Highly Time-Resolved Apportionment of Carbonaceous Aerosols from Wildfire Using the TC–BC Method: Camp Fire 2018 Case Study

The Camp Fire was one of California’s deadliest and most destructive wildfires, and its widespread smoke threatened human health over a large area in Northern California in November 2018. To analyze the Camp Fire influence on air quality on a 200 km distant site in Berkeley, highly time-resolved total carbon (TC), black carbon (BC), and organic carbon (OC) were measured using the Carbonaceous Aerosol Speciation System (CASS, Aerosol Magee Scientific), comprising two instruments, a Total Carbon Analyzer TCA08 in tandem with an Aethalometer AE33. During the period when the air quality was affected by wildfire smoke, the BC concentrations increased four times above the typical air pollution level presented in Berkeley before and after the event, and the OC increased approximately ten times. High-time-resolution measurements allow us to study the aging of OC and investigate how the characteristics of carbonaceous aerosols evolve over the course of the fire event. A higher fraction of secondary carbonaceous aerosols was observed in the later phase of the fire. At the same time, the amount of light-absorbing organic aerosol (brown carbon) declined with time.


Introduction
In recent years, the frequency of extreme wildfire events and the total burned area in California have increased due to climate-change-related temperature increases and precipitation decreases [1,2]. Besides the material damage and potential risk to human lives, the wildfires also emit an enormous amount of carbonaceous aerosols (CA), which can be transported to a few hundred kilometers away from the wildfire location, affecting the air quality and public health over a wider area [3,4]. Epidemiological and toxicological studies indicate that CA may be relatively more toxic than other PM 2.5 compounds, such as soluble salts, due to their high potential to inflict oxidative stress [5,6]. On the other hand, CA influence the Earth's radiation budget directly through light absorption and scattering or indirectly due to their role in cloud formation [7][8][9]. CA generally combine two chemically different species, black carbon (BC) and highly variable organic aerosols (OA) [10]. BC is an inert fraction that strongly absorbs light. BC is always a consequence of incomplete combustion and is emitted directly from sources as a particle. On the contrary, OA can be primarily emitted as particles (primary OA-POA) or formed in the atmosphere via the oxidation of gaseous precursors (secondary OA-SOA). Brown carbon (BrC) is a fraction of OA that absorbs light. The light absorption by BrC is usually enhanced near the ultraviolet part of the spectra [11,12]. The mass of carbon atoms in CA and OA is called total carbon (TC) and organic carbon (OC), respectively. Similarly, primary OC (POC) and secondary OC (SOC) represent the mass of carbon atoms in POA and SOA. Docherty et al. [13] reported typical POA/POC and SOA/SOC ratios of 1.2 and 1.8, respectively, for the average situation in Los Angeles, California, unaffected Similarly, primary OC (POC) and secondary OC (SOC) represent the mass of carbon atoms in POA and SOA. Docherty et al. [13] reported typical POA/POC and SOA/SOC ratios of 1.2 and 1.8, respectively, for the average situation in Los Angeles, California, unaffected by wildfires. During wildfires, the OA/OC ratio usually increases owing to the higher oxidation state of biomass-burning-related (BB) aerosols. For example, Zhou et al. [14] reported OA/OC values between 1.7 and 3.0 when wildfire plumes impacted the air, and similarly, Zhang et al. [7] reported OA/OC ratios between 1.62 and 2.58.
CA emitted from wildfires are mostly organic [7,15]. Like other BB aerosols, fresh wildfire emissions contain a high fraction of BrC [15]. Dark aging can enhance BrC absorption [15], while BrC absorption is decreased with aging during daylight owing to the photobleaching of chromophores [16][17][18].
There are two major pathways of biomass-burning-related SOA formation: SOA formation from either primarily emitted volatile organic compounds (VOCs) or dilutiondriven POA evaporation and subsequent SOA formation [19,20]. Amongst the VOCs, the main contributors to SOA formation from wildfires are phenol, benzene, and catechol [21], where gaseous catechol represents the intermediate phase of phenol oxidation to SOA.
Our study reports the extreme pollution of a large wildfire event characterized in detail using high-time-resolution TC and BC measurements. With the TC-BC method, we analyzed the influence of the Camp Fire smoke at a distant site in Berkeley in November 2018. A BC tracer and brown carbon models were used to characterize further the primary and secondary OA and the increased light absorption on BrC. The highly time-resolved measurements also offer to analyze the smoke aging.

