The multiwavelength Mie–Raman polarization lidar called LILAS (Lille Lidar Atmosphere Study) is the main instrument installed at the observation site. The lidar system has been operating in LOA (Laboratoire d'Optique Atmosphérique, Lille, France) since 2013 . During the DAO campaign, LILAS was transported from Lille to Kashi (and Beijing in the second session of the campaign) to perform observations. LILAS uses a Nd:YAG laser that emits at three wavelengths: 355, 532 and 1064 nm. The laser repetition rate is 20 Hz. A Glan prism is used to clean the polarization of the laser beam. The emitting power after the Glan prism cleaning is about 70, 90 and 100 mJ at 355, 532 and 1064 nm, respectively. The LILAS system has three Raman channels, including 387 (vibrational-rotational), 530 (rotational) and 408 nm (water vapor). The use of rotational Raman at 530 nm provides a stronger Raman signal and relieves the dependence of the derived extinction and backscatter coefficients on the assumption of the Ångström exponent . The backscattered light is collected by a 400 mm Newton telescope. The incomplete overlap range of the LILAS system is about 1000–1500 m in distance depending on the selected field of view angle. In the receiving optics, the three elastic channels are equipped with both a perpendicular and a parallel channel with respect to the polarization plane of the emitted linearly polarized laser light in order to measure the linear depolarization ratio at three wavelengths. LILAS can provide the profiles of the 2α+3β+3δ (α: extinction coefficient; β: backscatter coefficient; δ: particle linear depolarization ratio, PLDR) parameters. Benefiting from the coupled Raman channels, the extinction and backscatter coefficients at 355 and 532 nm are calculated using the Raman method proposed by . The Raman signal generated by the radiation at 1064 nm is not measured by LILAS; thus, the Raman method is not applicable. The backscatter coefficient at 1064 nm is calculated using the Klett method, in which a vertically constant lidar ratio (extinction-to-backscatter ratio) is assumed . The particle linear depolarization ratios are derived from Eq. (): 1δp=1+δmδvR-1+δvδm1+δmR-1+δv, where R represents the ratio of the total backscatter coefficient, involving molecules and particles, to the particle backscatter coefficient; δm represents the molecular depolarization ratio; and δv represents the volume linear depolarization ratio (VLDR), which equals the calibration coefficient multiplied by the ratio of the signal of the perpendicular channel to the parallel channel. The polarization calibration is performed following the ± 45∘ method . During the DAO campaign, the polarization calibration was performed at least once per day.

The Ångström exponent of the extinction coefficient and backscatter coefficient are calculated with Eq. (): 2Å=-log⁡p(λ1)-log⁡p(λ2)log⁡λ1-log⁡λ2, where p(λ) represents the optical parameters, such as aerosol optical depth (AOD), extinction or backscatter coefficient at wavelength λ, and Å represents the Ångström exponent of the corresponding parameters p(λ).

The statistical error of lidar derived parameters is estimated using the method presented in . The data presented in this study are recorded at nighttime, so the background radiation is negligible. The error in the extinction and backscatter coefficients is about 10 %, which leads to about a 15 % error in the lidar ratios at 355 and 532 nm. The error in the backscatter coefficient at 1064 nm is about 20 %. The error in PLDR is calculated in terms of the backscatter ratio, VLDR and molecular depolarization ratio. For the data presented in this study, the error in PLDR is no greater than 15 % at 532 and 1064 nm. Therefore, we conservatively use 15 % as the error in PLDR for 532 and 1064 nm. At 355 nm, the error of 15 % still holds when dust concentration is high enough, but when the concentration drops, the error could exceed 15 %. In the case study, errors at 355 nm are calculated separately. The errors for the water vapor mixing ratio (WVMR) and relative humidity (RH) are about 20 %.

Three sun–sky photometers are deployed in the Kashi observation site. One is affiliated with AERONET (AErosol RObotic NETwork; ), and the other two are affiliated with SONET (Sun–Sky Radiometer Observation Network). SONET is a ground-based sun photometer network with the extension of a multiwavelength polarization measurement capability to provide long-term columnar atmospheric aerosol properties over China . The three sun–sky photometers provide complementary measurements by following different measurement protocols. In all, they can measure daytime aerosol optical depth (denoted as AOD hereafter) at 340, 380, 440, 675, 870, 1020 and 1640 nm, polarized/unpolarized sky radiances at 440, 675, 870 and 1020 nm, and moon AOD as well. The succeeding data treatment and retrieval are performed following the protocols and standards of AERONET or SONET depending on the affiliation of the instruments.

