Coupled Aerosol-Cloud Systems over Northern Vietnam during 7-SEAS / BASELInE : A Radar and Modeling Perspective

The 2013 7-SEAS/BASELInE campaign over northern Southeast Asia (SEA) provided, for the first time ever, comprehensive ground-based W-band radar measurements of the low-level stratocumulus (Sc) systems that often exist during the spring over northern Vietnam in the presence of biomass-burning aerosols. Although spatially limited, groundbased remote sensing observations are generally free of the surface contamination and signal attenuation effects that often hinder space-borne measurements of these low-level cloud systems. Such observations permit detailed measurements of structures and lifecycles of these clouds as part of a broader effort to study potential impacts of these coupled aerosolcloud systems on local and regional weather and air quality. Introductory analyses of the W-band radar data show these Sc systems generally follow a diurnal cycle, with peak occurrences during the nighttime and early morning hours, often accompanied by light precipitation. Preliminary results from idealized simulations of Sc development over land based on the observations reveal the familiar response of increased numbers and smaller sizes of cloud droplets, along with suppressed drizzle formation, as aerosol concentrations increase. Slight reductions in simulated W-band reflectivity values also are seen with increasing aerosol concentrations and result primarily from decreased droplet sizes. As precipitation can play a large role in removing aerosol from the atmosphere, and thereby improving air quality locally, quantifying feedbacks between aerosols and cloud systems over this region are essential, particularly given the negative impacts of biomass burning on human health in SEA. Such an endeavor should involve improved modeling capabilities along with comprehensive measurements of time-dependent aerosol and cloud profiles.


BACKGROUND AND MOTIVATION
Aerosols are omnipresent in Earth's atmosphere and central to many environmental concerns such as air quality and climate change.During boreal spring in Southeast Asia (SEA), biomass-burning from natural forest fires and slashand-burn agricultural practices strongly modulates the atmospheric composition over northern SEA (Lin et al., 2013;Reid et al., 2013, Tsay et al., 2016, andreferences therein).The chemical and aerosol species associated with the burning degrade the regional air quality and trigger adverse health effects in the local population (cf.., Huang et al., 2013;Lin et al., 2014), and can additionally alter the lifecycles of clouds and distribution of precipitation through radiative and microphysical effects (Graf et al., 2009;Lee et al., 2014).
Satellite and ground-based observations of biomassburning (BB) aerosols over SEA exhibit consistent spatiotemporal distribution patterns, with considerable variability due primarily to factors ranging from large-scale climatic influences (e.g., ENSO) to small-scale meteorological events (Gautam et al., 2013;Reid et al., 2013;Sayer et al., 2016).Downwind from source regions, the transported and local BB aerosols often overlap with low-level stratocumulus (Sc) cloud decks (Hsu et al., 2003;Tsay et al., 2013) associated with the development of the region's pre-monsoon (February-April) cloud and rain system (Fig. 1(a)).Analyses of Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation/Cloud-Aerosol Lidar with Orthogonal Polarization (CALIPSO/CALIOP) vertical feature mask (Winker et al., 2009) Version 3 data for the months of March and April from 2007-2014 depict the vertical frequency distribution of aerosols and clouds within a 2 × 2° box over northern Vietnam (Fig. 1(b)).A persistent aerosol layer peaking near 3 km above sea level is evident, whereas cloud top heights range mainly between 2 and 3 km.Increased occurrences of low-level cloudiness and an apparent overlap of the cloud layer top with the aerosol layer at night compared to daytime are also evident for both months and imply a diurnal cycle.However, significant attenuation of the lidar signal by clouds and heavy aerosol loading can lead to biases in feature detection capability in the boundary layer (Winker et al., 2009).
Biomass-burning aerosols have also been recognized as effective cloud condensation nuclei (CCN, e.g., Crutzen and Andreae, 1990;Reid et al., 2005;Engelhart et al., 2012).They are particularly active at lower supersaturation levels, such as those occurring in Sc clouds, where aerosol-cloud interactions are most favored (e.g., Platnick and Twomey, 1994;Sorooshian et al., 2009).Cloud precipitation processes (i.e., droplet self-collection, collision-coalescence) are modulated by CCN size, and to a lesser extent composition (Dusek et al., 2006;Reutter et al., 2009;Feingold et al., 2013).In general, the impacts of increasing aerosol/CCN concentrations on clouds tend to decrease cloud particle size and increase particle number, and thereby potentially suppress precipitation (Twomey, 1977;Albrecht, 1989;Ackerman et al., 2004).However, BB aerosols located within the boundary layer over land may increase convective cloudiness associated with aerosol-induced boundary layer warming and could lead to stronger convection and greater rainfall rates (Feingold et al., 2005;Martins et al., 2009).
Aerosol impacts on low-level Sc over land remain unclear given that far less attention has been paid to these clouds compared to marine Sc (cf.Klein and Hartmann, 1993;Lin et al., 2014), presumably given the critical role of the latter in modulating radiant energy and latent heat that affect weather and climate (Stephens and Greenwald, 1991;Lohmann and Feichter, 2005;Wood, 2012).Indeed, the semi-persistent low-level Sc systems that exist over northern Vietnam from November through April offer a unique opportunity to investigate a myriad of possible feedbacks between these cloud systems and BB aerosols.During winter and early spring, low-level Sc cloud layers consistently form over the South China Sea/East Sea and Gulf of Tonkin under conditions of broad subsidence and lower-tropospheric stability associated with a local Hadley circulation (Klein and Hartmann, 1993).The cloud layers often extend into southeast China and northern Vietnam as the prevailing low-level flow southwest of the East Asian anticyclone advects moisture inland beneath a strong inversion.Drizzle from these clouds frequently occurs over north and north-central Vietnam from mid-winter into the premonsoon months (Nguyen and Nguyen, 2004).Decreasing trends in rainfall over the past 50+ years for this region (Fig. 2) are consistent with recent regional climate trends (Endo et al., 2009;Nguyen et al., 2010;Ho et al., 2011;Phan et al., 2014;Schmidt-Thomé et al., 2015), and it is believed that increased BB aerosol loading during the spring may be partially responsible.The removal of atmospheric aerosols by nucleation and precipitation scavenging processes mitigate local and regional air quality in addition to potential  effects on cloud and precipitation evolution downstream (Pruppacher and Klett, 1997;Levin and Cotton, 2009).Thus, it is probable that if precipitation suppression due to enhanced aerosol loading occurs over northern SEA, this could act to exacerbate the poor air quality conditions often observed over the region during the spring.
A better understanding of aerosol-cloud-precipitation interactions over SEA naturally entails detailed observations of the cloud systems in the region.Multi-wavelength radars, and cloud radars in particular, present the most relevant platforms for obtaining comprehensive 3-D cloud structure, as well as composition and evolution (cf.Clothiaux et al., 1995;Baab et al., 1999;Kollias et al., 2007, Tridon et al., 2013 andreferences therein).The Cloud Profiling Radar (CPR) aboard CloudSat (Stephens et al., 2002), for example, has a vertical resolution of about 500 m (240 m for oversampled bin centers) and is the first space-borne radar to measure the vertical structure of clouds and light precipitation.However, the most significant difficulty with the CloudSat/CPR is that surface clutter effectively reduces the radar sensitivity such that detection of moderate drizzle over oceans is limited to ~720 m above the surface (Stephens et al., 2008).Over land, surface clutter effects are not as easily removed resulting in greater uncertainty in detecting clouds and drizzle at lowlevels (Marchand et al., 2008;Christensen et al., 2013).Therefore, ground-based systems continue to provide key observations of clouds and precipitation over the lowest levels of the atmosphere (e.g., Feingold et al., 2003).
This paper provides a preliminary overview of the unique radar observations of cloud and precipitation systems made As these observations are the first of their kind for this region, a critical first step to understanding the potential impacts of aerosols on these cloud systems is to analyze the cloud systems themselves.Initial results from numerical cloud simulations with advanced microphysics examining aerosolcloud interactions over this region are also presented.

