously published studies, which are fully referenced in the manuscript.

Acknowledgements

This work was supported by National Key Research and Development Program of China (2023YFC3010602), National Natural Science Foundation of China (Grant No. 52074274), the Fundamental Research Funds for the Central Universities (Grant No. 2021YCPY0107).

References

1. T. F. Hatch, Distribution and deposition of inhaled particles in respiratory tract, Bacteriol. Rev., 1961, 25(3), 237–240.
2. H. Schulz, P. Brand and J. Heyder, Particle deposition in the respiratory tract, Lung Biol. Health Dis., 2000, 143, 229–277.
3. J. D. Brain and P. A. Valberg, Deposition of aerosol in the respiratory tract, Am. Rev. Respir. Dis., 1979, 120(6), 1325–1373.
4. C. Kleinstreuer and Z. Zhang, Airflow and particle transport in the human respiratory system, Annu. Rev. Fluid. Mech., 2010, 42(1), 301–334.
5. K. Donaldson, N. Mills, W. MacNee, S. Robinson and D. Newby, Role of inflammation in cardiopulmonary health effects of PM, Toxicol. Appl. Pharmacol., 2005, 207(2), 483–488.
6. B. Veronesi and M. Oortgiesen, Neurogenic inflammation and particulate matter (PM) air pollutants, Neurotoxicology, 2001, 22(6), 795–810.
7. J. O. Anderson, J. G. Thundiyil and A. Stolbach, Clearing the air: a review of the effects of particulate matter air pollution on human health, J. Med. Toxicol., 2012, 8, 166–175.
8. T. R. Nurkiewicz, D. W. Porter, M. Barger, L. Millecchia, K. M. K. Rao, P. J. Marvar, A. F. Hubbs, V. Castranova and M. A. Boegehold, Systemic microvascular dysfunction and inflammation after pulmonary particulate matter exposure, Environ. Health Perspect., 2006, 114(3), 412–419.
9. I. Kloog, B. Ridgway, P. Koutrakis, B. A. Coull and J. D. Schwartz, Long-and short-term exposure to PM2.5 and mortality: using novel exposure models, Epidemiology, 2013, 24(4), 555–561.
10. T. Liu and S. Liu, The impacts of coal dust on miners' health: a review, Environ. Res., 2020, 190, 109849.
11. E. Wouters, T. Jorna and M. Westenend, Respiratory effects of coal dust exposure: clinical effects and diagnosis, Exp. Lung Res., 1994, 20(5), 385–394.
12. E. National Academies of Sciences, Medicine, Health risks of indoor exposure to fine particulate matter and practical mitigation solutions, 2024.
13. R. Das, A. Acharyya and S. Majumdar, Influences of Human Presence on the Indoor Air Quality of Educational Institutions: Concurrent Multi-Pollutant Sensing Approach, IEEE Open J. Instrum. Meas., 2025, 4, 1–8.
14. H. E. Ayer, G. W. Sutton and I. H. Davis, Size-selective gravimetric sampling in dusty industries, Am. Ind. Hyg. Assoc. J., 1968, 29(4), 336–342.
15. E. Triantafyllou, E. Diapouli, E. Tsilibari, A. Adamopoulos, G. Biskos and K. Eleftheriadis, Assessment of factors influencing PM mass concentration measured by gravimetric & beta attenuation techniques at a suburban site, Atmos. Environ., 2016, 131, 409–417.
16. H. Patashnick and E. G. Rupprecht, Continuous PM-10 measurements using the tapered element oscillating microbalance, J. Air Waste Manage. Assoc., 1991, 41(8), 1079–1083.
17. J. O. Okeme, J. P. Koelmel, E. Johnson, E. Z. Lin, D. Gao and K. J. G. Pollitt, Wearable passive samplers for assessing environmental exposure to organic chemicals: current approaches and future directions, Curr. Environ. Health Rep., 2023, 10(2), 84–98.
18. I. Hayward, N. A. Martin, V. Ferracci, M. Kazemimanesh and P. Kumar, Low-Cost Air Quality Sensors: Biases, Corrections and Challenges in Their Comparability, Atmosphere, 2024, 15(12), 1523.
19. M. R. Giordano, C. Malings, S. N. Pandis, A. A. Presto, V. McNeill, D. M. Westervelt, M. Beekmann and R. Subramanian, From low-cost sensors to high-quality data: a summary of challenges and best practices for effectively calibrating low-cost particulate matter mass sensors, J. Aerosol Sci., 2021, 158, 105833.