Location and Measurements Setup
The Bay Area Air Quality Measurement District (BAAQMD) site in Berkeley (CA, USA) is located near the Aquatic Park ( Figure 1). Its micro-location is near the exit from a 10-lane highway (37.8648, −122.3028), so it is characterized as a traffic station. The Berkeley site is located approximately 200 km away from the Camp Fire center.  In this study, we used the Carbonaceous Aerosol Speciation System (CASS, Aerosol Magee Scientific, Ljubljana, Slovenia), comprising two instruments, a Total Carbon An- alyzer, model TCA08 [22], in tandem with an Aethalometer, model AE33 [23]. CASS measured total carbon (TC) and black carbon (BC) with high time resolution. The difference can be inferred as the organic carbon (OC): TCA08 used a simple thermal protocol: concentrations of TC were measured by the rapid combustion of particles collected on a quartz filter. The sample was heated almost instantaneously to 940 • C, at which the carbonaceous compounds were efficiently combusted to CO 2 . The pulse of CO 2 created during the combustion phase of the analysis was detected as a large transient increase above the background CO 2 level. TCA08 had two identical parallel chambers to ensure continuous measurements without dead time. Wherein one channel collected a sample, the other performed analysis of an already collected sample. The limit of detection (LoD) was 0.3 µg/m 3 . TCA was operated at a 30 min resolution, sampling the PM 2.5 fraction at 16.7 L min −1 .
The tandem Aethalometer measured the aerosol light absorption and corresponding equivalent BC concentration at seven wavelengths (370, 470, 520, 590, 660, 880, and 950 nm) with LoD equal to 0.03 µg/m 3 . The sample was collected on a glass fiber filter with a flow rate of 5 L min −1 and a PM 2.5 size-selective inlet. Light attenuation was measured at a time resolution of 1 min. The Aethalometer used a 'dual-spot' methodology to correct the filter-loading effect in real-time [23]. The absorption coefficient (b abs ) was calculated from the attenuation coefficient (b ATN ) using a multiple-scattering parameter C: The equivalent BC mass was calculated from b abs using mass absorption cross section (MAC): In Equations (2) and (3), the manufacturer default values were used for C and MAC (880 nm): 1.39 and 7.77 m 2 /g, respectively.
The advantages of using CASS compared to online and offline OC/EC analyzers include high time resolution, no sampling dead time, online filter-loading effect compensation for BC measurements, and low maintenance [23]. As the thermal measurements were performed without a fragile quartz sample oven, high-purity gases, and a catalyst, they are suitable for field campaigns [22].

BC Tracer Model and Brown Carbon Model
To split OC into primary OC (POC) and secondary OC (SOC), we used the BC tracer model [10,24]: where (OC/BC) prim is a time-independent ratio and is expected to be source-dependent. To find the optimal (OC/BC) prim ratio, the minimum R-squared (MRS) [25,26] method was used. Briefly, the hypothetical SOC(t) was first calculated for a wide range of hypothetical (OC/BC) prim ratios (for ratios 0.1 to 15 in 0.1 steps). Then, the R-squared value (R 2 ) between hypothetical SOC(t) and BC(t) was calculated for every hypothetical (OC/BC) prim ratio, and the optimal (OC/BC) prim ratio was chosen where the R 2 was minimal. Because the air quality was strongly influenced by the plume from the wildfire during the event and by traffic emissions otherwise, two different (OC/BC) prim ratios were used in our study.
(OC/BC) prim equal to 1.1 was determined as an optimal value for subperiods before and  Figure A1). The wavelength-dependent optical absorption was split between BC and brown carbon (BrC) using the two-component model [27][28][29][30] We assumed the absorption Ångström exponent of BC (AAE BC ) had a constant value of 1.15, as thoroughly discussed in a recent study from the Los Angeles basin [10]. Because AAE BrC (t) was allowed to vary with time, the two-component model (Equation (5)) did not have an analytical solution. To solve model (5), an assumption that BC was the only light-absorbing carbonaceous component at 880 nm was considered [10,28,29,31]: Using assumption (6), the optical absorption by BrC at a specific wavelength can be calculated from model (5) Spectral dependence of the absorption coefficient can be described by the bulk absorption Ångström exponent (AAE), which is calculated by fitting over all 7 wavelengths (AAE 7λ ). Besides the fraction of BrC absorption at 370 nm, the spectral dependence of b BrC abs can be described by the AAE BrC , where b BrC abs is fitted over 4 shortest wavelengths (370 nm to 590 nm).