OBSERVATIONAL ACTIVITIES
The ACHIEVE mobile laboratory operated at the regional meteorological station in Yen Bai, Vietnam (21.705°N, 104.87°E) from 25 March through 10 April 2013.Yen Bai is situated roughly 50 m above sea level and approximately 120 km west-northwest of Hanoi in northern Vietnam (Fig. 1).A small mountain range oriented approximately SE-NW with elevations ranging from about 1.5 to 2.5 km is located roughly 50 km to the west.The site lies within a BB aerosolcloud confluence region during the pre-monsoon months (cf.Tsay et al., 2013), and is thus a prime location for observing aerosols and clouds in an effort to better understand the fundamentals of their interactions.
The initial goal was to have ACHIEVE operating at Yen Bai beginning around 1 March 2013 and continue through early April to coincide with the transition period from dry to pre-monsoon season, as well as the onset of peak burning season over western SEA.However, storms in the Pacific Ocean delayed ACHIEVE's arrival in Vietnam by several weeks.Once ACHIEVE was set up at Yen Bai, power to the site was frequently interrupted due to ongoing upgrades of the local electrical transmission system resulting in data gaps for all instruments.Despite these setbacks, this inaugural deployment of ACHIEVE was a great success, capturing numerous low-level Sc events and creating a unique dataset of cloud systems over this underrepresented region.

