Correcting Black Carbon Absorption Measurements with Micro-aethalometer Model 200: Insights from Comparative Analysis

Black carbon (BC) absorbs light significantly in the visible spectrum and has a non-negligible warming effect on the Earth system ^[1]. BC is emitted to the atmosphere through incomplete combustion of fossil fuel and biomass ^[1]. BC is the third most important anthropogenic radiative forcing (RF) factor next to CO₂ and CH₄ ^[2]. However, the RF of BC still has great uncertainty. Bond et al. estimated BC RF of + 1.1 W m^–2 with an uncertainty range from + 0.9 to + 2.1 W m^{–2 [2]}. Szopa et al. estimated BC RF of + 0.1 W m^–2 with an uncertainty range from –0.2 to + 0.4 W m^–2, which even had negative value ^[3].

A lot of factors lead to this large uncertainty of BC radiative effects, such as the accuracy of the measurement of BC light absorption and the variation of BC in the vertical direction ^{[2, 4]}. An accurate measurement of BC light absorption coefficient (σ_ab) requires technical improvement of instrumentation. Many methods have been proposed to measure BC σ_ab based on various properties of BC. Photo-acoustic measurement, such as photoacoustic spectrometer (PAS) ^[5], measures σ_ab according to acoustic waves generated by the heated aerosol particles. Filter-based measurement, such as aethalometer ^[6], determines σ_ab according to light attenuation through the aerosol-laden filter. Thermal/ optical measurement, such as organic carbon/elemental carbon (OC/EC) aerosol analyzer ^[7], quantifies the mass concentration of OC (EC) based on its chemical behaviors in the absence (presence) of O₂ when it is heated. Light incandescent measurement, take single particle soot photometer (SP2) ^[8] as an example, quantifies the mass concentration of BC (m_BC) according to the incandescent light emitted by BC. Among various methods, the filter-based technique is widely used because filter-based instrument, such as aethalometer model 33 (AE33, Magee, USA), is relatively inexpensive, convenient and easy to maintain ^{[9, 10]}.

Several attempts have been made for vertical measurement of BC. For example, Schwarz et al. measured m_BC with SP2 on aircraft over the tropical atmosphere ^[11], and Safai et al. made onboard BC measurement with aethalometer over South India ^[12]. The excessively high cost of aircraft and corresponding human resources, rigorous meteorological conditions for flight and high temporal resolution required for instruments due to the fast speed (~100 m s^–1) of aircraft greatly limit the wide spread of aircraft measurement. Some alternative schemes have been proposed to avoid the abovementioned disadvantages. Ran et al. measured the vertical BC profile with a micro-aethalometer attached to a tethered balloon during its ascent and descent in the boundary layer over an observation site of North China Plain (NCP) ^[13]. Sun et al. deployed 5 micro-aethalometers on different altitudes of a 356-m mast to measure BC vertical profiles ^[4]. Liu et al. put a micro-aethalometer on an unmanned aerial vehicle (UAV) to profile the vertical variation of BC ^[14].

The weight and volume of the instrument are expected to be as light and small as possible for vertical profiling, considering the limited load capacity of the carrier platform. The micro-aethalometer is one potential solution for vertical BC profiling. For instance, the weight and size of the micro-aethalometer MA200 are 400 g and 13.7 × 8.5 × 3.6 cm³, respectively, much smaller and lighter than the widely used AE33. Another advantage MA200 owns is that it measures absorption over 5 wavelengths, namely 880, 625, 528, 470 and 375 nm. In contrast, another widely used micro-aethalometer, AE51, only measures absorption at 880 nm. The multi-wavelength measurement makes it possible for source apportionment ^[15]. These advantages make MA200 have great potential for vertical measurement of BC. However, the quality of σ_ab measured by MA200 is not well studied.

In this work, we evaluated MA200 by comparing the measured σ_ab from MA200 with that from the well-developed AE33 ^[16]. A detailed description of the experiments and methods is presented in section 2. The results and discussions are shown in section 3. Section 4 comes to a conclusion.

