Indoor air quality monitoring and source apportionment using low-cost sensors

Understanding of the various sources of indoor air pollution requires indoor air quality (IAQ) data that is usually lacking. Such data can be obtained using unobtrusive, low-cost sensors (LCS). The aim of this review is to examine the recent literature published on LCS for IAQ measurements and to determine whether these studies employed any methods to identify or quantify sources of indoor air pollution. Studies were reviewed in terms of whether any methods of source apportionment were employed, as well as the microenvironment type, geographical location, and several metrics relating to the contribution of outdoor pollutant ingress versus potential indoor pollutant sources. We found that out of 60 relevant studies, just four employed methods for source apportionment, all of which utilised receptor models. Most studies were undertaken in residential or educational environments. There is a lack of data on IAQ in other types of microenvironments and in locations outside of Europe and North America. There are inherent limitations with LCS in terms of producing data which can be utilised in source apportionment models. This applies to external pollution data, however IAQ can be even more challenging to measure due to its characteristics. The indoor environment is heterogeneous, with significant variability within the space as well as between different microenvironments and locations. Sensor placement, occupancy, and activity reports, as well as measurements in different microenvironments and locations, can contribute to understanding this variability. Outdoor pollutants can ingress into the space via the building envelope, however measurement of external pollution and environmental conditions, as well as recording details on the building fabric and ventilation conditions, can help apportion external contributions. Whether or not source apportionment models are employed on indoor data from LCS, there are parameters which, if carefully considered during measurement campaigns, can aid in source identification of pollutants.


Introduction
Pollutants in the atmosphere and their effects on ambient air quality are one of the primary concerns of governments around the world (UNEP 2021).The World Health Organisation estimates that seven million deaths per year can be attributed to the adverse health effects of ambient and household air pollution (WHO, 2021).Due attention has been paid by many researchers and governmental bodies to the issue of air pollution in the external environment.Limits have been set for various pollutants by the WHO (2021), European Commission (EC 2022 and2008), and Public Health England in the UK (PHE 2019), among others.However, far less investigation has been conducted into air quality inside buildings (Lewis et al 2022).The research conducted so far has often discovered higher levels of air pollutants indoors (Hoskins 2003, Bari et al 2015), which is concerning given that humans typically spend most of their days inside (Matz et al 2014).benefit to presenting a framework for IAQ measurement parameters which contribute to source identification, whether or not source apportionment methods are applied.As such we attempt to fill the research gap by synthesising recent studies of IAQ using LCS in terms of source apportionment methods and measurement parameters which facilitate source identification.
The primary objectives of this review are therefore: (i) to provide an overview of recent studies in which LCS have been used to measure IAQ in different microenvironments and across geographical locations, (ii) to assess whether they have employed any methods for source apportionment of indoor air pollution, and (iii) to examine the measurement methodology and reporting of these studies in terms of parameters which would facilitate better identification and understanding of indoor pollutants.

Methods, scope, and outline
Given the interest in global and local air quality and the lack of quantification of indoor pollutant levels, this is a fast-moving area of research and studies are continually being undertaken at locations around the world.These studies vary considerably in their approach and methodology, and this introduces uncertainties regarding the reliability of data from measurement using LCS.We undertook a state-of-the-art review of recent IAQ studies, which utilised LCS, the results of which are presented in the remaining sections.A literature search was conducted using Scopus and the terms 'indoor AND air AND quality AND low AND cost AND sensor', restricted to the previous five years (2018 to 2023) due to the pace at which research is being conducted in this field at present.The results were scanned for relevance by way of their abstracts and out of 359 remaining studies, 72 included location information.The study locations are marked on a map in figure 3 of section 3.4.2.Conference papers have been included in the location map but excluded from further analysis in the text.
A search of the literature containing all keywords 'source AND apportionment AND indoor AND air AND pollution AND low AND cost AND sensor' returned three relevant titles.Therefore, two additional literature searches were conducted using Scopus to identify additional studies relating to this topic.The first search employed the terms 'source AND apportionment AND indoor AND air AND pollution' and the second with the terms 'source AND apportionment AND low AND cost AND sensor', date range restricted to between 2018 and 2023.The first search returned 166 titles, the second returned 31 titles, these were filtered to include journal articles only, and the remaining abstracts were screened for relevance.Only four studies were found to have been conducted using data from LCS.Other relevant titles discovered via the search results were also included in the literature review.Detailed information on these studies is provided in the subsequent sections of this paper and tabulated in table 2 of section 3.4.A graphical representation of the review process is provided in figure 1(a) and (b).
A wealth of research has been undertaken in recent years on the use of LCS for measuring air quality in external environments to supplement fixed monitoring (Rai et al 2017, Karagulian et al 2019), and on the applicability of citizen science programmes using LCS to enhance the spatiotemporal availability of air quality data and raise public awareness of the issue of air pollution (Lu et al 2022, Mahajan et al 2020, Wesseling et al 2019).This review excludes these topics and focuses primarily on the use of LCS for measuring air pollution indoors, and specifically on their measurement methodology and reporting.Portable/wearable low-cost air quality sensors have also been excluded from this review.
Table 1.Review papers on the use of LCS for measuring IAQ, published between 2014 and 2022, with three studies published in 2021, two in 2022, 2018 and 2016, one in 2017 and one in 2014.Four were systematic reviews and eight were state-of-the-art reviews.The focus of the authors included assessing LCS against reference instruments, highlighting the strengths and limitations of LCS, utilising LCS for exposure assessments, LCS for energy management in buildings, and using LCS in Smart Homes.