Satellite data have complementary advantages due to their large spatial coverage compared to ground-based remote sensing techniques. In order to monitor dust activities in the Taklamakan desert, we use the UV aerosol index (UVAI hereafter) derived from the OMPS (Ozone Mapping Profiler Suite) on board the Suomi-NPP (National Polar-orbiting Partnership) satellite . OMPS provides full daily coverage data, and the overpass time for the Kashi region is around 06:30 UTC. The UVAI is calculated using the signal in the 340 and 380 nm channels . 3UVAI=-100×log⁡10I340I380meas-log⁡10I340I380calc, where I340 and I380 represent the backscattered radiance at 340 and 380 nm. The subscripts “meas” and “calc” respectively represent the real measurements and model simulation in a pure Rayleigh atmosphere. By the definition of UVAI, its positive values correspond to UV-absorptive aerosols such as desert dust and carbonaceous aerosols. Hence, the UVAI from OMPS is a good parameter for monitoring the activity in the Taklamakan desert.

A radiosonde station (39.47∘ N, 75.99∘ N) in Kashi is 6 km from the observation site. The data are accessible on the website of the Wyoming weather data website (see http://www.weather.uwyo.edu/upperair/sounding.html, last access: 1 April 2020). The radio sounding data are recorded at 00:00 and 12:00 local time every day. They provide the vertical temperature and pressure profiles for the calculation of molecule scattering parameters in lidar data processing. The HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory; ) model developed by the National Oceanic and Atmospheric Administration (NOAA) Air Resources Laboratory is used for the back trajectory of the air mass and for the air mass clustering. The HYSPLIT model is driven by the 0.5∘ gridded GDAS (Global Data Assimilation System) data and could produce the transport pathways of the air mass at different vertical levels.

Moreover, instruments measuring particulate matter (PM10 and PM2.5), gas concentration (SO2, O3 and NOx), particle size distribution, particle scattering and absorption coefficients, and solar radiation and a cloud monitor are also deployed in the field campaign. These data contribute to relevant air quality and solar radiation studies within the frame of the DAO campaign.

Dust plumes over the Taklamakan desert were detected on 7, 8 and 9 April, as shown in Fig. . The most intense plume in the 3 d appeared on 7 April with a maximum UVAI of about 4.0. On 9 April, a belt-like plume appeared in the north and northwest of the desert. Figure b shows the range-corrected lidar signal at 532 nm collected between 9 and 10 April. The boundary layer height slightly increases from 3000 to 4000 m during the night, and a strong backscattered lidar signal is seen below 2000 m.

Figure  shows the profiles of the optical properties, WVMR and RH averaged between 17:00 and 22:00 UTC, 9 April 2019. The extinction coefficients gradually decrease with height. At 1000 m, the extinction coefficients are greater than 0.5 km-1 and remain almost stable below 2000 m. The RH is no more than 40 ± 8 % below 2000 m and rises to 60 ± 12 % at 3800 m. The lidar ratio varies between 40 ± 6 and 48 ± 7 sr at 532 nm and between 55 ± 8 and 62 ± 9 sr at 355 nm. The PLDR is about 0.32 ± 0.05 at 355 nm and 0.36 ± 0.05 at 532 nm. The VLDR at 1064 nm is about 0.32 ± 0.03. The backscatter coefficient, as well as the PLDR, at 1064 nm is not available above 1800 m since the 1064 nm lidar signal was distorted in the upper boundary layer. We can expect that the VLDR is approximate to PLDR at 1064 nm in this case because the dust content is so high that molecular scattering at 1064 nm can be neglected. The EAE355-532 is about -0.10 ± 0.30 at 800 m and rises to 0.10 ± 0.30 at 3800 m. The BAE355-532 (Ångström exponent related to backscatter coefficient) is negative and varies from -0.7 ± 0.3 to -0.4 ± 0.3. Below 3000 m, the lidar ratios mildly decrease with height, while the PLDRs do not show obvious vertical variations. Above 3000 m, the vertical variations in the lidar ratios and PLDRs become more significant. The vertical variations in the lidar ratios and PLDRs are possibly the result of particle sedimentation and/or vertically dependent particle origins.

On 9 April, the Taklamakan desert was covered by a low-pressure zone with easterly and northeasterly winds prevailing over the western part of the desert. It is a favorable condition for the elevation of dust particles. Figure  shows the 48 h back trajectory ending at 20:00 UTC for air masses at 1000, 2000 and 3000 m. The air masses at the three vertical levels originated from the Taklamakan desert. They all passed over the area where dust plumes have been observed and then diverged when approaching the rim of the desert. In the end, the air masses at 1000, 2000 and 3000 m arrived at the observation site from the northeast, east and southeast, respectively. The particles observed by LILAS on 9 April are fresh desert dust without long-range transport. Therefore, they could contain a large fraction of coarse-mode particles, especially giant particles (radius > 20 µm). Moreover, the back trajectories in Fig.  show convective strong air flows arising from below 500 to 3000 m within 3 h, suggesting the possibility of large particles near the surface being lifted to higher levels.

