Estimation of Aerosol Extinction Coefficient Using Camera Images and Application in Mass Extinction Efficiency Retrieval

Abstract

In this study, we attempted to calculate the extinction parameters of PM_2.5 using images from a commercial camera. The photo pixels provided information on the characteristics of the objects (i.e., the reflectivity, transmittance, or extinction efficiency) and ambient brightness. Using the RGB values of pixels, we calculated the extinction coefficient and efficiency applied to the mass concentration of PM_2.5. The calculated extinction coefficient of PM_2.5 determined from the camera images had a higher correlation with the PM_2.5 mass concentration (R² = 0.7) than with the visibility data, despite the limited mass range. Finally, we identified that the method of calculating extinction parameters using the effective wavelength of RGB images could be applied to studies of changes in the atmosphere and aerosol characteristics. The mass extinction efficiency of PM_2.5, derived from images, and the mass concentration of PM_2.5 was (10.8 ± 6.9) m² g⁻¹, which was higher than the values obtained in Northeast Asia by previous studies. We also confirmed that the dry extinction efficiency of PM_2.5, applied with a DRH of 40%, was reduced to (6.9 ± 5.0) m² g⁻¹. The extinction efficiencies of PM_2.5, calculated in this study, were higher than those reported in previous other studies. We inferred that high extinction efficiency is related to changes in size or the composition of aerosols; therefore, an additional long-term study must be conducted.

1. Introduction

Atmospheric aerosols, one of the most critical pollutants suspended in the atmosphere, have received increasing attention owing to their adverse impacts on air quality and human health [1,2]. Suspended particulate matter with an aerodynamic diameter less than 2.5 μm (PM_2.5) in atmospheric aerosols are more closely associated with adverse health effects than the larger particles. Therefore, the 2006 World Health Organization Air Quality Guidelines recommended PM_2.5, rather than PM₁₀, as an indicator of air pollution from particulate matter, which has increasingly become a public concern worldwide [3]. The World Health Organization (WHO) recently decreased the standard limit, in the Air Quality Guidelines, of PM_2.5 from 10 to 5 μg m⁻³ due to the increasing impact of PM_2.5 on air pollution over time [4].

PM_2.5 can be distributed at various altitudes and over long distances in the air. Therefore, various remote sensing technologies, such as satellites, light detection and ranging (lidar), and sun photometers are used to measure the concentrations and distributions of particulate matter in the atmosphere as well as on the ground [5,6,7,8]. Remote sensing technologies have many spatiotemporal advantages, but they also have significant disadvantages. Satellites provide information over a wide area; however, confirming detailed data within a narrow region of several kilometers is challenging. Temporal resolutions also require long intervals, that is, daily (polar-orbiting satellites) and hourly (geostationary-orbiting satellites) [9,10,11,12]. The lidar technique is a valuable active remote sensing device that provides concentration data over a distance, but it mainly detects vertical distribution depending on altitude. Although lidar sometimes detects horizontal distributions, it has limitations due to eye safety and topographical factors. In general, optical measurement methods (lidar, optical particle counters, visibility systems, and nephelometers) calculate the optical concentration using transmittance or scattering methods, which require improved accuracy in the signal-to-noise ratio (SNR). Furthermore, remote sensing techniques have limited applications as they are cost and skill intensive.

Furthermore, aerosols in the atmosphere can change rapidly; hence, prediction or measurement methods must be developed for real-time data [13,14]. Therefore, there is a need for improving the spatial and temporal resolution of the national aerosol monitoring systems (for PM₁₀ and PM_2.5). From a scientific perspective, continuous monitoring of the mass extinction efficiency (MEE) is crucial as it depends on the type of aerosol irrespective of the mass quantities.

To overcome the limitations of remote sensing technology and obtain information on air quality effectively, we used landscape images to analyze visibility [15,16,17,18,19,20,21,22]. Many studies have attempted to measure visibility change [18,19,20,22] and aerosol optical depth [23] using the RGB information of the images; however, they have not calculated the extinction coefficient of aerosols. The RGB values of landscape or sky images change due to the extinction coefficient and scattering properties of aerosols, object position, the position between the sun and the camera, atmospheric cloud conditions, and other parameters such as the sensitivity and transmittance of the imaging optics in the measurement. Therefore, we aimed to confirm the applicability of monitoring aerosols using images from a commercial camera to provide a practical and uncomplicated method to overcome the limitations of remote sensing and existing measurement systems.

