A new multispectral photometer for monitoring aerosol microphysical, optical, and radiative propertiesEvaluation of aerosol microphysical, optical and radiative properties measured from a multiwavelength photometer

. AAn evaluation of aerosol microphysical, optical and radiative properties measured from a multiwavelength photometernew multispectral photometer, named CW193, was proposed in this study for monitoring aerosol microphysical, 20 optical, and radiative properties. The instrument has a highly integrated design, smart control performance, and is composed of three parts (an optical head, a robotic drive platform, and a stents system). Because of its low maintenance requirements, this instrument is appropriate for the deployment in remote and unpopulated regions. Based on the synchronous measurements, the CW193 products was validated using reference data from the AERONET CE318 photometer. The results show that the raw digital counts from CW193 agree well the counts from AERONET ( R 2 > 0.97), large as 10.0%, whereas they are 2.0−6.0% for the visible bands before 10:00 BJT and after 14:00 BJT. These results reveal that the digital counts of CW193 measurements fluctuate considerably during the morning and the afternoon, owing to the weak solar radiation and rapid and extensive changes of the solar altitudinal angle.

For these reasons, aerosol detection from ground-based observations is regarded as the most direct, accurate, and effective 70 technique to measure and study the columnar microphysical, optical, and radiative properties of atmospheric aerosols, and there are extensive ground-based monitoring networks across the world dedicated to aerosol detection, such as the Precision Filter Radiometer (PFR) network of The Global Atmosphere Watch program of the World Meteorological Organization (WMO-GAW; (Cuevas et al., 2019)), the China Aerosol Remote Sensing NETwork (CARSNET; (Che et al., 2015, the Aerosol Robotic Network (AERONET; (Holben et al., 1998), the PHOtomé trie pour le Traitement Opé rationnel de 75 Normalisation Satellitaire (PHOTONS; (Goloub et al., 2008), and the SKYrad Network (SKYNET; (Nakajima et al., 2020), all consisting of precise instruments with the 0.02 AOD accuracy suggested by the WMO (Che et al., 2009). Most of these observation networks are equipped with the CE318, an automatical multiband Sun photometer (Cimel Electronique, France), as the master and/or observation instrument, to provide long-term data on the aerosol microphysical, optical, and radiative characteristics on the global scale. These networks have an important role in determining the climatic and environmental 80 effects of aerosols, and the measurement results have been strictly verified under a wide range of conditions, such as in polar and plateau regions Eck et al., 1999;Xing et al., 2021a;Zhuang et al., 2017). However, on the global scale, detection sites in specific areas, such as desert regions, continental plateaus, and sea islands, are still insufficient. As these regions are important pathways for the long-range transportation of aerosols, this results in an unsatisfactory description of global aerosol cycles. There are three main reasons for the lack of detection sites in these regions. First, although these Sun 85 photometers are automatic, wired communication (for example, serial communication via  between the instruments and a personal computer is still necessary for most CE318-N photometers to conduct the data storage, which is difficult to realize in some remote regions. Second, the non-integrated instrument components, such as the control unit, external battery, protection box, and stents platform, not only cause most of the operational problems but also make the deployment and maintenance difficult for staff with inadequate training. Lastly, the relatively high cost of these multiband photometers and 90 their accessories constrains the expansion of aerosol monitoring stations for many developing countries. As reported by WMO-GAW's report No. 162, 207, 227 and228 (2004, 2012;, the multiwavelength aerosol optical depth (AOD) is still recommended as the long-term measurement variables at the implementation plan from 2016 to 2023. Particularly via ground based AOD attenuation observation, it is regarded as the highly accurate monitoring method to provide indispensable data for satellites validation and global modelling. According to this guideline, an absolute limit to the estimated uncertainty of 0.02 95 optical depths for acceptable data and <0.01 as a goal to be achieved in the near future. Additionally, the international coordination of AOD networks is inadequate and could be improved by a federated network under the WMO-GAW umbrella, and networks should become traceable and maintainable via intercomparisons and calibrations. These guidelines highlighted that data assessment is as important as the field observation. However, in China, due to the vast territory and various landform, there are still many observation gaps in aerosol optical properties monitoring. Furthermore, the complicated underlaying 100 surface and emission condition result in the distinct temporal and spatial variations of aerosol. Therefore, referring to WMO-GAW's criterion, conducting field observation and data evaluation is of great importance to reduce the uncertainties of aerosol optical properties, which will be a great assistance to combat climate change.
So far, except for CE318 and POM-02 (Nakajima et al., 2020), there are many photometers have realized the function of AOD 105 measurement in China, such as DTF-5 and PSR-2 (Li et al., 2012;Huang et al., 2019). However, we suggest that all the instruments and their products should meet the WMO-GAW's criterion and keep consistency with AERONET, providing comprehensive, comparable aerosol optical products. Here we present a new highly integrated multiwavelengthmultispectral photometer named CW193 (CW means Chinese device for World) for monitoring aerosol microphysical, optical, and radiative properties. It has a user-friendly instruction system, and most of the components are assembled in a robotic drive platform, 110 which makes the whole system efficient, secure, low cost and highly integrated. By using direct Sun and diffuse-sky radiation measurements, the CW193 not only provides the columnar instantaneous AOD but also gives detailed information on the aerosol characteristics, including, but not limited, to the volume size distribution (VSD), the single scattering albedo (SSA), the asymmetry factor (ASY), and the aerosol direct radiative forcing (ADRF), which are the key input parameters for numerical models (Miao et al., 2020;Stier et al., 2005;Wang et al., 2013). These features make the CW193 a particularly suitable 115 multiwavelengthmultispectral photometer for monitoring aerosol microphysical, optical, and radiative properties, especially in remote regions without computer availability to fill in the observation gaps. It is also suitable forwhich is contribute to verifying the satellite and modelling products in these tough environments.
For this study, synchronous measurements were conducted between CW193 and CE318s from AERONET and CARSNET at 120 CAMS (Chinese Academy of Meteorological Sciences), and the products of CW193 were evaluated and compared in detail with the reference of AEROENT, aiming at keeping consistency with it. Following this introduction, the observation site and ancillary information for this study are introduced in section 2. In section 3, a description of the new CW193 multiwavelengthmultispectral photometer is provided. Section 4 provides an evaluation and comparison of the aerosol microphysical, optical, and radiative properties from CW193. Finally, the main conclusions are presented in section 5. 125 5 the urban atmospheric conditions in China and a good test environment for CW193. The CAMS site is part of the AERONET observation network (named "Beijing-CAMS") and has provided the AOD and other inversion products since its establishment in 2012. In addition, Beijing-CAMS is a transfer Sun calibration site for CARSNET, with the master instruments sent to the 135 Izaña Observatory (Izaña,Canary Islands,Spain;28.3° N,16.5° W, 2373 m a.s.l.) for annual calibration.