On 24 April, the observation site was enclosed by floating dust. In the daytime, the sky radiance dropped below the detection limit of the sun–sky photometer, so the AERONET and SONET retrievals could not be applied. A large and intense plume was first detected in the morning of 23 April 2019 (Fig. ). On 24 April, a hot spot of UVAI appeared over the observation site. The daily average of AOD is 3.63 and the Ångström exponent is about -0.01 according to the daytime sun–sky photometer measurements. The lidar quicklook on 24 April in Fig.  shows that the boundary layer height rises from about 1200 to 2000 m from 14:00 to 24:00 UTC. Due to the high dust attenuation in the boundary layer, both sun–sky photometer and lidar can not detect whether clouds existed on 24 April.

Figure  plots the averaged parameters between 15:00 and 24:00 UTC, 24 April 2019. The dust layer was so thick that the laser beam could not penetrate it. The amplitude of the Raman signal dropped by 5–6 orders of magnitude in the lower 2000 m. In this condition, we could not find an aerosol-free zone for the calibration of lidar signal; therefore, the calculation of the backscatter coefficient using Raman method is not possible, but the extinction coefficient can be derived from the Raman signal . The extinction coefficients are 1.0 ± 0.1 km-1 at 800 m and increase to about 1.5 ± 0.2 km-1 at 1500 m. The extinction coefficient at 355 nm is removed above 1500 m because it starts to oscillate due to an insufficient signal-to-noise ratio. The extinction coefficient at 532 nm decreases to about 1.1 ± 0.1 km-1 at 2000 m. By assuming that the lidar ratios are about 55 and 45 sr at 355 and 532 nm, respectively, we obtain the backscatter coefficient from the extinction coefficient and then calculate the PLDRs (in Fig. c). The PLDR is about 0.32 at 355 nm and 0.37 at 532 nm, which are rather consistent with the results in Case 1. The uncertainties of the PLDRs are not accessible because the uncertainties of the assumption of the lidar ratio are not known.

The back trajectories (not shown) indicate that dust particles (at 1000 and 2000 m) originated from the northeast and east where intense dust plumes were observed on 23 and 24 April. Figure shows synoptic conditions at 00:00 UTC, 23 April, and 06:00 UTC, 24 April. The meteorological conditions on 23 and 24 April are favorable for dust emission, similar to Case 1. The Taklamakan desert is enclosed by a low-pressure zone (Fig. a and c). The plume observed by OMPS on 23 April was probably lofted during the local morning. In the eastern part of the Taklamakan desert (37–39∘ N, 83–88∘ E), the wind velocity at 10 m above ground level reaches more than 50 km h-1 (Fig. a) and at 850 hPa level the maximum wind velocity reaches 90 km h-1 (Fig. b). The high wind velocity near the surface and large vertical wind gradient help elevate dust particles from the surface into the atmosphere. On 23 and 24 April, easterly and northeasterly wind were prevalent in the desert region, thus blowing the lifted dust particles to the observation site. Case 2 is a more severe manifestation of Case 1 regarding the intensity of the dust loading. In both cases, the observed dust particles originated from nearby dust source. Compared to typical depolarization ratios in worldwide dust observations, which are 0.23–0.30 at 355 nm and 0.30–0.35 at 532 nm, the depolarization ratios we obtained in the two cases are relatively higher . While previous observation sites were mostly not as close to the dust source as in our campaign, the differences are probably due to the fraction of coarse-mode particles that remain in our dust observation. also observed dust particles near the source in North America and reported PLDR of 0.37 at 532 nm, which is consistent with our result. However, the PLDR at 355 nm measured by is about 0.24, which is lower than what we obtained.

On 15 April, the daily mean AOD on 15 April was 0.63, and the Ångström exponent was about 0.10. Compared to the previous two cases, the Ångström exponent increases obviously. The boundary layer height started to increase at 15:00 UTC and stayed at 3500 m during the night of 15 April. Cirrus clouds were continuously present during the period of lidar measurement (Fig. c). The lidar-derived profiles between 18:00 and 20:00 UTC are plotted in Fig. . The extinction coefficients in the boundary layer are about 0.15 km-1 and decrease to almost zero at 3500 m. The RH increases up to 60 % ± 12 % at 2200 m. Below this height, the lidar ratio, PLDR, EAE and BAE are almost stable. The lidar ratio is about 51 ± 8 sr at 355 nm and 45 ± 7 sr at 532 nm. The PLDRs at 355, 532 and 1064 nm are around 0.32 ± 0.07, 0.34 ± 0.05 and 0.31 ± 0.05, respectively. The EAE355-532 is about 0.02 ± 0.30, showing a gentle increase with height, and the BAE355-532 is about -0.29± 0.30. Above 2200 m, the RH starts to increase and reaches its maximum, i.e., 80 % ± 16 %, at 2800 m. The EAE355-532 and BAE355-532 increase to 0.10 ± 0.30 and -0.06 ± 0.30, respectively. In contrast, the lidar ratios and PLDRs decrease and reach their minima at about 2800 m. The lidar ratio is about 40 ± 6 sr at 2400–2800 m with a weak spectral dependence, and the PLDRs are about 0.23 ± 0.06 at 355 nm, 0.26 ± 0.04 at 532 nm and 0.24 ± 0.03 at 1064 nm. It should be noted that the backscatter coefficient at 1064 nm is performed using the Klett method with an assumption of the lidar ratio equal 40 sr .