In this study, we calculated the aerosol extinction coefficient using the images captured by a commercial camera and compared it with the mass concentration of PM_2.5, and the retrieved extinction coefficient from the visibility data, to validate data reliability. In addition, the extinction efficiency of PM_2.5 was calculated using the extinction coefficient and mass concentration of PM_2.5.

2. Materials and Methods

2.1. Calculation of the Extinction Coefficient from the Camera Images

We calculated the extinction coefficient from the camera-image files according to the method developed by Kim and Noh [24]. The imaged pixels include information on the reflectance of the target object, transmittance or extinction/scattering/absorption efficiency, and brightness in the atmosphere. Thus, the intensity, I (D_i), of the image incident on a given pixel can be described by Equation (1) [25,26,27]:

$$I\left(D_i\right)=C_1{e}^{-\alpha D_i}+C_2(1-e^{-\alpha D_i})=\left(C_1-C_2\right)e^{-\alpha D_i}+C_2$$

(1)

where D_i is the distance from the target object (i) to the camera, C₁ is the light intensity of the target object itself, α is the average extinction coefficient between the target object and the camera, and C₂ is the atmospheric scattering light scattered by the particulate matter. C₁ is only affected by the characteristics of the scattering objects (i.e., the reflectivity and sun–camera angle) and not by the distance. The first term in Equation (1) represents the intensities scattered from the objects that decrease exponentially depending on the distance between the objects. The second term describes the amount of scattered light from the particulate matter and air molecules, which increases with the distance of the target.

Equation (1) demonstrates the possibility of calculating the image-extinction coefficient (α_IMAGE) based on the following assumptions: (1) C₁ and C₂ are located at different distances in the same direction in a single image; (2) a known pixel value, I (D_i), of the target object exists with similar scattering characteristics; and (3) C₁ is constant regardless of distance because it is not sensitive to the scattering angle from the sun and camera. Therefore, mountains with similar characteristics and inclinations are the proper target objects.

The critical factor in determining C₂ is the scattering angle, which depends on the sun, object, and camera. In addition, C₂ is dependent on the scattering phase functions related to the properties of aerosol particles, such as size, shape, and refractive index. However, the scattering of particles is constant when the target objects are positioned in the same direction. Thus, the C₂ values of two pixels extracted from different objects in the same direction are similar.

Obtaining targets in the exact same direction as a single image, from actual observation, is challenging. In most of the cases in this study, we used target objects in similar directions because the differences in the object angles were within the range of systematic error. Finally, we calculated the extinction coefficient of the objects in a single image. The steps for selecting the pixel value, I (D_i), of the target object through the best analysis conditions were as follows: (1) the objects with similar directions (for example, a mountain in this study) were selected, (2) the specific areas at the same distance (D₁, D₂, …, D₅ in Figure 1a) were selected, and (3) the average or minimum values of all pixels in the selected area were determined. The sky was considered an object with a reflectance close to zero because no real object was present.

Figure 1a shows four mountains composed of similar ecosystems, and the distances to D₁, D₂, D₃, D₄, and D₅ are 0.4, 1.4, 2.4, 3.5, and 10 km, respectively. The red boxes denote the point pixels of the objects on the image of the mountain from the camera. We selected the smallest values among all of the pixels inside the boxes and considered them as I (D_i) from their average. The intensities of the RGB wavelengths were calculated from the pixels of the images. Kim and Noh [24] confirmed that the calculated and measured intensities of RGB are highly correlated and developed a method to calculate the effective RGB wavelength (λ_i) by considering the RGB sensor characteristics of a digital camera, as follows:

$${\lambda }_i=\frac{{\int }_{-\infty }^{\infty }\lambda S_i\left(\lambda \right)I\left(\lambda \right)\epsilon \left(\lambda \right)d\lambda }{{\int }_{-\infty }^{\infty }{S}_i\left(\lambda \right)I\left(\lambda \right)\epsilon \left(\lambda \right)d\lambda }, i=R, G, \mathrm{and} B$$

