Physical Mechanisms of Deep Convective Boundary Layer Leading to Dust Emission in the Taklimakan Desert

Deserts play an important role in the climate system, which is closely associated with the emission and transport of dust aerosols. Based on the intensive observation experiment in the Taklimakan Desert, the potential physical processes between the deep convective boundary layer (CBL) and dust emission are revealed in this study. Deep CBL enables the formation of clouds in the late afternoon, leading to significant cooling of surface. Large‐scale buoyant coherent structures thereby transform into the mechanical coherent structures confined near the surface. The responses promote the earlier occurrence of low‐level jet (LLJ) than in cloudless conditions, which allows the downward transport of LLJ momentum and substantially increases surface wind. Therefore, dust emission is initiated by strong wind at dusk and lasts for several hours. The results are useful to predict dust emissions and improve our understanding of distinctive boundary‐layer processes in desert regions.


Introduction
The Taklimakan Desert, the second-largest shifting sand desert in the world, is both highly influencing and highly sensitive to climate change (Huang et al., 2014(Huang et al., , 2017)).As an important dust source in Asia, the Taklimakan Desert directly impacts the local/regional air quality, radiative energy budget, and biogeochemical cycles via emissions, long-range transport, and deposition of dust aerosols (Chen et al., 2017;Ge et al., 2014;Huang et al., 2009).Furthermore, through coupling with adjacent Tibetan Plateau, the Taklimakan Desert and its dust aerosols have significant implications for downstream precipitation and regional climate (Huang et al., 2023;Liu et al., 2020Liu et al., , 2022)).These imply that clearly understanding atmospheric physical processes in the Taklimakan Desert, particularly the various mechanisms of dust emissions and transport, is crucial for weather forecasting and climate prediction.
Turbulence is responsible for the exchange of momentum, heat, and substances (e.g., dust particles) between the underlying surface and the atmosphere (Santanello et al., 2018).However, intense solar heating makes turbulent structures and boundary-layer processes rather distinctive in desert regions (Garcia-Carreras et al., 2015;M. Wang et al., 2019;L. Zhang et al., 2024).Specially, deep convective boundary layer (CBL), roughly 3,000-5,000 m, is frequently formed in summer with the role of detrainment of the warmest thermals (L.Zhang Supporting Information may be found in the online version of this article.et al., 2022).Considering that the deep CBL and dust activities mentioned above are two important characteristics of deserts, one might wonder if there are physical connections between them.The potential evidence is that the deep CBL developed in the afternoon is followed by dust emissions at dusk for several days during a 1-month intensive observation.Moreover, the deep CBL is often accompanied by the low-level jet (LLJ) on the following night (M.Wang et al., 2019), while the formation of LLJs is an important mechanism to generate strong surface winds for dust emissions (Fiedler et al., 2013;Ge et al., 2016).All these findings suggest a potential connection, while the physical mechanism remains unclear, which limits corresponding representation in numerical weather models.
This study aims to reveal potential physical processes connecting deep CBL and dust emission based on 1-month intensive observations in the hinterland of Taklimakan Desert.It involves the development of deep CBL, the formation of late-afternoon clouds, the transformation of turbulent coherent structures, the occurrence of LLJs, and finally the dust emission at dusk.Describing this physical picture will enrich our understanding of dust emission mechanisms and promote the prediction of dust emission in the desert.