All the experiments are conducted in a measurement container on the rooftop of a building at the School of Physics, Peking University, Beijing, China (39°59'N, 116°18'E). It lies in the northwest of Beijing and is surrounded by two busy streets, namely Chengfu Road and Zhongguancun North Street. Thus, it is a typical urban site. The temperature in the container is well controlled at around 25℃ by an air-conditioner. The detailed information of the measurement site can be found in Zhao et al. ^[17].

Before sampling, aerosol particles are first drawn through a PM_{1 0} impactor, where particles with aerodynamic diameters larger than 10 μm are removed. Then, the particles pass through a Nafion dryer, where the relative humidity is dried to below 40 %.

Two experiments are conducted to evaluate MA200. For the first experiment, particle-free airflow is drawn into the MA200 at 150 mL min^–1 to measure the background noise of the MA200. The corresponding σ_ab was measured at a time resolution of 1 min and 1 s on two different days.

For the second experiment, the ambient aerosol σ_ab is measured simultaneously by MA200 and AE33. The comparison is also made at two sampling intervals, namely 1 min and 1 s, the former from 2018.11.1 to 2018.11.5 for about 4 d, the latter from 2020.6.15 to 2020.6.17 for about 2 d. For 1-min sampling rate, the flow rate of AE33 is set at 3 L min^–1. For 1-s sampling rate, in order to reduce the noise of AE33, the flow rate of AE33 is increased to 4 L min^–1. As the maximum flow rate of MA200 is 150 mL min^–1, we do not regulate the flow rate of MA200 to reduce noise.

For filter-based instruments, aerosol particles are collected on a filter by suction of an internal pump at flow rate F₁ from the ambient environment. A light beam at a specific wavelength transmits through the aerosol-laden part of the filter, and the transmitted light intensity I₁ is recorded by an optical detector underneath the filter. Another light beam at the same wavelength transmits through the aerosol-free part of the filter, and the transmitted light intensity I₀ is recorded simultaneously as reference light intensity.

The light attenuation (ATN) is defined as

Because aerosol particles are continuously collected on the filter by suction of the internal pump, the aerosol loading on the filter increases with time, and less light transmits through the aerosol-laden part of the filter, leading to a decrease of I₁ and an increase of ATN with time. ATN serves as a proxy for aerosol loading. The change of ATN with time represents aerosol accumulation on the filter. When aerosol loading is large enough, the nonlinear effect of aerosol loading is not negligible anymore; at this point, the filter has to be changed to a clean one. This "advance of filter" is technically achieved by predefining a threshold ATN, which is 120 for AE33 and 100 for MA200 by default. When ATN reaches threshold ATN, the advance of the filter is triggered, and the subsequent aerosol particles are collected on a new particle-free filter.

The attenuation coefficient is defined as the time derivative of ATN:

where A is the area of the aerosol-laden part of the filter, whose shape is usually a round spot; Δt is the sampling interval of ATN, whose value is 1 s or 1 min in this work; ΔATN is the change of ATN during Δt.

Two kinds of correction need to be performed to convert σ_ATN into absorption coefficient σ_ab. The first is loading effect correction. The loading effect can be mathematically described as the nonlinear relationship between aerosol loading and ATN, namely for the same increase in aerosol loading, the corresponding increase in ATN is smaller when aerosol loading is heavy than that when aerosol loading is light. Both MA200 and AE33 use dual spot correlation to reduce the loading effect ^[18]. The basic idea of dual spot correction is that besides the original spot, which measures ATN₁ at flow rate F₁, another spot is added for measuring ATN₂ simultaneously at a lower flow rate F₂ (F₁ > F₂). ATN₁ (ATN₂) measured at F₁ (F₂) represents heavy (light) aerosol loading, and σ_{ATN, 1} (σ_{ATN, 2}) can be calculated from ATN₁ (ATN₂) representing attenuation coefficient measured at heavy (light) aerosol loading. The attenuation coefficient σ_ATN at zero loading, namely without the influence of aerosol loading, is derived based on σ_{ATN, 1} and σ_{ATN, 2}.