ACHIEVE Instrumentation
ACHIEVE contains an integrated suite of active and passive remote sensing instruments for measuring the structures and properties of clouds and precipitation, as well as retrieving columnar aerosol and atmospheric properties, from a ground-based platform.Fig. 3(a) shows the setup of ACHIEVE at Yen Bai.The active instruments include a scanning W-band (93.93 GHz) dual polarization Doppler cloud radar, a vertically-pointing K-band (24 GHz) Micro Rain Radar (MRR), and a 910 nm ceilometer.ACHIEVE's W-band radar exhibits great sensitivity (e.g., -53 dBZ at 1 km range for 25 m range resolution), and can be configured to transmit and receive in either short pulse mode or pulse compression mode, with the latter permitting increased sensitivity and range resolution.The MRR measures Doppler spectra and estimates radar reflectivity profiles, Doppler velocities, and path integrated attenuation.Additionally, retrievals of rain rate and liquid water content are possible with a stated minimum detectable threshold of 0.01 mm hr -1 for rain rate.The Vaisala CL51 ceilometer uses a 910 nm wavelength laser to measure vertical profiles of attenuated backscatter up to 15 km and infer cloud base height for up to three cloud layers.
Passive instruments include an Atmospheric Emitted Radiance Interferometer (AERI), and a sun photometer (part of the Aerosol Robotic Network, AERONET).AERI measures absolute radiances of downward thermal emissions over the spectral range of about 3 to 18 µm at 1 cm -1 resolution.Data products include sky brightness temperatures at measured wavenumbers, along with retrievals of aerosol and bulk cloud properties, and profiles of temperature, pressure and humidity under cloud-free sky conditions, as well as below cloud under cloudy conditions.Unforeseen problems with AERI's data processing software following 7-SEAS/BASELInE have unfortunately precluded use of this data at the time of this writing.An upgraded AERONET sun photometer measures polarized radiances at 9 channels ranging from 340 to 1640 nm and operates in both cloud-free and cloud mode.Data include spectral aerosol optical depth along with retrievals of aerosol size distributions in cloud-free conditions, and retrieved cloud optical depth when operating in cloud mode.Table 1 provides the specifications and data products for ACHIEVE instrumentation that operated during the 2013 7SEAS/BASELInE deployment.
The W-band radar operated mostly in vertically-pointing mode throughout the campaign in order to capture the evolution of cloud and precipitation system vertical structures as they passed overhead.The scanning strategy also included hourly Range Height Indicator (RHI) scans at E-W, SE-NW, and S-N azimuth angles and spanning 45-135° elevation, as well as Plan Position Indicator (PPI) scans at 60-85° elevation in increments of 5° in an effort to sample spatial variability of the systems.Full hemispheric scans were not  possible due to the presence of nearby buildings.The radar moment data (reflectivity, mean Doppler velocity, and spectral width) were computed from the range-dependent Doppler power spectra after subtracting out the mean noise power (i.e., average of sky noise samples taken during the spectra averaging period).In addition, moment data having signal to noise ratios less than 5 dB were excluded from analyses presented herein.Given the narrow beamwidth and high elevation angles (> 45°) of the radar above the horizon for this deployment, ground clutter echo contamination was negligible.Radar reflectivity was not corrected for gaseous attenuation, although this should be minimal for vertically-pointing mode.Attenuation by liquid water was also not performed as this requires information on the particle size distributions.On-site calibration of the Wband radar was performed at the beginning and mid-way through the deployment using a 16.25 cm trihedral corner cube reflector mounted atop a 20 m tall portable tower located 370 m to the west of ACHIEVE (Fig. 3(b)).