CE318 Sun photometer and its observation network
In this comparative observation campaign, the AOD data and their correlative aerosol inversions provided by AERONET and CARSNET were used to validate the results retrieved from the CW193 observations. AERONET is the biggest federated instrument network in the world, providing open-access data for aerosol microphysical, optical, and radiative properties (https://aeronet.gsfc.nasa.gov/). CARSNET is the largest ground-based aerosol remote-sensing network in China, with more 145 than 80 sites in China, of which 51 are currently operational. CARSNET uses the same similar algorithm as AERONET  and has a rigorous calibration process; therefore, the aerosol retrievals of CARSNET are of great importance for determining the temporal and spatial variations of aerosol optical properties in China Yu et al., 2015;Zhao et al., 2021b;Zheng et al., 2021).

150
The master instrument used in AERONET and CARSNET is the CE318 Sun photometer, which performs direct Sun and diffuse-sky radiation measurements according to set observation times. For the direct Sun measurements, the radiation is measured at 340, 380, 440, 500, 675, 870, 1020, and 1640 nm to calculate an accurate AOD and at 936 nm for water vapor (WV), with uncertainties within ±0.02 and ±0.10, respectively. The diffuse-sky measurements are conducted at 440, 500, 670, 870, 1020 and 1640 nm to retrieve the microphysical and optical properties of aerosols in different routines: the almucantar 155 (ALM) and the principal plane (PPL). The azimuth angle is varied while the zenith angle is kept constant for the ALM, and vice versa for the PPL. In this study, the CE318s and CW193 were set to perform intensive direct Sun observations every 3 minutes (otherwise every 15 minutes) to obtain enough data to evaluate the AOD accuracy.

CW193 multiwavelengthmultispectral photometer
The CW193 is an automatic photometer and designed to obtain AOD and other retrievals (such as microphysical, optical, and 160 radiative properties of aerosols) from Sun radiation and sky radiation monitoring, respectively. The instrument is mainly composed of three parts: an optical head, a robotic drive platform, and a stents system (as shown in the left part of Figure 2). These three parts can be easily connected together only by a few screws. Except for its highly integrated design, the cross weight of CW193 is about 12 kg, and this make it easier to transport. And Specifically, we presented the comparison of technical specifications between CE318-TN and CW193 in table 1. 165 The two collimators within a 1.30° full field-of-view are both screwed tightly to the optical head separately, making its disassembly and maintenance more convenient, to avoid the interference of stray light and reduce the measurement error originating from the non-parallel integrated collimators used in CE318. To compare the results with AEROENT, the detector in the optical head is designed to equip nine optical filters with nominal wavelengths centered at 340,380,440,500,675,870,936, 1020, and 1640 nm, which are precisely coated to delay the aging of their optical transmittance. There are sensors inside 180 the optical head for internal humidity and temperature monitoring, and this environmental information is used to conduct the temperature correction of the raw signal, minimizing the temperature dependence of the silicon detectors for 1020 nm and 1640 nm.
The robotic drive platform is the main dynamic system to make the optical head track the direct solar radiation, as well as in 185 the ALM scan routines. To avoid mechanical problems owing to excessive usage of robotic platform, the CW193 is designed to keep tracking the Sun all the time, unless the ALM routines are activated at specific integral local time (9:00, 10:00, 11:00, 12.00…). In addition, all the measurements routines will be suspended when the precipitation is detected by the wet sensor of platform, and then the optical head will turn down to avoid the rain contamination. On the whole, protection degrees of these two frames up to IP65, making its tough enough for the running under the humid or dusty environment. 190 The stents system, directly supported on the base of the robotic drive platform, consists of an adjustable-length tripod with a horizontal adjustment knob at each foot; therefore, it can be quickly deployed and fixed on flat and/or rigid surfaces and has a reliable anti-wind capacity (<25 m s −1 if not fixed on the ground). The instrument is powered by a 220 V alternating current, and is also equipped with a solar panel for remote locations and in case of temporary/moveable observation campaigns. As a 195 result, the design of CW193 is very robust, ensuring long-term steady operation in a wide range of temperature and humidity, between about −30°C and 60°C and between about 0 and 100%, respectively.