Author (year)
Review focus Sá et al (2022) LCS assessed against comparison instruments Ródenas García et al (2022) Strengths and limitations of LCS for different applications Omidvarborna et al (2021) Advantages and drawbacks of using LCS in Smart Homes Fanti et al (2021) Systematic review of LCS for exposure assessment Saini et al (2021) Systematic review of LCS which use IoT-based applications Chojer et al (2020) Systematic review of LCS for IAQ monitoring Morawska et al (2018) State-of-the-art review of LCS for monitoring indoor and outdoor air pollution Schieweck et al (2018) Evaluation of LCS for measuring IAQ in Smart Homes Borghi et al (2017) Systematic review of LCS for indoor or outdoor exposure Kumar et al (2016a) Review of IAQ sensing technology and energy management Kumar et al (2016b) Review of LCS technologies for IAQ monitoring in urban buildings Wang and Brauer (2014) Review of next generation LCS for indoor and outdoor monitoring Table 2. Tabulated details of 60 IAQ studies using LCS which were reviewed by the authors.Details include pollutants measured, sensing systems employed, whether the sensors were laboratory or field collocated, the type of environment in which the measurements were conducted, the building type and construction year, the floor of the building where the measurements were conducted, sensor locations, details of external doors and windows, the number of occupants present, whether activity logs were kept, details of the ventilation systems, whether the investigators considered the influence of external pollutant levels, and the location of the studies.
The remainder of this paper focuses on the following: section 3 presents the results of this review, in which section 3.1 provides an overview of the use of LCS for measuring IAQ, their limitations, and requirements for instrumentation when employing source apportionment methods, section 3.2 discusses calibration methods for LCS, section 3.3 includes relevant advice on LCS data quality management and utilisation, and section 3.4 provides a detailed assessment of the reviewed studies, first in relation to methods for source apportionment of air pollutants, and subsequently in terms of metrics which can influence (i) the levels of external pollutant ingress into the measurement space, and (ii) indoor pollutant generation and the influence of occupancy and activity.Section 4 presents the conclusions and recommendations for future measurements of IAQ using LCS.

Results and discussion
3.1.Principles of LCS Sensors for measuring gaseous pollutants function according to specific principles, which are different from those for measuring particulate matter (PM).Electrochemical (EC), non-dispersive infrared (NDIR), metal oxide semiconductor (MOS), and photo-ionization detection (PID) are all types of the low-cost gas sensors.EC sensors quantify the concentration of a specific gas in the air by way of the current generated from electrochemical interactions between that gas and the sensing material (Liang et al 2021).NDIR sensors detect decreases in infrared radiation that occur when the target gas passes through an active filter (AQEG, 2022).MOS sensors recognise their target gas via a gas-solid interaction that induces an electronic charge to the metal oxide surface, which is then transduced into an electrical signal.LCS designed to detect volatile organic compounds (VOCs)-off-gasses from common household products and materials-can be MOS, or they can operate as PID systems.PID sensors use ultraviolet rays to ionize molecules, with the positive ions then collected by electrodes and converted into a digital reading to quantify chemicals in the measured gas (Coelho Rezende et al 2019).Most LCS for PM function by way of light-scattering technology.Particles of various sizes are carried in air flow, and as these pass a focused beam of light within the sensor, that light is then scattered and recorded by a photodetector.In 'volume scattering' nephelometers, the total light scattered is recorded as a single output and is converted into a particle mass concentration by comparison to readings from a reference instrument.In 'optical particle counters (OPCs)' the size of the different particles is estimated, and a particle mass concentration is provided by assuming that they are: spherical, have a consistent bulk density, and have the same refractive index (Morawska et al 2018).

LCS for IAQ monitoring and source apportionment
Source apportionment methods fall under two main categories-source-oriented models and receptor-oriented models (Mircea et al 2020).Source-oriented models mimic the physical and chemical processes that are involved in pollutant emission (Mircea et al 2020).They use emissions inventories as input data and evaluate using chemistry, transport, and dispersion models, for example by correlating wind direction with levels of measured pollutants to identify source locations (Viana et al 2008).Emissions inventories are developed using referencegrade monitoring instruments, for various pollutant sources and in many locations.These inventories, however, can vary in accuracy and coherency and are often not available.This method of understanding the contribution of various sources to air pollutant levels can therefore be limited in its applicability (Viana et al 2008, Kumar et al 2013).
The most-often used source-oriented models can be categorised as one of the following: Gaussian and non-Gaussian parametrised models, Lagrangian puff and particle models, and Eulerian chemical transport models.Parametrised models assume uniform or simplified meteorology, parametrising diffusion of plumes according to meteorological and terrain types.They construct steady-state situations in sequence, and do not account or chemical or physical transformations.Lagrangian models compute the movement of pollutants in a threedimensional flow field.They do so with pollutants modelled either as puffs or particles, with different approaches to turbulent diffusion for each.These models consider pollutants to be unreactive but can include simplified chemical transformations.Eulerian chemical transport models simulate dispersion in a threedimensional grid, in which the concentrations in a cell represent the total amount of pollutant present, without determining individual sources.They can model pollutants as non-reactive but are useful for modelling physical processes and chemical reactions since these involve the total amount of different species at a location (Mircea et al 2020).Source-oriented models have advantages which include their ability to provide output with high temporal resolution, to predict changes in air quality which relate to changes in emissions and provide evaluations for sites which do not have monitoring data available.These models are however limited by the quality of their input data.Data quality with LCS is discussed in subsequent sections of this paper (Mircea et al 2020).
Receptor-based methods focus on properties of the ambient environment to calculate the contribution of sources at the receptor site at a given time.They assume mass and species conservation between the source of emissions and the receptor and identify sources by solving the mass balance equation.Receptor-oriented models are based on measured data and do not require additional datasets such as emissions, meteorological information, or air concentrations at boundaries.In terms of the concentration of the pollutant species at the receptor, these models either use measured 'source fingerprint' or 'source profile' data or derive this information by way of an iterative process using 'factor profiles'.Receptor models exist on a spectrum in terms of how much knowledge of pollutant sources is required.Chemical mass balance (CMB) models require almost complete source profile information, as comprehensive knowledge of all emissions relevant to the receptor must be included for evaluating uncertainty (Mircea et al 2020).CMB models of indoor pollution therefore require identification of all potential pollutant sources (e.g., fires, cooking activities, traffic pollution ingress, etc).
Multivariate factor analysis models including positive matrix factorisation (PMF) and non-negative matrix factorisation (NMF) do not require source composition information (Viana et al 2008).These models use pollutant species concentrations at the receptor site to resolve a weighted factorisation problem (Mircea et al 2020).They also rely on known experimental uncertainties, however, which may or may not be available for LCS data.The incremental or 'Lenschow' approach to receptor-oriented models uses calculated differences in concentrations between spatial locations to determine source contributions (Mircea et al 2020).This has potential applications for using LCS in indoor source apportionment studies, given an effective sensor deployment strategy.Receptor model outputs are for specific site and time windows, and they assume neither that source profiles change significantly, nor that pollutants change during transport from source to receptor (Mircea et al 2020).
A review by Saraga et al (2023) found that PMF was the most often used source apportionment method on indoor air pollution, with 31% of studies reviewed employing this method.They found that PCA was the next most frequently utilised method (27% of studies), and just 7% used CMB models.However, these studies used data from reference-grade instruments.Some recent outdoor studies have used data from LCS along with other supplementary data to provide an indication of possible pollution sources.For instance, Hagan et al (2019) used NMF, a receptor-based model, on data from LCS in Delhi, India, and compared it to data from reference-grade instruments to identify combustion as a source of CO.Hodoli et al (2023) used source-based apportionment methods with data on wind speed and direction and particle size distribution to infer locations and types of sources of atmospheric PM in Ghana.Bousiotis et al (2021) used receptor-based k-means clustering with concurrent measurements from regulatory-grade instruments and LCS to identify sources of pollution at a site in Birmingham, UK.Bousiotis et al (2022) then went on to apply PMF receptor-based analysis to apportion sources using data collected from LCS.This study also used data from reference-grade instruments and aerosol chemical speciation monitors to verify their analysis.
When selecting or implementing source apportionment models, consideration must be given to the following: the spatial scale of the assessment, i.e., a high-resolution model may be preferable in small areas with high contributions from nearby sources, the use of reliable input data, identification of the full inventory of sources; and model validation (Mircea et al 2020).A combination of both source-and receiver-oriented models can form a coherent picture of air pollution, given that at any one location the air quality is a result of dispersion of emissions and secondary processes which change primary pollutants into secondary ones (Mircea et al 2020).The combination of source and receptor-based models in IAQ studies warrants further research.Traditionally, source apportionment methods require a large amount of data, offline sampling, and analysis in a laboratory to determine the chemical make-up of the samples.The following sections provide an overview of the limitations of different LCS to measure IAQ and conduct source apportionment.