For a heavily polluted environment, dual spot correction may not well correct the loading effect. In this work, another correction scheme developed by Virkkula and Makela is adopted for further loading effect correction after the dual spot correction ^[19]. It is a linear correction with respect to ATN essentially and formulated as

where σ_{ATN, corrected} and σ_{ATN, uncorrected} represent corrected and uncorrected attenuation coefficient, respectively; k is the correction factor determined by matching the first attenuation coefficient measured at the spot i+1 and the last attenuation coefficient measured at the previous spot i:

The second correction for σ_ab is called a scattering correction. The decrease of light intensity from I₀ to I₁ comes from not only the absorption of aerosol particles lying on the filter matrices but also the scattering of the aerosol particles and filter matrices. σ_ATN is larger than σ_ab because σ_ATN takes both absorption and scattering into account. Zhao et al. compare σ_ATN measured by AE33 with σ_ab measured by a three-wavelength photoacoustic soot spectrometer ^[16]. The results indicate that σ_ATN is 2.9 times larger than σ_ab. In this work, a scattering correction factor of 2.9 is used to convert loading-effect-corrected σ_ATN to σ_ab for AE33:

For this study, σ_ab measured by AE33 is considered as the true value of the absorption coefficient. Two scattering correction schemes for MA200 imitating equation (5) are compared, namely

C_MA in equation (6) as well as C_MA and σ₀ in equation (7) are determined by regressing σ_ATN (from MA200) to σ_ab (from AE33) based on least-square fitting. Mean ± standard deviation (std) of relative difference (RD) between fitted σ_ab and true σ_ab is compared between scheme (6) and scheme (7) to select a better scheme for converting σ_ATN to σ_ab.

For AE33, the measured σ_ab corresponds to the wavelengths at 370, 470, 520, 590, 660, 880 and 950 nm, respectively. For MA200, the wavelengths are 375, 470, 528, 625 and 880 nm, respectively. The same wavelengths the two instruments both measured are only 470 and 880 nm. Interpolation is needed for better comparison at other wavelengths. For 1-min sampled data, interpolation is based on the definition of absorption Angstrom exponent (AAE) ^[13]. For example, σ_ab at 370 and 470nm (σ_{ab, 370} and σ_{ab, 470}) measured by AE33 is used to calculate AAE_{370, 470}:

Then σ_ab at 375 nm (σ_{ab, 375}) is calculated to compare with σ_{ab, 375} measured by MA200:

For 1-s sampled data, there are too many negative values of σ_ab for both AE33 and MA200. It is impossible to calculate AAE for negative values. Thus, linear interpolation is performed for 1-s sampled data. For example, σ_{ab, 370} and σ_{ab, 470} measured by AE33 are used to calculate slope k_{370, 470}:

Then σ_{ab, 375} is calculated to compare with σ_{ab, 375} by MA200:

When the noise inside the data time series is not negligible, especially for 1-s sampled data, denoising is highly necessary before further usage. In this work, weighted running averaging is adopted to reduce noise. Take 5-point weighted running averaging as an example, as shown in Fig. 1, the weighting scheme is that the weights of edge points, namely points i-2 and i+2, are 1. When points come closer to the center point, the weights increase linearly. The weights of points i-1 and i+1 are 2. The weight of the center point is 3 and the highest. The idea of unequal weights is out of consideration that if the weights of all points in the time window are equal, data points used in the time window are only the first and the last point essentially based on equation (2). If the weights of the points inside the time window are unequal, such as in the fashion defined above, all data points in the time window can be utilized to reduce the noise of the center point.

Figure 1.  Schematic of 5-point weighted running averaging.