Observations
An intensive atmospheric boundary layer experiment was conducted from 1 to 31 May 2022 at Tazhong meteorological station (39°00' N, 83°40' E), situated in the hinterland of the Taklimakan Desert.The station is surrounded by shifting sand and dunes (Yang et al., 2021).The intensive observations include an 80-m tower, a ceilometer, and the Global Positioning System (GPS) radiosonde.Mean temperature, humidity, and wind were measured at 10 levels (0.5, 1, 2, 4, 10, 20, 32, 47, 63, and 80 m) on the tower.Sonic anemometer (CSAT-3, Campbell Scientific, Inc., USA) turbulence measurements, that is, 20-Hz wind components and temperature, were obtained at 10 m on the tower.Radiation components were measured using radiometers (Hukseflux, Netherlands) above the surface.The ceilometer (CHM 15k, Lufft, Inc., Germany) receives backscatter signals of clouds and aerosols from the zenith direction, with a spatiotemporal resolution of 15 m × 15 s.Its laser wavelength is 1,064 nm.The applications of ceilometer on dust aerosols and clouds have been well confirmed (Bi et al., 2022;L. Zhanget al., 2022).GPS radiosondes were launched at 05:15, 11:15, 17:15, and 23:15 LST (UTC+6 hr) during the observation period, to obtain the profile of temperature, humidity, pressure, wind speed, and wind direction.This study focuses on the cases that the deep CBL is followed by dust emissions at dusk, and thereby 3 days (i.e., 16 May, 27 May, and 28 May) were selected for the following analysis.
The raw data observed from the sonic anemometer were processed over 30-min intervals using EddyPro v7.0.6 (LI-COR Inc., USA) software, to obtain fluctuations and turbulent statistics (L.Zhang et al., 2023).The depth of CBL was determined based on GPS radiosondes and ceilometer observations, respectively.For the former, the top of CBL is identified as the peak height of potential temperature gradient (Q.Li et al., 2021;L. Zhang et al., 2022), which is more accurate but only available at 11:15, 17:15 LST.For the latter, the top of CBL is defined as the height where the normalized range-corrected signals rapidly decrease, accompanied by increased noise (Bi et al., 2022).This approach is available throughout the fair-weather daytime, but for the cases of elevated aerosol layer or multi-layer aerosol structure, the empirical judgments based on the evolution rule of the CBL are needed.

Lifting Condensation Level and Its Estimation
When the unsaturated air parcel is adiabatically lifted, it follows the dry adiabatic process.The potential temperature and specific humidity of air parcel remain constant with height, while parcel temperature decreases at the dry adiabatic lapse rate, more rapidly than the lapse rate of dewpoint (Iribarne & Cho, 1980).The cooling effect leads the parcel to saturation at a given level, which is called the lifting condensation level (LCL).It is also considered as the level of cloud base.Under fair-weather conditions in the Taklimakan Desert, the air-parcel lifting is mainly achieved by thermals, a kind of large-scale coherent structures (Williams & Hacker, 1993).The vertical extent of these structures is roughly equal to the CBL depth.In other words, the CBL top represents the level to which air parcels can be systematically lifted, while the LCL is the minimum level required for condensation.Therefore, it is expected that boundary-layer clouds form when CBL grows to the LCL (Ek & Mahrt, 1994;Westra et al., 2012).
In this study, the LCL is estimated by iterating along the dry-adiabatic lifting process, based on the conservation of potential temperature and specific humidity (see Text S1 in Supporting Information S1 for more details).The iteration is initialized by 10-m mean temperature, specific humidity, and pressure.Owing to limited water vapor content, the LCL would be very high in the Taklimakan Desert, which means that the CBL should be deep enough for cloud formation.

Quadrant Analysis
Quadrant analysis is used to identify turbulent coherent structures (Shaw et al., 1983;L. Zhang et al., 2022).There are various definitions of quadrants, and this study follows the definition of L. Zhang et al. (2023).Each quadrant and corresponding motion are prescribed as follows: Quadrant I: a′ > 0, w′ > 0, ejections Quadrant II: a′ < 0, w′ > 0, outward interactions Quadrant III: a′<0, w′<0, sweeps Quadrant IV: a′ > 0, w′ < 0, inward interactions where w′ is vertical velocity fluctuation and a′ is the fluctuation of another variable.Under unstable (stable) conditions, a′ is settled by a′ = θ′ (a′ = θ′) for sensible heat and a′ = u′ for momentum.This practice ensures quadrants I and III always represent downgradient turbulent motions, whereas quadrants II and IV are countergradient motions.For a given 30-min interval, absolute-flux contribution is quantified as where i is quadrant order (i = 1, 2, 3, 4), N is the total number of records in the 30-min interval, and I i (n) is an indicator function.If the instantaneous flux w′a′ is located in quadrant i, then The relative magnitude of absolute-flux contributions from four quadrants is indicative in distinguishing the type of coherent structures (L.Wang et al., 2014;L. Zhang et al., 2023).