Summary of ACHIEVE Observations
Throughout ACHIEVE's deployment, low-level Sc cloud decks observed over the Yen Bai region tended to follow a diurnal cycle.The cloud layer usually developed in the mid to late evening, thickened and occasionally developed drizzle or light rain overnight.The low-level Sc would generally linger through the early to mid-morning and eventually dissipate by mid-day as the boundary layer warmed.This diurnal cycle is evident in Contoured Frequency with Altitude Diagrams (CFADs; Yuter and Houze, 1995) of ACHIEVE's W-band reflectivity data collected from 28 March through 9 April 2013 (Figs.4(a)-4(d)).These diagrams depict the probability density of a field as a function of height and were constructed by partitioning the observed reflectivity values into 5-dBZ bins beginning at -30 dBZ for each height.The data were further partitioned into four 6-hour time periods (07:00-13:00, 13:00-19:00, 19:00-01:00, and 01:00-07:00, all times local) and normalized by the number of profiles in each respective time period in order to depict the diurnal cycle.The greater probabilities of radar echos above -20 dBZ observed below 2 km for the nighttime periods (19:00-07:00 local; Figs.4(c) and 4(d)) point to more frequent occurrences of low-level cloudiness during the night versus daytime in agreement with the CALIPSO cloud mask feature frequency profiles (cf.Fig. 1(b)).The CFADs also depict well the development of nighttime precipitation as evident the sloped region of peak probabilities that increases from -15 dBZ near 1.8 km to between 0 and 10 dBZ at the lowest levels in Figs.4(c) and 4(d).Physically, this sloped feature results from the growth of drops as they fall through the cloud layer towards the surface (Li et al., 2008;Mechem et al., 2011).Mid-and high-level clouds were also more prevalent overnight as shown by increased probabilities of echoes above roughly 3 km relative to the daytime period.These echoes were mostly associated with elevated convective systems that passed over the region on several occasions, yet their contribution to total cloudiness during the deployment was minimal as evident by the low magnitudes of probability (~2.5%) relative to the peaks (~17.5%) at low levels.
Reflectivity CFADS constructed from all available CloudSat/CPR overpasses within 100 km of Yen Bai during March and April for the years 2007-2012 are also presented in Figs.4(e) and 4(f).The CPR's minimum detectability is around -28 dBZ (Stephens et al., 2002), although this generally corresponds to features above the moist atmospheric boundary layer.As such, reflectivity values below -25 dBZ were not included for this particular analysis due to increased noise in the CPR signal near the Earth's surface.Slightly greater probabilities of cloud echoes around 2 km, as well as more frequent mid-and high-level clouds, are seen for night versus daytime; however, nominal overpass times around 01:30 and 13:30 local suggest that the diurnal cycle of low-level clouds is not well represented (Mace and Zhang, 2014).Peak probabilities in the CloudSat/CPR CFADs occur around 1 km height and for values greater than 15 dBZ; however, these most likely result from surface contamination (Marchand et al., 2008) and are therefore associated with large uncertainties regarding the presence of low-level clouds.
An example of the radar reflectivity time series from ACHIEVE's W-band radar on 9 April 2013 (Fig. 5 (Sauvageot and Omar, 1987;Fox and Illingsworth, 1997;Löhnert et al., 2001).Given that such small particles have negligible fall velocities (primarily due to turbulent eddies in vertical motion), they can act as tracers within the environmental flow.Thus, the mean radial velocities shown in Fig. 7 represent the along-beam component of this horizontal flow within the cloud layer, and imply flow from the southeast to the northwest, perpendicular to the axis of zero (or minimal) radial velocity (Doviak and Zrnić, 1993).Weak radar echo returns (< -30 dBZ) below 1 km prior to 18:00 UTC are due primarily to insects.
By 18:00 UTC (01:00 local), formation of a secondary cloud deck is evident near 800 m.Development of light drizzle is evident in the RHI scan (Fig. 6(d)) between 0 and 1 km distance SE, and appears as increasing reflectivity downward from cloud top to near cloud base as growing droplets near cloud top subsequently fall and grow via coalescence, as described by Fox and Illingsworth (1997).The NW to SE slope of this "drizzle streak" is likely due to the fact that as the drizzle drops grow in size, they attain greater fall speeds and deviate more from the environmental flow.The RHI and PPI reflectivity scans (Figs.6(c) and 6(d)) depict considerable spatial variability of the cloud properties.A period of moderate drizzle occurs between 19:30 and 22:00 UTC (02:30-05:00 local) as depicted by increases in reflectivity magnitudes (Fig. 6(a)) and downward mean Doppler velocities (Fig. 6(b)), with peak magnitudes of 11 dBZ and -2 m s -1 , respectively, around 20:00 UTC (03:00 local).Assuming the environmental vertical velocity is near zero and neglecting turbulent contributions, these mean Doppler velocity values correspond to mean drop diameters around 0.1 to 0.2 mm based on diameter-fallspeed relationships from Mitchell (1996) thereby placing these drops in the large drizzle to small raindrop size range.It is important to note, however, that the Doppler velocities include vertical air motion and turbulence, both of which are unknown without additional measurements.Reflectivity time series from the vertically-pointing MRR indicated only a trace of precipitation (< 0.1 mm) actually reached the surface (not shown); however, given the lower frequency, wider beamwidth, and much lower power of the MRR relative to the W-band, the weak intensity of the drizzle is likely underdetected (Maahn and Kollias, 2012).Beyond 22:00 UTC (05:00 local), drizzle production ceases and the two cloud decks become increasingly distinct from one another with time.Both cloud layers begin to dissipate by 04:00 UTC (11:00 local).

PRELIMINARY MODEL WORK
Measurements from 7-SEAS/BASELInE provide insight into the microphysical, chemical, and radiative properties associated with aerosol-cloud interactions for Sc over land (Tsay et al., 2013).To enhance our understanding of the relative importance of the roles played by microphysical and macrophysical processes in the different cloud regimes and at different stages within the cloud's lifecycle, synergistic combination of observed aerosol/cloud properties with comprehensive modeling efforts are essential.A critical first step in this process is to employ high-resolution cloud models capable of simulating the underlying small-scale physical processes in aerosol-cloud interactions before attempting to interpret the measurements taken at different times and locations over the broader region.