Uncertainties
The detailed evaluation of uncertainties for TC-BC measurements and applied models was widely discussed in a recent paper [10]. Briefly, the uncertainty of TC concentrations measured by TCA08 was estimated to be up to 10%, while the uncertainty for BC concentration measured using AE33 was estimated to be 25%, where the largest part of uncertainty arose from the selected MAC BC and C values. The uncertainty of the BC tracer model was also estimated to be 25%, and the BrC model yielded a discrepancy of 5% in the absorption coefficient of BrC at 370 nm.

Complementary Data
The smoke from Camp Fire extended over a broader area and affected air quality in Berkeley and a considerable fraction of northern California. To analyze the spatial extent of the Camp Fire on air quality in a broader area, the data from 4 publicly available databases were added:  [33]. The GFS quarter-degree archive of meteorological data is used for the calculation.
Ground-based daily-averaged PM 2.5 concentrations were estimated by Purple Air sensors (PASs). Owing to their low cost and, consequently, widespread use, PASs are useful for spatially visualizing the extent of pollution. Regardless of different corrections algorithms [36], the PAS biases could not be reduced entirely, and therefore, PASs were not used for mass closure in our study.

Results and Discussion
Camp Fire started on 8 November 2018, at around 6:15 a.m., approximately 145 km north of Sacramento, California. It was ignited by a faulty electric transmission line. The plume reached the Berkeley site in the next 4 h ( Figure A4). The heavy rain on 21 November completely extinguished the fire ( Figure A17) and washed out the pollution from the atmosphere. Figure 2 shows the time series of measured TC and BC (Figure 2a  Berkeley site were calculated every 3 h by the HYSPLIT model for the last 72 h [33]. The GFS quarter-degree archive of meteorological data is used for the calculation. Ground-based daily-averaged PM2.5 concentrations were estimated by Purple Air sensors (PASs). Owing to their low cost and, consequently, widespread use, PASs are useful for spatially visualizing the extent of pollution. Regardless of different corrections algorithms [36], the PAS biases could not be reduced entirely, and therefore, PASs were not used for mass closure in our study.