20. U.S. EPA, Air Sensor Performance Targets and Testing Protocols for Non-regulatory Supplemental and Informational Monitoring (NSIM), 2022.
21. U.S. EPA, Enhanced Air Sensor Guidebook, Office of Research and Development, 2022/2023, available at EPA Air Sensor Toolbox.
22. A. Molnar, K. Imre, Z. Ferenczi, G. Kiss and A. Gelencser, Aerosol hygroscopicity: hygroscopic growth proxy based on visibility for low-cost PM monitoring, Atmos. Res., 2020, 236, 104815.
23. T. Zheng, M. H. Bergin, K. K. Johnson, S. N. Tripathi, S. Shirodkar, M. S. Landis, R. Sutaria and D. E. Carlson, Field evaluation of low-cost particulate matter sensors in high-and low-concentration environments, Atmos. Meas. Tech., 2018, 11(8), 4823–4846.
24. J. Tryner, J. Mehaffy, D. Miller-Lionberg and J. Volckens, Effects of aerosol type and simulated aging on performance of low-cost PM sensors, J. Aerosol Sci., 2020, 150, 105654.
25. C.-H. Huang, J. He, E. Austin, E. Seto and I. Novosselov, Assessing the value of complex refractive index and particle density for calibration of low-cost particle matter sensor for size-resolved particle count and PM2.5 measurements, PLoS One, 2021, 16(11), e0259745.
26. S.-C. C. Lung, W.-C. V. Wang, T.-Y. J. Wen, C.-H. Liu and S.-C. Hu, A versatile low-cost sensing device for assessing PM2.5 spatiotemporal variation and quantifying source contribution, Sci. Total Environ., 2020, 716, 137145.
27. T. Matos, C. L. Faria, M. S. Martins, R. Henriques, P. Gomes and L. M. Goncalves, Development of a cost-effective optical sensor for continuous monitoring of turbidity and suspended particulate matter in marine environment, Sensors, 2019, 19(20), 4439.
28. R. Jayaratne, X. Liu, P. Thai, M. Dunbabin and L. Morawska, The influence of humidity on the performance of a low-cost air particle mass sensor and the effect of atmospheric fog, Atmos. Meas. Tech., 2018, 11(8), 4883–4890.
29. L. R. Crilley, M. Shaw, R. Pound, L. J. Kramer, R. Price, S. Young, A. C. Lewis and F. D. Pope, Evaluation of a low-cost optical particle counter (Alphasense OPC-N2) for ambient air monitoring, Atmos. Meas. Tech., 2018, 11(2), 709–720.
30. M. Y. Patel, P. F. Vannucci, J. Kim, W. M. Berelson and R. C. Cohen, Towards a hygroscopic growth calibration for low-cost PM 2.5 sensors, Atmos. Meas. Tech., 2024, 17(3), 1051–1060.
31. N. A. Amoah, G. Xu, A. R. Kumar and Y. Wang, Calibration of low-cost particulate matter sensors for coal dust monitoring, Sci. Total Environ., 2023, 859, 160336.
32. Z. Feng, L. Zheng, X. Zhang, J. Liu, N. Xue and M. Wang, Evaluation and calibration of low-cost particulate matter sensors for respirable coal mine dust monitoring, Aerosol Sci. Technol., 2024, 58(2), 158–169.
33. C. Chen and B. Zhao, Review of relationship between indoor and outdoor particles: I/O ratio, infiltration factor and penetration factor, Atmos. Environ., 2011, 45(2), 275–288.
34. D. L. Mendoza, T. M. Benney and S. Boll, Long-term analysis of the relationships between indoor and outdoor fine particulate pollution: a case study using research grade sensors, Sci. Total Environ., 2021, 776, 145778.
35. M. Vogt, P. Schneider, N. Castell and P. Hamer, Assessment of low-cost particulate matter sensor systems against optical and gravimetric methods in a field co-location in Norway, Atmosphere, 2021, 12(8), 961.
36. M. Levy Zamora, C. Buehler, A. Datta, D. R. Gentner and K. Koehler, Optimizing co-location calibration periods for low-cost sensors, EGUsphere, 2022, 2022, 1–20.
37. S. Manu and A. Rysanek, A Co-Location Study of 87 Low-Cost Environmental Monitors: Assessing Outliers, Variability, and Uncertainty, Buildings, 2024, 14(9), 2965.
38. M. Badura, P. Batog, A. Drzeniecka-Osiadacz and P. Modzel, Regression methods in the calibration of low-cost sensors for ambient particulate matter measurements, SN Appl. Sci., 2019, 1(6), 622.