A HEPA filter is added to the inlet of MA200 to measure the background noise of MA200. The data sampling interval is set at 1 min and 1 s, respectively. The data are sampled for about 1 d for both sampling intervals. The flow rate F is set as 150 mL min^–1 for both sampling intervals. During sampling, the stability of the flow rate is first checked. Due to the dual spot setup inside MA200, F is split into two parts to provide aerosol flows for two spots, hereinafter denoted as F₁ and F₂ for spot 1 and spot 2, respectively. For 1-min sampling interval, the mean ± std of F, F₁ and F₂ is 150.0 ± 0.2, 104.8 ± 0.2 and 45.3 ± 0.2 mL min^–1. For 1-s sampling interval, the mean ± std of F, F₁ and F₂ is 150.0 ± 0.3, 104.5 ± 0.2 and 45.5 ± 0.2 mL min^–1. For both sampling intervals, the std of the flow is smaller than 1 % of the mean flow for F, F₁ and F₂, indicating that the flow rate inside MA200 is stable. The relationship F = F₁ + F₂ is valid considering the std of the flow, indicating that the internal flow circuit is well air-tight and leakage of the flow is negligible.

The background noise of MA200 is characterized by the mean and std of σ_ATN after dual spot correction in this work. For 1-min sampling interval, the background noise is 2 ± 4, 1 ± 5, 1 ± 5, 1 ± 4, 1 ± 3 Mm^-1 at 375, 470, 528, 625 and 880 nm, respectively. For 1-s sampling interval, the background noise is 3 ± 475, 1 ± 267, 1 ± 225, 0 ± 213, 2 ± 187 Mm^–1 at 375, 470, 528, 625 and 880 nm, respectively. For both sampling intervals, although the mean is not zero, it is acceptable considering the corresponding std. The std of 1-s sampled data is about 70 times larger than the std of 1-min sampled data. It is highly recommended to reduce the noise before usage, especially for 1-s sampled data. For 1-s sampling interval, the std decreases with increasing wavelength, where the std at 375 nm is about 2.5 times larger than the std at 880 nm.

To evaluate the performance of the dual spot loading effect correction of MA200, a typical pollution episode from 2018.10.30 to 2018.11.5 is investigated. In order to reduce the influence of noise and let error from the loading effect dominate the time series, the sampling interval during this period is set as 1 min. Because lower wavelength is more sensitive to loading effect, the time series of σ_{ATN, 1}, σ_{ATN, 2} (both σ_{ATN, 1} and σ_{ATN, 2} are calculated from Eq. (2)) and σ_ATN (calculated from dual spot correction developed by Drinovec et al. with σ_{ATN, 1} and σ_{ATN, 2} as input) at the shortest wavelength of MA200 ^[18], namely 375 nm, is presented deliberately, as shown in Fig. 2(a), (b) and (c). It can be seen that the discontinuity caused by the advance of the filter is obvious. For example, at about 12:00 2018.11.3, the advance of the filter is triggered because ATN reaches its threshold of 100. At the start of the new filter, σ_{ATN, 1} exceeds 300 Mm^–1, whereas at the last of the previous spot, σ_{ATN, 1} is lower than 100 Mm^–1. It is clearly caused by the fact that the response of ATN to aerosol loading is significantly weakened when aerosol loading is too heavy. Thus, a lower threshold ATN is highly recommended to reduce the influence of the loading effect, especially for a polluted environment. For σ_ATN after dual spot correction, it can be seen that the discontinuity after dual spot correction is greatly reduced than that without dual spot correction. For example, also at about 12:00 2018.11.3, the discontinuity is about 200 Mm^–1 for σ_{ATN, 1} before dual spot correction. After dual spot correction, the discontinuity at this time is reduced to about 100 Mm^–1. The entire evolution of the pollution episode can be identified after dual spot correction, namely a gradual accumulation since about 2018.10.30 and an abrupt ending at about 6:00 2018.11.4.

Figure 2.  Time series of (a) σ_{ATN, 1}, (b) σ_{ATN, 2}, and (c) σ_ATN at 375 nm of MA200 from 2018.10.30 to 2018.11.5 at the sampling interval of 1 min.

