2.1.

History of AD-Net and Station Locations

In 1996, NIES, in Tsukuba, Japan, developed an automated elastic scattering lidar (EL) system that could operate continuously without human intervention. At that time, only the total backscatter signal intensity at 532 nm ($I_{532}$) was recorded, and only aerosols and cloud layers could be identified. In 1999, the system was modified so that the polarization state of the backscattered light could also be recorded, allowing aerosol particle shapes to be investigated. By this modification, the NIES automated lidar system acquired the capability to distinguish mineral dust particles from anthropogenic particles and ice particles in cirrus clouds from water droplets in convective clouds. In 2001, an initial lidar network was constructed for the ACE-Asia campaign,⁶ an important aim of which was to investigate the distribution of Asian dust. For ACE-Asia, NIES operated lidar observatories at Tsukuba, Nagasaki, and Beijing and collaborated with other lidar observatories in Japan.^7,8 Since then, the number of NIES-type lidar observatories, which follow a mostly uniform operational strategy, has increased; at present, AD-Net comprises a total of 20 observatories in East Asia (i.e., in Japan, Korea, China, and Mongolia). Figure 1 shows the locations of AD-Net lidar observatories in operation as of July 2016. Nowadays some lidars were modified to Raman lidar (RL) or multiwavelength Raman lidar (MRL) systems (see Sec. 3). However, the data treatment for elastic scattering channels of such lidars is the same as that for EL. Thus, we describe EL system at first (Sec. 2.2) and then introduce the common data processing procedures on elastic scattering channels in Sec. 2.3.

Fig. 1

Locations of current AD-Net lidar observatories. Green, red, and blue symbols indicate locations of EL, RL, MRL systems, respectively.

2.2.

Hardware of an Elastic-Scattering Lidar System

The fundamental and common information about aerosol is retrieved from elastic scattering channels in all types of AD-Net lidars. Here, the specifications and configuration of the AD-Net elastic lidar (EL) system are described.

Table 1 and Fig. 2(a) summarize the specs and configurations of EL. The EL employs a flashlamp-pumped Q-switching Nd:YAG laser as the light source. Pulsed light is emitted with a fundamental wavelength of 1064 nm and a second harmonic of 532 nm, produced by second harmonics generation. A pulse with a mean power of 50 mJ (sum of 20 mJ at 1064 nm and 30 mJ at 532 nm) and a duration of 8 ns is expanded by a beam expander and emitted at a frequency of 10 Hz in the direction of the zenith. Backscattered light from the atmosphere is collected by a Schmidt–Cassegrain telescope with a diameter of 20 cm and a field of view of 1 mrad and separated by a dichroic mirror into $P_{1064}$ (light intensity at 1064 nm) and $P_{532}$ (light intensity at 532 nm). Then, $P_{532}$ is further separated by a beam splitter cube into two components: $P_{532\mathrm{p}}$ (the polarizing angle is parallel to that of the emitted laser beam) and $P_{532\mathrm{s}}$ (the polarizing angle is perpendicular). The extinction ratio of the beam splitter cube is 0.001. $P_{532\mathrm{p}}$ and $P_{532\mathrm{s}}$ are both detected by photomultiplier tubes (PMTs), and $P_{1064}$ is detected by an avalanche photodiode. Signal intensities are sampled at 25 MHz (corresponding to a range resolution of 6 m), digitized with 12- or 16-bit AD converters, and recorded by a personal computer as vertical profiles $P_{532\mathrm{p}}(z)$, $P_{532\mathrm{s}}(z)$, and $P_{1064}(z)$, where $z$ denotes the altitude above the lidar. To obtain a high signal-to-noise ratio, signals are accumulated for 5 min (i.e., 3000 laser shots). To reduce the flashlamp consumption, a 10-min rest period follows each 5-min emission period. Thus, the system obtains four profiles each hour.

Table 1

Specifications of the AD-Net lidars.

Fig. 2

Configurations of the lidars used in AD-Net. (a) EL and RL, (b) MRL, and (c) RHL.

The lidar system is installed in a room with a glass window in the roof. Because the observations are made through the glass, the system can be operated in all weather conditions, including rain and snow. Every hour, the observation results are transferred to an NIES data server via the Internet, the condition of the lidar system is checked, and an initial analysis is performed. An exception is the Beijing lidar, which by Chinese law is not allowed to send data to NIES in real time.