(2)

In Equation (2), S_i (λ) is the intensity of the digital camera for RGB sensors and I(λ) is the light intensity spectrum that varies with time and region. In this instance, ε(λ) is the change in extinction efficiency from aerosol properties, such as size, refractive index, and wavelength [28]. Although the effective wavelength of RGB depends on unknown parameters, such as the Mie scattering effect, three effective wavelengths can be obtained from a given solar intensity and sensitivity characteristic profile, S_i (λ). The effective wavelengths of camera RGB are 597 nm, 534 nm, and 459 nm, respectively; these have differences but are similar to the wavelength of 550 nm when we calculate the visual range. This study used a high-resolution monochromator to calculate the sensitivity profiles.

2.2. Calculation of the Extinction Coefficient and Efficiency Using Visibility Data

We compared the two extinction coefficient values calculated from the visibility data and camera images. The extinction coefficient (α_VIS) was derived from the Koschmieder equation [29].

$${\alpha }_{VIS}\left({\mathrm{km}}^{-1}\right)=-\frac{ln\epsilon }{VIS\left(\mathrm{km}\right)}$$

(3)

VIS indicates visibility and ε is the brightness contrast threshold. We set ε as 0.02, which is commonly used by the WHO. Equation (3) calculates the extinction coefficient of ambient conditions. To determine the extinction coefficient of PM_2.5, we used Equation (4), considering the mass loading of PM_2.5 and PM₁₀ (in μg m⁻³), and NO₂ concentrations (in ppb) [30].

$${\alpha }_{VIS2.5}\left({\mathrm{M}\mathrm{m}}^{-1}\right)=\frac{3912}{VIS\left(\mathrm{km}\right)}-0.6\left({\mathrm{PM}}_{10}-{\mathrm{PM}}_{2.5}\right)-10-0.33{\mathrm{NO}}_2$$

(4)

The first term represents the total extinction coefficient (at 550 nm) computed from the visual range (km) using Koschmieder’s formula. The other terms (subtracting factors not related to the PM_2.5 extinction coefficient) correspond to the extinction contribution of coarse particles, the Rayleigh scattering of air molecules, and the absorption of ambient NO₂ molecules. The constants were calculated using empirical studies [31,32,33]. We also applied Equation (4) to calculate the extinction coefficient of PM_2.5 (${\alpha }_{IMAGE2.5}$) from the camera images, excluding NO₂ concentration (ppb) and the effect of the mass loading of the particulate matter with a diameter of over 2.5 μm.

Finally, we calculated the mass extinction efficiency of PM_2.5 (${\alpha }_{eff2.5}$) by dividing the mass concentration (${\mathrm{PM}}_{2.5}$) from the extinction coefficient of PM_2.5 obtained from the camera images (${\alpha }_{IMAGE2.5}$) or visibility data (${\alpha }_{VIS2.5}$), based on the following equation:

$${\alpha }_{eff2.5}\left({\mathrm{m}}^2{\mathrm{g}}^{-1}\right)=\frac{{\alpha }_{VIS2.5 or IMAGE2.5}\left({\mathrm{Mm}}^{-1}\right)}{{\mathrm{PM}}_{2.5}\left({\mathsf{\mu }\mathrm{g}\mathrm{m}}^{-3}\right)}$$

(5)

2.3. Measurement Period and Site

We captured the images from the fourth floor of the School of Art and Design building at Hanbat National University, Daejeon, Republic of Korea (36.34°N, 127.30°E), in the west–southwest direction every day from February 7 to 28, 2021. This step was conducted at one-hour intervals from 08:00 to 18:00 h in local time (LT). In this study, all time values are in LT. We excluded data when the weather conditions seriously affected visibility, such as rainfall, snowfall, and fog. We obtained 193 images during the entire measurement period (22 days). The mass concentration of PM_2.5 was obtained from the nearest observatory (the rooftop of the Noeun 1-dong community service center; 36.37°N, 127.32°E) available on the national real-time air pollution level public website (www.airkorea.or.kr (retrieved 15 March 2021)) managed by the Korea Environment Corporation. The visibility and relative humidity data were obtained from the Daejeon Regional Meteorological Administration (36.37°N, 127.37°E), which was located 6.8 km to the northeast of Hanbat National University. We presented all measurement sites in Figure 2.