GCE Model Overview
The Goddard Cumulus Ensemble (GCE) model is a nonhydrostatic cloud-resolving model developed and maintained at NASA/GSFC as part of the larger Goddard Multi-scale Modeling System with Unified Physics (Tao et al., 2009(Tao et al., , 2014)).Simulations can be performed in either 2D or 3D and can be forced by soundings or large-scale tendencies such as from regional-scale model output or reanalysis data.Subgrid-scale turbulent transport of energy and mass associated with boundary-layer clouds (i.e., stratocumulus) is predicted via turbulent kinetic energy (TKE, Klemp and Wilhelmson, 1978;Soong and Ogura, 1980) or optional Smagorinsky dynamic (Kirkpatrick et al., 2006) schemes and is also coupled with land and ocean models to provide surface fluxes of heat and moisture.Short-and long-wave radiative transfer processes, as well as their interactions with aerosols and clouds, are computed for eight bands each across the solar and infrared spectrum (Chou and Suarez, 1999;Chou et al., 1999), respectively.
The GCE model's single-moment bulk microphysical schemes are based on Lin et al. (1983) with various modifications/improvements (Tao et al., 2003;Lang et al., 2014).The GCE model additionally incorporates the singlemoment (1M) and double-moment (2M) versions of Colorado State University's Regional Atmospheric Modeling System (RAMS) bulk microphysical scheme (Meyers et al., 1997;Saleeby andCotton, 2004, 2008) as well as a spectral bin microphysical scheme (Khain et al., 2004).Bin schemes are advantageous over bulk models in representing processes such as particle nucleation, collision-coalescence, and sedimentation by explicitly predicting the evolution of hydrometeors of discrete sizes, at considerable computational expense however.Bulk schemes employ analytical distribution functions to represent hydrometeor distributions and predict distribution characteristics such as mass mixing ratio q (related to the 3 rd moment), particle concentration N t (related to the 0 th moment), and/or spectral width (ν).The modified gamma distribution (Ulbrich, 1983) is commonly used in cloud studies since it adequately represents many cloud/rain drop size distributions (CDSD) measured by insitu probes and disdrometers (Miles et al., 2000;Brandes et al., 2003).Furthermore, this distribution function permits variability in CDSDs due to aerosols acting as CCN that affect cloud droplet sizes, number concentrations, and distribution spectral widths (e.g., Liu et al., 2008;Martins and Silva Dias, 2009).
One issue with 1M-and 2M-bulk schemes is the inability to account for changes in the spectral width (ν) of the size distributions resulting from microphysical processes such as droplet activation, collection, sedimentation, melting, and evaporation.With the exception of melting, these processes are critical in assessing the aerosol-cloud-precipitation interactions of stratocumulus clouds (e.g., Sundu et al., 2008;Wood, 2012).Milbrandt and Yau (2005) developed a triplemoment (3M) bulk microphysics scheme that predicts the sixth moment (related to the radar reflectivity factor) of precipitating hydrometeor size distributions in addition to q and N t in order to obtain time-dependent ν values.Loftus et al. (2014) developed and implemented a similar 3M scheme into the RAMS model, and an enhanced version of this scheme that extends 3M prediction to cloud and rain distributions has recently been integrated into the GCE.The prediction of ν for cloud and precipitating hydrometeor distributions allows for a more complete assessment of aerosol effects on droplet sizes, numbers, and the spectral width/shape of the distributions.In turn, the 3M scheme results in improved simulation of microphysical processes and allows for more accurate calculations of simulated active/passive signals (e.g., Dawson et al., 2010;Kollias et al., 2011;Loftus, 2012), which can then be used to gauge model performance against both ground-based and satellite observations as well as to provide insight into various physical processes (Matsui et al., 2014).
The bulk 3M microphysics scheme has also been incorporated into the Goddard Satellite Data Simulator Unit (G-SDSU) to account for variability in ν when computing radar reflectivity from the predicted hydrometeor fields.The G-SDSU is a forward radiation model coupled to the GCE that permits interfacing remote-sensing/in-situ measurements with model output (Matsui et al., 2009;Masunaga et al., 2010;Matsui et al., 2013).The resultant meteorological and geophysical parameters simulated by the GCE are used as inputs to G-SDSU to drive the individual instrument simulators for calculating radiance, reflectivity and backscattered Level-1b signals.

Simulations of 7 April 2013 Case
The 7 April 2013 case was selected as a 'golden scenario' case to gauge the ability of the recently implemented 3M scheme to simulate the early developmental stage of lowlevel Sc over land.Due to the processing data issues with retrieving atmospheric profiles from AERI, the 12:00 UTC atmospheric sounding from Hanoi (Fig. 7) was used as a proxy for the atmospheric state over Yen Bai just prior to the time at which the first cloud echoes were observed.A small increase in vapor mixing ratio was applied to the 850 to 820 mb layer to force a slightly supersaturated (0.05%) layer in which clouds could develop.The model domain was initialized horizontally homogenously and no additional forcing was applied.Because this was a nighttime case, the absence of shortwave radiative impacts of aerosols and clouds simplifies the interpretation of results.Table 2 lists the model parameter settings applied in the simulations.
Six simulations were performed in which number concentrations of aerosols (N a ) acting as potential CCN were systematically increased from surface maximum values of 100 to 2000 cm -3 (Fig. 8) to examine the impacts on the modeled clouds.The upper limit of this range was selected to correspond with in-situ surface aerosol measurements taken at Son La, Vietnam (about 110 km west-southwest of Yen Bai) prior to 7 April.Analyses of back trajectories using the NOAA HYSPLIT model (Draxler and Rolph, 2015) for the deployment period revealed air parcels in the 1.5 to 3 km layer above the Yen Bai region generally arrived from the southwest and west (not shown), such that they often traversed the Son La region.As Son La and Yen Bai are situated along the long-range transport path of BB aerosols (cf.Fig. 2 of Tsay et al., 2013), it is reasonably assumed that the aerosols were aged sufficiently (i.e., more hygroscopic) and hence suitable to serve as CCN (Reid et al., 2005).Aerosol distributions in all cases were assigned a dry mean mass diameter of 0.13 µm, within the range of values (0.12 to 0.18 µm) from sun photometer retrievals at Yen Bai as well as in-situ measurements upstream at Son La, and a fractional solubility of 50% as a proxy for aged BB aerosols.The highly idealized nature (i.e., no forcing beyond the imposed supersaturation) and identical environmental conditions in these simulations help ensure that differences in cloud microphysical responses in each case are due primarily to changes in N a .As neither in-situ measurements of aerosol/CCN nor profiles of these particles are available at Yen Bai during the 2013 7-SEAS/BASELInE campaign, results from these particular simulations are not intended to draw specific conclusions regarding the effects of BB aerosols on the observed cloud system, but rather to serve as a guideline for potential impacts of aerosols on these clouds and how the impacts might be 'observed' from a remote sensing standpoint.

