Outdoor PM2.5 air filtration: optimising indoor air quality and energy

TO CITE THIS ARTICLE: Belias, E., & Licina, D. (2022). Outdoor PM2.5 air filtration: optimising indoor air quality and energy. Buildings and Cities, 3(1), pp. 186–203. DOI: https:// doi.org/10.5334/bc.153 ABSTRACT Human inhalation exposure to fine particulate matter (PM2.5), including the PM2.5 of outdoor origin, predominantly occurs indoors. To limit outdoor PM2.5 penetration into buildings, ventilation standards often require the filtration of outdoor air with a minimum efficiency. Nevertheless, the PM2.5 filter selection recommended by the standards is based on the annual average outdoor concentrations without considering seasonal or diurnal fluctuations. This could result in a waste of energy or elevated indoor PM2.5 exposures. Representative outdoor PM2.5 data from 37 cities worldwide in conjunction with a simulated office building are used to examine the impact of filtration strategies on indoor PM2.5 levels and the fan’s energy consumption. Two energy-saving methods are tested: (1) the optimum filter selection that maintains the indoor PM2.5 below the World Health Organization’s (WHO) air quality guidelines; and (2) the baseline filter recommended by standards in combination with a filter bypass. Relative to a standard recommended baseline case, the two applied methods could reduce energy demand by between 4% and 17%. This indicates that the outdoor air is over-filtered in the majority of the investigated cities. In cities with low-to-moderate outdoor PM2.5 levels, using a filter bypass can be an effective energy conservation method without compromising PM2.5 exposures indoors.


INTRODUCTION
Outdoor particulate matter (PM) is listed as one of the top 10 risk factors for human health (Gakidou et al. 2017). Within the full-size spectrum of suspended particles, PM 2.5 (a particle mass with a diameter <2.5 μm) provoke more severe health outcomes because they can penetrate deeper into the lungs and reach the bloodstream (Feng et al. 2016;Kappos et al. 2004). The relevant adverse health effects include allergic reactions (Mimura et al. 2014), childhood asthma (Hua et al. 2014), strokes , chronic bronchitis, cardiovascular and respiratory diseases (Sarigiannis et al. 2015), acute pulmonary infections, heart diseases and lung cancers (Song et al. 2017), and premature deaths (Feng et al. 2016;Song et al. 2017). These health outcomes have significant economic impacts on societies (Dutton et al. 2013;Sarigiannis et al. 2015;Zhao et al. 2015). Thereby, the majority of inhalation exposures to PM 2.5 , including the PM 2.5 of outdoor origin, occur while being indoors (Nazaroff 2018). This highlights the importance of the proper design and management of building defence mechanisms against the outdoor PM 2.5 , including the provision of adequate air filtration systems.
The building envelope serves as a passive barrier that limits the penetration of outdoor PM 2.5 indoors (Chen & Zhao 2011;Martins & Carrilho da Graça 2018). Among active techniques used to limit the penetration of outdoor PM 2.5 , outdoor air (OA) filters are the most commonly used. Buildings equipped with adequate OA filters were found to have a significantly lower indoor-to-outdoor PM 2.5 ratio (Oh et al. 2014;Ren et al. 2017). However, arrestance efficiency and resistance to airflow vary across different filter grades Martins & Carrilho da Graça 2018). Filters with higher particle-capture efficiencies create higher pressure drops in the ventilation system, which leads to higher energy use in fans Montgomery et al. 2012;Zaatari et al. 2014).
The majority of building standards recognise the importance of outdoor PM 2.5 filtration. They propose minimum filtration efficiencies according to building type and outdoor air pollution levels (ASHRAE 2019;CIBSE 2020;EN 16798-3 2017). With the most comprehensive and rigorous methodology, EN 16798-3 (2017) classifies outdoor air quality into three outdoor air categories (ODA) based on the annual average or maximum diurnal PM 2.5 concentrations. Following the outdoor air evaluation based on the data provided by the governmental authorities, EN 16798-3 gives recommendations for the particle filter selection according to the desired supply air class and outdoor air pollution category. In regions where outdoor air pollution is episodic, the standard proposes bypassing filtration during the periods when the outdoor air is clean in order to economise energy and extend the filters' lifetime (EN 16798-3 2017).
Several research studies investigated the optimal OA particle filter selection according to the outdoor PM 2.5 levels. Stephens et al. (2016) proposed customisable filters selection based on annual average outdoor PM 2.5 levels to reduce occupant exposure and, at times, to economise energy. Zimmer (2019) presented a simple tool named Comfort IAQ, which aids designers in selecting the appropriate air filter to optimise indoor air quality (IAQ) by considering the annual mean outdoor PM 2.5 concentrations and solving a steady-state mass-balance equation. Yu et al. (2020) introduced a method that, besides the annual mean, includes the minimum-maximum and the probability distributions of the outdoor PM 2.5 concentrations. Ben-David & Waring (2018) probed the effect of various filter efficiencies and OA ventilation rates on indoor PM 2.5 levels and buildings' energy consumptions in several US cities. Beyond this work conducted in the context of several countries or regions, relatively little is known about optimising OA filter selection that mitigates indoor PM 2.5 exposures whilst optimising energy use. Additionally, the available knowledge is limited when it comes to the potential use of filter bypasses in the ventilation system. The optimal selection of air filtration merits close attention due to the role files have on particle penetration indoors (and the associated health implications) and their energy use.
To address the knowledge gap, this study evaluated the performance of outdoor PM 2.5 air filtration methods in relation to indoor PM 2.5 exposures and energy use. First, a large-scale outdoor PM 2.5 data collection and analysis was performed to identify the representative yearlong hourly concentrations for 37 selected cities across the globe. Two methods to reduce the fan's energy consumption were then investigated in relation to standard recommended baseline filtration: (1) the optimum filter selection that maintains the indoor PM 2.5 below the World Health Organization's (WHO) (2006) air quality guidelines; and (2) the baseline filter recommended by standards in combination with a filter bypass. A mass-balance model was applied to a theoretical office building. This quantified the indoor PM 2.5 concentrations and the potential energy savings for the examined filtration methods.
The paper is structured as follows. The next section presents the methods used, including the outdoor PM 2.5 data sourcing, the selection of representative data, the simulation model and the methods for energy saving. The results and discussion present the findings regarding the outdoor PM 2.5 penetration indoors and energy use as a result of different filtration methods. The limitations are then considered. The final section briefly restates the key findings and highlights the importance of this work.