Results and Discussion
Camp Fire started on 8 November 2018, at around 6:15 a.m., approximately 145 km north of Sacramento, California. It was ignited by a faulty electric transmission line. The plume reached the Berkeley site in the next 4 h ( Figure A4). The heavy rain on 21 November completely extinguished the fire ( Figure A17) and washed out the pollution from the atmosphere. Figure 2 shows the time series of measured TC and BC (Figure 2a   Before and after the Camp Fire event in November 2018, it is possible to observe the typical daily cycle of CA in Berkeley ( Figure 2 and diurnal profiles in Figure A2), comparable to the typical daily cycle from the Los Angeles basin [10]. BC represented approximately 40% of TC. The highest fraction of OC, up to 80% of TC, was observed during the nights (Figure 2b). During the day, primary CA (BC + POC) prevailed, and during the night, SOC contributed up to 70% of TC (Figure 2d). Before and after the Camp Fire event, the light absorption by BC strongly prevailed over BrC. BrC appeared only during the nights (Figure 2f) and, on average, represented approximately a 5% fraction of the light absorption at 370 nm in November 2018 (Figure 3e).
Toxics 2023, 11, 497 6 Before and after the Camp Fire event in November 2018, it is possible to observ typical daily cycle of CA in Berkeley ( Figure 2 and diurnal profiles in Figure A2), com rable to the typical daily cycle from the Los Angeles basin [10]. BC represented app mately 40% of TC. The highest fraction of OC, up to 80% of TC, was observed durin nights (Figure 2b). During the day, primary CA (BC + POC) prevailed, and during night, SOC contributed up to 70% of TC (Figure 2d). Before and after the Camp Fire e the light absorption by BC strongly prevailed over BrC. BrC appeared only during nights (Figure 2f) and, on average, represented approximately a 5% fraction of the absorption at 370 nm in November 2018 (Figure 3e). The air quality significantly deteriorated when the wildfire plume reached Berk (Figure 2a). We split this period into two phases: Fire Phase I and Fire Phase II (den with vertical green lines in Figure 2). In Fire Phase I, the fire was active due to The air quality significantly deteriorated when the wildfire plume reached Berkeley (Figure 2a). We split this period into two phases: Fire Phase I and Fire Phase II (denoted with vertical green lines in Figure 2). In Fire Phase I, the fire was active due to high vegetation fuel loading, and this period can be characterized by the flaming phase of the fire. The high plume rise and strong winds allowed the plume to be transported over long distances (Figures A4-A9). The smoldering and spot fires were more typical for Fire Phase II. The initial plume was already dispersed, while the intake of fresh smoke was lower for more distanced places such as the Berkeley site ( Figures A10-A16). The air masses were coming to the Berkeley site from the northeast direction most of the time when the fire was active (back trajectories in Figures A4-A16). The daily averaged PM 2.5 concentrations, measured by Purple Air sensors, exceeded 200 µg/m 3 in Berkeley and 300 µg/m 3 in places nearer the fire ( Figures A3-A17).
In Fire Phase I, the median TC concentrations increased from 3.5 µg/m 3 to 23.9 µg/m 3 with a maximum hourly and 24 h average value of 73.8 µg/m 3 and 37.1 µg/m 3 , respectively (Figure 2a). In Fire Phase II, even higher TC concentrations were observed: 37.0 µg/m 3 , 102.9 µg/m 3 , and 75.0 µg/m 3 for the median, hourly maximum, and maximum 24 h average TC, respectively. During both fire phases, the 24 h average TC exceeded the WHOrecommended PM 2.5 daily limit of 15 µg/m 3 [37] every day. In 7 of these 13 days, the 24 h average TC concentrations were two times higher than the limit. It is worth noting that the TC concentration only represents the carbon content in carbonaceous aerosols. In addition, the PM 2.5 may be even higher due to other non-carbonaceous particles in the aged wildfire plume. While the fractional increase in median BC concentrations against the median BC value before and after the event was comparable in both fire phases (3.7 times in Fire Phase I and 3.9 times in Fire Phase II- Figure 3a), the pattern of OC increase was different. Compared to periods before and after the Camp Fire event, median OC increased 8.4 and 13.9 times in Fire Phases I and II, respectively (Figure 3a). This agrees with the study of Zhang et al. [7], where they reported more than 10 times higher OA concentration during the period affected by the wildfire than the background. Increase in the OC fraction in Fire Phase II infers more OC was formed secondarily. We confirmed this with the BC tracer model: during Fire Phase I, 63% of TC was POC (Figure 3c). Similarly, Palm et al. [19] reported the dominant POA role during the fire's starting phase. In Fire Phase II, the most considerable fraction of TC was SOC (48%) (Figure 3d).
The fraction of the light absorbed by BrC increased significantly during the fire period compared to the typical situation before and after the event (Figure 2e,f). The maximum fraction of light absorbed by BrC at 370 nm was observed on 11 November when it reached 80% (Figure 2f). The median BC absorption coefficient at 370 nm was comparable between both fire phases-82.9 Mm −1 and 87.4 Mm −1 in Fire Phases I and II, respectively. The median BrC absorption coefficient at 370 nm decreased from 94.6 Mm −1 to 61.2 Mm −1 in Fire Phase II (Figure 3f,g). This resulted in a lower fraction of light absorbed by BrC at 370 nm-it decreased from 60% in Fire Phase I to 44% in Fire Phase II (Figure 3f,g). The decrease in BrC absorption is most probably connected to the photobleaching of BrC. The net decrease in BrC absorption with aging due to photobleaching is also consistent with other studies [15,17,18,38,39]. The lowest fraction of light absorbed by BrC was found between 17 and 19 November, with less than a 20% share (Figure 2f). The lower smoke intensity can also be seen in satellite images ( Figures A14 and A15). Figure 4 contains analyses of the AAE for bulk absorption (AAE 7λ ) and BrC (AAE BrC ). The mean AAE 7λ was equal to 1.15 before and after the air quality was affected by Camp Fire smoke (Figure 4a), and the mean value of AAE BrC for the same periods was equal to 3.78 (Figure 4d). Both AAE 7λ and AAE BrC increased when the air was affected by Camp Fire smoke (Figure 4b-c,e-f). The AAE BrC was slightly (statistically significant) higher in Fire Phase II compared to Fire Phase I-the mean AAE BrC increased from 5.63 to 5.90 with a p-value of 0.01 for Mood's median test (Figure 4e,f). The opposite pattern was observed for AAE 7λ -it decreased in Fire Phase II (Figure 4b,c) due to the higher fraction of light absorption by BC in Fire Phase II (Figure 3f,g). The decrease in AAE 7λ during the later wildfire stage is consistent with other wildfire studies of aged plumes in Northern America [38,39].

