1. Introduction

Clouds play an important role in weather and climate.Approximately 60%–70% of the earth’s surfaceis covered by clouds at any given time, and cloudsexert significant effects on the global energy balance(Baker, 1997). Clouds cool the climate system by reflectingincoming shortwave solar radiation back tospace and warm the climate system by absorbing and reemitting longwave radiation. Small variations incloud cover can cause large changes in incident solarradiation. Clouds and precipitation also regulatethe atmospheric hydrological cycle, redistributing waterboth within the atmosphere and among differentregions of the earth’s surface(Quante, 2004; Stephens, 2005). Many aerosols act as cloud condensation nuclei(CCN)or ice nuclei(IN), and alter the radiative and microphysical properties of clouds and precipitation incomplex ways. Recent research suggests that aerosoleffects can either trigger or restrain rainfall ocurrence, depending on meteorological conditions(Rosenfeld et al., 2008). Latent heating and cooling during the for-mation and evaporation of clouds and precipitationmodify the atmospheric circulation, while certaintypes of clouds can produce lighting(Tao et al., 1993;Lee, 2012). All of these processes are closely associatedwith cloud microphysics.

Numerical models for weather forecasting and climatesimulation employ a wide variety of explicitmicrophysical schemes; however, these microphysicalschemes remain highly uncertain due to the physicalcomplexity of cloud-precipitation processes(Kann et al., 2011; Angevine et al., 2012; Brown et al., 2012;Hamill, 2012). The Intergovernmental Panel on ClimateChange(IPCC)has confirmed that cloud processes and associated feedbacks represent one of thelargest sources of uncertainty in projections of futureclimate change(Houghton et al., 2001). The evolutionof clouds and precipitation is affected by dynamic and thermodynamic conditions in the atmosphere, which isin turn substantially influenced by cloud microphysicalprocesses(such as latent heat release and hydrometeordrag). Clouds and precipitation processes occuron scales from less than one micrometer to thesynoptic scale. This wide range of spatial scales and the complex interactions among them makes it practicallyimpossible to accurately describe the full rangeof cloud processes in numerical models. A numberof relevant processes must either be ignored or heavilyparameterized. The construction of reasonable parameterizationsfor cloud-precipitation processes is a vitalcomponent of model development.

A large number of numerical models have beenused for operational weather forecasting and climatesimulation in China since the 1980s, including modelsdeveloped both locally and abroad(e.g., Xu and Wang, 1985, 1990; Hu and He, 1987; Xiao et al., 1988; Kong et al., 1990; Liu et al., 1993; Wang and Zhou, 1996; Zhou and Wang, 1996; Hong, 1998; Guo et al., 2001; Sun et al., 2008). Within these models, almost all the microphysical parameterizations weredeveloped abroad. One of the major reasons for thisis a lack of in-depth and systematic analysis on thedomestic observations of microphysical processes and internal structures of clouds. In the past, domesticmeasurements of cloud microphysical properties havebeen conducted over a variety of places in China.Here, we present an overview of observations of insitucloud-precipitation physical properties in Chinaduring the past five decades, particularly those propertiesthat are directly relevant to the developmentof microphysical parameterizations. These observationsmay be helpful for verifying the physical processes and hydrometeor fields simulated by microphysicalparameterizations, improving the parameterizationschemes in numerical models, and underst and ingcloud-precipitation-radiation interactions.

2. In-situ measurement devices and methods

Table 1 lists information about in-situ observationaldevices used in past measurement campaigns.Triplex droplet collectors(TDCs) and imitation SovietTMP-1 type probes were widely used for measurementsof cloud microphysical properties in the early1960s. TDCs are designed to detect cloud droplets, salt nuclei, and aerosols with three different detectors.These instruments are lightweight and portable(Zhang, 1963). Early versions of the TMP-1 typeprobe took samples using magnesia glass slides. Theseinstruments were later modified to use iodine and starch film. TMP-1 type probes took measurementsin sampling intervals of 0.12 s with a measurementpath of 7.2 m and a ventilation cross-section of 200cm2. The Particle Measuring System(PMS)was introducedfrom abroad in the early 1980s and has beenwidely used for measurements of cloud-precipitationmicrophysical properties. Considerable improvementsto the PMS since its introduction made the instrumentsuitable for the collection of cloud-precipitationmeasurements over East Asia(You, 1994). The PMSrecords observational data on magnetic tapes in realtime, which is convenient for later data processing and analysis. The PMS contains four sensors(PCASP-100X, FSSP-100-ER, OAP-2D-GA2, and OAP-2DGB2)that can measure particles in size from 0.1 to6200.0 μm(Liu et al., 2001, 2003). All of these instrumentshave advantages and disadvantages. The TDCallows the detection of different cloud types, but isonly used at high-altitude surface stations(e.g., HengMountain). The TMP-1 type probe can be mountedin an aircraft to enable the detection of microphysicalproperties through the full vertical extent of a cloud;however, strong turbulence in cumuliform clouds typicallylimits the collection of measurements from aircraft.

Both oil and filter paper methods were used tostudy the properties of raindrop before the 1990s(e.g., Gu, 1962; Ruan, 1962; He, 1965; He et al., 1985;Jiang, 1988). The oil method involves letting raindropsfall onto oil film and then measuring the number and size of raindrops using transparent grid-paper underthe film. The filter method involves putting filterpaper daubed with dyes(such as a mixture of rhodamine and talcum powder)into rainfall. The number and size of raindrops can then be calculated byinspecting the raindrop spots left on the paper. Onedisadvantage of the oil method is that it is difficultto gauge the importance of collision-coalescence processes(Chen, 1963). The filter paper method is alsooften inaccurate, with large differences among inspectioncurves when different filter papers are used(Chen et al., 1989; Xu and Lei, 1989). GBPP-100 probeshave been widely used to measure raindrop size distributionssince the mid 1980s. These instruments areoptical array probes. Particle properties are recordedby sensing the shadowing across a linear array of photodiodesas particles pass through the probe field-ofview.The GBPP-100 probe is sensitive to raindropswith sizes from 0.2 to 12.4 mm, and provides continuous and automatic recording of raindrop information.However, the GBPP-100 may overestimate the contributionof large drops due to drop distortion(Liu and You, 1994; Chen and Ma, 1995; Wang Jun, 1997). ThePruppacher and Pitter(1971)method has been extensively used to correct for these distortion effects.