CO 2 sensors
The measurement accuracy of many NDIR LCS for CO 2 is in the range of ±30-50 ppm for an atmospheric concentration of around 400 ppm when compared with reference instruments, however, this is widely considered acceptable for many applications (Müller et al 2020).A study conducted by Madureira et al in 2016 utilised CO 2 data from NDIR sensors in a PCA to investigate relationships between classroom characteristics, occupant behaviour and CO 2 .The fluctuations in CO 2 were relatively large given that it was measured in classrooms which are high-density spaces, as such the measurement variation of the sensors may not have been a major concern.However, in cases where measurements of small CO 2 concentration differences, relative to the accuracy of 30-50 ppm, is required, NDIR data may not be usable.EC and MOS sensors are sensitive to changes in atmospheric conditions, and they have high cross-sensitivity between different gases (AQEG 2022).A susceptibility to interference from other pollutants may mean that data from these LCS is inappropriate for source apportionment models.3.2.2.VOC sensors LCS which measure a metric of 'total VOC' or 'tVOC' output a level which can include a large (between 100 to 200) individual substances and cannot provide chemical speciation.As a result, the concentrations reported by these sensors cannot be attributed to all compounds present in the atmosphere.Low-cost VOC sensors can detect some but not all VOCs and are more sensitive to specific compounds (Ródenas García et al 2022).Like many other LCS, sensors to detect VOCs do not have a high degree of selectivity, which is a limitation given that compounds are often found at low concentrations, but some are very toxic, e.g., formaldehyde and benzene (Kumar et al (2016b)).

PM sensors
Low-cost PM sensors cannot detect ultrafine particles (UFPs)-particles less than 0.1 μm in diameter (Kumar et al 2014), as these are not large enough to sufficiently scatter light (Morawska et al 2018).Challenges with PM sensors include cheaper pumps that may not adequately control flow rate and lead to errors, impacts of environmental factors such as relative humidity (RH) and cross-contamination of other pollutants (Rodenas Garcia et al 2022).High RH (above 85%) can cause PM sensors to overestimate concentrations.The inlet efficiency depends on the particle size and on the airflow in the environment, which is typically low indoors, and as such particle losses between the inlet and the sensing zone can be significant (AQEG 2022).Source apportionment methods are applied either on the chemical composition of particles, which requires impactor sampling using a gravimetric filter-based method and off-line analysis in the laboratory for major chemical components, or on their size distribution, which uses particle number concentration (PNC) and particle number size distribution (Saraga et al 2023).Neither is currently possible with low-cost PM sensors.

Limitations common to all LCS
A limitation inherent in all instruments is that they can suffer from measurement drift over time and require frequent calibration to ensure accuracy (Chojer et al 2020).Calibration of LCS is discussed in section 3.2.In addition, there are several features inherent in LCS which need to be considered when using them for IAQ measurements.These include their power supply and consumption, sensitivity, selectivity, speed, stability, durability, reliability, and detection limits (Morawska et al 2018).Kumar et al (2016a and2015) summarised that LCS should ideally be powered by long-life batteries, as using mains power in the field is often not feasible.Preferably they should be miniaturised and with a low sound power level so as not to cause a nuisance to occupants of adjacent spaces.A major challenge for LCS is the ability to detect sufficiently low concentrations of pollutants when measuring indoors (Kumar et al 2016b).

