Real-Time Monitoring of SO2 Emissions Using a UV Camera with Built-in NO2 and Aerosol Corrections

Nitrogen dioxide (NO2) absorption correction of the sulfur dioxide (SO2) camera was demonstrated for the first time. The key to improving the measurement accuracy is to combine a differential optical absorption spectroscopy (DOAS) instrument with the SO2 camera for the real-time NO2 absorption correction and aerosol scattering correction. This method performs NO2 absorption correction by the correlation between the NO2 column density measurement of the DOAS and the NO2 optical depth of the corresponding channel from the SO2 camera at a narrow wavelength window around 310 and 310 nm. The error of correction method is estimated through comparison with only using the second channel of the traditional SO2 camera to correct for aerosol scattering and it can be reduced by 11.3% after NO2 absorption corrections. We validate the correction method through experiments and demonstrate it to be of greatly improved accuracy. The result shows that the ultraviolet (UV) SO2 camera system with NO2 absorption corrections appears to have great application prospects as a technology for visualized real-time monitoring of SO2 emissions.


Introduction
Emissions of polluting gases from industries and ships have brought severe air pollution, particularly in developed and coastal areas. Atmospheric pollutants from industries and ships are mainly generated from fuel combustion. The main product of combustion includes sulfur dioxide (SO 2 ), nitrogen dioxide (NO 2 ), carbon dioxide (CO 2 ), and particles [1][2][3]. SO 2 is of special importance and interest. As a toxic gas, SO 2 is responsible for many deleterious effects on human health, the environment, and the climate. SO 2 emissions contribute to the formation of sulfate aerosols and small particles, which may penetrate deeply into the lungs, and, in sufficient quantity, can contribute to health problems. Increased SO 2 in the Earth's atmosphere can alter the radiation balance by intercepting scattered light. However, its far distance and low concentration make SO 2 emissions difficult to measure by using the existing techniques.
The existing optical techniques for SO 2 concentration measurements are an effective tool for high concentrations or fixed sources or near distances. Raman scattering lidar and differential absorption lidar are active detection methods [4,5] that can realize stereoscopic detection of pollutant space and have distance resolution capability. However, their spatial resolution is low and cannot meet the requirements of portable and mobile applications. For the sake of portability and engineering, passive detection methods have emerged. Fourier transform infrared spectroscopy and Fourier function analysis are characterized by multi-component analysis, wide measurement range, and fast analysis speed [6][7][8]. Differential optical absorption spectroscopy (DOAS) provides fast response and real-time monitoring to characterize and obtain data [9][10][11]. Recent advancements in spectroscopic techniques allow remote analysis of many species. Non-imaging measurement techniques allow us to measure the total column density of the trace gas along a single direction within the plume. Imaging DOAS (I-DOAS) combines the advantages of DOAS with imaging capabilities [12,13]. This technique allows the spectroscopic measurement of 2D pollutant gas distributions and improves the accuracy of quantitative measurement. A drawback of scanning DOAS systems is a relatively long time to acquire a 2D image of the trace gas distribution.
The development of remote sensing techniques based on a novel ultraviolet (UV) SO 2 camera system [14] for trace gas measurement has emerged. The SO 2 camera is increasingly used in SO 2 emission due to its ability to remotely measure the 2D distribution of SO 2 path concentrations in space and the emission rate over time in real-time. The measurement of 2D SO 2 distribution in volcanic plumes has been realized by using a single UV camera. UV cameras have been increasingly established [15,16]. The University of Heidelberg discussed the theoretical basis of UV cameras and calibration issues, including the measurement principle, data evaluation, and solar zenith angle [17]. They used the 3D backward Monte Carlo radiative transfer model to describe the spectral radiance transmitted through the filter [18]. Christoph Kern et al. used seven different UV cameras and four different filters to continuously monitor the volcanic plumes in real-time [19]. Osorio, M et al. proposed a new two-image method (2-IM) to acquire background images and quantify SO 2 emissions from industrial sources [20]. However, UV cameras are mostly used to measure higher concentrations of SO 2 emissions from volcanic sources. Few measurements of emissions with low concentrations from industries and ships have been performed with UV cameras. Some persistent problems may arise when UV camera systems are used to measure the SO 2 emissions of industrial and ship sources.
UV SO 2 cameras are used to measure the SO 2 absorption in a narrow wavelength (310 nm), and the effect of aerosol scattering can be partially corrected by measuring the weak absorption of SO 2 around 330 nm. The measured signal is the integral intensity of the incident spectrum over the filter transmittance window. SO 2 cameras are certainly limited by the fact that only one trace gas can be measured and the other spectral interferences and the effect of aerosol scattering on the particulate matter can potentially lead to an inaccurate measurement.
This work aims to correct NO 2 absorption for real-time measurement of SO 2 emissions with UV cameras. This proposed method performs NO 2 absorption correction by the correlation between the NO 2 column density measurement of the DOAS and the NO 2 optical depth of the corresponding channel from the SO 2 camera at a narrow wavelength window around 310 and 310 nm. Theoretical simulation analysis shows that different NO 2 column densities have a relatively large influence on the error of SO 2 column density measured by the UV camera. In this paper, we briefly described the theoretical basis for pertinent aspects of spectral consideration. The basic principles of UV SO 2 camera methodology and the simulation of error analysis are described. The main experimental results are obtained from a plant in Wuhan. In order to further enhance the measurement accuracy, the Mie scattering corrections are made. The following discusses the measurement uncertainty of SO 2 column density. Then, we show the SO 2 error on NO 2 absorption corrections.