The concentrations and properties of atmosphericIN may be detected by using Millipore filter techniquesor Bigg-mixing cloud chambers(You and Shi, 1964;Wang et al., 1965; Zhao et al., 1965; Chen, 1994;Yang et al., 2004; Shi et al., 2006). The structureof a Bigg-mixing cloud chamber is similar to the instrumentdescribed by Isono et al.(1959), with minordifferences in the volume of the cloud chamber.Similarly, the Millipore filter technique is largely analogousto the method first introduced by Bigg et al.(1961), with some differences in pore size. Both methodshave advantages and disadvantages. Sampling ofIN using Bigg-mixing cloud chambers is convenient, but the accuracy of these measurements is limited byan inability to properly control for the relationshipbetween IN activation and supersaturation(Huffman, 1973; Hussain and Shaukat, 1989). The Millipore filtertechnique allows large volumes of air to be sample;however, this advantage is partially vitiated by the vaporcompetition effect, in which the measured concentrationof IN decreases with sampled volume(Mossop and Thorndike, 1966; Wang, 1983).

The properties of cloud ice and snow particlesmay be measured by using aluminium foil samplers, Mee 120-type Ice Counters or the PMS. The principlebehind the aluminium foil sampler is that strikemarks where ice particles hit the foil can be read and interpreted using a microscope. The PMS can retrievetwo-dimensional images of ice particles with sizes between25 and 800 μm using two optical array greyprobes with a variety of channels. Cloud ice particlesgenerally have maximum dimension less than 300μm, while snow crystals can have a minimum dimensionlarger than 300 μm(even up to several centimeters).Hail properties can be evaluated using a varietyof techniques, such as the hail pit method, thescreened hail spectrometer method, the aluminium foilhailpad method, and the camera method, among others.These methods can be divided into two main categories.The first involves the evaluation of hail spotscreated when hailstones fall onto soft matter(such asaluminium foil or soft soils). The second involves therecording of hail properties using a camera as hailstonespass through the probe field-of-view. Methodsbelonging to the former category are cheap, easy tocarry out, and do not require constant supervision;however, these methods are unable to record the starting and ending time of hailstorms or the evolutionof hailstorms with time. Camera-based methods aremore accurate than other methods(Xu, 1979).

Aerosol measurements commenced around themiddle of the 1970s and have been conducted at anumber of locations in China since 1980. These measurementshave been conducted with a variety of instrumentswith unequal size ranges. Examples ofthese instruments include the LG-83 counter, the PM-730 counter, the DL-8138 counter, the APS-3310Acounter, the Laser-Air Kinetic Particle Spectrometer, the PMS, and the Shao-Erci counter. These instrumentscover several different aerosol size ranges.

The accuracy of the microphysical measurementscollected over the past several decades in China dependssignificantly on the errors associated with differentinstruments and techniques. The developmentof new and improved measurement technologies and instruments has reduced these errors substantially, but complicates the comparison of measurementstaken using different instruments. Comparativeanalyses of instruments and methods for observingcloud-precipitation microphysical properties wereconducted as early as 1964, and researchers attemptedto develop unified st and ards for these measurementsby the end of the 1970s. A number of error analyses and comparative studies have been presented, butfew studies have provided detailed comparisons amongresults obtained using different instruments.

3. Characteristics of liquid particles 3.1 Cloud droplets

The initial measurements of cloud-precipitationmicrophysical properties in China commenced in 1958 and were targeted for weather modification(ActaMeteorologica Sinica Editorial Board, 1959; ArtificialPrecipitation Working Group of Gansu Province, 1959; Cheng, 1959). No measurements of cloudprecipitationmicrophysical properties were collectedin China before the 1950s. At the beginning of the1960s, Gu(1962)measured cloud droplet size distributions(CDSD)in cumulus congestus, fair weathercumulus, cumulonimbus, altostratus and nimbostratusover the Nanyue area of Hunan Province. Despitelimitations in instrument technology and workingst and ards, these observations have proven useful fortheoretical studies of cloud-precipitation microphysics, weather modification, and other topics. These observationspioneered the study of cloud-precipitation microphysicsin China and promoted the subsequent developmentof research into cloud-precipitation physics and weather modification(Gu, 1963).

Comprehensive observations of clouds and fogwere conducted over Nanyue Mountain and Qin Mountainin Hunan Province from spring to summer of1962. These measurements focused on the microphysical and macrophysical characteristics of cumuliform and frontal clouds. Zhu et al.(1965)discussed thecharacteristics of stratus and stratocumulus clouds inprefrontal and frontal areas. They observed narrowcloud droplet size distributions with a single peak inprefrontal clouds, along with small liquid water content(LWC) and large cloud droplet number concentration(Nc). By contrast, they found large LWC, smallNc, and broad-spectrum cloud droplet size distributionwith double or multiple peaks in frontal clouds.Stratiform clouds tended to form far from frontal areas and their cloud droplet size distributions were stablewith narrow spectral widths. Hong and Huang(1965)used a large number of continuous observations to describetwo characteristics of clouds with double-peakcloud droplet size distributions. First, double-peakcloud droplet size distributions had broad spectralwidths. Shifts to double-peak distributions were consistentlyaccompanied by increases in the cloud spectralwidth, while disappearance of one of the peakswas associated with a narrowing of the spectrum. Second, the appearance of a double-peak cloud dropletsize distribution was associated with an increase inthe number of large cloud droplets and a decrease inthe number of small ones. The opposite changes occurredwhen the second peak disappeared. Xu(1964)identified fluctuations in continuously sampled microphysicalvariables(e.g., Nc and LWC), and reportedthat these fluctuations contributed to cloud dropletgrowth. The measured fluctuations were smaller in thepresence of medium-sized cloud droplets and larger inthe presence of large-sized ones(Zhan et al., 1965).