Preliminary Simulation Results
Time-height plots of cloud droplet number concentrations (N c ), cloud droplet effective radius (r e ) and simulated Wband attenuated reflectivity (Z) at the domain mid-point represent the evolution of the modeled cloud layer for selected aerosol sensitivity simulations (Fig. 9).Because the supersaturated layer spans roughly 1.4 and 1.7 km, cloud droplets nucleate within this layer upon model startup in all cases.The cloud layer slowly thickens over roughly the next two hours, with cloud tops extending to near 2 km by 16 UTC in each simulation (Fig. 9).As expected, increases in N a lead to greater values of N c (Figs. 9(a)-9(d)) and smaller r e (Figs. 9(e)-9(h)) in accordance with the 'first aerosol indirect effect' (Twomey, 1977;Rosenfeld and Lensky, 1998;Andreae et al., 2004).Values of N c even at high N a conditions are small (< 50 cm -3 ) relative to typical observed values for Sc over land (Miles et al., 2000) most likely due to lower supersaturation values in the absence of any organized vertical motion.The evolution of Z in the cases shown (Figs.9(i)-9(l)) matches well with the early stages of the observed Sc deck (Fig. 6, prior to ~18 UTC) in terms of magnitudes and vertical extent, although the simulated cloud decks are shifted upwards by 200 to 300 m relative to the observations.Development of a second, lower cloud layer does not occur in the simulations due to the fixed environmental conditions.A decrease in Z of approximately 2.5 dBZ is seen as N a increases from 300 to 2000 cm -3 during the initial development stage (Figs.9(i)-9(l), 14-16

Cloud droplet N c
Cloud r e W-band Z e UTC), with a weaker trend of decreasing reflectivity with increasing N a evident beyond about 16 UTC.The negative correlation between N a and Z results largely from decreases in droplet sizes with increasing N a .Very light drizzle (rain water contents of order 10 -4 to 10 -5 g m -3 ) develops in the lower aerosol loading cases (c300 and c600) beyond roughly 17 UTC associated with Z ≥ -18 dBZ (Figs.9(i) and 9(j)), within the -20 to -15 dBZ range typically used to distinguish drizzling from nondrizzling Sc (e.g., Sauvageot and Omar, 1987;Löhnert et al., 2001;Kogan et al., 2005).While such low rain water content values seem negligible, they are non-trivial from a microphysical modeling standpoint as they represent a small, yet quantifiable transition from a non-precipitating to a precipitating category.Shallow regions of weak reflectivity values (< -37.5 dBZ) immediately below cloud base in these cases depict virga, whereas reduced rain water contents and reflectivity values in cases c1000 and c2000 (Figs.9(k) and 9(l)) suggest drizzle formation is suppressed.It is noted that drizzle production is inefficient even at low N a values.Such inefficiency results from shallow cloud depth, low cloud LWP values (cf.Fig. 10(a)), and lack of large scale forcing (e.g., upward vertical motion within the cloud layer does not exceed 0.1 m s -1 in all simulations, not shown).As vertical motion is absent initially, local buoyancy perturbations due to condensation and evaporation of droplets create turbulent motions that maintain the cloud layer.Lee et al., (2009) also noted dominance of condensation and evaporation over droplet self-collection and accretion in clouds with low LWP under weakly forced conditions.Although aerosol removal processes are not included in these simulations, the suppression of drizzle at greater N a values infers little to no impact on the boundary layer air quality in these particular scenarios given that precipitation scavenging of aerosols below cloud base is proportional to precipitation rate (Pruppacher and Klett, 1997;Seinfeld and Pandis, 1998).
Time series of domain-averaged total LWP and cloud optical depth (τ c ) for all simulations (Fig. 10) show that LWP and τ c values increase steadily as the cloud deck develops, reach a plateau signifying a more or less quasi-equilibrium state for roughly two hours, then begin to increase again.There is an increasing delay in reaching quasi-equilibrium state as N a increases.The reason for this behavior is linked to the growth of the droplets within a limited reservoir of water vapor.The more numerous cloud droplets, activated under higher N a conditions, leads to faster reduction in supersaturation due to more rapid growth rates of smaller droplets and more vigorous competition for available water vapor (Yum and Hudson, 2005).The plots also reveal an increase in these radiatively important parameters as N a increases in agreement with the original model of Albrecht (1989), although there is little noticeable difference in cloud thickness as has been suggested for marine Sc (e.g., Pincus and Baker, 1994).This result agrees with the initial increases in LWP and τ c with increasing N a seen for simulations of non-drizzling nocturnal marine Sc by Sundu et al. (2008), but is opposite to decreases in these cloud parameters reported by Jiang et al. (2002), Wang et al. (2003), and Hill et al. (2008, 2009).Ackerman et al. (2004) and Lee et al. (2009) showed that decreased LWP under higher aerosol loading scenarios were largely due to increased entrainment of overlying dry air and evaporation of droplets within updrafts near cloud top.Given the extremely weak vertical motion in the current simulations, a plausible explanation for increasing LWP with increasing aerosol loading is the apparent lack of entrainment of dry air from above cloud top.