Spectral Consideration
When scattered solar radiation passes through industrial and ship emission plumes, light is scattered and absorbed along the path. This condition is caused by absorption within the band due to other gases (such as NO 2 , O 3 ), scattering of particles, and multiple scattering. In the UV region, the O 3 absorption [blue line in Figure 1a] blocks part of the scattered solar spectra and the UV cannot break through the ozone at a wavelength shorter than 300 nm. In Figure 1a (red line), the relative scattered solar spectral intensity reaches the ground at a spectral resolution of 0.1 nm, which is calculated on MODTRAN [21]. The UV cameras are used to measure the SO 2 emissions in a narrow wavelength band centered at 310 nm, which is the region of the scattered solar spectral intensity with an increasing magnitude as the wavelength increases. The spectral interference from other major emission products (such as NO 2 ) should be considered for measuring SO 2 emissions. The measurement of SO 2 emission from industries and ships cannot be ignored due to the extremely high NO 2 absorption interference. Figure 1b shows the SO 2 absorption spectrum (blue line) and NO 2 absorption spectrum (red line) in the spectral range of 280-360 nm. As shown in Figure 1b, the absorption band of SO 2 overlaps with the absorption band of NO 2 , and the absorption cross-sections of SO 2 and NO 2 are comparable in the chosen spectral window. We calculated the SO 2 column density (CD) error caused by the interference of NO 2 absorption. In accordance with the Beer-Lambert law to resolve the SO 2 signal, the signal channel and the reference channel can be expressed as: where the incident spectral radiation intensity I and radiation intensity I 0 are the intensity images after and before traversing the plume at wavelength λ, σ SO 2 (λ on ) is the absorption cross-section of SO 2 , S SO 2 is the SO 2 CD or the integral of SO 2 concentration along the effective optical path, σ NO 2 (λ on ) is the absorption cross-section of NO 2 , and S NO 2 is its CD. τ SO 2 is the apparent absorbance, which is the difference between the optical depth (OD) for λ on and λ o f f . No aerosols are assumed to be present in the emission plume. We obtain an expression for the SO 2 optical depth: where is the ratio between the integral of NO 2 absorption cross-section at two region wavelengths of narrow-band filters. In this work, λ on is 310 nm, and λ o f f is 330 nm with narrow-band filters in the center wavelength. Figure 2 shows the simulated SO 2 CD measurement error under the influence of NO 2 CD. With the significant increase in NO 2 CD towards high concentration, the signal channel (λ on ) SO 2 OD is particularly influenced by the NO 2 CD. This error is the main source for measuring SO 2 emissions. It is important to eliminate the influence of NO 2 absorption in the plume.  The error is approximately the monotone function of NO 2 CD, and the relation between them is particularly affected by the NO 2 CD. The error clearly increases with the increase in NO 2 CD. Therefore, the error of measuring SO 2 CD can be effectively eliminated if the ratio of the two NO 2 ODs of two channels is known.