The China Meteorological Administration(CMA) and the Guangdong Provincial Meteorological Bureaulaunched an observational project in April and May1971 to observe the LWC and Nc of warm cumulostratusin the Xinfeng River basin. The resultsshowed that the LWC of cumulus clouds over theXinfeng River basin was larger than that of cumulusclouds over northern China(Wu et al., 1988). Theobserved cumulus clouds over Guangdong Provinceshared many characteristics with maritime clouds.

The Northern Research Group of Artificial PrecipitationExperiment for Stratiform Clouds(1991)conducted long-term observations of the particle sizedistributions of cloud droplets, ice crystals, snow crystals, and raindrops over northern China. Liu et al.(1995)reported that Nc values were typically around100–150 cm^-3 and LWC values were often less than0.3 g m^-3 in clouds over northern China. They alsoshowed that LWC values decreased exponentially withincreasing temperature and that liquid droplets generallyexisted when temperatures were warmer than–20℃. Wu(1987)indicated that small amounts of liquidwater could exist in clouds at temperatures lessthan –20℃. Wang and Yang(2003)measured verticaldistributions of the sizes, number concentrations, and mass densities of various particles in Meiyu-front rainfallusing a balloon-borne precipitation particle imagesensor.

Nc is a fundamental microphysical property ofclouds, which is currently not a predicted variable inmost large-scale or single-moment mesoscale models, and is typically assumed to be constant. Accurate and representative estimates of Nc are particularly importantin this case. Deng et al.(2009)measured theaverage Nc in warm clouds over Beijing, and reportedthat Nc values were 376 cm^-3 for cumulus clouds, 257cm^-3 for stratocumulus clouds, 147 cm^-3 for altocumulusclouds, 60 cm^-3 for altostratus clouds, and 60cm^-3 for nimbostratus clouds, with an overall averageof 193 cm^-3. Yin et al.(2011)performed a statisticalanalysis based on long term in-situ measurementsover East Asia, and reported that typical values of Ncin stratiform clouds ranged from 0.04 to 426.4 cm^-3, with an overall average of 120.9 cm^-3.

Cloud LWC plays an important role in both cloudphysics and dynamics. The amount of liquid waterprovides information about the extent of condensation, entrainment, and precipitation occurring withina cloud. LWC is also an important factor for weathermodification. Deng et al.(2009)estimated the typicalvalue of LWC of approximately 0.055 g m^-3 forwarm clouds over the Beijing region, while Zhang etal.(2011a)reported values of LWC between 0.0024 and 0.093 g m^-3 over Sh and ong Province. Long-termmeasurements over East Asia indicate that the averageLWC of stratiform clouds varied from 0.0002 to 0.520g m^-3 with an average of 0.140 g m^-3, while the averageLWC of cumuliform clouds varied from 0.005 to2.000 g m^-3 with an average of 0.875 g m^-3(Yin et al., 2011). Yin et al.(2014a)provided an empiricalrelationship between LWC and Nc based on a largenumber of in-situ measurements collected over northernChina. The number of measurements available forcumuliform clouds is relatively small because it is difficultto take measurements in cumuliform clouds due tostrong turbulence. The results for cumuliform cloudsmay therefore contain significant sampling biases.

The CDSD is another fundamental microphysicalproperty of clouds. Figure 1 shows a variety of observedcloud droplet size distributions for stratiformclouds in China. Cloud droplet size distributions cangenerally be divided into two groups. One group hasa wide spectrum and a small change rate, while theother has a narrow spectrum and a larger change rate.The Nc values observed in these cases vary widely, with differences of up to two orders of magnitude. Althoughonly a few datasets are available for cloudsover southern China, the available observations indicatethat the characteristics of stratiform clouds oversouthern China are similar to those of marine stratocumulusclouds. In particular, strati-form cloudsover southern China have spectra that are wide and smooth. Some of the differences in the spectral widthsof the observed CDSDs may be attributable to differencesin the probes used to sample the clouds.

The CDSDs are often described with gamma distributions, which represent reasonable approximations to observed spectra. The general gamma distributionis defined as

where N(D)is the cloud droplet number concentration(also called Nc in this paper), D is the cloud dropletdiameter, γ is a shape parameter, and λ is the slopeparameter. Positive values of γ imply that gammadistributions are concave downward in log-linear coordinates, while negative values indicate that gammadistributions are concave upward. N₀ has no substantialphysical meaning and varies widely, with an apparentdependence on γ. The gamma function hasbeen widely adopted to represent CDSDs in stratiformclouds(Table 2). Statistical analyses indicate that thevalue of γ varies between 0 and 12, N₀ ranges from10⁻⁹ to 102 cm⁻³ μm⁻¹, and λ varies between 0.0019 and 1.254 μm⁻¹. These results suggest that the rangeof CDSDs cannot be represented using a fixed slope, as is often done in bulk microphysics schemes. Predictingor diagnosing the shape γ as a function of thepredicted moments is a necessary step in improvingthe realism of bulk microphysics schemes(Milbr and t and Yau, 2005a; Thompson, 2006).