Calibration of LCS
There is currently no recognised standard procedure for the calibration of LCS (Russell et al 2022).The sensors can be calibrated in the laboratory, prior to deployment, or in the field by comparing their results to those from reference-grade instruments.There is a need to perform both field and laboratory calibration for LCS, as good correlation with reference instruments in the lab does not always correspond to equally good performance under field conditions (Castell et al 2017).LCS are sensitive to fluctuations in environmental conditions such as changes in temperature, RH, wind direction and speed, and the presence of other interacting pollutants.A systematic review by Chojer et al (2020) found that only 16 out of 35 studies had laboratory-calibrated their LCS, and even fewer of these collocated them with reference instruments.Kumar et al (2015) remarked that a good deal of measurement uncertainty is introduced in field calibration.It is necessary to conduct collocation measurements with LCS, using reference-grade instruments and under a range of environmental conditions, to verify performance and determine any correction factors which need to be applied to the LCS data.The EPA provides guidance on different collocation strategies and cites the benefits and drawbacks of each (Clements et al 2022).The EPA has also developed Federal Reference Methods (FRMs) to measure air pollutants accurately and reliably (EPA website, accessed December 2023).The Air Quality Sensor Performance Evaluation Center (AQ-SPEC) at the South Coast Air Quality Management District in the State of California has proposed a protocol whereby sensors must be tested under field conditions at two different fixed monitoring stations.If the results of the field tests are promising, subsequent laboratory evaluation can be undertaken (South Coast AQMD 2017).
Limitations are inherent to field collocation of LCS with reference instruments.Despite best intentions of researchers, it is likely impossible to conduct collocation under a wide-enough range of atmospheric and operational conditions.Omidvarborna et al (2020) developed a custom environmental-pollution ('Envilution TM ') chamber to generate a controlled environment for temperature and relative humidity, along with different concentrations of particles, to simulate a variety of real-world conditions to performance evaluate low-cost PM sensors.Another relatively new method has been developed for calibrating low-cost gas sensors, which has been shown to equal that of laboratory and field calibration, while taking a fraction of the time to conduct.The methodology has been termed enhanced ambient sensing environment (EASE) and uses a mixture of ambient air plus added pollutants to measure in-duct.Russell et al detail the methodology and its use for MOS and EC sensors (specifically for NO 2 and O 3 ) in their Russell (2022) study.
In addition to traditional laboratory calibration and field collocation, machine learning algorithms have been developed which can be applied to LCS data to account for drift over the measurement period.This method is becoming increasingly popular (e.g., Wang et al 2023).Several of these algorithms are outlined in Morawska et al (2018): the LCS can be collocated with a reference instrument for a designated 'training period' to develop a machine learning algorithm, the data from the LCS is corrected based on its expected agreement with a reference instrument that is not adjacent to the sensor, or calibration algorithms are developed by the LCS manufacturer and applied to the sensor itself or utilised in the cloud.It should be noted, however, that these calibration algorithms may not always be reliable, and it is often, if not always, necessary to frequently calibrate LCS in the field over longer periods of deployment (De Vito et al 2020).Wallace (2022) analysed the proprietary algorithms employed by Plantower for their PMS5003 sensors, used in the PurpleAir sensing system which has been widely used, and found them to be 'seriously flawed'.
In the 60 studies using LCS to measure IAQ reviewed in this paper, 36 reported collocating the LCS with reference instruments in the field, 14 undertook laboratory calibration of the sensors prior to deployment, and just seven reported doing both.This information is presented graphically in figure 2. Nine studies stated that the LCS used had been factory calibrated and therefore did not require further laboratory calibration by the users.

Data quality, management, and utilisation
Data quality from LCS for both outdoor and indoor air pollution measurement is often a concern.Any source apportionment model requires accurate input data to produce accurate outputs.The EU Directive 2008/50/EC stipulates that data from monitoring stations should be made available in standardised form to facilitate effective handling and comparison of pollutant levels.In contrast, data collected by LCS in supplementary monitoring projects tends to be of low quality not standardised in format (Kumar et al 2015).Without adequate calibration, along with frequent review and analysis of data and environmental conditions which can affect LCS response, the amount of accurate, usable data can decrease.EPA guidance stipulates that an acceptable data 'completeness level'-i.e., data collected by LCS out of all possible data which could have been collected-is 75% (Clements et al 2022).The EC has produced guides on the use of both receptor-oriented and source-oriented apportionment methods for outdoor pollution, and states that the inclusion of these methods should be planned for in advance of a field study, given the need for large amounts of high-quality data (Viana et al 2014).
This data also requires either local storage or transmission via network connectivity, both of which have implications for power consumption.Transmitting to a cloud-based storage facility may have implications in terms of data privacy and security compliance (Sun et al 2014).Stakeholders and end users should be factored into the design of any LCS system, to appropriately match the technology capability and data output/ visualisation methods.In addition, an important part of collecting IAQ data inside homes, schools, offices, and other spaces is to inform the occupants of the results and provide them with clear guidance on, where necessary, improving conditions, which involves obtaining knowledge about sources of pollutants and their impact.The authors of this paper are keen to stress the importance of parameters other than LCS performance which can influence the data that is collected and in turn affect knowledge-gathering around pollutant sources.These parameters are outlined in the subsequent sections.

IAQ study methodology
The 60 studies which comprise this review were examined in relation to several metrics: geographical location, the type of microenvironment the measurements took place in, the pollutants measured, use of any source apportionment methods, sensing systems used, whether they undertook laboratory calibration or field collocation of sensors, the building construction year, floor of the building, details of external doors, windows and ventilation openings, details of the ventilation strategy and operation during measurement, sensor placement/location, number of occupants present, and whether activity logs were kept.The reported metrics and parameters are presented in table 2. Locations of the studies, as well as an additional 23 conference papers which were excluded from further analysis in the text, are marked on a map in figure 3.
When considering the concentration of an air pollutant indoors, several factors come into play: outdoor pollutant concentration, ventilation rate or the rate of air exchange between outdoors and indoors, location and strength of indoor emission source(s), whether there are any sinks which remove the pollutant, mixing conditions inside the space, i.e., the indoor airflow and turbulence; and the dimensions of the space (AQEG 2022).By extension, there are several parameters relating to the above which should be considered when monitoring air quality indoors.The UK's Chartered Institute of Building Services Engineers (CIBSE) produced a guidance document on IAQ and associated assessment, monitoring, modelling, and mitigation (CIBSE 2021).Regarding IAQ measurements, the guide stresses the importance of considering the following parameters: construction year and/or time since the refurbishment of the building (which can provide an indication on construction type and materials used), cleaning or decorating activities undertaken, ventilation and heating or cooling of the spaces, and cooking or other occupant activities with the potential to generate pollutants.From their review, the authors of this paper found that these parameters were frequently overlooked or lacking in the reporting of studies employing LCS.The specific parameters are discussed in sections 3.4.1 to 3.4.3.