200
The main circuit board is in the head of the robotic drive platform, with the integration of operation control, data acquisition, data storage, transmission communication, and status diagnosis. The control unit is designed to conduct observations automatically under the default state, once the geographic information of the observation site is confirmed by the built-in BDS (BeiDou Navigation Satellite System) module. The data unit comprises an internal data logger and a 32 GB memory, considered as life-time storage with a daily data size of ~150 KB. Data transmission to a computer can be realized in two ways: 205 serial communication via RS-232 or the 4G network. The diagnostic module checks the whole system when the instrument is powered on, and the running state can be easily recognized by the indicator light in the optical head.
The system provides a friendly user interface on a computer, which makes the CW193 easy to operate, convenient to maintain, and highly functional. As shown in Figure 3, the functional area and monitoring area are clearly presented in the left and right 210 part of the interface, respectively. It is very convenient to receiving data via 4G network when the serial communication is unavailable in some remote regions, and also in this mode, multiple device control is achievable (device 003, 005 and 006 are on-line and controllable in Figure 3). In data download part, the history data can be easy download by selecting the start and end time via drop-down menu. In the control commands area, all the observation instructions are provided and could be sent to device in the dialog box. In the monitoring area in right half, the plot and its specific data are located in the top and bottom 215 windows, respectively, making it convenient for monitoring the device's status. In summary, we presented the comparison of functional specifications between CE318-TN and CW193 in table 2.

Calibration and data processing
In this work, the direct Sun calibration of CW193 was conducted at the atmospheric composition observation platform of CAMS (one of the calibration centers of CARSNET), using the method of coefficient transfer (inter-comparison) with the reference master instruments of AERONET (Che et al., 2009(Che et al., , 2019cZheng et al., 2021). The sphere calibration was performed at the optical calibration laboratory (CAMS, Beijing) of CARSNET by integrating the sphere. We conducted 50 measurements 230 of the sphere's radiance and found extremely small fluctuations in the CW193 digital counts (<1‰), indicating excellent detection stability and accurate sphere calibration coefficients (Tao et al., 2014).
In this work, wWe calculated the cloud-screened AOD and columnar water vapor of CW193 via the similar algorithm as AERONET. As the algorithm has been used multiple times in many observation campaigns, numerical modelling, and satellite 235 verification for CARSNET, it is suitable and reliable to evaluate the AOD performance of CW193 using this method (Wang et al., 2010;Xia et al., 2021;Yu et al., 2015;Zhao et al., 2021c;Zheng et al., 2021). The algorithm verification is provided in the Supplementary Information to guarantee the accuracy in this campaign (Figures S1 and S2). As for the inversions of VSD and SSA in this campaign, they were retrieved from the observational data from the diffuse-sky measurements of the CW193 at 440, 670, 870, and 1020 nm using the algorithms of Dubovik et al. (2002Dubovik et al. ( , 2006. The ADRF was calculated by the radiative 240 transfer module, which is similar to the inversion of AERONET (Garcí a et al., 2008(Garcí a et al., , 2012. Because the introduction, validation and application of these inversions and their algorithms have been presented in many previous studies based on CARSNENT observation, we did not repeat these again in this paper (Che et al., , 2019cZhao et al., 2018;Zheng et al., 2021). In general, the AODs' uncertainty was 0.01 to 0.02 . The VSD accuracy was 15 % to 25 % between 0.1 µm ≤ r ≤7.0 µm while 25 % to 100 % for other radius (Dubovik et al., 2002). The SSA accuracy was 0.03 when its was 245 calculated under the condition of AOD440 nm >0.50 with a solar zenith angle >50 ° (Dubovik et al., 2002). The bias for measured radiation at the surface was about 9±12 W m −2 , affected by the dominant aerosol type (Garcí a et al., 2008).

Results and discussion
In this work, synchronous measurements with five instruments were conducted at the CAMS observation platform during 1 to 250 11 November, 2020. Specifically, photometers #543 and #746 of the CE318-N mode and photometers #1043 and #1046 of the CE318-T mode are the four master instruments for the "Beijing-CAMS" site in AERONET, the raw data of which are transmitted in real time to AERONET. The AODs and other inversion products from these four instruments can be downloaded from the AERONET website. Furthermore, these four instruments are also the reference instruments of CARSNET, and have an important role in the operational observations and annual calibration of CARSNET. 255

Raw digital counts evaluation
The raw digital counts are the deciding factor for the precision of the calculation and retrieval results, reflecting the running status and stability of the instrument. In Table 3, we show the observed signal with the least squares method, presenting a basic statistical intercomparison at the coincident spectral wavelengths. It is noted that these instruments measure three times within ~30 seconds in one scenario, and we calculated the average values of the digital counts for each observation in this comparison. 260 Furthermore, the results from the AERONET webpage during this campaign were mainly derived from photometers #1043 and #1046 according to the "Instrument Number" in the downloaded files; therefore, we used the corresponding observation signals of these two master instruments to carry out the performance evaluation of CW193. In addition, to avoid the effect of instantaneous atmospheric disturbance, only the values of which the observation time's difference within 20 s compared to the master instruments, were selected and considered as effective data in this work. 265 From Table 3, it can be seen that the digital counts measured by CW193 and the master instruments are highly correlated for these specific bands, with correlation coefficients (R) and coefficients of determination (R 2 ) higher than 0.98 and 0.97, respectively, suggesting high linear consistency rather than possible nonlinearities of CW193 in the selected measurement was generally more consistent with photometer #1046, with all the R 2 values exceeding 0.9988, whereas the R 2 values for photometer #1043 varied from ~0.9792 to 0.9994. In practical terms, the CW193 performs better in the ultraviolet (UV) bands (340 nm and 380 nm) and visible bands (440 nm to 870 nm), with R 2 values larger than 0.9971 and 0.0.9985 with photometers #1043 and #1046, respectively. The R 2 values were relatively low in the infrared bands of 1020 nm and 1640 nm. The minimum R 2 values were ~0.9792 with photometer #1043 at 1020 nm and ~0.9988 with photometer #1046 at 1640 nm, indicating a 275 greater variation in these two bands than in the other bands owing to its temperature sensitivity (Che et al., 2011;Tao et al., 2014). For the WV channel at 936 nm, the R 2 values were ~0.9977 and ~0.9997 for photometers #1043 and #1046, respectively; hence, the CW193 showed good detection ability for columnar WV. To obtain a better description of the stability of the instrument and atmospheric conditions, a triplet value is a more effective parameter, which is defined as (maximum − minimum)/mean × 100%. Thus, we calculated the triplets for each band and present its diurnal variation in Figure 4. It can be clearly seen that the triplets show a typical diurnal distribution in this study, 285 as found in many previous studies (Barreto et al., 2016;Che et al., 2011;Estellé s et al., 2012), characterized by increasing dispersion with increasing airmass. However, cloud contamination is also an important factor affecting the triplets variation.
Using the weather record and the cloud-screening results of AERONET as a reference, we found that the atmospheric conditions on 7 and 11 November were greatly influenced by cloud processes. As a result, the dispersion of the triplets on these two days was larger than that on the other days, with almost all values exceeding 2.0% at all times. The observation 290 conditions on the other days were less affected by cloud, and it can be seen that the values reduced to a relatively low level, with most values <2.0% between 10:00 and 14:00 BJT (Beijing local time) for all cases. The triplets of the UV bands are as large as 10.0%, whereas they are 2.0−6.0% for the visible bands before 10:00 BJT and after 14:00 BJT. These results reveal that the digital counts of CW193 measurements fluctuate considerably during the morning and the afternoon, owing to the weak solar radiation and rapid and extensive changes of the solar altitudinal angle. 295