2.3.

Data Processing

To obtain the optical properties of aerosols, the following analyses are applied to the AD-Net observation results in elastic-scattering channels of EL, RL, and MRL.

The background removed and range corrected signal intensities of the three channels ($I_{532\mathrm{p}}$, $I_{532\mathrm{s}}$, and $I_{1064}$) must be calibrated before physical quantities such as the attenuated backscatter coefficient can be determined. First, total backscatter intensity $I_{532}$ is calculated as

Fig. 3

Examples of time–height sections of (a) attenuated backscatter coefficient at 532 nm (${\beta }_{532}^{\prime }$), (b) attenuated backscatter coefficient at 1064 nm (${\beta }_{1064}^{\prime }$), and (c) volume depolarization ratio at 532 nm ($\delta v_{532}$) and with cloud/rain detection results, (d) dust extinction coefficient at 532 nm (${\alpha }_{532\mathrm{d}}$) and (e) spherical particle extinction coefficient at 532 nm (${\alpha }_{532\mathrm{s}}$). The observations were made at Osaka in March 2016. The height range for (a)–(c) is from the surface to 18 km and for (d) and (e) is from the surface to 9 km. Black areas in (d) and (e) indicate cloud layers, and gray areas indicate regions above clouds, rain or snow, or no observation data.

Fig. 4

Flowchart of the backscattering intensities at 532 nm ($I_{532}$) and at 1064 nm ($I_{1064}$) calibration procedures.

The depolarization ratio, a measure of the irregularity of the scatterer shape,⁹ is the most important property of Asian dust measured by lidar systems. To calculate the depolarization ratio, it is important to first calibrate the signal intensities of the $I_{532\mathrm{p}}$ and $I_{532\mathrm{s}}$ channels. In AD-Net, the difference in sensitivity between the two PMTs used to detect these components is checked routinely by the following method. A sheet polarizer whose polarizing direction is set at 45 deg to the polarizing plane of emitted light is inserted in front of the beam splitter cube, and the backscatter signal from the sky is recorded as a reference signal. In this reference record, the light intensities of the two channels are equal after the sheet polarizer, so the calibration constant $C\mathrm{d}$ can be obtained by comparing the recorded values of $I_{532\mathrm{p}}$ and $I_{532\mathrm{s}}$. Next, the sheet polarizer is rotated 90 deg to set the polarizing angle at $-45\mathrm{deg}$, and another reference signal is recorded. This pair of reference signals reduces any error caused by poor positioning of the sheet polarizer.¹⁰ The reference signals are usually recorded once per year for each lidar.

The cloud base height must be determined first because the aerosol layer analysis cannot be applied to cloud layers. In AD-Net, the cloud base height is determined from the vertical gradient of ${\beta }_{1064}^{\prime }$ and the peak value of ${\beta }_{1064}^{\prime }$ between the cloud base and the apparent cloud top, where ${\beta }_{1064}^{\prime }$ is equal to that of the cloud base. The threshold values are determined empirically because they depend on the vertical resolution and signal accumulation time of the lidar measurements and there are no true cloud base height reference data with equivalent lidar time and height resolutions at each observatory. Therefore, the thresholds are determined subjectively by estimating the cloud distribution on time–height sections [see Figs. 3(d) and 3(e)].

Scattering by rain droplets or snowflakes obscures the signal from aerosols. Thus, data recorded during rainy or snowy conditions must be eliminated before the analysis. Moreover, the affected profiles must be selected using the lidar data without ancillary data because not all observatories have a surface rain gauge or an equivalent rain or snow detection system. AD-Net uses the color ratio (${\gamma }^{\prime }$, the ratio of ${\beta }_{1064}^{\prime }$ to ${\beta }_{532}^{\prime }$) to distinguish rainy and clear (no rain) regions. Large droplets have a large ${\gamma }^{\prime }$ value, so once ${\gamma }^{\prime }$ exceeds a threshold (1.1) over a certain vertical interval in the lower atmosphere, the profile is classified as a rain or snow profile and not used for further analysis of aerosols.