Experiment Instrument
To verify the accuracy of the theoretical analysis and simulation, we used SO 2 cameras to acquire images of SO 2 emissions from the plant near Wuhan on 16 July 2021. The SO 2 camera system is shown in Figure 3. Figure 3a shows a photograph of the SO 2 camera system and the DOAS system. The main part of the instrument (see green dotted line area in Figure 3a) is the DOAS system, a lens, and spectrometer-connected fiber. As well as two cameras (blue dotted line area in Figure 3a), each with a lens and a bandpass filter (310 nm and 330 nm). From outside the plant, we can observe one stack from which SO 2 emissions are monitored (red dotted line area in Figure 3a). Figure 3b shows a schematic diagram of the SO 2 camera system (correspond to the blue dotted line in Figure 3a). The SO 2 cameras (Prime 95B Blue) used for these experiments have the most sensitive scientific backilluminated CMOS sensor (1200 × 1200 pixels, 11 µm × 11 µm pixel size) manufactured by Photometrics (Tucson, AZ, USA). The quantum efficiency of approximately 50-55% at 310 and 330 nm of each SO 2 camera is high from 200-1000 nm. The UV lens (UV1054B) of 105 mm focal length (Universe Kogaku America Inc., Oyster Bay, NY, USA) with a total field of view (FOV) of 9.8 • and F-number of 4.0 were mounted in front of each camera. Two UV narrow-band filters (ASAHI SPECTRA, Torrance, CA, USA), with a central wavelength of 310 (XBPA310) and 330 nm (XBPA330) and full width at half maximum of 10 nm, were used for the measurements. Two narrow-band filters were mounted between the lens and the cameras. This setup was chosen to reduce the influences from the illumination angle. Two cameras were fixed side by side to simultaneously capture the current polluting gases of industrial emissions by using a function generator (Tektronix AFG 3022B, Beaverton, OR, USA) with a frequency of 1-5 Hz. For comparison, a DOAS system was installed on the two UV cameras. The DOAS system consisted of an Ocean Optics Maya (Dunedin, FL, USA) 2000Pro spectrometer with a spectral range between 247 and 390 nm and a resolution of 0.035 nm, a 600 µm optical fiber, and a telescope with the same quartz lens as the UV camera. The SO 2 camera system is portable. The location (approximately 500 m from the plant through a laser rangefinder) was chosen to be as close as possible to the plant plume emissions (red dotted line area in Figure 3a) to reduce light dilution effects. The exposure times (approximately 0.1 s for 310 nm and 0.06 s for 330 nm) were used to acquire images at approximately 50-60% of the intensity saturation level. The camera with a 310 nm narrowband filter required long exposure times due to the low intensity of solar radiation. The UV cameras and DOAS system were controlled through panning and tilting. The exposure time of the camera system should be multiple times, especially for the measurement of industrial emissions, to overcome the mechanical complexity of the camera system and the influence of the zenith angle. The images were captured with acquisition rates of 5 Hz, with an additional collection of dark images prior to the capturing of plume sequences. The acquisition and subtraction of a dark frame are required to correct the dark current and electronic noise on the UV camera system. Figure 4a,b shows a pair of the raw images (λ on and λ o f f ) of an industrial plume. The raw images were processed following the protocols, which are already described in the literature [20], including the subtraction of a dark current and image matching. Simultaneously, the background sky images (λ on and λ o f f ) of the plume can be constructed from the raw plume images by plume segmentation and interpolation.  The measurement accuracy of the SO 2 CD from industrial emissions is affected by the NO 2 absorption. Industrial emissions were measured when the plume entered the camera FOV within the plant. The camera system can be calibrated by using the real-time continuous calibration method of Wu [22,23]. In the particular case of assuming the absence of aerosols in the plume, the SO 2 CD density from industrial emissions obtained from the real-time continuous calibration method is shown in Figure 5. Figure 5a shows the SO 2 image ignoring the effect of NO 2 absorption, and Figure 5b shows the SO 2 image using the second channel to correct the NO 2 absorption. Ignoring the NO 2 absorption leads to an overestimation of the emissions.

Mie Scattering Corrections
As illustrated in Figure 5, the effect of NO 2 absorption can lead to an inaccurate SO 2 image measurement. The result of the SO 2 image obtained by the two methods differed twice. This finding may be the result of the NO 2 absorption correction based on the use of a second channel. However, when scattered solar radiation reaches a camera after passing through a plume from a stack, it contains several trace gases (SO 2 and NO 2 ) and aerosol. Assuming that aerosol scattering is independent of wavelength, the second channel of the traditional camera is mostly used to correct plume aerosols in the stack. Therefore, a second channel is used to correct the aerosol or NO 2 absorption. Although a second channel certainly reduces the influence of aerosol scattering and NO 2 absorption on SO 2 camera measurements, it does not completely remove it. This condition is a limitation of the SO 2 camera. However, this problem can be overcome by correcting NO 2 absorption and aerosol scattering. The OD of SO 2 by applying two band-pass filters can be expressed as: is the ratio between the scattering cross-section at two different wavelengths λ on and λ o f f . τ Mie (λ on ) and τ Mie (λ o f f ) are the plume aerosol ODs (AODs). Data analysis was conducted and is summarized in Figure 6. For two band-pass filters, correcting the effects of NO 2 absorption and aerosol scattering at the same time is impossible. The corresponding signal channel (λ on ) and the corresponding reference channel (λ o f f ) are influenced by NO 2 absorption and aerosol scattering. Here, the spectral data were analyzed by using DOAS to obtain the NO 2 CD. The two cameras simultaneously capture the plume with the same FOV, and the spectrometer measures the spectral information of the plume area with the same FOV of the camera. This process was to accurately determine the area in which the DOAS optical FOV is directed and obtains the best correlation between the corresponding channel NO 2 OD from the SO 2 camera and the NO 2 CD measured by DOAS. Using Equation (1), the relationship between the column density of NO 2 and the optical density τ NO 2 is obtained. Figure 7 shows the calculated relationship curve between the signal and reference channel NO 2 CD of the SO 2 cameras and the NO 2 OD. The signal channel (red line) and the reference channel (blue line) NO 2 OD of the SO 2 cameras are derived from the NO 2 CD measurement by DOAS. The NO 2 CD is converted to NO 2 OD by multiplying with the factor obtained from the relationship curve. The subtraction of the NO 2 OD is required to correct the NO 2 absorption in the two channels. Figure 8b shows the concentration image of the SO 2 plume obtained by UV cameras with the NO 2 absorption and aerosol scattering considered. For comparison, Figure 8a shows the SO 2 CD from UV cameras with only aerosol scattering considered. A second channel is used to correct the plume aerosols because the SO 2 CD is underestimated because of the presence of NO 2 absorption in the plumes. Therefore, this method using DOAS with the SO 2 camera makes the real-time NO 2 absorption correction. The accuracy of this method is significantly improved by the NO 2 absorption correction.  The accuracy of the SO 2 CD depends on the difference between the NO 2 OD measured by DOAS and the NO 2 OD of the corresponding channels on the SO 2 camera. Simultaneously, it can be found that only using the second channel from the SO 2 camera to correct for aerosol scattering results in an underestimation of the SO 2 measurement. However, this method is practical for emission plume monitoring from industrial sources and ships, especially for SO 2 CD measurement with NO 2 absorption correction.