Nov 11
The daily average triplets were also calculated in this intercomparison ( Figure 5). We found that the daily average triplets for the UV bands showed the largest fluctuation amplitude range, which were ~1.5−3.0% for 340 nm and 1.2−2.5% for 380 nm.
For the visible bands from 440 nm to 870 nm, it can be clearly seen that the variation trend of daily average triplets decreases with increasing wavelength. With the exception of 7 November, which was greatly affected by cloud processes, the daily average triplets in the visible bands were all less than 2.0%. Relatively weak fluctuation amplitudes were observed in the 305 infrared bands from 1020 nm to 1640 nm in all cases, with daily average triplets mostly lower than 1.0%, except on 7 November, and showing less variation with wavelength. The fluctuation for the WV channel at 936 nm was moderate compared with the other bands, and the daily average triplets were slightly higher than those in the infrared bands from 1020 nm to 1640 nm, but much lower than the UV band. In general, the WV had a similar variation range to the 870 nm band, which was ~0.5−2.5%.
As can be seen from Figure 5, the lowest daily fluctuations were found on 3 November, with a variation range of ~1.4−1.8% 310 for the UV bands and ~0.4−0.8% for the other bands. Using the meteorological and environmental records as a reference (no cloud contamination and daily PM2.5 ~11 μm m −3 ; Table 4), these results indicate that the dispersion of diurnal triplets is quite small under clear and stable weather conditions, suggesting the reliable measurement capability of CW193.