At present, in the AD-Net lidar systems, the overlap of the laser beam and the field of view of the telescope is insufficient for near-field observation. Typically, the full overlap is achieved at around 500 to 600 m altitude. The compensation function $Y(z)$ is therefore inferred from vertical profiles obtained on a day when the planetary boundary layer (PBL) is well developed, when the aerosol distribution is expected to be homogeneous near the surface. $Y(z)$ is determined such that the slope of the compensated signal $[Y(z)\times I_{532}(z)]$ is constant near the surface. $Y(z)$ is redetermined after routine maintenance of the lidar equipment is carried out. With this compensation, the optical properties are provided above 120 m altitude. Recently, a small telescope with a wide field of view has been deployed at several lidar observatories in AD-Net, allowing the signal to be measured near the ground (above 60 m altitude). At these observatories, $Y(z)$ can be determined without assuming homogeneous mixing near the surface.

The lidar equation for elastic scattering is resolved into two components (particles and molecules) by the method described by Fernald.¹¹ The ratio of the extinction coefficient to the backscatter coefficient for aerosols (lidar ratio, $S_1$) is assumed to be 50 sr. This value was initially determined from values reported in the literature on Asian dust because originally the main target of this lidar network was Asian dust. The vertical profile of molecular densities is obtained from the CIRA-86 global climatology of atmospheric parameters.¹² Usually, the maximum height of Fernald’s inversion is set to an altitude of 9 km (6 km before 2012) if the signal-to-noise ratio within that interval is enough to solve the equation. Because 9 km is usually in the troposphere, where the aerosol extinction value is unknown, the lidar equation is first solved with the initial extinction value set to zero. If the resulting aerosol extinction value is negative anywhere between 0 and 9 km height, the extinction at 9 km is increased slightly and the lidar equation is solved again. Although this method is not rigorous, the obtained extinction is sometimes validated by independent measurements made near the surface or by columnar optical depth observations. Finally, the particulate depolarization ratio ($\delta p_{532}$) is calculated from the volume depolarization ratio ($\delta v_{532}$) and the particulate backscatter coefficient. These procedures are explained in detail in Ref. 7 and outlined in Fig. 5.

Fig. 5

Flowchart of the cloud and rain detection procedure and Fernald’s inversion.

In East Asia, the aerosols being analyzed are assumed to be an external mixture of two components. One component is mineral dust, mainly Asian dust that has been transported a long distance, which has a particulate depolarization ratio ($\delta p_{532}$) of 35%.⁷ The other component consists of spherical particles and is assumed to be a mixture of sulfate, nitrate, organic carbon, elemental carbon particles, and sea-salt droplets. The $\delta p_{532}$ of the latter component is zero because the particles are spherical and Mie scattering theory is applicable. The observed particulate extinction coefficient and particulate linear depolarization values are the linear combination of these two components. Thus, it is possible to use the observed ${\delta }_{\mathrm{p}532}$ to separate the total particulate extinction coefficient (${\alpha }_{532}$) into two components, the dust extinction coefficient (${\alpha }_{532\mathrm{d}}$) and the spherical particle extinction coefficient (${\alpha }_{532\mathrm{s}}$), by calculating the optical dust mixing ratio ($R$) by following equation:¹³

Although ${\beta }_{532}^{\prime }$ was initially estimated from a histogram of all $I_{532}$ data, a more precise estimation is made after inversion. The $I_{532}$ time series at 600 m height is compared with the total backscatter coefficient (sum of molecular backscatter from CIRA-86 and particulate backscatter obtained by inversion) at 600 m height to evaluate the system calibration constant $C_{532}$ again. The revised $C_{532}$ is then utilized to estimate ${\beta }_{532}^{\prime }$ from $I_{532}$. ${\beta }_{1064}^{\prime }$ is similarly recalculated.

Finally, the AD-Net server generates numerical files in netCDF format that contain time–height sections of ${\beta }_{532}^{\prime }$, ${\beta }_{1064}^{\prime }$, $\delta v_{532}$, ${\alpha }_{532}$, $\delta p_{532}$, ${\alpha }_{532\mathrm{d}}$, and ${\alpha }_{532\mathrm{s}}$. Quantities below 120 m altitude are eliminated from published data because their reliability is not high owing to the uncertainty of the overlap correction function $Y(z)$. Currently, files are generated hourly for the month and archived one per month. Symbols for two types of missing values ($-9999$ for cloud layers and −999 for regions above clouds, with rainy or snowy conditions, or with no observations) are embedded into the ${\alpha }_{532}$, ${\alpha }_{532\mathrm{d}}$, and ${\alpha }_{532\mathrm{s}}$. Simultaneously, figures of time–height sections for these parameters are plotted and both the numerical files and the figures are uploaded onto the AD-Net web page every hour. The procedure is outlined in Fig. 6.