39. M. Si, Y. Xiong, S. Du and K. Du, Evaluation and calibration of a low-cost particle sensor in ambient conditions using machine-learning methods, Atmos. Meas. Tech., 2020, 13(4), 1693–1707.
40. Y. Zou, J. D. Clark and A. A. May, A systematic investigation on the effects of temperature and relative humidity on the performance of eight low-cost particle sensors and devices, J. Aerosol Sci., 2021, 152, 105715.
41. D. Park, G.-W. Yoo, S.-H. Park and J.-H. Lee, Assessment and calibration of a low-cost PM2.5 sensor using machine learning (hybridlSTM neural network): feasibility study to build an air quality monitoring system, Atmosphere, 2021, 12(10), 1306.
42. P.-Y. Yin, C.-C. Tsai, R.-F. Day, C.-Y. Tung and B. Bhanu, Ensemble learning of model hyperparameters and spatiotemporal data for calibration of low-cost PM2.5 sensors, Mathematical Biosciences and Engineering, 2019, 16(6), 6858–6873.
43. R. Kalaivani, K. P. Sampoornam, G. S. Mathi and M. Nivetha, AI-based IoT Real Time Air Monitoring System, in 2025 5th International Conference on Trends in Material Science and Inventive Materials (ICTMIM), IEEE, 2025, pp. 1273–1278.
44. C. Mai, Z. Wang, L. Chen, Y. Huang, M. Li, A. Shirazi, A. Altaee and J. L. Zhou, Field-based calibration and operation of low-cost sensors for particulate matter by linear and nonlinear methods, Atmos. Pollut. Res., 2025, 102676.
45. G. Kumar, D. P. Kumar, J. Dhariwal and S. Srirangarajan, Reliability assessment of low-cost PM sensors under high humidity and high PM level outdoor conditions, IEEE Sens. J., 2025, 25(17), 32746–32758.
46. J. Dong, N. Goodman, A. Carre and P. Rajagopalan, Calibration and validation-based assessment of low-cost air quality sensors, Sci. Total Environ., 2025, 977, 179364.
47. Z. Feng, L. Zheng, L. Liu and W. Zhang, Real-Time PM2.5 Monitoring in a Diesel Generator Workshop Using Low-Cost Sensors, Atmosphere, 2022, 13(11), 1766.
48. D. Suriano and M. Penza, Assessment of the performance of a low-cost air quality monitor in an indoor environment through different calibration models, Atmosphere, 2022, 13(4), 567.
49. J. Zhang, L. Bai, N. Li, Y. Wang, Y. Lv, Y. Shi, C. Yang and C. Xu, Low-Cost Particulate Matter Mass Sensors: Review of the Status, Challenges, and Opportunities for Single-Instrument and Network Calibration, ACS Sens., 2025, 10(5), 3207–3221.
50. J. Ouimette, W. P. Arnott, P. Laven, R. Whitwell, N. Radhakrishnan, S. Dhaniyala, M. Sandink, J. Tryner and J. Volckens, Fundamentals of low-cost aerosol sensor design and operation, Aerosol Sci. Technol., 2024, 58(1), 1–15.
51. A. Di Antonio, O. A. Popoola, B. Ouyang, J. Saffell and R. L. Jones, Developing a relative humidity correction for low-cost sensors measuring ambient particulate matter, Sensors, 2018, 18(9), 2790.
52. I. Demanega, I. Mujan, B. C. Singer, A. S. Anđelković, F. Babich and D. Licina, Performance assessment of low-cost environmental monitors and single sensors under variable indoor air quality and thermal conditions, Building and Environment, 2021, 187, 107415.
53. J. L. Hand and S. M. Kreidenweis, A new method for retrieving particle refractive index and effective density from aerosol size distribution data, Aerosol Sci. Technol., 2002, 36(10), 1012–1026.
54. R. H. Moore, E. B. Wiggins, A. T. Ahern, S. Zimmerman, L. Montgomery, P. Campuzano Jost, C. E. Robinson, L. D. Ziemba, E. L. Winstead and B. E. Anderson, Sizing response of the Ultra-High Sensitivity Aerosol Spectrometer (UHSAS) and Laser Aerosol Spectrometer (LAS) to changes in submicron aerosol composition and refractive index, Atmos. Meas. Tech., 2021, 14(6), 4517–4542.