AOD evaluation
The AOD performance of the CW193 was tested at the Beijing-CAMS site, using CE318s as the reference, as the instrument has been widely verified under a wide tange of conditions (Che et al., 2015Holben et al., 2001;Xia et al., 2016). In this work, we calculated the cloud-screened AOD of CW193 via the similar algorithm as AERONET. As the algorithm has been 320 used multiple times in many observation campaigns, numerical modeling, and satellite verification for CARSNET, it is suitable and reliable to evaluate the AOD performance of CW193 using this method (Wang et al., 2010;Xia et al., 2021;Yu et al., 2015;Zhao et al., 2021c;Zheng et al., 2021). The algorithm verification is provided in the Supplementary Information to guarantee the accuracy in this campaign (Figures S1 and S2).  Table 4. Figure 6 shows the diurnal variation of cloud-screened AOD (only from daytime observation) for each band from CW193 during this campaign. An obvious decreasing trend in AOD with increasing wavelength can be seen, which is in agreement with many previous studies (Che et al., 2019c;Holben et al., 1998;Liang et al., 2019). Consequently, under weak pollution conditions, the high AOD made the characteristics of the wavelength dependance more apparent. On the most 340 polluted day (Level Ⅲ, PM2.5 ~104 μg m −3 , AOD440 ~1.32−1.47, 11 November), the diurnal AOD was distributed in an orderly pattern and had a similar variation trend at each wavelength, with each curve clearly visible and not intersecting with others.
This distribution was also found under the Level Ⅱ situation on 4 and 10 November. Although AOD440 (~0.20−0.60) was relatively smaller than that at Level Ⅲ, the diurnal AOD curves for each wavelength were more continuous and can be recognized more easily, which is partly attributed to the reduced impact of cloud contamination. In terms of AOD evaluation, 345 the key point is that the performance under quite low aerosol loading is largely affected by the instrument accuracy and stability (Campanelli et al., 2007;Che et al., 2009;Tao et al., 2014). From Figure 4, it can be seen that, with the exception of 7 November when severe cloud contamination occurred, the variation of AOD curves can be easily identified owing to its wavelength dependence. Under the cleanest conditions (Level Ⅰ, PM2.5 ~6 μg m −3 , AOD440 ~0.08−0.15, 11 November), despite the cloud contamination in the afternoon, the AOD variation of each band consistently showed a gradually increasing trend, 350 strictly following the rule of decreasing AOD with increasing wavelength. Therefore, in summary, the CW193 showeds high stabilitygood ability of AOD's wavelength dependence under both high and low aerosol loadings; hence, the excellent detection ability makes it a reliable instrument for aerosol monitoring.  In the next step, the precision performance of CW193 was validated in detail using the AOD from AERONET as a reference. Figure 7 shows a comparison of the AODs from CW193 with the AODs from AERONET at coincident spectral wavelengths.
In general, the AODs from CW193 agree well with AERONET results, with correlation coefficients (R) of ~1.000 for 340−675 nm, ~0.999 for 870 nm, and ~0.997 for 1020 nm and 1640 nm, which indicates that the AODs from CW193 were similarly distributed on both sides of y = x line. From the R values, we can see that the correlation tends to slightly decrease with 380 increasing wavelength. This result can be explained by the temperature sensitivity of the instrument to some degree. As reported by Campanelli et al (2007), the AOD in the near-infrared bands is susceptible to the ambient temperature, and the retrieval accuracy could be improved if the data for the 870 nm and 1020 nm wavelengths were corrected for temperature effects. In addition, although the CW193 is equipped with the same type of temperature sensor in the optical head, there are still many factors that influence the temperature readings, such as mechanical structure and coating color, which could be the 385 main reasons for the temperature uncertainty and the larger AOD deviation at the longer wavelengths of 70 nm, 1020 nm, and 1640 nm.
From this linear regression figure, it can be seen that the slopes for the 340 nm and 1020 nm bands are ~0.993 and 0.966, respectively, whereas those for the other bands were all larger than 1, varying from ~1.001 to 1.021. This indicates that the 390 AOD from CW193 tends to be higher than that from AERONET. As in many previous AOD validation studies, expected error (EE) analyses were also conducted in this study. We set the envelopes as ±(0.05 + 10%). It was found that the AODs from CW193 for each band were all able to achieve a satisfactory performance with 100% retrievals within the EE, much higher than the standard deviation of ~70% (Che et al., 2019b;Levy et al., 2010). The root mean square errors (RMSEs) were all less than 0.05 for all bands, which revealed that the AODs from CW193 are all highly concentrated in the reference AOD range. 395 In addition, these extremely small deviations could also highlight the stability and accuracy of CW193. To evaluate the AOD accuracy further, the relative mean bias (RMB) for each linear regression equation was calculated. As mentioned above, the AOD uncertainties for the near-infrared bands are obviously larger than those for the other bands in this campaign. Specifically, the AODs in the 1020 nm band were underestimated by ~7.8% (RMB = 0.922), whereas they were overestimated by ~11.2% (RMB = 1.112) in the 1640 nm band. The AODs from CW193 in the other bands were all slightly overestimated (~1.6%−4.4%) 400 with the RMB varying in a relatively narrow range of ~1.016−1.044. This indicates that, from the perspective of stability and accuracy, the AODs derived from CW193 have a better performance in the UV bands (340 nm and 380 nm) and visible bands (440 nm to 870 nm) than in the near-infrared band from 1020 nm to 1640 nm. Further studies and experiments still need to be conducted in the future, aimed at algorithm and mechanical structure optimization, to improve the retrieval accuracy.  AODs were sorted in ascending order, and then sampled with 20 bins. From the bias boxplots, it can be seen that the mean biases (red dots) have similar trends in the 340 nm to 870 nm bands with a narrow range from about −0.02 to 0.03, characterized by an initial increase, followed by a decrease, and then a slight increase at high AOD. This indicates that the AODs in these bands from CW193 are overestimated at low AOD (for example, AOD440 ~0.10 to 0.40). Then under moderate AOD levels (for example, AOD440 ~0.50 to 0.90), these biases become smaller or almost equal to zero (even little bit negative) in the range 420 of about −0.01 to 0.01, indicating that the calculations were more consistent with reference values and a high accuracy. At high AOD levels (for example, AOD440 ~1.30 to 1.50), a slight increase in bias was observed in this campaign, with mean values varying from about 0 to 0.02. However, the bias performance for the 1020 nm and 1640 nm bands were quite different.
For the 1020 nm band, the mean biases decreased from zero to −0.02, consistent with AOD varying from ~0.05 to 0.20, and remained relatively constant at about −0.02 when the AOD continually increased to ~0.50. For the biases at 1640 nm, the mean 425 values of each bin showed a roughly parabolic distribution varying from ~0.01 to 0.02, consistent with the AOD varying from ~0.02 to 0.36. Similar to the results mentioned above, the different distribution of the bias boxes for the near-infrared bands suggests that an improvement in accuracy is in needed. Although the linear regression and bias showed fluctuations to some degree, the AOD performance of CW193 was outstanding with high accuracy and stability based on the comprehensive analysis above, characterized by a bias concentrated within ~0.02 for the visible and near-infrared bands and within ~0.03 for 430 the UV bands, which meets the accuracy requirements for AERONET (Holben et al., 1998). represent the mean, median of the AOD bias, and 25% and 75% percentiles, respectively.