3. Results and Discussion

3.1. Extinction Coefficient

The mass concentration and extinction coefficient, measured by the camera images and visibility data, of PM_2.5 at one-hour intervals during the measurement period are shown in Figure 3. We ignored the effect of coarse particles because their contribution to extinction is relatively lower than fine particles [34,35,36]. Measurements were obtained from basin regions where urban pollutants dominate; therefore, we assumed that the contribution of coarse particles to extinction would be less than 30% [37]. The PM_2.5 mass concentration was high on 7, 11–15 February, and was over 30 μg m⁻³ on the 19–21. The maximum value was 63.5 μg m⁻³ at 11:00 h on 7 February. The ${\alpha }_{IMAGE2.5}$ values show similar trends with the mass concentration of PM_2.5. On 7 February, when the PM_2.5 mass concentration was the highest, ${\alpha }_{IMAGE2.5}$ was also high (0.22–0.95 km⁻¹). In addition, ${\alpha }_{IMAGE2.5}$ was low (0.006–0.22 km⁻¹) on 8–9 and 16–18 February. The maximum and minimum values of ${\alpha }_{IMAGE2.5}$ were 0.95 and 0.002 km⁻¹, respectively. The ${\alpha }_{VIS2.5}$ values had few similarities with the change in PM_2.5 concentration, except on 7 February. The maximum and minimum values of ${\alpha }_{VIS2.5}$ were 1.09 and 0.031 km⁻¹, respectively, and were higher than those of ${\alpha }_{IMAGE2.5}$.

The PM_2.5 mass concentrations were 6.0–21.5 μg m⁻³ in the maximum visibility range of 60 km for this measurement. The calculated ${\alpha }_{VIS2.5}$ on the maximum visibility varied from 0.031 to 0.041 km⁻¹ depending on the atmospheric pressure and NO₂ concentration. Moreover, the range of ${\alpha }_{IMAGE2.5}$ was 0.004–0.15 km⁻¹. The minimum values of ${\alpha }_{IMAGE2.5}$ were considerably lower than those of ${\alpha }_{VIS2.5}$. Visibility can reach up to 300 km, when affected only by Rayleigh scattering and the gas absorption of air molecules [38], but the maximum visibility in this study was limited to 60 km. Therefore, we expected the ${\alpha }_{VIS2.5}$ to be overestimated on clear days.

Figure 4 shows the correlation between the mass concentration of PM_2.5, ${\alpha }_{IMAGE2.5}$, and ${\alpha }_{VIS2.5}$. The mass concentration of PM_2.5 has a high correlation with ${\alpha }_{IMAGE2.5}$ (R² = 0.70) (Figure 4a), which is larger than its correlation with ${\alpha }_{VIS2.5}$ (R² = 0.54 and 0.64 for linear and polynomial regression, respectively). In previous studies, the R² values between the extinction coefficients from lidar, satellite-based lidar, or visibility data and the mass concentration of PM_2.5, measured by the in situ instrument, ranged from 0.50 to 0.96 [39,40,41]. The extinction coefficient calculated from mobile lidar on the vehicle was highly related to PM_2.5, with a relative humidity of 90% or lower (R² = 0.8) [39]. The mass concentration of PM_2.5 measured by the in situ instrument also has a significant relationship with the calculated value, and is classified into seven types (i.e., clean marine, dust, polluted, continental/smoke, clean continental, polluted dust, elevated smoke, and dusty marine) derived from the data of Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), which is a satellite-based lidar [40]. Compared with previous studies, the correlation coefficient of 0.7 is low, which may be due to the short measurement period and relative humidity. This study had a limited measurement period of 22 days; therefore, the mass concentration range was narrow, ranging from 4.5 to 63.5 μg m⁻³. The other studies, as mentioned above, measured the mass concentration of PM_2.5 for extended periods [40] or identified a varied mass concentration range of 200 μg m⁻³ [39]. In addition, we concluded that correction by relative humidity should be considered for a good correlation, as obtained in other studies.