Conclusions
Wildfires are an uncontrollable source of carbonaceous aerosols. Wildfire smoke cause high levels of PM2.5 in urban areas, exceeding the WHO guidelines for several d Exposure to such high levels of PM2.5 can adversely affect respiratory health, especi for vulnerable groups such as children, the elderly, and people with chronic respira diseases. Different, non-typical measures must be applied to reduce the health risks of population, such as closing schools and other public objects. People are usually adv to spend most of their time indoors and reduce the natural ventilation of the buildi they occupy. Real-time and highly time-resolved measurements are required to make propriate decisions about the measures and risks.
In our study, the CASS, a combined unit of a TCA and an Aethalometer, was use measure the influence of the smoke from the distant Camp Fire on local air qualit Berkeley. High-time-resolution measurements, in combination with numerical mo such as the BC tracer and BrC model, allow us to apportion the pollution in real-time. 24 h average TC concentration exceeded the recommended WHO PM2.5 limit every when the Berkeley site was affected by the wildfire smoke from Camp Fire, and the 2 average concentrations were two times higher than the limit in 7 of these 13 days. Med BC concentrations increased approximately four times above the pollution level typ for Berkeley. The median OC increased by 8.4 in the first, more active fire phase w flaming is expected to dominate. Later, when more smoldering is expected, the med OC increase was even higher, 13.9 times above the background concentrations, sugges a more intense formation of SOA during the later fire phase.
The results of apportionment models that can be run in real-time can be an essen input for short-term air quality prediction models. On the other hand, we showed an creased contribution of BrC to total light absorption at short wavelengths during the ev Therefore, the fraction of the absorbed light by BrC can be used as a marker for detec the influence of wildfire plumes. While the fractional contribution of BrC to l

Conclusions
Wildfires are an uncontrollable source of carbonaceous aerosols. Wildfire smoke can cause high levels of PM 2.5 in urban areas, exceeding the WHO guidelines for several days. Exposure to such high levels of PM 2.5 can adversely affect respiratory health, especially for vulnerable groups such as children, the elderly, and people with chronic respiratory diseases. Different, non-typical measures must be applied to reduce the health risks of the population, such as closing schools and other public objects. People are usually advised to spend most of their time indoors and reduce the natural ventilation of the buildings they occupy. Real-time and highly time-resolved measurements are required to make appropriate decisions about the measures and risks.
In our study, the CASS, a combined unit of a TCA and an Aethalometer, was used to measure the influence of the smoke from the distant Camp Fire on local air quality in Berkeley. High-time-resolution measurements, in combination with numerical models such as the BC tracer and BrC model, allow us to apportion the pollution in real-time. The 24 h average TC concentration exceeded the recommended WHO PM 2.5 limit every day when the Berkeley site was affected by the wildfire smoke from Camp Fire, and the 24 h average concentrations were two times higher than the limit in 7 of these 13 days. Median BC concentrations increased approximately four times above the pollution level typical for Berkeley. The median OC increased by 8.4 in the first, more active fire phase when flaming is expected to dominate. Later, when more smoldering is expected, the median OC increase was even higher, 13.9 times above the background concentrations, suggesting a more intense formation of SOA during the later fire phase.
The results of apportionment models that can be run in real-time can be an essential input for short-term air quality prediction models. On the other hand, we showed an increased contribution of BrC to total light absorption at short wavelengths during the event. Therefore, the fraction of the absorbed light by BrC can be used as a marker for detecting the influence of wildfire plumes. While the fractional contribution of BrC to light absorption before and after the wildfire event was low (approximately 5%), it increased up to 50% during the event. Considering that the magnitude and variability of BrC absorption are still uncertain and poorly represented in climate models, we believe the multiwavelength measurements of aerosol optical properties can help improve the parameterization of BrC absorption in climate models.