Figure 2 shows a selection of CDSDs observed incumuliform clouds in China, with representative samplesfrom(left to right)fair-weather cumulus, cumuluscongestus, cumulonimbus, and bimodal cumulusclouds, respectively. The CDSDs associated with fairweathercumulus are significantly different from thoseassociated with cumulus congestus and cumulonimbusclouds. The spectral width of the selected fair-weathercumulus CDSD is 54 μm, and Nc changes slowly withincreasing diameter(Fig. 2a). By contrast, the spectralwidths of the selected cumulus congestus and cumulonimbusCDSDs are 80 μm or even larger, withsharp declines in number concentration as the diameterapproaches 20 μm(Figs. 2b and 2c). The first and larger peaks in the bimodal cumulus CDSDs occurnear 8 μm, while the second peaks occur at diametersbetween 14 and 30 μm(Fig. 2d).

3.2 Raindrops

Raindrops represent the integrated effects of cloud microphysical, macrophysical, and dynamical processes, as well as interactions among these processes.Raindrop size distributions(RSDs)describe the number and size of precipitation particles. Reliable measurementsof RSDs are important for underst and ingthe development and evolution of precipitation. Gu(1962)classified RSDs into four basic types. The firsttype is characterized by an exponential decrease inraindrop number concentration with increasing raindropdiameter. This type of RSD has a narrow spectralwidth and a large number concentration of smallraindrops. The maximum raindrop number concentrationin this type of RSD may sometimes be locatedat a diameter that is slightly larger than the smallestraindrop diameter. The second type of RSD is characterizedby a large number concentration of small raindrops and a second spectral peak at a slightly largerdiameter. The third type of RSD also has multiplepeaks, but the first peak is at a diameter of approximately250–350 μm and the second peak is located faraway from the first peak. The fourth type of RSD ischaracterized by multiple spectral peaks at small diameters and large fluctuations in raindrop numberconcentrations at larger diameters. The first and second types are usually observed in the rain originatedfrom stratocumulus, altostratus, and altocumulusclouds. The third type occurs in the rain from cumuliformclouds, while the fourth type occurs in the rain from thunderclouds(Ruan, 1962).

Ruan(1962) and Wu(1989)observed that theRSDs associated with thunderstorms often had multiplepeaks. These multiple peaks might be attributedto the strong convection that occurred in thunderclouds.Bian et al.(1984), Jiang et al.(1986), and Xu et al.(1987)observed that the RSDs associatedwith Meiyu frontal precipitation had less small raindrops and more large raindrops than the classic exponentialsize distribution proposed by Marshall and Palmer(1948). Meiyu-front rainstorms had large rainwater contents but small raindrop number concentrations, leading to RSDs with wide spectral widths and large number concentrations of raindrops at large diameters.However, these observations were collectedby using oil and filter paper methods, which were unableto account for collision and coalescence amongraindrops. Several types of RSDs are associated withfrontal precipitation. Most of these RSDs contain multiplepeaks, while very few contain no peaks(Chen et al., 2013).

Chen et al.(2012)reported significant differencesin RSDs between the outer rainb and s and the eyewallregion of a typhoon system. The eyewall precipitationhad a broader size distribution than the outerrainb and s and eye region. Jiang(1988) and Chen and Gu(1989, 1990)suggested that RSDs in heavyrain can be best fit with gamma functions. Each ofthese studies focused on a single precipitation type.Gong et al.(1997a)provided a detailed analysis ofthe RSDs in the rainfall originated from three differentcloud types: stratiform, cumuliform, and mixedstratiform-cumuliform. The RSDs of rainfall fromthese three types of clouds exhibited some systematicdifferences. The RSDs associated with stratiformclouds had narrower spectral widths than those associatedwith cumuliform or mixed stratiform-cumuliformclouds. Furthermore, the RSDs associated with cumuliformclouds and mixed clouds had several spectralpeaks at large diameters.

Several other studies also suggested to separateprecipitation events into convective and stratiformcloud types(e.g., Liu and Lei, 2006). Figure 3 showsobserved RSDs in the rain originated from stratiformstratiformclouds, cumuliform clouds, and mixed stratiformcumuliformclouds(Fig. 3). The decrease in raindropnumber concentration with increasing diameteris fastest in the rains from stratiform clouds, so thestratiform RSDs have the narrowest spectral widthamong these three types. By contrast, the decreasein raindrop number concentration with increasing diameteris slowest in the rains from cumuliform clouds, so cumuliform clouds have the largest spectral width.All of these RSDs are concave downward in log-linearcoordinates. Raindrop number concentration in con-vective rainfall fluctuate significantly at large raindropdiameters, often leading to multiple peaks.

Raindrop size distributions are typically modeledas exponential distributions. The most widely usedRSD in the scientific literature is the distribution proposedby Marshall and Palmer(1948):

where N(D) is the raindrop number concentration and D is the raindrop diameter. The classic M-P distributiondefines N₀ as a constant equal to 8000 m^-3mm^-1 and λ as a function of the rainrate R(λ =4.1R^−0.21). Subsequent research has indicated thatthe “constant” N₀ is not constant at all, and alsodepends on R(Sekhon and Srivastava, 1971; Willis, 1984). Waldvogel(1974)pointed out that a suddenincrease or decrease in the value of N₀(an “N₀ jump”)indicates a transition from one type of rainfall to another, even if the intensity of the rainfall remains thesame. Chen et al.(1998)showed that N₀ increaseswith rain rate while λ decreases with rainn rate, and proposed the formulae N₀ = 517.9R^0.368 m^-3 mm^-1 and λ = 2.841R^−0.274 mm^-1. The results of thesestudies indicate that treating N₀ as a constant introducesimportant limitations, implying that both N₀ and λ should be allowed to vary across different RSDs.Yin et al.(2011)reported variations in N₀ between432.9 and 3036.4 m^-3 mm^-1(with an average valueof 1366.4 m^-3 mm^-1) and variations in λ between 1.7 and 3.13 mm^-1(with an average value of 2.82 mm^-1).