The discontinuity is not completely eliminated, as shown in Fig. 2(c). The loading effect correction scheme developed by Virkkula et al. as described in section 2.2.2 is adopted for further loading effect correction, and the results are presented in Fig. 3 ^[19]. The mean relative difference (after and before correction) is 24.5% at 375 nm and –7.2% at 880 nm, respectively, which is the largest and smallest among other wavelengths, exhibiting an inverse proportional wavelength dependence. When the ambient environment is relatively clean (polluted), such as < 100 Mm^–1 (> 100 Mm^–1) at 375 nm, the correction is not obvious (obvious). Therefore, besides dual spot correction, further loading effect correction is recommended, especially for shorter wavelengths in a polluted environment.

Figure 3.  Correcting σ_ATN using a method developed by Virkkula et al. ^[19]. (a), (b), (c), (d), and (e) represent attenuation coefficient measured at 375, 470, 528, 625, and 880 nm, respectively. Red dots and blue dots represent σ_ATN before and after correcting using a method developed by Virkkula et al. ^[19].

σ_ATN after two times of loading effect correction in section 3.2 is used for scattering correction. Another reason for choosing data in section 3.2 for scattering correction is that data in section 3.2 record a complete pollution episode, during which the absorption coefficient varies over a wide range from < 5 Mm^–1 to > 60 Mm^–1 at 880 nm. Whereas for a typical clean episode, the absorption coefficient is usually < 20 Mm^–1 at 880 nm. A wider data range is more beneficial for a more comprehensive correction since correction based on a limited data range is insufficient and may lead to discrepancies for data out of that limited range.

As mentioned in section 2.2.3, two schemes, namely without and with a compensation to σ_{ATN, MA200}, are compared to select a better presentation of scattering correction for MA200. For equation (5), as shown in Fig. 4, unlike AE33 whose scattering correction coefficient does not exhibit significant wavelength dependence ^[16], the scattering correction factor of MA200, namely C_MA, is inversely proportional to wavelength, namely larger C_MA for shorter wavelength, varying from 2.37 at 880 nm to 3.07 at 375 nm. The wavelength dependence of C_MA possibly results from difference in filter material. The filter of MA200 is made from polytetrafluoroethylene (PTFE) (https://aethlabs.com/microaeth/accessories). The filter of AE33 is made from glass fibers coated with PTFE and polyethylene terephthalate (PET) (https://mageesci.com/mproducts/model-ae33-aethalometer/). The mean RD indicates that MA200 overall underestimates σ_ab from 3.6 % to 18.8 % if equation (5) is adopted for scattering correction. The large std of RD is also not negligible, especially at 880 nm whose std is as large as 32.2 %. Whether the absorption coefficient at 880 nm is well corrected or not is of great importance because "equivalent BC mass concentration" (EBC) is derived from the absorption coefficient at 880 nm ^[20]. The reason for choosing 880 nm to derive EBC is that aerosol absorption is mainly from BC at this wavelength ^[21]. At shorter wavelengths, absorption of organic carbon is not negligible any more, leading to difficulty in extracting BC absorption from total aerosol absorption.

Figure 4.  Comparison of σ_ab between AE33 and MA200. σ_ab is derived from σ_ATN based on scheme (5). (a), (b), (c), (d), and (e) correspond to σ_ab at 375, 470, 528, 625, and 880 nm, respectively. The black solid line in each subplot is the 1:1 line. The scattering correction factor of MA200, as well as the mean ± std of RD between σ_ab measured by MA200 and AE33 at different wavelengths, are also presented in the corresponding subplot.