CONCLUDING REMARKS AND FUTURE WORK
The 2013 7-SEAS/BASELInE campaign involved the first ever deployment of a ground-based millimeter cloud radar to the northern Vietnam region in an effort to capture detailed evolution of low-level Sc layers over land in the presence of BB aerosols.As current satellite observations of this low-level Sc layer generally lack details on the cloud structure due to surface clutter effects and signal attenuation, ground-based observations continue to provide crucial measurements over the lower layers of the atmosphere.Despite a reduced data collection period (17 days versus the expected 5-6 weeks) and unplanned power interruptions, data collected by ACHIEVE's W-band radar and ancillary instrumentation provided the firstever comprehensive ground-based measurements of low-level Sc systems over this particular region.Analyses of the Wband radar data revealed these cloud systems were generally confined to below 2 km and tended to follow a diurnal cycle, with peak occurrences during the nighttime and early morning hours, often accompanied by light precipitation.Future analyses will incorporate retrieved aerosol properties and profiles of atmospheric thermodynamic variables (T, RH, p) from the AERI data once the processing issue has been resolved.These data, along with regional reanalysis data, will permit more thorough understanding of the role of meteorology on the cloud system lifecycle, help determine the degree of coupling of the aerosol and cloud layer, and aid in estimations of precipitation impacts on local air quality via the aerosol washout effect.
A lack of concurrent aerosol data during cloudy episodes precludes a detailed examination of aerosol-cloud interactions from these particular observations, and thus, modeling efforts constrained by the available data are required to better understand potential impacts BB aerosols may have on the cloud system evolution.Previous investigations of aerosol impacts on low-level Sc have focused overwhelmingly on marine cloud systems, which are generally controlled by very different dynamical processes compared to systems over land (Mechem et al., 2010;Wood, 2012).The preliminary results from aerosol sensitivity simulations presented herein revealed the familiar response of increased numbers and smaller sizes of cloud droplets with increasing aerosol concentrations, whereas increases in LWP and cloud optical depth were mostly opposite to findings from previous studies on marine Sc.Slight reductions in simulated Wband reflectivity values were seen with increasing aerosol concentrations and resulted primarily from decreased droplet sizes.Additionally, drizzle formation, albeit extremely small, was evident only for low aerosol loading cases in agreement with previous studies.
There are certainly shortcomings of the highly idealized simulations presented herein, and efforts are currently underway to alleviate some of the model deficiencies with respect to the representation of aerosol processes within the GCE model.In particular, the RAMS bulk aerosol module (Saleeby and van den Heever, 2013) simulates aerosol removal via cloud droplet nucleation and wet/dry deposition, and aerosol regeneration upon hydrometeor evaporation.Aerosol mass is also tracked within hydrometeors, and non-activated aerosols are advected with the model-predicted flow fields.This aerosol module was recently implemented into the GCE model and is currently undergoing compatibility tests within the larger GCE framework.The removal of atmospheric aerosols by nucleation and precipitation scavenging, as well as cloud processing of aerosols, are processes that impact local and regional air quality in addition to potential effects on cloud and precipitation evolution downstream (Levin and Cotton, 2009;Flossman and Wobrock, 2010).Wet scavenging below cloud increases with rain rate and is most efficient for aerosol particles with diameters D a ≤ ~0.05 µm, as well those with D a ≥ ~2-3 µm, whereas aerosols within the "Greenfield Gap" (0.1 ≤ D a ≤ 1 µm) are most effectively removed via cloud droplet nucleation (Seinfeld and Pandis, 1998;Andronache, 2003;Feng, 2007).Relationships between below-cloud scavenging efficiency and radar reflectivity (e.g., Jylhä, 1999), as well as dependencies of scavenging efficiency on rain drop distribution characteristics (e.g., Jung et al., 2003), also suggest the need for accurate representation of hydrometeor distributions such as with the 3M scheme used herein.Another key aspect of this module is the use of a species-dependent soluble fraction parameter, ε, which accounts for aerosol hygroscopicity in a manner analogous to the κ-parameter (Petters andKreidenweis, 2007, 2008).For example, smaller (larger) values of ε can be applied to represent freshly emitted (aged) biomass-burning aerosols.Future simulations will also incorporate time-dependent meteorological forcing based output from week-long simulations over the SEA region using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) (Grell et al., 2005;Fast et al., 2006).Aerosol emissions, transport, and aging processes predicted by WRF-CHEM and constrained by collaborative 7-SEAS/BASELInE measurements (e.g., Lin et al., 2013;Tsay et al., 2013;Lin et al., 2014) will further permit more realistic aerosol input into the GCE.
The knowledge and experience gained regarding instrument capabilities and sampling requirements for obtaining cloud and aerosol measurements during 7-SEAS/BASELInE and subsequent campaigns have led the SMARTLabs team to develop a more comprehensive approach for addressing aerosol-cloud interactions.Planning is currently underway for a follow-on experiment in the coming years as part of the ongoing 7-SEAS campaign to deploy SMARTLabs' ACHIEVE and Chemical, Optical, and Microphysical Measurements of In-situ Troposphere (COMMIT, cf.Pantina et al., 2016) mobile laboratories to the Yen Bai region in conjunction with AERONET, Micro Pulse Lidar Network (MPLNET) and regional contributing instruments.The addition of an Unmanned Aerial Vehicle (UAV) equipped with microphysics and aerosol probes for sampling in-cloud and near-cloud environments, particularly cloud-clear air interface regions, is also possible (Tsay et al., 2016).This multinational collaborative effort will permit co-located remote sensing and in-situ measurements of thermodynamic, aerosol and cloud profiles that will greatly aid in synergistic analyses of the observational data, as well as provide critical constraints for continued development and evaluation of models and retrievals aimed at improving our understanding of aerosol-cloud interactions.