Source apportionment in IAQ studies using LCS
There are inherent limitations with LCS in terms of producing data which can be utilised in source apportionment models.This applies both to ambient pollution data as well as data collected indoors, however IAQ has specific characteristics which render it even more challenging to measure.Firstly, the indoor environment is largely heterogeneous, with significant variability within the space as well as between different microenvironments and locations.Indoor pollutants can ingress into the space via the building envelope, but can also transform via interaction into secondary pollutants, as well as via photochemistry and surface boundary layer reactions.Without identifying potential sources of indoor pollution, one cannot easily choose the appropriate factors or model for source apportionment.Most models assume stable source emissions which, as outlined above, is likely untrue within internal spaces (Saraga et al 2023).
Although not strictly a source apportionment methodology, Xiu et al (2022) utilised data from low-cost PM 2.5 and carbon monoxide (CO) sensors to develop a simplified method for determining local pollutant source contribution.The method used the mass concentration ratios of PM 2.5 to CO from measured vehicle emissions, bushfires, household biomass, cigarette smoke, incense, and mosquito coils.Results showed lower ratios in all indoor pollutant sources compared to outdoor pollutant sources, highlighting the complexity of identifying indoor sources.They did find that some ratios were clearly different, for example between petrol and diesel road vehicles.As such they concluded that CO/PM 2.5 ratio can be used to identify some sources of pollution, in advance of or in lieu of source apportionment models.
As discussed in section 3.1.1,Madureira et al (2016) used data from LCS in a principal component analysis model to investigate relationships between classroom characteristics, student behaviour and pollutants.The authors measured in 20 naturally ventilated classrooms in Porto, Portugal, and the PCA highlighted CO 2 , PM 10 and total VOCs as components for further analysis.They found that ceiling height, window area and the number of windows open had the highest influence on CO 2 .The number of windows open, presence of visible damp spots or mould, ceiling surface material, the presence of art materials, as well as external PM sources had the highest influence on internal PM 10 .And finally, the number of windows open, floor surface material, and the presence of art materials influenced the level of VOCs (Madureira et al 2016).
Just four of the studies included in table 2 utilised source apportionment methods as a means of identifying and quantifying indoor pollutant sources.Bousiotis et al (2023) placed LCS inside the bedroom, kitchen and office of a residential home, as well as outside the home, and used positive matrix factorisation to apportion PM contributions.They measured PM 1 , PM 2.5 and PM 10 and found highest concentrations of PM 2.5 and PM 10 in the bedroom due to activities and the presence of carpets and soft furnishings.While the kitchen had the highest spikes of PM, the office had the highest PM 1 concentration due to its increased ventilation, with infiltration of external PM a significant contributor.Overall, the PMF model showed 95% of ultrafine particles resulted from outdoor sources.
Ainiwaer et al (2022) measured PM 2.5 concentration at eight different heights in a residential kitchen and bedroom using LCS with high temporal resolution.The use of sensor elevation during both cooking and noncooking periods enabled source identification of PM 2.5 .From occupant activity reports the authors determined that there was no incense burning or heating devices operating during the sampling period and surmised that cooking and dust resuspension the only two significant indoor sources of PM.PCA was used to characterise source patterns and contributions from cooking events, as well as the contributions of indoor versus outdoor (from a reference monitoring site) sources to vertical PM 2.5 distributions.They measured variations of between 18 μg m −3 and 28 μg m −3 at different vertical gradients in the kitchen, underlining the importance of sensor placement when measuring pollutants from cooking activities.
Liu et al (2023) measured PM 2.5 in over 100 houses and outdoors in China, over a six-month period, and evaluated inter-and intra-household PM 2.5 variations.The authors first used a mixed model to identify factors influencing intrahousehold and interhousehold variations in PM 2.5 concentration.Intrahousehold influencing factors included outdoor PM 2.5 , outdoor temperature, wind speed and direction, temperature difference, season, and the past-day indoor PM 2.5 .Interhousehold factors of influence were found to be fuel type, open windows, and temperature difference, as well as resident education and age.The authors then used source apportionment method to identify PM 2.5 concentrations resulting from three main sources -outdoor infiltration, solid fuel use for heating or cooking, and other unidentified sources, and found that solid fuel burning contributed 31%−55% of PM 2.5 .They recommended substituting biomass pellets for solid fuel to significantly reduce PM 2.5 .Zhang et al (2022) utilised data from the Air Quality Egg V2 and their custom-built LCS system using sensors from the same manufacturer.The authors' intention was to identify relationships between concentrations of indoor air pollutants and architectural-environmental factors of university buildings.They ran a PCA to identify factors influencing concentrations of PM, nitrogen dioxide (NO 2 ), and ozone (O 3 ).The indoor ventilation conditions were controlled in all buildings, with air conditioning, mechanical ventilation, HEPA with MERV13 filters, and enclosed windows a requirement.They found strong associations between indoor PM 2.5 /PM 10 /O 3 values and their corresponding outdoor concentrations, as well as distance of the building from the major traffic, cracks in the building structure, outdoor temperature, and humidity.NO 2 concentrations inside were also affected by distance of the building from the major traffic, as well as indoor relative humidity, O 3 , air grilles installed, room volume, and window-to-wall ratio.
As far as the authors of this paper can tell few, if any, other studies have been undertaken using data LCS to undertake source apportionment indoors, although a discussion on the topic was published at least 35 years ago (Sexton and Hayward 1987).The inherent requirements for source apportionment methods may explain the lack of implementation in IAQ studies so far.There are several determinants which influence indoor air pollutant concentration and must be included in any model of IAQ: outdoor pollutant concentration, ventilation rate, indoor source strength, mixing conditions in the space, and the specific chemical reactivity, amount of adsorption and other characteristics which relate to how long the pollutant stays in the atmosphere (Sexton and Hayward 1987).Knowledge of all these determinants allows IAQ to be modelled using the mass balance equation, assuming a well-mixed compartmentalised space.However, several, if not all, of these determinants are difficult to measure in field studies employing LCS, and assuming a well-mixed zone may be inappropriate (Sexton and Hayward 1987).The reporting or consideration of these determinants within the studies reviewed in this paper are discussed in the subsequent sections.