Figure 5 presents a comparison of σ_ab between AE33 and MA200. σ_ab is derived from σ_ATN based on equation (6). The mean RD shows that after correction based on equation (6), MA200 slightly overestimates σ_ab. The mean RD varies from 1.2 % at 375 nm to 3.5 % at 880 nm, which is significantly smaller than that of equation (5) and closer to zero. Although the std of mean RD is still large, it is overall smaller than that of equation (5). R² at each wavelength is 0.97, which means that MA200 is able to depict the general variation of BC. The large std of RD means that discrepancy still exists between MA200 and AE33 with respect to variation of BC at a smaller time scale. The discrepancy can possibly be reduced by increasing the flow rate in the hardware upgrade of future MA200 or attaching an external vacuum pump to MA200 ^[22]. The maximum flow rate of MA200 is 150 mL min^–1, which is 20 times smaller than that of AE33 used in this work. A small sampling flow rate may not well capture BC variation inside the air mass and lead to artefacts due to the low signal-to-noise ratio. The scattering correction factor also exhibits wavelength dependency ranging from 2.65 at 880 nm to 3.21 at 375 nm. For σ₀, it varies from – 6.4 Mm^–1 to – 15.3 Mm^–1 without wavelength dependence and is not negligible, especially for a clean environment.

Figure 5.  Same as Fig. 4, except that scheme (6) is adopted for scattering correction to MA200.

Equation (6) performs better than equation (5) based on the above discussion. Mathematically, equation (6) is essentially a linear fitting equation with two fitting parameters. Equation (5) is also a linear fitting equation, except that the zero-order term is omitted, so equation (5) has only one fitting parameter. The fact that Equation (6) performs better is within expectation since one more fitting parameter usually leads to a better regression result. Physically, the nonnegligible zero-order term may indicate that a systematic difference exists between MA200 and AE33. The systematic difference may be an integral result of differences in filter material, flow rate, and even optical detectors.

The above discussion may be biased since the data used for the above analysis are only based on one episode measured at one urban site. Data from other episodes or sites characterized by different sources of BC and polluted levels may result in different conclusions. By contrast, the idea expressed in this work should be reliable and unbiased. That is, before MA200 is used for measurement, comparison with other reliable instruments, such as AE33, is highly recommended. The correction scheme based on the comparison should be applied to the subsequent measurement rather than directly using uncorrected MA200 data.

The difference between this section and section 3.1.2 is that MA200 in this section is directly used to measure ambient aerosol particles rather than connected to a filter to measure particle-free airflow. AE33 is also used to measure ambient aerosol particles simultaneously. Δσ_ab, the difference between σ_ab measured by MA200 and AE33, is used to represent the noise of absorption coefficient measured by MA200 from true value. Δσ_ab is investigated at 1-min and 1-s sampling rate. Weighted running averaging is used to reduce the noise of MA200 at different time windows. Mean and std of Δσ_ab, as well as R² between MA200 and AE33, are used to evaluate the effectiveness of weighted running averaging on reducing the noise of MA200, as shown in Fig. 6. It should be noted that σ_ab measured by AE33 is not weighted running averaged along with MA200 at different time windows since σ_ab measured by AE33 is considered as true value and σ_ab measured by MA200 averaged at different time window is compared to the true value for evaluation. σ_ab measured by MA200 is derived from σ_ATN based on section 3.2 and 3.3, except that for 1-s sampling rate, the loading effect is only corrected by dual spot correction, not further corrected by the method developed by Virkkula et al. because discontinuity is not distinguishable in presence of excessively large noise ^[19].

Figure 6.  Evaluation of weighted running averaging on reducing the noise of σ_ab, namely Δσ_ab. (a), (b) and (c) are based on a 1-min sampling interval; (d), (e) and (f) are based on a 1-s sampling interval. (a) and (d), (b) and (e), as well as (c) and (f), are the mean of Δσ_ab, std of Δσ_ab and R² between MA200 and AE33 as a function of the average time window.