SELECTION OF CITIES AND DATA SOURCING
For the study sample, 37 cities were selected from six continents based on their populations, coverage of different geographical locations and climatic conditions (Figure 1). Where available, the data were collected from multiple measuring stations in a selected city. Hourly PM 2.5 data were accessed and retrieved from the website of official, governmental air quality measurement stations, as listed in the Table S1 in the supplemental data online. In total, data were collected from 86 different outdoor PM 2.5 measurement stations. The locations and names of the investigated stations are presented in Table S2 online.

REPRESENTATIVE HOURLY PM 2.5 DATA SELECTION
The hourly data were collected over a five-year period  to create a sufficient sample of representative concentrations. Data from 2020 were excluded on purpose because they cannot be characterised as representative owing to the Covid-19 pandemic (Dutheil et al. 2020).
Currently, the most comprehensive methodology that selects year-long hourly representative data is presented by Chen et al. (2019). Their method selects representative data with the goal to create a complete hourly time series of a typical year. The present study adopts this methodology as a baseline and creates an expanded database with representative PM 2.5 concentrations on a global scale. To address the issue of missing data and provide complete time series, a linear interpolation was used, because its performance in filling short gaps of missing air quality data is satisfactory. Monthly time series with more than 10% of missing data or more than 12 consecutive missing records were excluded from further analyses, because linear interpolation cannot efficiently impute data gaps of this size (Junninen et al. 2004). The next step was to select the representative hourly concentrations. The Sandia method (Hall et al. 1978) used to create typical meteorological years (TMY) was applied to select the representative PM 2.5 data. This method selects specific full-month datasets from different years based on the comparison of their cumulative distribution function (CDF) with the CDF of the long-term data for each targeted month. According to this method, the year-long hourly data are clustered by month. For each month, hourly data are used to create the CDFs for the long-term concentrations, which are further used to select the representative data according to the long-term concentrations and validate the representativeness of the selected data.
While the selection process for the TMYs takes into consideration various meteorological variables (e.g. air temperature, humidity, wind and solar radiation), the study used only the outdoor PM 2.5 concentration data. The statistical indicator to compare the CDFs used in the Sandia method is Finkelstein-Schafer (FS) statistics (Finkelstein & Schafer 1971), which measures the difference between the two CDFs: where δ i is the absolute difference between the CDF of the targeted month and the long-term data CDF at each point X(i) (i = 1, 2, …, n); and n is the number of hourly recordings in the specific month.
The FS statistics calculate the difference between the CDF of each month's long-term data and the CDF of each targeted full month. The smaller is the FS index, the closer are the two compared CDFs. Hence, the FS index can evaluate how close/far is the selected dataset from the long-term (five-year) data trend, and if the selected data are sufficiently representative. A 5% limit for the FS statistic, which is an indicator of good data representativeness, is adopted (Chen et al. 2019(Chen et al. , 2020. Finally, the selected months were merged in order to create a year-long dataset with the representative PM 2.5 values.