CBL Development and the Formation of Boundary-Layer Clouds
Figures 1a-1c present the normalized range-corrected signals of ceilometer for three cases, respectively.The altitude of LCL and the top of CBL are superimposed.The CBL begins to form at approximately 07:00 LST and slowly deepens until it has nearly incorporated the nocturnal stable layer at 10:00 LST.Then, the CBL rapidly grows during the late morning to the early afternoon with the presence of near-neutral residual layer.Finally, the CBL reaches the maximum depth, more than 4,000 m, and maintains the depth in the late afternoon.Meanwhile, the CBL depths derived from GPS radiosondes and from ceilometer are rather consistent, suggesting the reliability of ceilometer in the CBL depth retrieval under fair-weather conditions.
The LCL starts with a relatively low altitude owing to lower temperature and higher humidity at sunrise.Thereafter, it gradually rises in the morning and maintains at higher altitudes in the afternoon, approximately 4,700, 4,200, and 4,300 m for 16 May, 27 May, and 28 May, respectively.Until late afternoon, the top of CBL crosses the LCL, which is considered as the moment boundary-layer clouds begin to form.At the same time, surface net radiation shows a drop (denoted by black arrows in Figure 1), suggesting the formation of clouds.
There are also drops in surface net radiation in earlier time, their distinction in effects will be discussed in Section 3.2.For the cases of 16 May and 27 May, boundary-layer clouds are well captured by the ceilometer, shown by the enhanced signals above the LCL after the crossing (dark red in Figures 1a and 1b).However, the clouds are not directly observed on 28 May (Figure 1c).This is because the ceilometer only emits laser and receives signals in the zenith direction, while boundary-layer clouds do not necessarily appear right above the ceilometer.Although the altitude of cloud base can be defined using the LCL, the horizontal location of clouds is rather random, which depends on the thermals (Stull, 1988).Instead, solar radiometers aggregate the information from the whole sky, so surface net radiation is very sensitive to the formation of clouds.After cloud formation, a series of responses is performed in the lower atmosphere because of the irreversible drop of surface net radiation (Sections 3.2 and 3.3).Finally, dust emission is generated at 19:15, 16:40, and 19:35 LST for 16 May, 27 May, and 28 May, respectively, indicated by intense signals below 500 m and sharp extinction above (Figures 1a-1c).
Moreover, there are noteworthy characteristics in each case.For the first case, an elevated dust layer is observed from 07:00 to 14:00 LST, with a core of 3,000 m (Figure 1a).This phenomenon is unique to the Taklimakan Desert and caused by basin topography (Ge et al., 2014;Huang et al., 2009).The elevated dust layer is finally mixed into the CBL at 15:00 LST.For the second case, there are sharply increased signals from the surface to the crossing of the LCL and CBL top, approximately between 13:10 and 13:20 LST (Figure 1b), which is caused by precipitation.For the third case, it is interesting that there are some weak clouds on the LCL, marked by cyan circles in Figure 1c, while the top of CBL is lower than the LCL.Nevertheless, the top of the residual layer, derived from the radiosonde at 11:15 LST, roughly coincides with the LCL at this time.L. Zhang et al. (2022) reported that with the joint roles of intense surface heating and near-neutral residual layer, the warmest thermals can be detrained from the CBL and overshoot the top of residual layer in the Taklimakan Desert, shown by the process numbered 5 in Figure 4.This distinctive boundary-layer process leads to the unexpected clouds at the top of residual layer.Aforementioned characteristics reflect the uniqueness of atmospheric physical processes in the Taklimakan Desert, which deserves more attention.