A number of previous studies(e.g., Takeuchi, 1978; Ulbrich, 1981; Chen and Gu, 1989, 1990)haveproposed that the RSDs for stratiform rainfall may bereliably modeled as gamma distributions(E_q.(1)).Gamma functions have been shown to provide goodfits for RSDs in the rains originated from stratiform, cumuliform, and mixed stratiform-cumuliform clouds, and have therefore been extensively used to describeRSDs. Long-term observations indicate that N₀ maytake values between 32.4 and 38248 m⁻³ mm⁻¹, whileλ varies between –0.19 and 3.62 mm⁻¹ and the value ofγ varies between –4.06 and 2.79. Several studies haveproposed that λ may be closely related to γ(Br and es et al., 2003, 2007; Cao et al., 2008; Yang et al., 2010;Chen et al., 2013). Analysis of long-term observationsindicates that the relationship between λ and γ can beapproximated with the following second-order polynomial

Numerous studies have suggested that the shapeparameter γ reflects important aspects of the microphysicalprocesses related to raindrops(e.g., Milbrandt and Yau, 2005b; Mansell et al., 2010). Mostcurrent microphysical schemes only calculate the intercept and slope parameter when predicting rain massmixing ratio and number concentration. The value ofγ is generally assumed to be constant. A new predictionfunction is required to predict the value ofγ. Milbrandt and Yau(2005b)developed a closurefor γ using radar reflectivity: given predictive equationsfor mass content, total number concentration, and radar reflectivity, γ becomes a fully prognosticvariable. However, radar reflectivity is not a dependentvariable, and is closely linked to mass content and total number concentration. As a result, this methodis not an effective means of predicting γ. To the bestof our knowledge, no current bulk microphysics model includes an explicit formulation of the relationship betweenγ and λ. The development of a prognostic formulationfor γ is recognized as a key requirement forfuture weather and climate models( Milbrandt and Yau, 2005a). The shape parameter γ plays importantroles in the computation of sedimentation and instantaneous growth rates in bulk microphysics parameterization.It may therefore be useful to allow γto vary as a diagnostic function of the predicted valueof λ rather than to use a fixed value. Seifert(2005)reportedthat the γ–λ relationship is reliable for strongconvective precipitation, but suggested that this relationshipmight not be appropriate for weak or stratiformprecipitation. Yang et al.(2010) and Zhang etal.(2011b)analyzed observations collected over China and reported similar conclusions.

4. Characteristics of solid particles 4.1 Ice nuclei

Ice nuclei(IN)acts as seeds for the formation ofice crystals in cold clouds. IN plays a significant rolein precipitation microphysics, including the Wegener-Bergeron-Findeisen process. Ice particles can oftenform via ice nucleation and multiplication processesat temperatures much warmer than the homogeneousfreezing threshold of approximately –40℃. The numberconcentration, size, shape, and other properties ofIN have significant effects on precipitation efficiencyin clouds. The results of recent cloud-resolving modelsimulations suggest that changes in IN can significantlyaffect cloud ensembles; these changes in cloudensembles can in turn affect the radiation budget and alter cloud feedbacks to climate change(Zeng et al., 2009).

the early 1960s. You and Shi(1964)observeda low concentration of IN(approximately 4.8 L^-1)at–20℃ over the suburbs of Beijing. Zhao et al.(1965)measured IN at three locations near Lanzhou, Xi’an, and Dalian. Their measurements indicated that theconcentration of IN is strongly influenced by both location and weather conditions. Wang et al.(1965)confirmed the influence of weather conditions on INconcentration and further proposed that IN concentrationconcentrationmay also be affected by sources of the air mass.The Laboratory of Cloud Physics in Hebei Province(1980)analyzed the vertical distribution of IN usingmeasurements collected at Taihang Mountain and duringaircraft campaigns. IN measurements were alsocarried out over southern China(e.g., Wang, 1983;Huang et al., 1986; Chen, 1994). You et al.(2002)presented observations of IN over the suburbs of Beijingusing the same Bigg-mixing cold-cloud-chamberused by You and Shi(1964). They concluded thatthe concentration of IN at –20℃ had increased by approximately15 times between 1964 and 1996. Recentefforts have collected numerous measurements in XinjiangRegion and the upper valley of the Yellow River(e.g., Li and Huang, 2001; Li and Du, 2003; Shi et al., 2006).

A large number of IN measurements have beencollected over China since 1963. Yin et al.(2012)reported that typical IN concentrations over Chinarange from 3.6 to 78.9 L^-1 at –20℃, with an overallaverage of 22.9 L^-1. Most past observations ofIN were collected over northern China, and relativelyfew over southern China. Huang et al.(1986) and Wu et al.(1986)collected IN measurements at ShitaMountain in Fujian Province and compared their resultswith measurements collected in other locations.They found that IN concentrations over Fujian weresimilar to those observed over Dalian, Washington, and Tokyo. Yang et al.(2013a, b)reported thatIN concentrations over Nanjing were lower than thoseover Beijing or Qinghai Province, and observed thataerosol particles with diameters greater than 0.5 μmcontributed a substantial fraction of all IN. Zhou etal.(2012b)recently observed a very high average INconcentration of 99.5 L^-1 at –20℃ in Shenyang. Theirobservations indicated a decrease in IN concentrationwith increasing height.

