HTML
-
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 PM1 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 σab was measured at a time resolution of 1 min and 1 s on two different days.
For the second experiment, the ambient aerosol σab 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
$$ \mathrm{ATN}=-100 \cdot \ln \left(\frac{I_0}{I_1}\right) $$ (1) 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:
$$ \sigma_{\mathrm{ATN}}=\frac{A \cdot \Delta \mathrm{ATN} / 100}{F_1 \cdot \Delta t} $$ (2) 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 σATN into absorption coefficient σab. 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 σATN, 1 (σATN, 2) can be calculated from ATN1 (ATN2) representing attenuation coefficient measured at heavy (light) aerosol loading. The attenuation coefficient σATN at zero loading, namely without the influence of aerosol loading, is derived based on σATN, 1 and σATN, 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
$$ \sigma_{\mathrm{ATN}, \text { corrected }}=\sigma_{\mathrm{ATN}, \text { uncorrected }}(1+k \cdot \mathrm{ATN}) $$ (3) where σATN, corrected and σATN, 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:
$$ \sigma_{\mathrm{ATN}}\left(t_{i, \text { last }}\right)=\sigma_{\mathrm{ATN}}\left(t_{i+1, \text { first }}\right) $$ (4) -
The second correction for σab 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. σATN is larger than σab because σATN takes both absorption and scattering into account. Zhao et al. compare σATN measured by AE33 with σab measured by a three-wavelength photoacoustic soot spectrometer [16]. The results indicate that σATN is 2.9 times larger than σab. In this work, a scattering correction factor of 2.9 is used to convert loading-effect-corrected σATN to σab for AE33:
$$ \sigma_{\mathrm{ab}}=\frac{\sigma_{\mathrm{ATN}}}{C_{\mathrm{AE}}}, C_{\mathrm{AE}}=2.9 $$ (5) For this study, σab 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
$$ \sigma_{\mathrm{ab}}=\frac{\sigma_{\mathrm{ATN}}}{C_{\mathrm{MA}}} $$ (6) $$ \sigma_{\mathrm{ab}}=\frac{\sigma_{\mathrm{ATN}}-\sigma_0}{C_{\mathrm{MA}}} $$ (7) CMA in equation (6) as well as CMA and σ0 in equation (7) are determined by regressing σATN (from MA200) to σab (from AE33) based on least-square fitting. Mean ± standard deviation (std) of relative difference (RD) between fitted σab and true σab is compared between scheme (6) and scheme (7) to select a better scheme for converting σATN to σab.
-
For AE33, the measured σab 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, σab at 370 and 470nm (σab, 370 and σab, 470) measured by AE33 is used to calculate AAE370, 470:
$$ \mathrm{AAE}_{370, 470}=-\frac{\ln \left(\sigma_{\mathrm{ab}, 370}\right)-\ln \left(\sigma_{\mathrm{ab}, 470}\right)}{\ln (370)-\ln (470)}. $$ Then σab at 375 nm (σab, 375) is calculated to compare with σab, 375 measured by MA200:
$$ \sigma_{\mathrm{ab}, 375}=\sigma_{\mathrm{ab}, 370, \mathrm{AE} 33} \cdot\left(\frac{375}{370}\right)^{-\mathrm{AAE}_{370, 470}} . $$ For 1-s sampled data, there are too many negative values of σab 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, σab, 370 and σab, 470 measured by AE33 are used to calculate slope k370, 470:
$$ k_{370, 470}=\frac{\sigma_{\mathrm{ab}, 370}-\sigma_{\mathrm{ab}, 470}}{370-470} . $$ Then σab, 375 is calculated to compare with σab, 375 by MA200:
$$ \sigma_{\mathrm{ab}, 375}=\sigma_{\mathrm{ab}, 370}+k_{370, 470}(375-370). $$ -
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.