BUILDING SIMULATION MODEL
A theoretical reference open-space office building model was created for the present study. The floor-to-ceiling height of the model was selected to be 2.8 m to represent the typical floor-toceiling heights of European and North American office buildings (EN 15265 2007;DOE 2021). The total floor area was 849 m 2 . For this floor area, according to EN 16798-1 (2019) and for ventilation Category II and low-polluting buildings, the OA ventilation rate during occupied hours was 944 l/s (3398.4 m 3 /h). This is aligned with the nominal airflow for the energy efficiency evaluation of the air filters (Eurovent 4/11 2014; Eurovent 4/21 2018). During the unoccupied hours, as required by EN 16798-1 (2019), the OA ventilation rate was 127.4 l/s (458.5 m 3 /h) to dilute the emissions from building materials. The OA ventilation system of the building was a dedicated outdoor air system (DOAS). The speed of the DOAS's supply fan was variable in order to supply the defined OA rates according to the schedule. The study focused only on the OA ventilation system; therefore, the heating/cooling systems were not considered.
The EN 16798-1 (2019) building standard was also used to define the occupancy density and schedule. The adopted occupancy density was 17 m 2 /person, and the building was occupied from 07.00 to 18.00 hours during the workdays. The infiltration air change rate of the theoretical building was assumed to be constant and equal to 0.1/h in order to represent the infiltration rates of energy-efficient office buildings (Ben-David & Waring 2016). Additional simulations with different infiltration rates were conducted to reveal how this parameter influenced the results.

METHODS FOR ENERGY SAVINGS
The study examined two common DOAS fan energy-saving methods in order to explore their impact on energy savings and indoor PM 2.5 levels. Together with the baseline scenario based on the EN-16798-3 (2017) standard procedure, the total energy consumption of the fan and the indoor PM 2.5 levels were compared for the following three scenarios (Figure 2): • Baseline scenario: the DOAS filters were selected for each city according to the EN 16798-3 (2017) procedure. Specifically, an F7 filter was used when the annual mean outdoor PM 2.5 concentration was <10 μg/m 3 (in compliance with WHO 2006 guidelines; ODA 1), a combination of M5 + F7 filters when the annual mean outdoor PM 2.5 concentration was between 10 and 15 μg/m 3 (exceedance of the reference values by a factor up to 1.5; ODA 2), and a combination of F7 + F7 filters when the annual mean outdoor PM 2.5 concentration was >15 μg/m 3 (exceedance of the reference values by a factor >1.5; ODA 3). This scenario was used to define the baseline for the fan's energy consumption and the indoor PM 2.5 levels for each selected city.
• Optimum scenario: the simulation model was programmed to test seven different filter grades ( Table 1) and to select the filter with the minimum arrestance efficiency and pressure drop, which can maintain the indoor PM 2.5 levels in compliance with WHO (2006) air quality guidelines (annual mean = 10 μg/m 3 ; 24-h mean = 25 μg/m 3 ).
• Bypass scenario: it used the baseline filters (required by EN 16798-3 2017 as described in the baseline scenario) in combination with the filter bypass. During the hours when the outdoor PM 2.5 concentration was <10 μg/m 3 , the filter's bypass was used. When the bypass was in use, no filtration was applied, meaning that both the PM 2.5 arrestance efficiency and filter's pressure drop were zero.