External pollutant ingress
The WHO (2005) stresses the importance of geographical location and the spatial distribution of sources when considering external air pollution, with local, urban, regional, and global distribution depending on the atmospheric lifetime of different pollutants.Given that IAQ results from a combination of both internal generation and external pollutant ingress, it is imperative to conduct IAQ studies in as many locations as feasible.It is clear from the map in figure 3 that most studies have been conducted in the northern hemisphere, with Africa and South America particularly underrepresented.More information is required on levels of air pollution indoors in all locations, however this can be of particular importance across the developing world, where there is widespread reliance on coal and biomass burning for cooking and heating (Bruce et al 2000).
Just over half of the studies reviewed took advantage of fixed monitoring stations in the locality and compared external pollution levels to those measured indoors (Kumar et al 2023a, 2023b, He et al 2022), a useful initiative as the variance between outdoor and indoor air quality needs to be considered on a case-by-case basis.Levels of radon and carbon monoxide, for example, can vary greatly depending on location (Miles andAppleton 2005, Holloway et al 2000).External pollutant ingress to buildings depends on the proximity of the building to outdoor sources but also the building construction (specifically the penetration fraction of the building envelope) and ventilation rate (WHO 2021).Twenty percent of the articles in this review cited the building construction or renovation year, 25% the building type, and 50% the level of the building on which the measurements were conducted.Thirty-seven percent provided details of the external windows, doors, and ventilation openings of the measurement room, but none stated the glazing type or façade construction.This information is presented graphically in figure 4. Sixty-seven percent discussed the ventilation strategy and operation, a topic which will be further discussed in section 3.4.3.
Given the need for better source identification in IAQ monitoring, this information could all form part of a comprehensive LCS field study methodology and provide valuable insight into the contribution from outdoor versus indoor sources of pollutants.For instance, Salthammer et al (2022) emphasised the need to look at building conditions when assessing IAQ in schools and the resultant impacts on the health, well-being and learning capability of children.Air tightness testing-the practice of measuring the amount of air that escapes through gaps in the building fabric-could also be a useful addition to any IAQ measurement campaign (d'Ambrosio et al 2016).
3.5.3.Indoor pollutant generation and the influence of occupancy and activity IAQ is influenced not just by the ingress of pollutants via the building envelope, but also by materials and pollutant sources located inside and by the type and duration of human activity (Pacitto et al 2018).The type of microenvironment, occupancy, and by extension the activities inside the spaces, are therefore important factors to consider when conducting IAQ studies with LCS. Figure 5(a) presents the proportion of studies reviewed per microenvironment category.The majority of studies were conducted inside homes and educational buildings.Out of the four studies that used source apportionment methods, three were conducted in homes and one was conducted in university buildings.The 'other' microenvironment category includes three studies: one in a museum (Dzullkiflli et al 2018), one which measured agricultural pollutants in milking parlours and greenhouses (Tugnolo et al 2022), and one in a library (Wheeler et al 2021).There are several types of microenvironments which have scarcely been measured with LCS, as far as the authors can tell, and these include restaurants and bars, gyms and fitness facilities, concert venues, galleries, and other cultural spaces.To identify various types of pollutant-generating activities and substances-for instance, theatrical smoke/fog in concert venues and clubs (Varughese et al 2005)-and quantify their contributions to IAQ, more representative LCS studies in these microenvironments would be useful.
Figure 5(b) reports the pollutants measured in the studies, with PM 2.5 and CO 2 the most-measured pollutants, followed by PM 10 and VOCs.Of the 60 studies reviewed, just 14 measured NO 2 , four SO 2 , seven CH 2 O and only one measured C 6 H 6 .The source apportionment methods in Bousiotis et al (2023), Liu et al (2023), Ainiwaer et al (2022) and Zhang et al (2022) used PM 2.5 data.Zhang et al (2022) also employed NO 2 and O 3 data.There is a prevalence of CO meters in homes in many countries with internal combustion sources such as open fires, however these are designed primarily as alert systems, with a low sensitivity, and are incapable of monitoring lower levels of the gas.O 3 , although not directly harmful to human health, is highly interactive and a common means of secondary formation of other, more toxic, pollutants.No studies measured naphthalene, PAHs, radon, trichloroethylene, tetrachloroethylene, acetaldehyde, alpha-Pinene, D-Limonene, styrene, toluene, or xylenes.Surface finishes and furniture in rooms can be sources of VOCs and are therefore important to consider when assessing IAQ.Dust resuspension varies significantly between hard flooring and carpeted floors (Yuan et al 2023).Hardboard and woodchip boards can have formaldehyde emission rates of 0.51 mg h −1 m −2 and 1.69 mg h −1 m −2 , respectively.These rates depend on various conditions including material temperature, age after installation and local ventilation rate (CIBSE, 2021).Currently consumer products such as paints with high VOC content are regulated in the EU by way of its Directives (2004/42/EC), and manufacturers should be able to provide emissions information.Information on surfaces and finishes was included in three of the reviewed studies.
Individual behaviour is an important determinant of IAQ, and this effect increases in complexity with additional occupants (AQEG, 2022).However, just half of the 60 studies reported occupancy levels during the measurements.Twenty-three studies included activity logs which were completed by the occupants, and this was most prevalent in cooking emissions studies (e.g., Kumar et al 2022a).Information on cleaning procedures was included in some of the school studies, but otherwise, no information was provided on other potential  The proportion of studies per microenvironment category, out of 60 studies reviewed.The proportions are as follows: 31 studies were in residential accommodation, 16 were in educational facilities, which includes schools and universities, five were in healthcare facilities, two in offices, three inside different forms of transport including cars, trams, trains, buses, and ships.Three studies were grouped together in the 'other' microenvironment category, and these studies were conducted inside a library, a museum and agricultural facilities (milking parlours and greenhouses).(b) The number of studies, out of a total of 60 reviewed, which measured each pollutant.The most commonly measured pollutants were PM 2.5 , with 48 studies having measured this, and CO 2 , which was measured in 36 studies.Next were PM 10 (26 studies), VOCs (20 studies), and NO 2 (14 studies).Nine measured O 3 , seven measured CH 2 O, four SO 2 , and just one study each measured C 6 H 6 , H 2 S and NH 3 .
pollutant sources in the rooms, such as printers (Cacho et al 2013).Four studies in homes included information on the presence of pets (e.g., Justo Alonso et al 2022) which in themselves can be a source of pollutant generation and resuspension (Vardoulakis et al 2020).Moisture in the air, although not a pollutant, increases relative humidity indoors which can promote mould growth and affect the amounts of VOCs released.Water vapour is released in human respiration and at differing rates, depending on activity, for instance as adult will emit 0.04 kg h −1 when sleeping but 0.05 kg h −1 when active, and activities such as bathing or dishwashing can produce 0.2 kg h −1 and 0.4 kg h −1 , respectively (CIBSE, 2021).
A primary determinant of indoor air pollution is the ventilation strategy and operation.There is often an improvement in IAQ achieved by ventilation in terms of removing pollutants from a space, but this depends on the pollutant type and source (Kumar et al (2016a)).Standards for ventilation rates and design in the UK are found in the Building Regulations (HM Government 2021), but these will be designed for estimated occupancy levels and activities, which may differ once the space is occupied.Flowrates and by extension rates of external pollutants delivered to indoor spaces in naturally ventilated buildings depend on several factors: temperatures inside and outside the building, wind speeds, direction, and pressure coefficients, location, size and type of ventilation opening, flow paths within the space, and the flow regime (CIBSE 2021).Mechanical ventilation will often also be used to moderate the internal temperature, and flow rates for heating or cooling will differ from those for ventilation only.There will be pressure differences between spaces which are managed by mechanical ventilation, but which will also cause mixing of air (and potentially pollutants) in adjacent spaces (CIBSE 2021).Occupants will utilise ventilation to adjust the temperature within in a space to increase their thermal comfort levels or turn off/down mechanical ventilation when adversely affected by airflows, often without consideration for the impact on air quality.
Out of the 60 studies covered in this review, 40 included details of the ventilation strategy and/or operation during measurement, and 44 provided information on sensor placement or location inside the monitoring space.If a sensor is located closed to an open door/window or air inlet, pollutant levels measured by the sensor might not be representative of levels in other areas of the space.Whilst models of the indoor environment most often operate under the premise of a well-mixed space, this may not always be true (Wang and Chen 2007, Furtaw et al 1996), and deploying sensors at different locations within the space could provide valuable indication of pollutant sources.For example, Kumar et al (2023aKumar et al ( , 2023b) ) conducted a study of microcharacteristics in naturally ventilated classrooms with varying sensor locations and used this information to determine appropriate placement and operation of air purifiers.Another study by Kumar et al (2023) looked at spatial distribution of PM and ventilation efficacy on the London Underground.Omidvarborna et al (2021) discussed the two types of strategy employed when deploying LCS indoors: engineering methods, which are based on previous experience and rules of thumb, often with a uniform deployment of sensors, and an optimisation method, recently developed to consider indoor airflow patterns.Optimisation methods utilise other tools such as computational fluid dynamics (CFD), statistical models and algorithms to optimise number and placement of sensors in advance.The authors of that paper then propose their own example deployment strategy which does not require development of an optimisation model in advance, with sensors deployed based on evidence-based rationale for likely pollutant types and levels in different indoor spaces (Omidvarborna et al 2021).
The final stage of the CIBSE IAQ assessment strategy presented in their 2021 guidance document is the identification and effective mitigation or reduction of indoor air pollution.Schieweck et al (2018) stressed the need for smart home sensing systems to provide instruction on actions to improve IAQ.With the exception of CO 2 monitoring in school studies, in which alert systems are often utilised to notify teaching staff once a threshold is reached, this type of alert system is absent from the majority of studies reporting IAQ measurements with LCS.Attempts to identify and quantify internal pollutant sources can be facilitated by a good understanding of the internal space.This includes characteristics such as room geometry and layouts, locations of potential pollutant sources, surface finishes and furniture, the number of occupants and their activities, and whether any pets are present.Deployment of sensors for IAQ measurement should consider these characteristics, however out of the 60 studies reviewed in this paper, none reported all of them.The percentage of studies which reported each metric is presented graphically in figure 6.