Fig. 6

Flowchart showing the procedures for updating calibration constant ($C_{532}$) and data recording.

2.4.

Validation of Dust Extinction and Comparison with Other Instruments

It is not a simple task to validate dust extinction coefficients obtained by lidar because this parameter cannot be obtained directly by other instruments. However, the surface mass concentration of particles consists mainly of mineral dust during periods with heavy loading of Asian dust. Thus, dust extinction near the surface can be compared with filter-sampled mass concentrations by making some assumptions about vertical mixing near the surface. This comparison was first done in Beijing, China.¹³ In Ref. 13, it is reported that the dust extinction coefficient retrieved by lidar is almost proportional to the mass concentration of total suspended particles. A similar comparison has been conducted at many stations in Japan during several Asian dust events.¹⁴ In addition, in Ref. 15, the iron concentration in daily PM2.5 samples is compared with the dust extinction coefficient. The results of all these comparisons suggest that dust extinction coefficients determined by lidar are usable as an index of surface dust loading.

2.5.

Utilization of the Retrieved Optical Parameters

2.5.3.

Climatology of aerosols

The climatology of aerosols in East Asia can be derived from AD-Net data. Horizontal, seasonal, and interannual variations of extinction coefficients are calculated for each of the two components.

In Fig. 7, mean vertical profiles of dust and spherical particle extinction coefficients are plotted. The exponential decrement with height is apparent in the dust extinction coefficient in most stations; however, a boundary layer structure is found at around 1.8 to 2.3 km in the spherical particle extinction coefficient for many stations. To depict horizontal, annual, and interannual variations of spherical particle extinction coefficients more clearly, the probability density of a time series at a certain height is examined by calculating the 5, 25, 50, 75, and 95 percentiles at the selected height and comparing the results among observatories or among months or years.

Fig. 7

Averaged vertical profiles of (a) dust extinction coefficient (${\alpha }_{532\mathrm{d}}$) during March to May in 2010 to 2015, and (b) spherical particles extinction coefficient (${\alpha }_{532\mathrm{s}}$) for all seasons in 2010 to 2015 at 16 stations.

The spatial distribution of extinction coefficients of spherical particles among observatories is shown in Fig. 8. An apparent gradient from west to east can be seen except for two stations in Mongolia, but this effect disappears at higher altitudes.

Fig. 8

Distribution of spherical particle extinction coefficients at (b) 500 m altitude and (a) 2500 m altitude among observatories. The boxes and the horizontal line show the 25 to 75 percentile range and the 50 percentile, respectively, and the whiskers show the 5 to 95 percentile range, based on all data collected during 2010 to 2015.

Intra-annual changes differ depending on altitude and among stations. Figure 9 shows monthly percentiles based on data from 2010 to 2015. In the PBL (500 m), seasonal changes are apparent in large cities (Beijing, Seoul, and Tokyo), whereas the annual cycle of changes is different in the lower troposphere (2500 m). These results are consistent with the findings of a comprehensive analysis of anthropogenic particles in this region.¹⁶

Fig. 9

Same as Fig. 8 but showing monthly variations of the spherical extinction coefficient at (b) 500 m and (a) 2500 m at Beijing, Seoul, and Tokyo, from left to right.

${\alpha }_{532\mathrm{s}}$ has shown a slight interannual decrease in these cities from 2008 to 2015 (Fig. 10). These results confirm the findings of several studies that have reported a decrease in emissions of anthropogenic gaseous pollutants in this region since 2006.²⁴

Fig. 10

Same as Fig. 8 but showing year-to-year variations during 2008 to 2015.

3.1.

Use of Raman Lidar Techniques

The AD-Net uses the nitrogen RL technique. Details on the RL system, data analysis method, calibration method, measurement uncertainties, and observation results are given in Ref. 27. Here, a summary on the RL observation of AD-Net is given.