The correlation coefficient between the mass concentration of PM_2.5 and ${\alpha }_{VIS2.5}$ was low (0.54) compared with the correlation coefficient between the mass concentration of PM_2.5 and ${\alpha }_{IMAGE2.5}$ (0.70) using linear regression (Figure 4b). The correlation of ${\alpha }_{VIS2.5}$ and ${\alpha }_{IMAGE2.5}$ was also low (0.55); therefore, we could estimate light extinction by PM_2.5 with a higher accuracy using camera images than visibility data.

The differences between ${\alpha }_{VIS2.5}$ and ${\alpha }_{IMAGE2.5}$ might be due to several reasons. First, the error due to the large proportion of low PM_2.5 cases (52 cases among 193 cases) and the limitation of visual range at 60 km might cause these differences. Therefore, we believe that the visibility deduced from the extinction coefficient, that was calculated from the camera images, is more effective than that deduced using the visual method in person. Second, we did not consider the effects of relative humidity. Many studies confirmed that relative humidity has a high impact on visibility [30,35,36,42,43,44,45,46]. The correlation coefficient of the mass concentration of PM_2.5 and visibility data increased from 0.81 to 0.89 when correcting for the relative humidity [47]. Finally, deviations may occur due to the distance between the measurement site and the visibility observatory.

3.2. Mass Extinction Efficiency

The mass extinction efficiency (MEE) of PM_2.5 (${\alpha }_{eff2.5}$) is a key parameter for confirming the extinction intensity per unit mass concentration. The light extinction by particulate matter is mainly related to hygroscopic deliquescence [48,49] while MEE is proportional to the relative humidity due to the increase in the water absorption capacity of aerosols [50,51].

In terms of the time-dependent changes in ${\alpha }_{eff2.5}$ and relative humidity (Figure 5), the calculated average values of ${\alpha }_{eff2.5}$ were (10.8 ± 6.9) m² g⁻¹. The correlation between ${\alpha }_{eff2.5}$ and relative humidity was as low as 0.27 but it increased with the relative humidity (Figure 6). The ${\alpha }_{eff2.5}$ decreased on the dates of 10–11, 17–19, and 24–25 February, when the PM_2.5 concentrations were high. The ${\alpha }_{eff2.5}$ is an important factor that changes with various factors such as chemical composition, particle-size distribution, and refractive index. We found the particle-size distribution changed through PM_2.5/PM₁₀ ratio variations. The average ${\alpha }_{eff2.5}$ was higher than the measured values in previous other studies (

Table 1. Mass extinction efficiency (MEE) of PM_2.5.

Location	Year	MEE $({\mathbf{m}}^2 {\mathbf{g}}^{-1})$	Reference
24 cities of China	2013–2014	4.4 ± 0.84	Cheng et al. (2017) [30]
YRD	2011–2012	4.1	Cheng et al. (2013) [35]
Eastern China	2014	5	He et al. (2016) [36]
Beijing	2006	4.3	Jung et al. (2009) [43]
Lin’an	1999	5	Xu et al. (2002) [42]
Beijing	2003–2008	4.7	Zhang et al. (2010) [34]
Nanjing	2013 2018	7.1 9.3	Liu et al. (2020) [45]
Busan, Korea	2015–2019	12.1 ± 8.3	Joo et al. (2021) [46]
Daejeon, Korea	2021	10.8 ± 6.9 (6.9 ± 5.0) 1	This research

¹ denotes the limited cases in which the deliquescence relative humidity (DRH) is lower than 40%.

) primarily conducted in China. One study measured the ${\alpha }_{eff2.5}$ in Busan, Korea, as 4.1–12.1 m² g⁻¹ [46]. We attributed these differences to the influence of relative humidity. Water-soluble aerosol particles exist in solid form until the ambient relative humidity reaches the deliquescence relative humidity (DRH) threshold. The solid particles spontaneously absorb atmospheric moisture and their particle size rapidly increases on reaching the DRH, gradually converting them to a saturated aqueous solution. The DRH values vary with the ambient temperature and chemical species of aerosols [52,53]. A typical DRH value is 40% relative humidity [54]. In this study, the ${\alpha }_{eff2.5}$ decreased to (6.9 ± 5.0) m² g⁻¹ for a DRH of 40%; however, it was higher than the values obtained in previous studies. Overall, the MEE of China and Korea is significantly different, which reflects that MEE is also an important parameter to determine the characteristics of aerosols.