FAN OPERATION: ENERGY CALCULATION
The energy for the DOAS's fan operation was calculated based on the evaluation procedure applied by Eurovent for the air filter's energy efficiency (Eurovent 4/21 2018). The fan's energy consumption in a DOAS is a function of airflow rate, operation time, total pressure difference and the fan's total efficiency: where E (kWh) is the total energy consumption of the fan to overcome the system's pressure drop; Q v (m 3 /h) is the ventilation airflow rate; t (h) is the operation time; Δ p (Pa) is the average pressure drop of the system; and η (-) is the total energy efficiency of the fan (efficiency of the fan and efficiency of the fan motor). The η was considered equal to 0.5 (-) for all calculations, because this number well represents the efficiency of the fans used in the heating, ventilation and airconditioning (HVAC) systems (Eurovent 4/21 2018). For the fan's energy consumption calculation,  the pressure drop provoked by the filter and the other elements of the ventilation system was considered. The pressure drops induced by the other elements of the ventilation system (air ducts, heat recovery unit, humidifier, cooling/heating coils, terminal device, air inlet and outlet) were assumed based on the recommendations of CEN/TR 16798-4 (CEN/TR 2017) to be 640 (Pa) for an airflow of 0.944 m 3 /s (during occupied hours).
In this study, the pressure drop of the filters was considered to be static. For the OA ventilation airflow during occupied hours, the average pressure drops of each filter grade were assumed based on Eurovent's energy efficiency classification table (Eurovent 4/11 2014), as summarised in Table 1. It was considered that the used filters belong to energy class C, because this class is the most common among commercially available air filters (Vadoudi & Kelijian 2019).
As the OA ventilation rates were lower during unoccupied hours, the pressure drops were also reduced. The reduced pressure drop was calculated using equation (3), as the static pressure drop can be considered propositional to the square of the airflow and to a proportionality constant C (Stephens et al. 2010 where Δ p (Pa) is the average pressure drop of the system; C (-) is the proportionality constant; and Q v (m 3 /h) is the ventilation airflow rate.