55. P. Salimifard, D. Rim and J. D. Freihaut, Evaluation of low-cost optical particle counters for monitoring individual indoor aerosol sources, Aerosol Sci. Technol., 2020, 54(2), 217–231.
56. S. C. C. Lung, M. C. M. Tsou, S. C. Hu, Y. H. Hsieh, W. C. V. Wang, C. K. Shui and C. H. Tan, Concurrent assessment of personal, indoor, and outdoor PM2.5 and PM1 levels and source contributions using novel low-cost sensing devices, Indoor Air, 2021, 31(3), 755–768.
57. Z. Idrees, Z. Zou and L. Zheng, Edge computing based IoT architecture for low cost air pollution monitoring systems: a comprehensive system analysis, design considerations & development, Sensors, 2018, 18(9), 3021.
58. G. Coulby, A. K. Clear, O. Jones and A. Godfrey, Low-cost, multimodal environmental monitoring based on the Internet of Things, Build. Sci., 2021, 203, 108014.
59. G. W. Thomas, S. Sousan, M. Tatum, X. Liu, C. Zuidema, M. Fitzpatrick, K. A. Koehler and T. M. Peters, Low-cost, distributed environmental monitors for factory worker health, Sensors, 2018, 18(5), 1411.
60. L. Spinelle, M. Gerboles, M. G. Villani, M. Aleixandre and F. Bonavitacola, Field calibration of a cluster of low-cost commercially available sensors for air quality monitoring. Part B: NO, CO and CO₂, Sens. Actuators, B, 2017, 238, 706–715.
61. C. Malings, R. Tanzer, A. Hauryliuk, P. K. Saha, A. L. Robinson, A. A. Presto and R. Subramanian, Fine particle mass monitoring with low-cost sensors: corrections and long-term performance evaluation, Aerosol Sci. Technol., 2020, 54(2), 160–174.
62. B. Feenstra, V. Papapostolou, S. Hasheminassab, H. Zhang, B. Der Boghossian, D. Cocker and A. Polidori, Performance evaluation of twelve low-cost PM2.5 sensors at an ambient air monitoring site, Atmos. Environ., 2019, 216, 116946.
63. H. Yiğitler, B. Badihi and R. Jäntti, Overview of time synchronization for IoT deployments: clock discipline algorithms and protocols, Sensors, 2020, 20(20), 5928.
64. J. K. Bean, Evaluation methods for low-cost particulate matter sensors, Atmos. Meas. Tech., 2021, 14(11), 7369–7379.
65. H. Chojer, P. Branco, F. Martins, M. Alvim-Ferraz and S. Sousa, Development of low-cost indoor air quality monitoring devices: recent advancements, Sci. Total Environ., 2020, 727, 138385.
66. M. Ródenas García, A. Spinazzé, P. T. Branco, F. Borghi, G. Villena, A. Cattaneo, A. Di Gilio, V. G. Mihucz, E. Gómez Álvarez and S. I. Lopes, Review of low-cost sensors for indoor air quality: features and applications, Appl. Spectrosc. Rev., 2022, 57(9–10), 747–779.
67. A. Tiele, S. Esfahani and J. Covington, Design and Development of a Low-Cost, Portable Monitoring Device for Indoor Environment Quality, J. Sens., 2018, 2018(1), 5353816.
68. T. W. Hnat, V. Srinivasan, J. Lu, T. I. Sookoor, R. Dawson, J. Stankovic and K. Whitehouse, The hitchhiker's guide to successful residential sensing deployments, in Proceedings of the 9th ACM Conference on Embedded Networked Sensor Systems, 2011, pp. 232–245.
69. E.-S. Apostol, C.-O. Truică, F. Pop and C. Esposito, Change point enhanced anomaly detection for IoT time series data, Water, 2021, 13(12), 1633.
70. D. M. Lunderberg, Y. Liang, B. C. Singer, J. S. Apte, W. W. Nazaroff and A. H. Goldstein, Assessing residential PM2.5 concentrations and infiltration factors with high spatiotemporal resolution using crowdsourced sensors, Proc. Natl. Acad. Sci. U. S. A., 2023, 120(50), e2308832120.
71. J. Xiang, C.-H. Huang, J. Shirai, Y. Liu, N. Carmona, C. Zuidema, E. Austin, T. Gould, T. Larson and E. Seto, Field measurements of PM2.5 infiltration factor and portable air cleaner effectiveness during wildfire episodes in US residences, Sci. Total Environ., 2021, 773, 145642.