SO 2 CD Error
The error measured by the SO 2 camera is an extremely important quantity for accuracy. The measurement uncertainty of the SO 2 CD can be determined through propagation of error analysis. The dominant sources of uncertainty during the measurements are the system error (∆S 1 , uncertainty of 11.3%) and the random error (∆S 2 , uncertainty of 7.48%). The SO 2 CD error ∆S can be expressed as: The absorption in the second channel is due to NO 2 and aerosols. However, the correction procedure for the second channel is only used to correct for plume aerosols. The influence of NO 2 absorption on SO 2 camera measurements is not completely removed. The system error of the SO 2 CD measurement is calculated under the influence of NO 2 CD. The random error of SO 2 CD is derived from light intensities I(λ on ), I 0 (λ on ), I(λ o f f ), and I 0 (λ o f f ), which are referred to as the apparent quantities. The random error of the SO 2 OD ∆τ SO 2 is obtained from the relation.
where ∂τ/∂I(λ on ), ∂τ/∂I 0 (λ on ), ∂τ/∂I(λ o f f ), and ∂τ/∂I 0 (λ o f f ) represent the polarization of OD to light intensity, and ∆I is the amount of change in light intensity. The random errors of the SO 2 OD are due to: (1) photon noise, which can be reduced by combining pixels. This precision can be improved to approximately twice by averaging 4 adjacent pixels and reducing the image size to 0.25 pixels.
(2) the light dilution effect and the main source of the random error on UV SO 2 camera measurements is related to the scattering of ambient photons [24]. The time series of the error of the SO 2 CD corrected for aerosol scattering using the second channel of the UV SO 2 camera is shown in Figure 9. The error of the SO 2 CD is under 14%. The variation of SO 2 CD error is mostly due to the change of NO 2 CD in the SO 2 emission plume.

Conclusions
In this work, we first put forward an SO 2 camera for the real-time NO 2 absorption corrections to improve the measurement accuracy of SO 2 emissions. The instrument was designed with NO 2 absorption corrections and aerosol scattering corrections to improve the accuracy of the SO 2 CD. The theoretical analysis for the spectral interference of industrial SO 2 emissions is given in detail, and the error of NO 2 absorption interference is simulated. The UV SO 2 camera system with NO 2 absorption corrections is developed to measure the SO 2 CD of an industrial plume in Wuhan, and a series of experimental results are obtained to verify the accuracy of this theoretical analysis. The 2-IM method is utilized to obtain the artificial background sky images, SO 2 CD, and the error of the SO 2 CD. The traditional camera system only used the second channel to correct the interference of aerosol scattering, which underestimates the SO 2 emissions due to the interference of NO 2 absorption in the plume. We compare the experimental results with and without the effect of NO 2 absorption correction. The SO 2 CD errors between them are effectively reduced by 11.3% after NO 2 absorption corrections. Therefore, an effective means to improve the accuracy of the SO 2 CD is to combine the UV SO 2 cameras and a spectrometer to measure the plume at the same time, which can effectively overcome the interference of NO 2 absorption and aerosol scattering. This new method can provide a good temporal resolution for real-time NO 2 absorption corrections, especially for measuring the plumes from industries and ships. This promising method may greatly improve the measurement accuracy of SO 2 emissions and provide the most convenient option for rapid measurements of industrial and ship SO 2 emissions in the foreseeable future.