The Contrast on Turbulent Coherent Structures Before and After Cloud Formation
Figures 2a-2f compare the quadrant maps of turbulent fluxes before cloud formation, after cloud formation, and during dust emission, and Figures 2g-2i show their normalized power spectra of variables.Since the results are consistent among three cases, only the results of the first case are shown here and that of the other two cases can be seen in Figures S1 and S2 in Supporting Information S1.Before cloud formation, the quadrant map of sensible heat flux is characterized by the maximum of the joint probability density functions (JPDFs) located in quadrant III whereas the tail of JPDFs skewing toward quadrant I (Figure 2a).This suggests that even sweeps are numerous, ejections are stronger and dominate the transport of heat (with an absolute-flux contribution of 0.51), which is the signature of thermals in the CBL (Mahrt & Paumier, 1984).In this situation, turbulent transport efficiency, quantified by the correlation coefficient (Schmutz & Vogt, 2019), is 0.50 for sensible heat whereas only 0.20 for momentum.Additionally, the transport of momentum is equivalently contributed by ejections and sweeps, which is partially offset by the countergradient transport (Figure 2d).The transport dissimilarity between heat and momentum well reflects the spatial evolution of coherent structures.With the enhancement of thermals, convective circulation is formed and vertically extends throughout the CBL.The horizontal divergence/ convergence of convective circulation near the surface disturbs horizontal wind (i.e., the process numbered 6 in Figure 4) and thereby reduces turbulent transport efficiency of momentum (L.Zhang et al., 2023).Moreover, u spectrum peaks at the low-frequency range (Figure 2g), toward the horizontal scale of convective circulations (Kaimal et al., 1976).Therefore, boundary-layer-scale coherent structures (i.e., thermals and convective circulations) prevail and thus the whole CBL is closely communicated with the surface before the formation of clouds.
After cloud formation, surface net radiation drops to negative values (Figure 1d).Boundary-layer-scale coherent structures are difficult to maintain due to surface cooling.The signature of thermals disappears in the quadrant map of sensible heat flux, instead by more symmetric both JPDFs and absolute-flux contribution between ejections and sweeps (Figure 2b).This becomes similar to that of momentum (Figure 2e).The symmetry suggests the role of mechanical coherent structures, for example, hairpin vortices and packets (D.Li & Bou-Zeid, 2011;L. Zhang et al., 2023).Specially, w spectrum indicates that dominant eddies scales with the observation height, that is, f max z/ U ≈ 1 where the wind speed U is 5 m/s, further confirming turbulence is generated by bulk shear at this time (Sun et al., 2012).Correspondingly, the peak of u spectrum shifts toward higher frequency, and θ spectrum shows a cave-in at middle-frequency range (Figures 2g and 2h).Therefore, as buoyant coherent structures at boundary-layer scale are replaced by mechanical coherent structures in the surface layer, only the lower layer is directly related to the surface whereas the upper layer is gradually released from the turbulent coupling.
The above results emphasize that, the time of boundary-layer cloud emergence is critical for the transformation of coherent structures before dust emission.For example, for the clouds formed in the early afternoon, although there is a drop in the surface net radiation, it will return to its original values after the clouds have been advected away (Figure 1d).Turbulent coherent structures cannot transform in this situation.Only when boundary-layer clouds form at approaching sunset, the decrease in surface net radiation is irreversible.Thus, surface cooling is strengthened and subsequently the transformation of coherent structures occurs.Furthermore, mechanical coherent structures are enhanced during dust emission, shown by the increased transport efficiency of momentum (Figure 2f).Intense turbulent mixing substantially reduces the vertical gradient of temperature in the surface layer, leading to near-neutral conditions (Sun et al., 2012).Hence, turbulent transport efficiency of heat decreases to approximately zero and even is controlled by countergradient transport (Figure 2c).Meanwhile, θ spectrum collapses, with notable noise in the high-frequency range (Figure 2h).This enhanced mechanical turbulence provides sufficient dynamic for dust emission, so dust emission can last more than 5 hr (Figures 1a-1c).