IAQ MODEL
In order to calculate the indoor PM 2.5 concentration at each hourly time step, a mass-balance model was applied. The indoor PM 2.5 concentrations were calculated using: where dC in (t) (μg/m 3 ) is the time-dependent indoor PM 2.5 concentration, C out (t) (μg/m 3 ) is the outdoor hourly representative PM 2.5 concentration; λ (/h) is the infiltration/exfiltration rate; P (-) is the outdoor PM 2.5 penetration coefficient; η v (-) is the PM 2.5 arrestance efficiency of the filter; Q v (t) (m 3 /h) is the OA ventilation airflow rate; S r (t) (μg/h) is the PM 2.5 emission rate from resuspension; k (/h) is the PM 2.5 deposition rate; and t (h) is the time.
For the simulated theoretical office building, the only indoor PM 2.5 source considered was the resuspension due to the occupants' walking, because it was assumed that there were no other significant PM 2.5 sources. The particle emission rate from resuspension can differ significantly according to the floor coverings and occupants' activities Thatcher 1995;Tian et al. 2014). Based on occupancy activity detection studies, each office occupant walks on average  (Anderson et al. 2019;Brakenridge et al. 2016;Jancey et al. 2014). With the assumption that the floor was covered by a hard material, the emission rate from resuspension (S r ) was 0.01 mg/min/walking person Thatcher 1995).
The PM 2.5 penetration coefficient (P) can differ significantly from building to building because it depends on the envelope's characteristics and nature of its cracks. For the purpose of this study, P was assumed to be 0.8 (-) based on previous studies (Chen et al. 2012;Chen & Zhao 2011;Li et al. 2017). The PM 2.5 deposition rate (k) was assumed to be 0.5/h based on previous studies (Ruan & Rim 2019;Zhao & Stephens 2017).
Unlike the EN ISO 16890-1 (2016) standard, the ASHRAE 52.2 (ASHRAE 2017) and BS EN 779 (CEN 2012) standards for filter grade classifications do not explicitly report the PM 2.5 arrestance efficiencies. Thus, the PM 2.5 arrestance efficiencies were assigned by the authors and were based on Eurovent's recommendations (Eurovent 4/23 2018) for BS EN 779 and on Azimi et al. (2014) for ASHRAE 52.2 (ASHRAE 2017) filter grades accordingly. The PM 2.5 arrestance efficiencies of the investigated filter grades are presented in Table 1. As the EN 16798-3 (2017) standard requires a combination of air filters for the ODA 2 and ODA 3 outdoor air quality categories, the total PM 2.5 arrestance efficiencies for these combinations were calculated using: where η vt is the total PM 2.5 arrestance efficiency; and η v1 and η v2 are the arrestance efficiencies of the first and second filters, respectively. The PM 2.5 arrestance efficiencies of the M5 + F7 and F7 + F7 filter combinations used in this study are presented in Table 2.
All the simulations of the IAQ and fan's energy consumption model were conducted using scripts in MATLAB programming language. The inputs and outputs of the IAQ and fan's energy modeling are summarised in Figure 3.     Figure 4, the outdoor PM 2.5 levels differed across the cities, so the selected filters also varied according to the procedure described in EN 16798-3 (2017). Figure 5 presents the average outdoor and indoor PM 2.5 levels categorised based on the outdoor air classification categories and corresponding filters used in baseline scenario.   For the baseline scenario, it was observed that filters recommended by EN 16798-3 (2017) were able to maintain the indoor PM 2.5 levels in compliance with the WHO's (2006) air quality guidelines for the majority of the investigated cities. As shown in Figure 5, the average indoor PM 2.5 concentrations were 1.7 μg/m 3 for ODA 1, 2.2 μg/m 3 for ODA 2, and 4.8 μg/m 3 for ODA 3 areas. These results were expected, given that the EN 16798-3 (2017) is the most stringent available standard for filter selection. The compliance with the indoor PM 2.5 levels was not achieved only in Delhi (India), Ulaanbaatar (Mongolia) and Beijing (China). For Ulaanbaatar and Beijing, the maximum 24-h limit of 25 μg/m 3 was exceeded more than three times. In the case where only annual average outdoor PM 2.5 data were used, it would be impossible to detect this non-compliance. In Delhi, both the annual mean and the 24-h PM 2.5 limits were not compliant with the WHO (2006) guidelines. These results reveal that the process for the OA filters selection proposed by EN 16798-3 (2017) is able to select filters that can sufficiently protect the occupants of non-residential buildings from high exposures to outdoor originating PM 2.5 for all European cities and in the majority of non-European cities. However, this does not fully apply to cities with severe continuous or temporary air pollution episodes. Figure 6 summarises the annual indoor PM 2.5 concentrations in the selected cities resulting from the three different outdoor air filtration scenarios. The optimum scenario, which aims to balance indoor PM 2.5 exposures and energy use, increased the indoor PM 2.5 concentrations relative to the baseline scenario in almost every city. Exceptions were Kuwait City, Beijing, Ulaanbaatar and Delhi, where higher grade filters were selected to maintain compliance with the WHO's air  quality guideline. The annual average indoor PM 2.5 concentrations in the optimum scenario were 3.2 μg/m 3 for ODA 1, 5.2 μg/m 3 for ODA 2, and 6.4 μg/m 3 for ODA 3 areas. Despite the optimum scenario increasing the indoor PM 2.5 levels relative to the baseline scenario, the indoor air usually remained compliant with the WHO (2006) guidelines. Exceptions again were Beijing, Ulaanbaatar and Delhi, where even the high-efficiency particulate arrestance (HEPA) filter (99.7% PM 2.5 arrestance efficiency) failed to prevent the excess penetration of outdoor PM 2.5 . In these cases, airflow infiltration through cracks in the building envelope played a critical role. This highlights that in severely polluted cities, the selection of an OA filter alone cannot eliminate the problem of outdoor PM 2.5 penetration, and that filtration should be combined with other measures, such as the appropriate airtightness of the building.