We inferred that the high values of ${\alpha }_{eff2.5}$ were caused by the changes in the size and composition of aerosols. Liu et al. [45] confirmed that the MEE in Nanjing, China, increased from 7.1 m² g⁻¹ in 2013 to 9.3 m² g⁻¹ in 2018. Joo et al. [46] also stated that the MEE of PM_2.5 increased by 0.72 m² g⁻¹ annually during 2015–2019.

3.3. Diurnal Variation Patterns

The relative humidity exhibited regular variation patterns with maximum values from 5:00 to 8:00 h at approximately sunrise, then gradually decreased until approximately 15:00 h, and then increased again (Figure 5). The hourly change in the average mass concentration and the extinction coefficient of PM_2.5 depended on diurnal patterns (Figure 7). The mass concentration of PM_2.5, which had low values at 8:00 h, increased steadily, peaked at noon, and then decreased, indicating a minimum value at 16:00 h. In contrast, the extinction coefficient showed a maximum value at 08:00 h, then decreased until 16:00 h.

The daily change in the average values of ${\alpha }_{eff2.5},$ and the relative humidity for each time, are shown in Figure 8. The ${\alpha }_{eff2.5}$ exhibited a maximum value of (15.8 ± 5.9) m² g⁻¹ at 8:00 h, decreased sharply to (11.3 ± 6.9) m² g⁻¹ at 9:00 h, and then decreased slowly until 16:00 h. The relative humidity also had the highest value at 8:00 h, gradually decreased until 15:00 h, and increased from 16:00 h. From Figure 7 and Figure 8, we confirmed that relative humidity had a significant effect on ${\alpha }_{eff2.5}$.

To verify the mass concentration of PM_2.5 calculated using ${\alpha }_{IMAGE2.5}$ and ${\alpha }_{eff2.5}$ applied from the camera images, we compared it with the in-situ measurement of PM_2.5. The ${\alpha }_{eff2.5}$ was applied as the overall average value of 10.8 m² g⁻¹ in Figure 5 and the time average values in Figure 8. The correlation coefficient slightly increased from 0.71 to 0.73 when using the hourly data, but statistically significant results could not be evaluated (Figure 9). This might be because of the limited number of measurements; therefore, long-term observations and studies of seasonal characteristics must be conducted to obtain effective results in the future.

4. Conclusions

In this study, we calculated the extinction coefficient of PM_2.5 using the images captured by a commercial camera for the first time and confirmed its validity by comparing it with the extinction coefficient obtained from visibility data and the in-situ mass concentration of PM_2.5 measured instrumentally. The ${\alpha }_{IMAGE2.5}$ varied from 0.002 to 0.95 km⁻¹ for PM_2.5 mass concentrations of 4.2–63.5 μg m⁻³. The minimum and maximum values of ${\alpha }_{VIS2.5}$ were 1.09 km⁻¹ and 0.031 km⁻¹, which were higher than those of ${\alpha }_{IMAGE2.5}$. The mass concentration of PM_2.5 was more correlated with ${\alpha }_{IMAGE2.5}$ (R² = 0.7) than with ${\alpha }_{VIS2.5}$ (R² = 0.54). We confirmed that the ${\alpha }_{eff2.5}$ was (10.8 ± 6.9) m² g⁻¹, derived from the ${\alpha }_{IMAGE2.5,}$ and the PM_2.5 mass concentration was higher than the values from previous studies from Northeast Asia. Additionally, dry ${\alpha }_{eff2.5}$ (40 % DRH) was reduced to (6.9 ± 5.0) m² g⁻¹.

We proved that the calculation of the PM_2.5 extinction coefficient using the images captured by a commercial camera had a high accuracy and could be used for research on changes in the atmosphere and the characteristics of aerosols. This implies that researchers can use camera images to identify atmospheric changes and particle characteristics caused by PM_2.5, not limited to visibility studies, by replacing the existing expensive optical measuring instruments. Moreover, the extinction coefficient and visibility can be calculated more effectively from the camera images than by using conventional methods. However, we only analyzed limited data for one month. Therefore, long-term and continuous observations must be conducted to verify the accuracy of the proposed method and to identify the independent properties of MEE.