We improved the $2\beta +1\delta$ ELs at five sites of the AD-Net (Tsukuba, Matsue, Fukue, Seoul, and Beijing) by adding a nitrogen Raman scatter measurement channel at 607 nm and have conducted continuous observations since 2009. As a result, this RL system can provide extinction coefficient ($\alpha$), backscatter coefficient ($\beta$), and depolarization ratio ($\delta$) of particles at 532 nm and attenuated backscatter coefficient (${\beta }^{\prime }$) at 1064 nm. The configuration and specifications are given in Fig. 2(a) and Table 1. Furthermore, we built a multiwavelength RL system (MRL) providing ${\alpha }_{355}$, ${\alpha }_{532}$, ${\beta }_{355}$, ${\beta }_{532}$, ${\delta }_{355}$, ${\delta }_{532}$, and ${\beta }_{1064}^{\prime }$ at Fukuoka, Toyama, and Hedo sites of the AD-Net. The MRL observations started from 2013 at Fukuoka, from 2014 at Hedo, and from 2015 at Toyama. The configuration and specifications are given in Fig. 2(b) and Table 1. For the RL and MRL measurements, we conduct photon-counting measurement for Raman backscatter and analog measurement for elastic backscatter. No Raman channel data are available in the daytime due to strong sunlight.

After reducing signal noises using wavelet transform analysis and moving average, the uncertainties of $\alpha$, $\beta$, $\delta$, and lidar ratio $S_1$ ($\alpha /\beta$) of aerosols in the PBL derived from the RL and MRL measurements are evaluated to be less than 5%, which indicates that the RL and MRL have enough accuracy to understand aerosol optical properties in the PBL.

3.2.

Use of High Spectral Resolution Lidar Techniques

The HSR lidar is a more highly sensitive technique than the RL, indicating that the HSR lidar can provide measurements with enough signal-to-noise ratio in the daytime as well as the nighttime to retrieve $\alpha$, $\beta$, and $\delta$. We developed a Raman-HSR lidar (RHL) implementing both the nitrogen RL technique at 387 nm and the HSR lidar technique at 532 nm [Fig. 2(c) and Table 1] at the Tsukuba site from 2014. This RHL provides the same products as the MRL (i.e., $\alpha$, $\beta$, and $\delta$ at 355 and 532 nm, respectively, and ${\beta }^{\prime }$ at 1064 nm). This system uses the iodine absorption filter²⁸ to introduce the HSR lidar technique, indicating that the laser wavelength must be tuned to the center of the iodine absorption line to block the elastic scattering light and transmit the Rayleigh scattering light efficiently. We developed and implemented an automatical feedback control system using two acousto optic modulators²⁹ to tune the laser wavelength to the absorption line stably for a long period. We conduct photon-counting measurement for Raman backscatter and analog measurement for elastic backscatter including the Rayleigh scatter. No Raman channel data are available in the daytime.

After reducing signal noises in a similar manner as the data analysis of the MRL, the uncertainties of $\alpha$, $\beta$, $\delta$, and $S_1$ of aerosols in the PBL are evaluated to be comparable to or less than the uncertainties of MRL. Those uncertainties for nighttime are less than half of uncertainties for daytime.

3.3.

Aerosol Component Retrieval

A two-component (i.e., mineral dust and spherical particles) algorithm¹⁴ using the EL data was implemented in the standard data processing, and the component products as well as the other products have been provided in public (see Sec. 2.3). Furthermore, as another algorithm using the EL data, we developed a three-component (i.e., mineral dust, sea-salt, and air pollution particles) algorithm^30,31 to estimate a vertical distribution of extinction coefficient for the three aerosol components. This algorithm uses the difference of $\gamma$ due to particle size and the difference of $\delta$ due to particle shape among the aerosol components. For the RL, MRL, and RHL measurements, a four-component (i.e., mineral dust, sea-salt, black carbon, and air pollution particles except black carbon) algorithm^27,32 that further uses the difference of $S_1$ due to light absorption property among the aerosol components was developed.

The aerosol component products by three- and four-aerosol component algorithms as well as the aerosol and cloud optical property data (i.e., $\alpha$, $\beta$, $\delta$, and $S_1$) will be made public in the future. Since the developed three- and four-aerosol component retrieval algorithms do not use all the data of MRL and RHL, an aerosol component retrieval algorithm using all the data of MRL and RHL more effectively is being developed to provide more detailed optical and microphysical properties of each aerosol component.