For 1-min sampling rate, the mean Δσ_ab is around zero and does not vary significantly with the time window. Std of Δσ_ab decreases (dose not decrease) significantly with increasing time window when the time window is < (> ) about 5 min. When the time window continues to increase larger than 30 min, the std of Δσ_ab starts to increase slightly. R² between MA200 and AE33 also exhibits a similar behavior. R² increases rapidly with the increasing average window when the average window is < (> ) about 5 min. When the time window continues to increase to larger than 30 min, R² starts to decrease slightly. Therefore, 1-min sampled data is recommended to run average at a 5-min time window since a larger average window does not reduce noise obviously and consistency between MA200 and true value, on the contrary, is worsened. After 5-min averaging, the noise can be reduced by about 20 %, and R² is increased slightly by about 0.1.

For 1-s sampling rate, the mean Δσ_ab is about 3 Mm^–1, which is acceptable given that the std of Δσ_ab without averaging is as large as about 140 Mm^–1. The mean Δσ_ab also does not vary significantly with the time window. The std of Δσ_ab decreases exponentially with increasing time window. The std of Δσ_ab decreases from about 140 Mm^–1 without averaging to less than 6 Mm^–1 when the time window is larger than 30 min. If σ_ab is not time-averaged, there is no correlation between MA200 and AE33, which means that 1-s sampled σ_ab measured by MA200 is highly noisy and cannot be used if noise is not reduced. Even when σ_ab measured by MA200 is averaged to 20 s, R² is overall less than 0.4. When the average window is < (> ) 30 s, R² increases (does not increase) significantly with an increasing time window. R² increases to larger than 0.7 at 375 nm and to 0.5 at 880 nm when the average window is larger than 50 s. Therefore, for 1-s sampled data, σ_ab is recommended to running averaged to about 30 s considering that std of Δσ_ab and R² does not decrease and increase significantly if the time window is wider.

In order to minimize uncertainty in the radiative forcing (RF) of black carbon (BC), conducting comprehensive BC measurements, particularly in the vertical direction, is crucial. The micro-aethalometer MA200, known for its lightweight and compact design, holds immense potential for aerial measurements. This study aims to evaluate the performance of MA200, providing valuable insights and references for future users of this instrument.

The flow rate of MA200 is meticulously controlled, demonstrating exceptional stability with a noise of less than 1% when maintained at 150 mL min^-1. The built-in dual spot loading effect correction of MA200 has undergone evaluation. The findings indicate that dual spot correction effectively mitigates the discontinuity caused by tape advancement. However, it is worth noting that in polluted ambient environments, the discontinuity remains present and cannot be disregarded. Consequently, it is recommended to implement further loading effect correction techniques. In this study, the method developed by Virkkula et al. is adopted for additional correction purposes ^[19].

The scattering correction of MA200 is accomplished by comparing the attenuation coefficient (σ_ATN) after loading effect correction with the absorption coefficient (σ_ab) measured by well-studied and widely used AE33. The AE33 is considered as the reference for true values. Two scattering correction schemes are compared, and the results demonstrate that the scheme formulated as σ_ab = (σ_ATN – σ₀)/C_MA is particularly suitable for typical urban environments. The σ₀ value ranges from –15.3 to –6.4 Mm^–1 and it should not be overlooked, especially during clean episodes. A nonzero σ₀ value may rise from factors such as the low flow rate of MA200, variations in filter material and disparities in optical detectors compared to AE33. Furthermore, the wavelength dependence of C_MA was observed, ranging from 2.65 at 880 nm to 3.21 at 375 nm. The wavelength dependence is likely a consequence of the characteristics of the filter material used.

To minimize noise in the measured σ_ab values obtained by MA200, a weighted running averaging technique is employed. The σ_ab values concurrently measured by AE33 are considered as the true representation of σ_ab. For 1-min sampled data, it is recommended to apply a 5-min running averaging approach to effectively reduce noise without sacrificing the correlation with AE33. For data sampled at 1 s, a running average with a 30-s window is recommended. Using a window size lower than this range results in a poor correlation with AE33 and excessively large noise in the measured σ_ab values obtained by MA200. Conversely, using a window size greater than this range does not significantly enhance the reduction in σ_ab noise or improve the correlation with AE33.