72. J. He, C.-H. Huang, N. Yuan, E. Austin, E. Seto and I. Novosselov, Network of low-cost air quality sensors for monitoring indoor, outdoor, and personal PM2.5 exposure in Seattle during the 2020 wildfire season, Atmos. Environ., 2022, 285, 119244.
73. O. Safety and H. Administration, Indoor Air Quality in Commercial and Institutional Buildings, Maroon Ebooks, 2015.
74. A.-K. Viitanen, S. Uuksulainen, A. J. Koivisto, K. Hämeri and T. Kauppinen, Workplace measurements of ultrafine particles—a literature review, Ann. Work Exposures Health, 2017, 61(7), 749–758.
75. S. Cho, G. Lee, D. Park and M. Kim, Study on characteristics of particulate matter resuspension in school classroom through experiments using a simulation chamber: influence of humidity, Int. J. Environ. Res. Public Health, 2021, 18(6), 2856.
76. H. H. Jo, J. Nam, J. Y. Choi, T. Kim and S. Kim, Particulate matter in school buildings: influence of influx control of outdoor PM particles and building renovation in summer, Environ. Pollut., 2023, 329, 121582.
77. R. E. Connolly, Q. Yu, Z. Wang, Y.-H. Chen, J. Z. Liu, A. Collier-Oxandale, V. Papapostolou, A. Polidori and Y. Zhu, Long-term evaluation of a low-cost air sensor network for monitoring indoor and outdoor air quality at the community scale, Sci. Total Environ., 2022, 807, 150797.
78. S. v. Ratingen, J. Vonk, C. Blokhuis, J. Wesseling, E. Tielemans and E. Weijers, Seasonal influence on the performance of low-cost NO₂ sensor calibrations, Sensors, 2021, 21(23), 7919.
79. H.-J. Chu, M. Z. Ali and Y.-C. He, Spatial calibration and PM2.5 mapping of low-cost air quality sensors, Sci. Rep., 2020, 10(1), 22079.
80. H. Zhang, S. Zhang, W. Pan and Z. Long, Low-cost sensor system for monitoring the oil mist concentration in a workshop, Environ. Sci. Pollut. Res., 2021, 28(12), 14943–14956.
81. U.S Department of Health and Human Services Centers for Disease Control and Prevention, Record of the Proceedings: Annual Meeting of the Advisory Committee on Childhood Lead Poisoning Prevention, National Center for Environmental Health/Agency for Toxic Substances and Disease Registry, Atlanta, Georgia, 2012.
82. M. Binuya, E. Engelhardt, W. Schats, M. Schmidt and E. Steyerberg, Methodological guidance for the evaluation and updating of clinical prediction models: a systematic review, BMC Med. Res. Methodol., 2022, 22(1), 316.
83. B. Van Calster, E. W. Steyerberg, L. Wynants and M. Van Smeden, There is no such thing as a validated prediction model, BMC Med., 2023, 21(1), 70.
84. N. Zimmerman, A. A. Presto, S. P. Kumar, J. Gu, A. Hauryliuk, E. S. Robinson and A. L. Robinson, A machine learning calibration model using random forests to improve sensor performance for lower-cost air quality monitoring, Atmos. Meas. Tech., 2018, 11(1), 291–313.
85. R. Huerta, T. Mosqueiro, J. Fonollosa, N. F. Rulkov and I. Rodriguez-Lujan, Online decorrelation of humidity and temperature in chemical sensors for continuous monitoring, Chemom. Intell. Lab. Syst., 2016, 157, 169–176.
86. S. Zheng, H. Shen, G. Shen, Y. Chen, J. Ma, H. Cheng and S. Tao, Vertically-resolved indoor measurements of air pollution during Chinese cooking, Environ. Sci. Ecotechnology, 2022, 12, 100200.
87. A. N. Patton, K. Medvedovsky, C. Zuidema, T. M. Peters and K. Koehler, Probabilistic machine learning with low-cost sensor networks for occupational exposure assessment and industrial hygiene decision making, Ann. Work Exposures Health, 2022, 66(5), 580–590.
88. A. Preston and K.-L. Ma, Communicating uncertainty and risk in air quality maps, IEEE Trans. Visualization Comput. Graphics, 2022, 29(9), 3746–3757.
89. L. Morawska, G. Ayoko, G. Bae, G. Buonanno, C. Chao, S. Clifford, S. C. Fu, O. Hänninen, C. He and C. Isaxon, Airborne particles in indoor environment of homes, schools, offices and aged care facilities: the main routes of exposure, Environ. Int., 2017, 108, 75–83.

---