The Occurrence of LLJ and Subsequent Dust Emission
Figures 3a and 3b show the evolution of temperature and wind speed profiles on the 80-m tower before dust emission.Temperature increasingly decreases after cloud formation, with the most remarkable response near the surface; Conversely, the response of wind speed starts from a higher level, showing continuous increases with time (denoted by black arrows).The potential linkage between the two can be explained by the inertial oscillations (Blackadar, 1957).With the cooling of surface, buoyant coherent structures at boundary-layer scale transform into mechanical coherent structures confined in the lower layer (Section 3.2).The upper layer gradually releases from the surface frictional restraint, and the wind therein undergoes an undamped oscillation around the geostrophic wind, starting with an increase (Blackadar, 1957;Van de Wiel et al., 2010).Nevertheless, the wind speed in the lower layer is limited by frictional surface.Consequently, wind maxima appear at the junction of the two layers, constituting the LLJs.The occurrence of LLJs is demonstrated by radiosondes in all three cases, with a jet core of 400-600 m (Figure 3c).The downward transport of LLJ momentum causes the wind speed to increase sequentially from top to bottom (Figure 3b).
It is noteworthy that the timing of LLJ formation is also crucial for dust emission.Actually, LLJs are frequently observed at 23:15/05:15 LST during the intensive observation period.However, the formation of LLJs is significantly advanced for the cases of dust emission at dusk, for example, the LLJ appearing at 17:15 LST on 27 May (Figure 3c).In other words, even without the presence of late-afternoon clouds, LLJs can still form in the Taklimakan Desert, but is later than midnight because the natural surface cooling is relatively slow (Ge et al., 2016).The downward transfer of LLJ momentum would be largely blocked in cloudless conditions, because turbulence almost disappears at fair-weather midnight (L.Zhang et al., 2024).In contrast, for the cases of cloud formation in the late afternoon, surface cooling is accelerated, and the formation of LLJs would be advanced to the time when turbulent momentum transfer is efficient (Section 3.2), which is an important condition for dust emission.
The hodographs at the top of 80-m tower clearly show that within 30 min after cloud formation, winds begin to rotate clockwise (Figures 3d-3f), which reflects the development of LLJs (Van de Wiel et al., 2010).As the momentum of advanced LLJs is transported downwards by mechanical coherent structures (Figures 2e and 2f), the wind speed near the surface rapidly increases and exceeds the threshold (10 m/s at 80 m, see Text S2 in Supporting Information S1 for details).Therefore, dust emission is eventually triggered, corresponding to the sharply enhanced signals at lower levels in Figures 1a-1c.
A large number of dust particles are emitted into the atmosphere, marking the beginning of their weather and climate effects.Thereafter, smaller particles remain suspended during the night.In the next daytime, thermal convections will mix suspended particles vertically and transport them aloft, similar to that of 16 May (Figure 1a).Once dust aerosols are uplifted above 4,000 m, they can be entrained in westerlies and begin a long-term transport (Chen et al., 2017), leading to weather/climate implications for East Asia and wider regions (e.g., Huang et al., 2014;Liu et al., 2020).Furthermore, based on ceilometer observation, the emission mechanism revealed in this study is associated with 37.5% of dust events in the Taklamakan Desert from May to August 2022 (Table S1 in Supporting Information S1).These well demonstrate the significance and representativeness of this emission mechanism in the Taklamakan Desert.