Acknowledgments:
We would like to acknowledge the support of the measurement campaign provided by Jonathan P. Bower, Jack Connor, and Charity Garland from the Bay Area Air Quality Management District (BAAQMD). This work has not been reviewed for technical accuracy by the BAAQMD. The views expressed in this document are solely those of the authors, and the BAAQMD does not endorse any products or commercial services mentioned in this publication.  Figure A1 contains the results of the BC tracer model. The minimum R-squared method is used to estimate the optimal (OC/BC) prim ratio [25,26].

Appendix A
Toxics 2023, 11, 497 9 absorption before and after the wildfire event was low (approximately 5%), it increa up to 50% during the event. Considering that the magnitude and variability of BrC sorption are still uncertain and poorly represented in climate models, we believe the m tiwavelength measurements of aerosol optical properties can help improve the para terization of BrC absorption in climate models. Appendix A Figure A1 contains the results of the BC tracer model. The minimum R-squa method is used to estimate the optimal (OC/BC)prim ratio [25,26]. Figure A1. BC tracer model before/after Camp Fire event (a) and when the air was affected by C Fire smoke (b). OC/BC probability distribution is presented with blue and cumulative OC/BC tribution with a green line. The orange curve shows the R 2 score between the hypothetical SOC BC for different hypothetical (OC/BC)prim ratios. The vertical dashed orange line marks the m mum of the R 2 function (optimal value). The optimal value is also written in the title with the c lated percentile in brackets.
A significant shift to higher OC/BC ratios can be observed when the air quality strongly influenced by the smoke from Camp Fire in Figure A1. Before and after the Ca Fire event (Figure A1a), 75% of all OC/BC ratios were lower than 4.1, with a median v of 2.9, and 75% of all OC/BC ratios were higher than 4.4 during the Camp Fire event ( ure A1b), with a median value of 5.5. Lower OC/BC (OC/EC) ratios are typical for tr Figure A1. BC tracer model before/after Camp Fire event (a) and when the air was affected by Camp Fire smoke (b). OC/BC probability distribution is presented with blue and cumulative OC/BC distribution with a green line. The orange curve shows the R 2 score between the hypothetical SOC and BC for different hypothetical (OC/BC) prim ratios. The vertical dashed orange line marks the minimum of the R 2 function (optimal value). The optimal value is also written in the title with the calculated percentile in brackets.
A significant shift to higher OC/BC ratios can be observed when the air quality was strongly influenced by the smoke from Camp Fire in Figure A1. Before and after the Camp Fire event (Figure A1a), 75% of all OC/BC ratios were lower than 4.1, with a median value of 2.9, and 75% of all OC/BC ratios were higher than 4.4 during the Camp Fire event (Figure A1b), with a median value of 5.5. Lower OC/BC (OC/EC) ratios are typical for traffic sites. For example, Zhou et al. [40] collected typical OC/EC ratios for tunnel experiments between 0.3 and 1.6. Higher OC/EC ratios from biomass burning are expected-for example, in work by Tian et al. [41], the OC/EC ratios from different biomass-burning experiments were between 2.4 and 9.1. Figure A2 contains typical diurnal profiles of BC, SOC, POC, and b BrC abs , averaged for periods before and after the Camp Fire event.
Toxics 2023, 11,497 10 o sites. For example, Zhou et al. [40] collected typical OC/EC ratios for tunnel experime between 0.3 and 1.6. Higher OC/EC ratios from biomass burning are expected-for ex ple, in work by Tian et al. [41], the OC/EC ratios from different biomass-burning exp ments were between 2.4 and 9.1. Figure A2 contains typical diurnal profiles of BC, SOC, POC, and b abs BrC , averaged periods before and after the Camp Fire event.     sites. For example, Zhou et al. [40] collected typical OC/EC ratios for tunnel experiments between 0.3 and 1.6. Higher OC/EC ratios from biomass burning are expected-for example, in work by Tian et al. [41], the OC/EC ratios from different biomass-burning experiments were between 2.4 and 9.1. Figure A2 contains typical diurnal profiles of BC, SOC, POC, and b abs BrC , averaged for periods before and after the Camp Fire event.                      [34] daily averaged PM2.5 concentrations measured by Purple Air sensors [35] are on the right.