Inversions evaluation
According to the algorithm, the aerosol inversions, including microphysical, optical, and radiative properties, are retrieved from the aureole and sky radiance measurements. Similar to CE318, the CW193 conducts the ALM routine at a specific time 440 related to airmass, which is performed in two wings in the 440, 675, 870, and 1020 nm bands sequentially: right (azimuth angle displaced towards the right of the Sun position) and left (azimuth angle displaced towards the left of the Sun position).
In this study, we chose the VSD, SSA, and ADRF to represent the microphysical, optical, and radiative properties of aerosols, as they are not only widely used parameters in current research but also the most important factors influencing the radiative budget of the Earth-atmosphere system (Wang et al., 2013;Zhang et al., 2018). However, it is noted that the uncertainties of 445 these inversions are more difficult to ascertain. As the aureole and sky radiance measurements constitute only single observations (from one ALM routine) and the observation time of each sequence at a specific wavelength is largely subject to the mechanical design and instrument version (for example, the CE318-T mode has faster robotic movements than the N mode).
Furthermore, there is no absolute self-calibration procedure between the sphere calibrations; therefore, the uncertainty in the sky radiance at the time of calibration is assumed to be <5% for these four channels (Holben et al., 1998). As reported by Tao 450 et al (2014), the sphere calibration results of CARSNET differed by 3.12−5.24% in the 870 nm and 1020 nm bands, whereas is differed within 3% in the other two bands compared with the original values from Cimel. As a result, we suggest that an uncertainty of <10% is acceptable for the discussion in this section. In addition, to avoid disturbance from transient atmospheric processes, only the results with an observation time deviation of less than 10 minutes from those of AERONET were selected and the related inversions of CARSNET were also retrieved and presented to show a more detailed comparison. 455 Figure 9 shows a comparison of the VSD for four selected cases in this campaign. It can be seen that the results from CW193 can accurately present the variation pattern of aerosols: the typical bimodal distribution on 6 and 10 November and the nearly unimodal distribution for the two cases on 7 November. For fine-mode particles (radius <1.00 μm), the variations were apparently observed on 6 and 10 November. For the reference PM concentrations, the ratio of PM2.5/PM10 was ~53.1−57.1%, 460

Volume size distribution
suggesting a certain amount of small particles, which agrees with the distribution pattern from CW193 and AERONET. The maximum volume of fine-mode particles varied in the range of ~0.03−0.05 μm 3 μm −2 and ~0.07−0.08 μm 3 μm −2 for 6 and 10 November, respectively. Specifically, the largest deviations of the maximum for fine-mode particles occurred on 6 November, ~77.6% and ~57.1% for CW193 and CARSNET compared with AERONET, respectively. Despite the large volume deviations for fine-mode particles, the variation trends were consistent with those of AERONET, characterized by a maximum peak at a 465 radius of 0.15 μm. Hence, these patterns can be attributed to the different observation times to some degree. The time deviation varied from ~3 to 4 minutes compared with AERONET in this case, while the fine-mode volumes showed a gradually decreasing trend from CW193, followed by CARSNET and AERONET, which agreed with the time series. In contrast, the small deviations of the maximum for fine-mode particles occurred on 10 November, ~8.9% and ~6.8% for CW193 and CARSNET compared with AERONET, respectively. The peak of CW193 and AERONET occurred at a radius of 0.11 μm, 470 and the peak of CARSNEET was observed at 0.15 μm, indicating that both CW193 and CARSNET show good consistency with AERONET.
For coarse-mode particles (radius >1.00 μm), the variations were clearly detected for the four cases, especially on 7 November when the ratio of PM2.5/PM10 was ~21.1%, suggesting that large aerosols were dominant. In these four cases, the peak volumes of coarse-mode particles varied in the range ~0.09−0.13 μm 3 μm −2 , ~0.11−0.14 μm 3 μm −2 , ~0.18−0.25 μm 3 μm −2 , and 475 ~0.05−0.07 μm 3 μm −2 , respectively. It can be seen that the high deviations of the peak volume from CW193 were observed for the cases of 6 and 7 November, with values of ~29.2%, ~19.1% (the case around 8 AM), and 22.2% (the case around 12 AM) compared with AERONET, respectively. However, the performance of CARSNET was better than that of CW193 in these three cases, with deviations of ~5.7%, ~20.4%, and ~6.7%, respectively. As mentioned above, except for the calibration and algorithm uncertainties, these large deviations could be explained by the influence of instantaneous atmospheric disturbances 480 on the retrievals, although the time difference of the measurements between CW193 and AERONET were within ~4−8 minutes (~3−4 minutes for CARSNET). A narrow variation range was found for the 10 November case, characterized by a relatively small time difference among these three retrievals (~2−4 minutes). Consequently, the deviation of the peak volume for CW193 was ~13.1% compared with AERONET, while a larger difference of ~16.8% was found for CARSNET. In summary, the difference in the VSD showed an obvious time-correlation regularity-the smaller the deviation with time, the better the 485 retrieval consistency with AERONET. The SSA represents the scattering proportion affected by aerosol particles in the total extinction and is one of the key variables in assessing the effects of aerosols on the climate (Che et al., 2019c;Zhao et al., 2018). The variation of SSA at four spectral wavelengths for the four cases (6 and 10 November and two on 7 November) is shown in Figure 10. First, we examined the wavelength dependence of SSA, revealing the different scattering capacity for aerosols at specific bands, which is largely influenced by the aerosol chemical composition and can be regarded as an indicator of the dominant aerosol type (Eck et al., 495 1999;Zheng et al., 2021). It can be seen from Figure 10 that the SSA on the three days showed different variation trends.
Specifically, for the 6 November case, the SSA increased from 440 nm to 675 nm and showed a roughly decreasing trend from 675 nm to 1020 nm, indicating a relatively strong aerosol absorbance at shorter wavelengths in the visible bands. The SSA showed an increasing trend with wavelength for the two cases on 7 November, whereas a decreasing trend was observed on 10 November. This indicates that the aerosol absorptive ability was attenuated with increasing wavelength on 7 November, 500 whereas enhanced aerosol absorbance with wavelength was found on 10 November. From the discussion above, we can see that the wavelength dependence of SSA from CW193 and CARSNET were both highly consistent with that from AERONET, indicating the good performance of the retrieval for aerosol optical properties.
To elaborate the SSA assessment, we present a comprehensive comparison of the accuracy in detail here. On 6 November, the SSA peaked in the 675 nm band, with values of ~0.848, 0.857, and 0.853 for CW193, CARSNET, and AERONET, respectively. 505 The deviations of these maximums for CW193 and CARSNET were ~0.1% and 0.3% compared with AERONET, respectively.
In this case, the SSA of CW193 varied within a narrow range of ~0.834−0.848, whereas that of AERONET was ~0.836−0.853.
The highest deviation for a specific wavelength of CW193 was found in the 1020 nm band, with a value of ~1.7%, and the lowest was found in the 440 nm and 870 nm bands, with a value ~0.1%. As mentioned above, the SSA shows an increasing trend with wavelength for the two cases on 7 November. The smallest SSA values were all observed in the 440 nm bands, with 510 values varying in the range of ~0.858−0.861 and ~0. 840−0.859, respectively. For the case at around 08:00, the maximum of CW193 was found in the 870 nm band, with a value of ~0.899, whereas that of AERONET was found in the 1020 nm band, with a value of ~0.911, which suggests a maximum deviation of ~1.3%. The largest deviation for a specific wavelength of CW193 occurred in the 1020 nm band and was ~2.1% compared with AERONET, followed by 1.9% at 675 nm, 1.0% at 870 nm, and 0.6% at 440 nm. For the case at around 12:00, although the SSA was relatively low in the 440 nm band (~0.840−0.859), 515 it remained almost constant from 675 to 1020 nm for CW193 and AERONET, characterized by a small fluctuation amplitude of ~0.930−0.935 for the former and ~0.926−0.931 for the latter. The highest deviation for a specific wavelength, ~1.8%, was measured in the 440 nm band, followed by 0.8% at 1020 nm, 0.6% at 675 nm, and 0.1% at 870 nm. The SSA showed more obvious fluctuations for the 10 November case. Specifically, the peak SSA for CW193 and AERONET were both observed in the 440 nm band, with values of ~0.844 and 0.832, respectively. Likewise, the lowest values of ~0.733 and 0.708 for these two 520 were measured in the 1020 nm band. However, the variation of deviation at a specific wavelength showed no regular pattern compared with the SSA. The largest deviation of ~3.5% was found in the 1020 nm band, followed by ~2.6% at 870 nm, ~1.4% at 440 nm, and ~0.7% at 765 nm. In a conclusion, the SSA deviation between CW193 and AERONET varied in the range of ~0.1−1.8%, ~0.6−1.9%, ~0.1−2.6%, and ~0.8−3.5% for the 440, 675, 870, and 1020 nm bands, respectively, indicating a high consistency with AERONET. 525 Figure 10. Comparison of retrieved SSA for CW193, CARSNET, and AERONET for four selected cases.