A number of studies have reported that IN concentrationincreases exponentially as temperature decreases(e.g., You and Shi, 1964; Huang et al., 1986;Shi et al., 2006). Fletcher(1962)proposed a parameterizationof IN number concentration N(T)that dependssolely on air temperature T:

where N₀ and b are empirical regression parameters and ΔT(℃)is the supercooled temperature. This parameterizationremains widely used in simulations ofcloud microphysics over northern China. Long-termmeasurements of IN over China indicate that appropriatevalues of N₀ range from 3.68 × 10⁻⁵ to 5.01L⁻¹, while values of b range from 0.02 to 0.51℃⁻¹.

Figure 5 shows observed variations in IN concentrationwith temperature. IN concentration consistentlyincreases with decreasing temperature. Therates of this increase with decreasing temperature aresimilar across the set of observations even though theobserved IN number concentrations vary substantially.This result implies small variations in b even as N₀ varies over several orders of magnitude, and can beattributed to strong variations in the spatio-temporaldistribution of IN. Yin et al.(2012, 2013)reportedthat average IN concentrations over China are larger and less sensitive to changes in temperature than thoseobserved over other parts of the world. Simulations ofcloud and precipitation processes over China shouldtherefore include representations of IN activation thataccount for these unique conditions.

In-situ measurements help us to underst and theproperties of IN and variations of IN with temperature, and therefore provide critical guidance for the improvementof cloud microphysical parameterizations(e.g., Hu and Yan, 1986; Hu and He, 1987; Xu et al., 2004). However, these measurements remain limitedin fundamental ways. First, previous in-situ measurementswere primarily conducted at and immediatelyabove the ground surface. Relatively few measurementsof IN have been conducted in the upper atmosphere, and the variation of IN concentration withheight remains highly uncertain. Second, previousmeasurements of IN have been concentrated in a fewlocations. Additional measurements with similar instruments and methods should be conducted at newlocations. Third, long-term continuous measurementssuitable for studying the effects and variations of INon climatic timescales do not yet exist.

4.2 Ice and snow crystals

Ice particles are an important part of cold clouds.The precipitation efficiencies of cold clouds are largelydependent on ice crystal concentrations(Zhao and Ding, 1963). Sun and You(1965)conducted the firstmeasurements of ice and snow crystals in stratiformclouds in Jilin Province over China. They reported amaximum ice crystal concentration of 221.9 L⁻¹ and an average of 26.2 L⁻¹. The properties of the observedcloud ice particles depended on the verticalrange of the cloud, and the concentrations of cloudice varied significantly with height. You et al.(1965)found that cloud ice concentrations increased with decreasingcloud top temperature, and reported cloudice concentrations one order of magnitude higher thanthe associated IN concentrations. This latter observationhas important implications for cloud microphysicsparameterizations, as it suggests that cloudice results not only from IN activation but also fromother ice multiplication processes(such as freezing ofliquid cloud droplets and collection of cloud dropletsby snow crystals). Wang et al.(1982)showed thatcloud ice concentrations differed significantly betweentwo nearby observational points at the same altitude, suggesting that cloud ice was distributed unequally inspace.

Ice particle size distributions are fundamentalcharacteristics of ice and mixed-phase clouds. Severalmathematical expressions for ice particle size distributions have been proposed. Heymsfield and Platt(1984)suggested that the variation of ice crystal numberconcentration between 10 and 100 μm could bemodeled as a power law distribution, while Ryan(2000)proposed using an exponential distribution.These two studies focused on single-peak ice particlesize spectra because it is difficult to find simple functionsthat represent bimodal and multimodal distributions.Mitchell(1994)used the sum of two gammadistributions to represent a bimodal spectrum, whilePlatt(1997)used the sum of a power law and a simplegamma distribution for the same purpose. Exponentialdistributions have been widely used to representice particle size distributions over China(e.g., Chen and Yan, 1987; Chen and Wang, 2001); however, N₀can vary by up to two orders of magnitude. The valuesof N₀ and λ both increase with decreasing in-cloudtemperature. The relationship between N₀ and λ canbe modeled as a power law, as suggested by Heymsfield and Platt(1984) and Ryan(2000).

The shapes of snow crystals are complex and determinedby a variety of factors, including humidity and temperature. You et al.(1965)calculated thespectra of different snow crystal shapes and comparedthese spectra with the overall spectrum. They classifiedsnow crystals into plates, columns, other shapeswithout needles, and dendrites(which were primarilylocated in the upper layers of clouds). Snow crystalconcentrations decreased exponentially with increasingparticle diameter. They showed that the multimodalspectrum of snow crystal size represented thesum of different spectra associated with different snowcrystal shapes: the size distributions of needles and dendrites could be separately modeled as lognormaldistributions. The majority of studies conducted duringthe past several decades have adopted exponentialsize distributions to describe snow crystal size distributionsover China(Chen and Yan, 1987; NorthernResearch Group of Artificial Precipitation Experimentfor Stratiform Clouds, 1991; Feng, 1993; WangXiangguo, 1997; Xiang and Niu, 2008).

4.3 Hail

Ice-phase precipitation in the form of hail can cause severe disasters. Hailstones result from a combinationof dynamic and electrical processes, and theirformation is a cornerstone topic in cloud physics. Themicrophysical properties of hail include the numberconcentration, inner structure, density, and size distributionof hailstones. Xu et al.(1965)classified hailembryos into four basic types and six sub-types. Theyalso used the properties of hailstone embryo chipsto infer the conditions of hailstone growth. Shi and Wang(1983)found that air bubbles in hailstone embryoscan reach a maximum size of at least 0.4 mm, with an average air bubble size of 0.1 mm. They reportedthat graupel contributed 73% of hail embryos, while frozen drops contributed the remaining 27%. Shi(1987)showed that the number of dry-wet layers inhailstones could exceed 20. Long-term observationsindicated that hailstone concentrations varied between0.1 and 5.1 m⁻³(with an average of 1.4 m⁻³) and hailstonediameters varied between 2 and 75 mm. Theminimum observed hailstone diameter was less than 1mm, while the maximum was greater than 10 cm.