Fig. 1 .
Fig. 1. a) Aqua/MODIS RGB image on 9 April 2013 covering the region 10-30°N, 96-116°E superimposed with A-Train overpass (yellow line) track, and b) CALIPSO/CALIOP Version 3 aerosol and cloud vertical frequency distributions within a 2 × 2 degree sub-domain (grey box in panel a) for March (top row) and April (bottom row) partitioned by daytime (left column) and nighttime (right column) overpasses for the years 2007-2014.

Fig. 2 .
Fig. 2. Annual rainfall amounts (blue line) recorded at the Yen Bai meteorological station for the years 1960-2013.The bold red line depicts the trend over the same time period.Data courtesy of The Ministry of Natural Resources and Environment, Socialist Republic of Vietnam.

Fig. 3 .
Fig. 3. a) Set up of ACHIEVE at Yen Bai regional meteorological station, Yen Bai, Vietnam.Red arrow indicates orientation.AERONET sun photometer was located in a grassy area roughly 40 m north of ACHIEVE.b) Photo of the 10 m aluminum calibration tower with an additional 10 m carbon fiber mast atop which sits a 16.25 cm trihedral corner reflector.

Fig. 4 .
Fig. 4. Contoured Frequency with Altitude Diagrams (CFADs) computed from ACHIEVE's W-band radar reflectivity for the time period spanning 28 March through 9 April 2013 (panels a-d) and from CloudSat/CPR overpasses within 100 km of Yen Bai, Vietnam for the months of March and April for the years 2007-2012 (panels e and f).Data from both instruments is further partitioned into daytime (left column) and nighttime (right column), with ACHIEVE W-band CFADs further partitioned into 6-hour increments as listed in each panel (UTC times given in parentheses) whose sequence is denoted by the circular red arrow.Note different height and frequency scales are used for ACHIEVE and CloudSat CFADs.
(b)) shows in detail a persistent low-level Sc cloud deck between roughly 1.1-1.6 km height in addition to cumulus underneath stratus as revealed by the occasional gaps in reflectivity ~1 km.Variability in the ceilometer-indicated cloud-base height arises primarily from scattered cumulus clouds passing over the measurement site.Radar reflectivity values below the ceilometer-indicated cloud-base denote very light precipitation (virga) falling from the cloud, whereas sudden dips in the ceilometer-indicated cloud-base height (e.g., ~05:00-05:13) may indicate pannus (or stratus fractus), which can form in association with localized precipitation below clouds.For this particular time period, no clouds were present above the Sc deck, no surface precipitation was observed on site, surface temperatures ranged between ~18-20°C, and near-surface winds were out of the east and less than 5 m s -1 .Light rain had fallen earlier in the day between roughly 02:30-03:00 UTC, although the Sc deck existed both prior to and after this brief rain episode.Concurrent measurements made by CALIPSO and CloudSat/ CPR satellites during a nearby overpass around 06:30 UTC reveal a layer consisting of fine mode aerosols existed above the low-level Sc deck (Figs.5(c) and 5(d)), yet the CPR missed the majority of low-level Sc over the Yen Bai region (Fig. 5(a)).This particular example highlights the advantages of ground-based over space-borne radar for studying detailed cloud and precipitation structures and processes in low-level clouds.Fig. 6 shows an excellent example of the typical diurnal evolution of low-level Sc over Yen Bai as observed by ACHIEVE's W-band cloud radar on 7-8 April 2013.The cloud deck first develops around 1.5 km between 13:00-18:00 UTC (20:00 7 April-01:00 8 April local time), thickening to several hundred meters by 16:00 UTC.The evolving cloud deck is dominated by cloud-sized particles (diameter, D < 50 µm, reflectivity values below -15 dBZ) and downward Doppler velocities are mostly less than 0.7 m s -1 , typical thresholds used to delineate non-drizzling from drizzling Sc

Fig. 7 .
Fig. 7. Skew-T diagram for 1200 UTC sounding for Hanoi, Vietnam on 7 April 2013 used to initialize aerosol sensitivity simulations.

Table 2 .Fig. 8 .
Fig. 8. Names and corresponding aerosol number concentration profiles for the six sensitivity simulations performed.Naming convention is associated with maximum aerosol concentration at the surface.

Fig. 10 .
Fig. 10.Time series of domain-averaged a) total liquid water path (LWP, g m -2 ) and b) cloud optical depth τ for aerosol sensitivity simulations of 7 April 2013 case.