As shown in
As shown in Figure 6, the bypass scenario expectedly increased the indoor PM 2.5 concentration relative to the baseline scenario. The PM 2.5 levels in the bypass scenario were on average 2.7 μg/m 3 for ODA 1, 3.2 μg/m 3 for ODA 2, and 5.2 μg/m 3 for ODA 3 areas.
The compliance with the WHO's (2006) air quality guidelines remained equivalent across the baseline, optimum and bypass scenarios. Relative to the baseline scenario, on average, the optimum scenario led to greater increases in the indoor PM 2.5 levels compared with the bypass scenario. Notably, indoor PM 2.5 concentrations remained substantially below the WHO limits also for the optimum and bypass scenarios. This clearly indicates that when the filtration recommended by EN 16798-3 (2017) is applied, the air is unnecessarily over-filtered for the majority of the time and in the majority of cities.
The created typical outdoor air pollution profiles were validated regarding their ability to select representative data and maintain the variations of outdoor PM 2.5 . Figure S1 in the supplemental data online presents the values of the FS statistics, which were used for this validation. Figure S2 online presents the PM 2.5 concentrations of a selected location as a time-series plot, where seasonal/diurnal/hourly PM 2.5 variations can be observed.
The presented results were calculated assuming that the infiltration rate was 0.1/h. Section 3 in the supplemental data online summarises the information regarding the contribution of the infiltration in the indoor PM 2.5 levels. In particular, Figure S3 online presents the annual average PM 2.5 indoor levels for the baseline scenario by highlighting the contribution of infiltration. It was observed the infiltration becomes more critical as the outdoor pollution levels and the filter's efficiency increase. Furthermore, Tables S3, S4 and S6 online present the indoor mean annual PM 2.5 concentrations for the baseline, optimum and bypass scenarios, respectively, when different infiltration air change rates were applied. For the baseline and bypass scenarios, the higher infiltration rates led to higher indoor concentrations, while in the optimum scenario, the indoor concentrations were increased in the cases where the selected optimum filter remained the same. Table S5 online summarises the selected optimum filter grade for the optimum scenario per examined location under different infiltration air change rates. Figure 7 shows the calculated energy consumption required by the fan to overcome the pressure drops of the ventilation system in the three analysed scenarios. For the baseline scenario, the annual energy consumptions were 5.74, 6.3 and 6.79 kWh/m 2 year when F7, M5 + F7, and F7 + F7 filters were used, respectively.

FAN'S ENERGY CONSUMPTION
With the exception of the severely polluted cities of Delhi, Ulaanbaatar, Beijing and Kuwait City, where HEPA filters were selected, the optimum scenario had a substantially lower energy consumption by fans relative to the baseline scenario in all the investigated cities. In addition, the bypass scenario also reduced energy consumption relative to the baseline scenario in all cases worldwide, except for Kampala (Uganda), where the outdoor PM 2.5 concertation never dropped below 10 μg/m 3 .

196
Belias & Licina Buildings and Cities DOI: 10.5334/bc.153 As presented in Figure 8, the relative reduction in the annual energy consumption by fans for the optimum scenario was on average 12% for ODA 3, 17% for ODA 2, and 9% for areas with ODA 1. The bypass scenario achieved an average relative reduction in the energy consumption of 4% for ODA 3, 14% for ODA 2, and 14% for ODA 1 areas. This reveals that the bypass scenario was more efficient for energy savings in areas with low PM 2.5 pollution levels (ODA 1), while the optimum scenario was more efficient for the areas with moderate or high PM 2.5 pollution levels. In the cities with low PM 2.5 concentrations, the duration of time when the filter bypass can be used was significantly higher compared with the cities with higher PM 2.5 levels (Figure 9), which led to higher energy savings in these cases. Without access to continuous outdoor PM 2.5 data, it would be impossible to reveal this potential for energy savings.
The demonstrated energy-saving potential could be significant in nearly zero-energy buildings with a primary energy demand of 80-100 kWh/m 2 year (Economidou et al. 2020). In these buildings, the reduction in the total primary energy demand can reach up to 3.7% when the electricity is not produced on-site by renewables, and if the proposed primary energy factor for Europe, which is 2.1 (-), is considered (European Commission 2021). Moreover, the use of bypasses can also extend  the filter's lifetime and reduce maintenance costs (EN 16798-3 2017). Hence, the designers should give greater attention to the selection of OA filters in order to design more energy-efficient and healthy buildings.

TRADE-OFFS BETWEEN ENERGY SAVINGS AND INDOOR PM 2.5 CONCENTRATIONS
As the energy savings resulting from less efficient particle filtration are associated with the greater penetration of outdoor PM 2.5 , the optimum and bypass scenarios were examined to ascertain their ability to save energy with the minimum increase in indoor PM 2.5 while remaining compliant with the WHO (2006) air quality guidelines. The ratio of the annual fan's energy consumption reduction (ΔE) to the indoor PM 2.5 increase (ΔPM 2.5 ) was calculated. The results (Figure 10) reveal that the bypass scenario was more efficient in reducing the fan's energy consumption without significantly increasing the indoor PM 2.5 levels, as the values of this ratio were significantly higher in all the cities except one.  198 Belias & Licina Buildings and Cities DOI: 10.5334/bc.153