Discussion of results
Recent studies utilising LCS to measure IAQ have been reviewed in relation to their use of source apportionment methods, as well as their measurement methodology and reporting in relation to identifying potential pollutant sources.We found that just four out of 60 studies reviewed utilised source apportionment models on their LCS data.Other parameters exist which can aid in source identification, and these must be considered when designing and implementing IAQ measurement campaigns using LCS.The key conclusions of this review are as follows: • Receptor models of source apportionment are most appropriate for IAQ measurement since they rely primarily on measured concentration data and do not require emissions or meteorological data, as sourceoriented models do.
• CMB-based receptor models require identification of all potential pollutant sources relevant to the receptor (e.g., fires, cooking activities, traffic pollution ingress, etc for indoor studies) to be able to evaluate the model uncertainty.
• Multivariate models such as PMF do not require source composition information but do require known experimental uncertainties-as such these need to be determined for IAQ/LCS data.
• Incremental or 'Lenschow' approach models could potentially be applied to IAQ studies using LCS data, since they use differences in pollutant concentrations between different spatial locations-which could be obtained with an effective sensor deployment strategy.
• There are inherent limitations, however, regarding the use of data from LCS in source apportionment, summarised in the following two points.
• A lack of granularity and a susceptibility to interference from other pollutants in gaseous sensors may preclude using these for source apportionment modelling.
• Requirements for either the chemical composition of particulate matter, or on their size distribution, may also preclude the use of many PM LCS for source apportionment.
• Metrics which relate to quantifying external pollutant ingress into the measurement space include measurement of external pollutant levels and meteorology, details of the building type, floor level of measurement, year of construction or renovation of the building, surface areas and construction of the façade and external windows, doors, and ventilation openings.
• Metrics relating to indoor pollutant generation include the type of microenvironment, number of occupants present, their activities, the presence of pets, internal surface finishes, and other potential sources of pollutants.
• There is a lack of data on IAQ in microenvironments other than residential, educational, and healthcare facilities.
• The ventilation strategy and operation influences both the level of external pollutant ingress and the levels of internal pollutant build-up.
• PM 2.5 and CO 2 were the most-often measured pollutants, followed by PM 10 and VOCs.
Figure 6.Bar graph showing the percentage of studies out of the 60 reviewed which reported various metrics relating to indoor pollutant sources.These metrics and the corresponding percentage of studies which reported them are as follows: 73% reported the sensor deployment strategy or where the sensors were placed; 50% of the studies provided the number of occupants present during the measurements, 38% kept activity logs, 7% reported whether or not the households included pets, 5% provided details on internal surface finishes and furniture, and none discussed other potential sources of pollutants except fires/stoves/cooking activities.Further information on the microenvironments is provided in table 2 of section 3.4.
• Conducting both laboratory calibration according to traceable gas standards and field collocation with reference instruments is recommended to ensure data accuracy.
• Sensor deployment strategy is an important factor in terms of identifying pollutant sources.