Aerosol direct radiative forcing
The ADRF is a key factor influencing the radiation budget of the Earth-atmosphere system, in which any small perturbation to this global energy balance can cause a profound change in the climate (Garcí a et al., 2012). In this context, much progress 530 had been made in this field to provide insight into the climate effects of aerosols. A previous study estimated the total anthropogenic radiative effect on a global scale to be +1.6 (−1.0 to +0.8) W m −2 , of which −0. Nov 10 with the direct radiative forcing of aerosols (Garcí a et al., 2008). However, it can be seen that there remains huge uncertainty in the evaluation of the ADFR. For this reason, we selected it as a main product of CW193 to examine the accuracy of the radiative retrieval. 535 In Figure 11, we show a comparison of ADRF for the four cases (6 and 10 November and two on 7 November) between the CW193, CARSNET, and AERONET. As reported by Zheng et al. (2019), the ADRF at Earth's surface (BOA) varies from −86±31 to −132±50 W m −2 , whereas the ARDF at the top of the atmosphere (TOA) varies from −35±18 to −55±26 W m −2 based on a five-year observation campaign in urban Beijing. Therefore, it can be seen that the BOA and TOA retrievals of 540 CW193 and CARSNET all show a reasonable range of values in this campaign. Specifically, the BOAs of CW193 were −127.1, −65.6, −108.4, and -105.6 W m −2 for the four cases in chronological order, respectively. Correspondingly, the BOAs from AERONET were −113.2, −58.4, −103.5, and -95.0 W m −2 . Thus, the deviation of BOAs in these cases was ~12.2%, 12.3%, 4.8%, and 11.2%, respectively, suggesting an overestimation of BOA compared with AERONET. For the TOAs, the CW193 retrievals for these cases were −22.8, −25.6 −34.3, and -16.5 W m −2 , whereas the reference values from AERONET were 545 −25.3, −22.1, −32.6, and -15.3 W m −2 , respectively. That is, the TOA deviation found in these cases was ~9.8%, 15.9%, 5.4%, and 7.4%, respectively. In summary, the deviation of the retrieval BOA was ~5%−12%, whereas it was ~5%−16% for the TOA. As shown above, the relatively larger uncertainties can be partly explained by the inherent algorithm error, as well as the difference in observation time.

Water vapor evaluation
Water vapor (WV) is a key atmospheric component for studies of climate change, because it not only has an important role in aerosol aging but also can influence the energy budget of the Earth-atmosphere system by absorbing and scattering solar 555 energy. Therefore, in this study, the precision performance of WV from CW193 was validated in detail using AERONET as a reference. Figure 12 shows a comparison of WV from CW193 with the results from AERONET. In Figure 12  Nov 10 the slope was ~0.941, suggesting that the WV from CW193 tends to be lower than that from AERONET. In terms of RMB values, it is found that the WV from CW193 is underestimated by ~2.1% (RMB = 0.979). The EE analysis showed that the retrieved columnar WV (100%) was within the EE. In addition, the small RMSE (~0.020) also reflected that the CW193 WV was highly concentrated in the reference AERONET range. 565 Figure 12 (b) shows the CW193 WV bias compared with equal frequency bins of WV from AERONET. From this boxplot, it can be seen that the biases vary in the range of −0.04 to 0.04, whereas its mean values (red dots) are concentrated in a narrower range from −0.02 to 0.02. As reported by Holben et al. (1998), the uncertainty of the WV retrieval is limited to less than 12%, based on an intercomparison with radiosonde results. In this study, the overall WV biases of CW193 was roughly lower than 4%, demonstrating the accurate measurement capability for columnar WV. However, it is noted that these biases, especially 570 the mean values, show an increasing trend (about −0.01 to 0.03) with increasing WV values (~0.24 to 0.80 cm). Gui et al (2017) revealed that the monthly WV for November was ~0.74 cm in urban Beijing, whereas that for the summer exceeded 2.00 cm.
In this campaign, the CW193 WV varied from ~0.26 to 1.08, indicating that whether this bias increasing trend exists still needs to be further tested in future, especially for humid summer days.