Research into hailstone growth requires detailedknowledge of hail size distributions. Xu et al.(1965)proposed that hail size distributions be best modeledas exponential distributions. This conclusion hasbeen echoed by several subsequent studies(e.g., Yang et al., 1981; Zhao, 1982), although other studies reportedthat most observed hail size distributions weremore reliably modeled as power law distributions(e.g., Huang and Li, 1982). Shi et al.(1989)suggested thathail size distributions be classified into six characteristicpatterns, including exponential functions, compoundfunctions A and B, logarithmic functions, hyperbolicfunctions, and gamma functions. They arguedthat the exponential type was dominant amongthese types and accounted for approximately 60% ofall hail size distributions. Wang and Guo(1989)suggestedthat some hail size distributions be best representedas the sum of two or more of the functionaltypes identified by Shi et al.(1989). These studieshave focused almost exclusively on single-peak hailstonesize spectra, but the observed hailstone size spectraare often bimodal or multimodal(e.g., Wang and Wei, 1981; Zhao, 1982; Zhang and Sun, 2007). Table 3 lists functional fits to a variety of observed hail sizedistributions. The majority of these observed distributionshave been modeled with exponential fits. Theempirical regression parameter N₀ ranges from 0.0007to 13.68 m⁻³ mm⁻¹, and λ ranges from 0.10 to 0.8mm⁻¹.

4.4 Aerosol

Aerosol particles(AP)play significant roles inthe formation and development of clouds and precipitation, and modulate the role of clouds in the radiationbudget(Chen et al., 2011; Tao et al., 2012).Aerosol particles affect weather and climate in severalways. The aerosol direct effect decreases the amountof solar radiation reaching the surface by increasingthe reflection and absorption of solar radiation in theatmosphere(Haywood and Boucher, 2000). In additionto the aerosol direct effect, aerosols modulate theformation and development of clouds and precipitationthrough influences on cloud droplet nucleation.Aerosol particles act as cloud condensation nuclei, sothat an increase in the concentration of aerosols typicallyincreases the number of small cloud droplets;this tendency toward smaller cloud droplets slows thedevelopment of precipitation that depends on the formationof large drops(Ackerman et al., 2000). Cloudsin the aerosol-rich environments tend to be brighter, thicker, longer-lived, and more extensive than cloudsin clean environments, thereby reflecting more sunlight and partially mitigating the warming effects of increasinggreenhouse gases(Rosenfeld, 2000). Aerosols haveimportant effects on clouds, precipitation, and climate.Recognition of these effects has motivated extensiveresearch into aerosols and cloud-precipitation-aerosolinteractions.

The World Meteorological Organization(WMO)began to measure AP in the early 1970s, as part of acampaign for monitoring atmospheric pollution. Measurementsof AP in China have been conducted sincethe mid 1970s(Zhou et al., 1978). The introductionof the PMS enabled measurements of the vertical distributionof AP. He(1987)reported that aerosol concentrationgenerally decreased with increasing height, although the existence of cirrus clouds in upper layersappeared to invert this relationship(Fan et al., 2007). Temperature inversions limited the upwardtransport of aerosol particles and created discontinuitiesof aerosol concentration occurrences in the verticaldirection. This situation could also result in profilesof aerosol concentration that increased with increasingheight(Sun et al., 1996).

Changes in aerosol concentration depend primarilyon weather conditions. He(1987)reported thatconcentrations of aerosols in the size range 0.5–8 μmwere low(between 4 and 38 cm⁻³)under fair weatherconditions. These fair weather concentrations werecomparable to the values of 5.68 cm⁻³ observed byShi and Yang(1993) and 2.47 cm⁻³ observed by Wang(1994). The PMS device can observe APs with sizes assmall as 0.1 μm in diameter. As a result, aerosol concentrationsmeasured with the PMS are much higherthan those measured with other instruments. Fan etal.(2007)observed a maximum aerosol concentrationof 12169 cm⁻³ and an average aerosol concentrationof 6850.5 cm⁻³ in the atmosphere over Beijing and itssurrounding areas. They also found that aerosol con centrations varied substantially(with fluctuations of5%–18%)between nearby locations at the same altitude.Yin et al.(2011)reported that concentrationsof aerosols larger than 0.3 μm in diameter over EastAsia ranged from 73.1 to 353.0 cm⁻³, with an averagevalue of 166.9 cm⁻³.

The AP size distribution is one of the most importantmicrophysical properties of aerosols. Figure 6shows a variety of observed AP size distributions. Althoughaerosol number concentration consistently decreaseswith increasing diameter, aerosol number con-centration differs by up to four orders of magnitudefrom case to case.

The complexity and variety of observed aerosolsize distributions makes it difficult to identify a simplefunctional model that can be applied to the generalcase. Previous studies have proposed a variety ofempirical formulae to describe AP size distributions, including modified gamma functions and Junge powerlaw functions(e.g., Wang et al., 1984; You and Ren, 1990; Zhang et al., 1990). Numerous studies havesuggested that using piecewise combinations of thesetwo functional forms to describe AP size distributionsreduces errors associated with using either of the functionalforms alone(e.g., Zhu, 1982; Chen et al., 1996).The modified gamma function is generally expressedas

where D is the AP diameter(or equivalent diameterfor non-spherical large APs) and N(D)is the AP concentrationwithin a given diameter range. The fourparameters N₀, γ, λ, and μ are all positive, and arenot mutually independent. The parameter N₀ rangesfrom 1.4×10⁴ to 6.5×10⁷ cm⁻³ μm⁻¹. The parametersλ, γ, and μ are 8–20 μm⁻¹, 0–8.8, and 0.5–1, respectively. The Junge power law function is generallyexpressed aswhere N(r)is the total number of particles per unitvolume in the radius range from r to r+dr, and c is anempirical regression parameter. The value of v varieswith aerosol composition but is generally between 1.17 and 8.237. The parameter c typically takes values between0.01 and 29.12 cm⁻³.