Dust activities were observed by the OMPS on 13 and 15 April 2019 (Fig. ), while the intensity was less strong than in cases 1 and 2, and the distance between the dust plume and the observation site was farther. The 48 h back trajectories ending at 19:00 UTC are shown in Fig. . Air masses at the three vertical levels (1000, 2000 and 3000 m) originated from the eastern part of the Taklamakan desert, where no intense dust activities had been observed by OMPS in the recent 3 d. It explains the decrease in dust content in the boundary layer. When dust loading decreases, the impact of fine mode particles emerges. The changes in EAE, BAE, PLDR and lidar ratios above 2200 m are clear evidence of polluted dust. The pollution could be lifted up from the ground in the local area by convection or be transported from another area. Additionally, the RH above 2500 m is about 60 % ± 12 %–80 % ± 16 %, which could lead to the hygroscopic growth of some aerosol species. Pure dust is regarded as hydrophobic aerosols because its compounds are insoluble, but when mixed with hygroscopic aerosol species, for example, nitrate, the ensemble of aerosol mixture could become hygroscopic. The fine mode particles can be hydrophobic or hydroscopic depending on their chemical compositions . In this case, there is no clear evidence indicating the occurrence of hygroscopic growth or the mixing state of dust and pollution particles.

The daily mean AOD on 3 April is 0.16, and the Ångström exponent is about 0.11. The boundary layer height is about 3000 to 4000 m, rising slightly during the night of 3–4 April. Starting from 16:30 UTC, some liquid cloud layers occurred at the top of the boundary layer (Fig. a). Figure  shows the profiles derived from lidar observations at 14:00–16:00 UTC, 3 April. The extinction coefficients decrease from about 0.28 ± 0.03 km-1 at 1000 m to about 0.10 ± 0.01 km-1 at 3000 m, with EAE355-532 (BAE355-532) increasing from 0.01 ± 0.30 (-0.38 ± 0.3) to 0.28 ± 0.30 (0.02 ± 0.30). Below 2100 m, the lidar ratios are about 45 ± 7 sr at 532 nm and 51 ± 8 sr at 355 nm. The PLDRs are about 0.35 ± 0.05 at 532 nm and 0.32 ± 0.05 at 1064 nm and 0.28–0.32 ± 0.07 at 355 nm. Between 2100 and 3000 m, the variation in lidar ratios is not monotonic. At 2500 m, the lidar ratios reach the minimum of 38±6 sr at 532 nm and 42±6 sr at 355 nm, and the PLDRs are about 0.27 ± 0.06 at 355 nm and 0.33 ± 0.05 at 532 nm. Both the lidar ratios and PLDRs in the 2400–2800 m range are very consistent with the properties of Central Asian dust reported by and . Above 2500 m, the lidar ratios and the RH (as well as WVMR) increase again, and PLDRs decrease. At 3000 m, the PLDR at 532 nm drops below 0.30, suggesting that aerosols are different from those in the lower boundary layer. The signatures of lidar ratio, RH and PLDR are possibly linked to the occurrence of polluted dust particles. In addition, long-range transported dust could also possess such PLDRs due to the deposition of big particles in the transport. This case is classified as polluted dust because the PLDR below 0.30 at 532 nm and the increase in WVMR (as well as RH) and EAE355-532 at the boundary layer top fit the characteristics of polluted dust better.

Figure  plots the 72 h back trajectories for 1000, 2000 and 3000 m. Air masses at 1000 and 2000 m are from the Taklamakan desert, while the air mass at 3000 m is from Central Asia. It corroborates the similarities of the lidar ratios and PLDRs between the measurements in Hofer's studies and in our study. When extending the trajectory duration to 96 h, the results (not plotted) suggest that the air mass at 3000 m originated from northern Africa. This result suggests that dust particles observed in Kashi may have a long-transport aerosol component. The air mass clustering based on 24 h back trajectories (not shown) indicates that about 52 % of the air mass at 3000 m is from northern Africa and the Arabian Peninsula. At 3500 m, this proportion increases to 74 %, and there is also a fraction of an air mass coming from Europe. The complexity in the aerosol sources in the transport pathways explains the variability of aerosol properties in the upper boundary layer in Case 4.