SCOPE AND LIMITATIONS
In interpreting the results of the present study, several limitations should be acknowledged. This exploratory simulation study demonstrated that there is a potential for energy savings in the ventilation fans when the outdoor air pollution is considered more comprehensively. The results from the simulations were cross-validated, as the results for the indoor PM 2.5 concentrations in North American cities are comparable with the previously reported results in the US when the F7 (minimum efficiency reporting values (MERV) 12) filter was applied (Ben-David & Waring 2018). In addition, the total energy consumption of the fan was in the range of the energy consumption of ventilation systems reported in previous studies (Bonato et al. 2020;Flourentzou & Pantet 2015). However, only one building type, infiltration rate and occupancy profile with a specific OA ventilation system were simulated using the two different energy-saving methods. Therefore, the results could differ for different building types, ventilation systems, indoor sources and airflow rates. Future studies should explore other energy-saving methods in OA filtration. It is recommended that designers carefully investigate and provide an ad hoc solution for OA filtration of each building case.
Additionally, the methodology used in this study for the representative data selection (Chen et al. 2019(Chen et al. , 2020 excluded months with extreme concentrations (either too low or too high), as they are not considered typical. The goal of the study was to reveal the opportunities for energy savings under typical conditions. Thus, extreme PM 2.5 pollution events such as wildfires (Brambilla et al. 2021) were not specifically targeted by the analyses.
In addition, the present study focused on only one air pollutant: PM 2.5 . Particles with different aerodynamic diameters, such as coarse particles (PM 10 ), are known to have adverse health effects, but are less significant compared with PM 2.5 (Kappos et al. 2004). Furthermore, ultrafine particles are also known to provoke deleterious health outcomes (Chen et al. 2016), but since the available Note: Data in the city of Kampala were partially presented as the ΔE/ΔPM 2.5 ratio and cannot be defined for the bypass scenario. The outdoor PM 2.5 concentration was never <10 μg/m 3 , and the bypass was never activated. Hence, both ΔE and ΔPM 2.5 were equal to zero.

199
Belias & Licina Buildings and Cities DOI: 10.5334/bc.153 data are sparse, this particle size range was not analysed. Other gaseous pollutants with wellknown effects on human health (Logue et al. 2012) should also be considered in future studies. Furthermore, the present study considered that if the indoor PM 2.5 levels comply with the WHO (2006) guidelines, there are no significant adverse health effects. In fact, some studies suggest that there is no clear lower limit for PM 2.5 exposures that can be considered completely safe (WHO 2013). In addition, the last update of the WHO (2021) air quality guidelines recommends a lower limit of 5 μg/m 3 for the annual mean PM 2.5 concentration.

CONCLUSIONS
The present study investigated the impact of three air filtration scenarios on energy consumption and indoor particulate matter PM 2.5 levels. The application of the outdoor air (OA) filter selection methodology recommended by the EN 16798-3 (2017) building standard (the baseline scenario) was found to provide acceptable indoor PM 2.5 concentrations for all the investigated areas, except for the cities of Beijing, Ulaanbaatar and Delhi.
It was found that energy savings, on average from 4% to 14%, for the fan can be achieved if a bypass process is used that combines the baseline filter with a filter bypass. This maintains the indoor PM 2.5 below the WHO's limits.
The energy-saving potential of the bypass scenario was greater in areas with low annual average PM 2.5 concentrations (<10 μg/m 3 ), where it achieved average energy reductions of up to 14%, whereas the optimum scenario could save on average 9% of the total fan's energy. This indicates that in areas with clean or periodically polluted outdoor air, the filter bypass offers significant potential for energy savings without compromising inhalation exposures indoors. In cities with high PM 2.5 levels, the average energy savings were 12% for the optimum scenario and 4% for the bypass scenario. The achieved energy savings indicate that the filters proposed by even the most advanced existing building standards result in unnecessary over-filtration of the OA and lead to a waste of energy by the fans in the majority of the investigated cities around the world. Overall, in 34 out of 37 investigated cities, the bypass scenario offered the best trade-offs between energy conservation and limiting indoor PM 2.5 exposures.
The present study highlights the need for building designers to consider a dynamic variation of local outdoor PM 2.5 when selecting the OA filtration in order to reduce energy demand from the mechanical ventilation fans without compromising indoor air quality. Future investigations should encompass a broader range of building typologies, ventilation systems and indoor sources in order to develop the improved selection matrix between indoor PM 2.5 exposures and energy-efficient filtration techniques. Such efforts promise to aid designers and engineers to implement efficient ventilation systems that can protect occupants from unwanted outdoor PM 2.5 exposures while, in parallel, cutting energy consumption.