Conclusions
In this study, we have presented a new multiwavelengthmultispectral photometer named CW193 for monitoring aerosol microphysical, optical, and radiative properties. The CW193 is highly integrated and is composed of three main parts: an optical head, a robotic drive platform, and a stents system. It has a user-friendly interface and all commands can be sent to the instrument via serial communication or the 4G network, which makes data acquisition and operation monitoring easier. A performance evaluation of CW193 was presented and discussed in detail, based on an intercomparison with the reference AERONET results. The main conclusions of this study are as follows.
1. The comparison of raw digital counts from CW193 and CE318s (two AERONET master instruments, photometers #1043 585 and #1046) showed a high coefficient of determination (R 2 ) for all wavelengths, which were >0.97 and >0.99, respectively.
Apart from the cloud contamination, the diurnal triplets for these 9 bands were mostly lower than 2.0% during 10:00 to 14:00 BJT. Daily average triplets for the UV bands (340 nm and 380 nm) varied from about 1.2% to 3.0%, whereas it was <2.0% for the visible and infrared bands (440 nm to 1640 nm). 590 2. Using reference PM concentrations, the wavelength dependance of AODs was examined. The AOD curves were nonintersecting and could be easily identified (AOD440 ~0.08 to 1.47) under the air quality Level Ⅰ to Level Ⅲ (PM2.5 ~6 to 104 μg m −3 ), showing a decreasing trend with increasing wavelength. From the regression analysis, a good AOD agreement (R > 0.99) and RMSE values from 0.006 (870 nm) to 0.016 (440 nm) were observed. The AODs from CW193 achieved a satisfactory performance with 100% of the retrievals within the EE (0.05 + 10%) and a RMB varying from 0.922 to 1.112. 595 The AOD bias analysis showed an overall deviation that varied within ±0.04, and within 0.02 for the mean values.
3. The variation of inversions was subject to the time of the measurement in this study. From the perspective of VSD retrievals, the deviations of the maximum for fine-mode particles varied from ~8.9% to 77.6%, whereas it varied from ~13.1% to 29.1% for coarse-mode particles. The wavelength dependance of SSA from CW193 showed a similar trend to the AERONET SSA, 600 and the variation range of the deviations was ~0.1−1.8%, ~0.6−1.9%, ~0.1−2.6%, and ~0.8−3.5% for the 440, 675, 870, and 1020 nm bands, respectively. For the ADRF, the BOA and TOA deviations of ~4.8%−12.3% and ~5.4%−15.9% were observed in this study, respectively. 4. A good WV agreement was found, characterized by a high R (~0.997), small RMSE (~0.020), and satisfactory EE 605 distribution (100% within EE). The RMB showed that the WV was underestimated by ~2.1% (RMB = 0.979). The biases mostly varied within ±0.04, whereas its mean values were concentrated within ±0.02.
The results of this preliminary study evaluation indicate that the CW193 is appropriate for monitoring aerosol microphysical, optical, and radiative properties, with the overall AOD (including WV) biases within ±0.02 for the 500 nm to 870 nm bands 610 and within ±0.04 for the other bands. Considering the uncertainty inherent in the algorithm (±0.02) and the AOD uncertainty of AERONET (±0.02), the direct Sun measurements seem reasonable and reliable for the AOD and WV calculations (uncertainty within ±0.04). However, its performance under extreme heavy aerosol loading still needs to be assessed in future, especially during severe haze and/or dust episodes when the AOD exceeds 2.00. Although the results for SSA and ADRF showed good agreement with AERONET, the VSD deviations were relatively larger than these two parameters. In fact, owing 615 to the joint influence of the sphere calibration's uncertainty and the measurement time difference, the evaluation of these inversions was difficult under the short period of the observation campaign. Consequently, the instruments still need to be further tested under different environment conditions, including long-term observations in mountainous, coastal, and desert regions. As a resultHowever, the CW193 retrievals in this study showed high precision for SSA and ADRF, and comparable results for VSD, indicating the stability and accuracy of the CW193 productsgood comparability and consistency with 620 AERONET.
Above all, the aim of this new multispectral photometer is to complement the observation gaps of CARSNET. Especially in remote locations, where the deployment of Sun photometers is still a challenge for poor infrastructure and logistics, the highly integrated design and smart control performance make CW193 more convenient and suitable for the aerosol monitoring 625 microphysical, optical, and radiative properties of aerosol, providing similar aerosol optical properties to AERONET. Due to its smart control performance and optional observation schedule, such as ALM mode, the CW193 could meet the different requirement of the aerosol microphysical, optical, and radiative properties. When the VSD and SSA is in great demand for the modification of numerical model and the verification of satellite inversion products, these inversions could be obtained about 2 to 3 times in an hour, while for once in default observation schedule. In addition, owing to the built-in 4G communication 630 module, CW193 could be used to create networks in an inexpensive and simple way. As a result, this instrument could have an important rolebe regarded as a contributor in regional and climate model data assimilation, satellite modification, and improving knowledge of the temporal and spatial variations of aerosols.

Data availability
Datasets used in the present study are available from the corresponding author on reasonable request. 635