Recommendations and implications for future IAQ studies
Many of the factors detailed in the previous section were overlooked in recent studies, and therefore as a recommendation, going forward the planning, implementation, and reporting of IAQ using LCS should ideally include the following, regardless of whether source apportionment methods are employed: • There should be careful consideration regarding the methodology and study design, with a particular focus on the following: sensor type, number, and placement indoors, room geometry, furniture, surfaces, and finishes, appliances and other pollutant-generating equipment, façade construction and air tightness, including external windows, doors, and ventilation openings, the ventilation design and operation, occupants, and activities.
• Considering the factors highlighted in the previous point, as well as the potential for implementing 'Lenschow' based source apportionment methods, a comprehensive, well-thought-out sensor deployment strategy should be conceived and implemented.
• Ideally the reporting should include marked-up floor plans illustrating the sensor deployment strategy, internal room volume and dimensions, and potential pollutant sources identified.
• Although there are inherent limitations in using LCS, it would be beneficial, where possible, to measure a wider range of pollutants in IAQ studies.
• Measurement across a wide range of geographical locations and in different types of microenvironments outside of residential and educational facilities would help to fill the current gap in knowledge.
• Measurements of IAQ should include as standard an assessment of the building ventilation strategy and operation during measurement, and its influence on airflow and external versus internal pollutant sources affecting the space.
• Both laboratory collocation prior to deployment and field collocation during, or under similar conditions to, deployment of sensors should be conducted to maximise LCS data reliability.
• A greater focus on source identification for IAQ should be foundational to the planning and implementation of studies.
• There is an opportunity for future work to, where feasible, employ methods for source apportionment using IAQ LCS data.
This review highlights the current gaps in IAQ measurement using LCS while outlining several opportunities for future work.As an overarching theme, a greater focus on source identification for IAQ should be foundational to the planning and implementation of studies, with or without the use of source apportionment methods on the survey data.LCS can provide an effective alternative to reference-grade instruments for data collection indoors if a comprehensive measurement strategy is designed and adhered to.Data from these studies can provide valuable information on IAQ, which is urgently needed worldwide.

Figure 1 .
Figure 1.(a) Flow diagram showing review process, which comprised: the initial identification search using Scopus which returned n = 440 items, screening process to restrict the date range to between 2018 and 2023 and exclude the Review document type, which returned n = 359 items, eligibility scanning of abstracts for relevance and location information then returned n = 72 once duplicates were removed, and finally n = 72 conference papers and journal articles were included on the map in figure 3, and n = 56 journal articles were included for further analysis.(b) Flow diagram showing second part of the review process, which comprised: two subsequent searches using Scopus which returned n = 166 and n = 31 items, combining of these results, screening process to restrict date range to between 2018 and 2023 and document type first to review and second to article, which returned n = 21 and n = 140 items, respectively, eligibility scanning of review abstracts for relevance returned n = 16 items, which were themselves screened for additional paper references, resulting in a final list of n = 4 articles on the topic of source apportionment of IAQ using LCS included for further analysis.

Figure 2 .
Figure 2. Percentage of the 60 reviewed studies which employed different calibration methods.Further information on the calibration of sensors in each study is provided in table 2 of section 3.4.

Figure 3 .
Figure3.Map showing locations of IAQ field studies using LCS.Most studies have been conducted in the northern hemisphere, with sparse information available on IAQ using LCS in Africa, South America, and Australia/New Zealand.The legend at the top right of the graph shows the number of studies per continent, with 37 in Europe, 21 in North America, 19 in Asia, nine in Australia and New Zealand, seven in Africa, and three in South America.

Figure 4 .
Figure4.Building and measurement methodology metrics reported by the reviewed studies.Further information on the metrics reported by each study is provided in table 2 of section 3.4.

Figure 5 .
Figure 5. (a)The proportion of studies per microenvironment category, out of 60 studies reviewed.The proportions are as follows: 31 studies were in residential accommodation, 16 were in educational facilities, which includes schools and universities, five were in healthcare facilities, two in offices, three inside different forms of transport including cars, trams, trains, buses, and ships.Three studies were grouped together in the 'other' microenvironment category, and these studies were conducted inside a library, a museum and agricultural facilities (milking parlours and greenhouses).(b) The number of studies, out of a total of 60 reviewed, which measured each pollutant.The most commonly measured pollutants were PM 2.5 , with 48 studies having measured this, and CO 2 , which was measured in 36 studies.Next were PM 10 (26 studies), VOCs (20 studies), and NO 2 (14 studies).Nine measured O 3 , seven measured CH 2 O, four SO 2 , and just one study each measured C 6 H 6 , H 2 S and NH 3 .