Conclusions
The potential physical processes between the deep CBL and subsequent dust emission are investigated in this study, based on the intensive atmospheric boundary layer experiment conducted in May 2022 in the hinterland of Taklimakan Desert.As shown in Figure 4, initially well-developed deep CBL enables the formation of clouds in the late afternoon even in the desert.After cloud formation, surface becomes cooling.Buoyant coherent structures at boundary-layer scale disintegrate, and mechanical coherent structures become dominant.Meanwhile, LLJs start to develop.As the momentum of LLJs is transported downwards, surface wind substantially increases, which prepares the dynamic for dust emission.Eventually, dust emission is initiated at dusk when wind speed exceeds the threshold, and lasts for several hours.It is noteworthy that the timing of cloud formation and LLJ occurrence is crucial during the processes.Too early clouds or too late LLJs are ineffective for dust emission since lacking of joint role of special turbulence.The results in this study can enrich our understanding of dust emission mechanisms and are of great significance to reasonably represent distinctive boundary-layer processes of desert (i.e., key climate region) in numerical weather models.

•
Deep convective boundary layer enables the lifting condensation of moisture and the formation of boundary-layer clouds in desert regions • The emergence of late-afternoon clouds strengthens the surface cooling and thus advances the development of low-level jet • Downward momentum transfer from low-level jet to near-surface wind is crucial to initiate dust emission at dusk Supporting Information:

Figure 1 .
Figure 1.Evolution of CBL characteristics and surface net radiation for three cases.(a-c), normalized range-corrected signals in logarithmic scale measured by the ceilometer, log 10 (RCS), superimposed by the time series of the LCL (blue-dot line), the top of CBL derived from GPS radiosondes (magenta diamonds) and from the ceilometer (black crosses), and the top of residual layer (blue squares), respectively.(d), the time series of surface net radiation, in which there are vertical shifts to avoid overlapping.Black arrows indicate the moment when surface net radiation drops for each case.Cyan circles mark the clouds caused by the detrainment of the warmest thermals.

Figure 2 .
Figure 2. Characteristics of turbulent coherent structures before cloud formation, after cloud formation, and during dust emission, which are sampled from the segments 16:30-17:00, 18:00-18:30, and 19:30-20:00 LST, respectively, on 16 May 2022.(a-c), quadrant maps of sensible heat flux before cloud formation (a), after cloud formation (b), and during dust emission (c).The joint probability density functions JPDFs (contour) are superposed on the instantaneous fluxes (gray scatters).The red numbers at corners of panels denote the absolute-flux contribution of each quadrant, and the numbers at the top of panels indicate turbulent transport efficiency.The coordinate is normalized by standard deviations.(d-f), same as (a-c) but for momentum flux.(g-i), normalized power spectra of streamwise velocity u (g), potential temperature θ (h), and vertical velocity w (i).Gray oblique lines plot the slope of 2/3.

Figure 3 .
Figure 3. Response of temperature and wind speed to the formation of late-afternoon clouds.(a and b), typical profiles of mean temperature (a) and wind speed (b) on the 80-m tower before dust emission, measured on 16 May 2022.(c), profiles of wind speed showing the presence of LLJs for each case, observed from the GPS radiosonde.(d-f), hodographs at 80 m after cloud formation on 16 May (c), 27 May (d), and 28 May (e), respectively.Each symbol represents a 30-min value, and corresponding hour is marked by Arabic numerals.Black dashed arcs indicate the wind speed threshold (i.e., 10 m/s at 80 m) for dust emissions in the Taklimakan Desert.

Figure 4 .
Figure 4. Schematic describing the physical processes associated with the development of CBL and the initiation of dust emission at dusk in the Taklimakan Desert, including (1) the formation of boundary-layer clouds in the late afternoon and enhanced surface cooling, (2) mechanical coherent structures replacing the buoyant ones (e.g., thermals and convective circulations), (3) the occurrence of LLJs and downward momentum transfer, and finally (4) dust emission at dusk.There are also (5) weak clouds induced by the detrainment of the warmest thermals and (6) horizontal convergence/divergence generated by convective circulations.