In addition to their environmental impact, aerosols represent a difficult challenge in weather and climate simulation and prediction. Networksfor aerosol observation, such as the Global AtmosphereWatch(GAW)Program and China Aerosol Net(CAeroNet), have been designed and built in China and all over the world(Yan et al., 2006). Scientistsin China have paid considerable attention to aerosols and their effects on weather and climate, and substantialprogress has been made in these areas(Duan and Mao, 2009; Yang et al., 2011; Yue et al., 2011). However, existing aerosol observations have a number ofimportant limitations. In particular, many of the existingmeasurements have been collected in only a fewlocations, and long-term continuous measurements donot yet exist. A large fraction of the measurementshave focused on dust and aerosol optical depth, whileonly a few of these measurements have been applied tostudies of weather and climate. More measurementsare needed for future studies of weather and climate, including investigations into the effects of aerosols onprecipitation and cloud microphysics.

5. Concluding remarks

Observations of cloud-precipitation microphysicsare an effective tool for underst and ing cloud microphysicalprocesses, improving microphysical parameterizations, and exploring the interactions amongclouds, precipitation, and radiation. A large numberof observations of cloud and precipitation microphysicalproperties have been collected over China duringthe past 50 years. These observations provide a basisfor our underst and ing of cloud-precipitation microphysicsin China. They are used in this study to highlightthe large variations in the particle number concentration and LWC in the atmosphere over China, and to identify the likely ranges of these variations.

Size distributions of cloud droplets in stratiformclouds have often been described as gamma distributions, with the shape parameter γ taking the values of0–12. Size distributions of raindrops originated fromstratiform clouds have been modeled with both exponential and gamma functions, while size distributionsof raindrops originated from cumuliform and mixedstratiform-cumuliform clouds have generally(thoughnot exclusively)been modeled with gamma functions.The relationship between λ and γ across a range ofraindrop size distributions can be modeled by usingthe second order polynomial fit λ = −0.02701γ² +0.6166γ + 2.655.

The maximum observed concentration of IN at –20℃ was 99.5 L⁻¹. Observed aerosol particle concentrationsvary widely. Profiles of aerosol concentrationgenerally decrease with increasing height, withexceptions in the presence of temperature inversionsor upper-level cirrus clouds. Aerosol particle size distributionscan be most reliably described as piecewisecombinations of modified gamma functions and Jungepower-law distributions. The size distributions of icecrystals, snow crystals, and hailstones can generallybe modeled as exponential functions, but the best-fitparameters vary considerably from case to case.

Increasing evidence of fundamental deficiencies inthe bulk microphysical parameterizations used in numericalweather models has emerged over the past severaldecades. The current version of the Weather Research and Forecasting(WRF)model includes morethan 20 choices of cloud microphysical scheme(Skamarock et al., 2008). However, these cloud microphysicalschemes have been developed and tested basedon observations collected over region beyond China, and often provide contrasting results as shown in numerousreports evaluating the performance of differentWRF cloud microphysical schemes over China(Yin et al., 2014b). Advanced studies are needed to validatethese schemes and modify them to better representcloud-precipitation microphysical processes overthe East Asian monsoon region.

Measurements collected over China during thepast five decades suggest at least five potential avenuesfor improving current cloud microphysical schemes.(1)Statistical results should be used to evaluate differentcloud microphysical schemes.(2)Parameters(such as the intercept N₀ of raindrop and snow particlesize distributions)should be adjusted for better consistencywith existing observations.(3)Three-momentmicrophysical parameterizations should be improvedor introduced in models to better represent the apparentthree-parameter dependence of observed particlesize spectra.(4)The assumption that γ is constantshould be relaxed by allowing γ to vary as a diagnosticfunction of the predicted variable λ.(5)IN schemesshould be reconstructed to reflect the higher IN concentrations and weaker temperature dependence observedover China relative to other parts of the world.

Despite substantial progress in scientific underst and ingof cloud microphysics and continued improvementof cloud microphysical schemes, the representationof cloud-precipitation processes in weather and climate models remains poor. Cloud and precipitationmicrophysics are well known to be complex, and interactwith processes on spatial scales ranging from lessthan one micrometer(e.g., the size of an aerosol particle)to more than 1000 km(e.g., the scale of Rossbywaves and other synoptic features). Full underst and ing and reliable parameterization of the mechanismsunderlying these processes will require huge investmentsof both time and effort. Detailed observationalinvestigations are one important means of promotingour underst and ing of the complicated interactionsamong these processes. Improvements in observationalmethods directly strengthen efforts to develop and improve microphysical parameterizations. Giventhe requirement for microphysical parameterizationdevelopment, we offer the following recommendationsto improve the design of observational systems, withthe hope that these suggestions may provide guidancefor future measurements.

(1)Balanced measurements. The geographicaldistribution of past and existing measurements ishighly imbalanced, with a much larger fraction ofcloud-precipitation microphysical observations collectedover northern China than over southern China.We note a particular lack of microphysical observationsover the seas around China. In addition, the majorityof measurements were conducted in and aroundstratiform clouds, with few measurements taken in and around non-precipitating and convective clouds.It is vital to design a comprehensive observationalcampaign that can cover the whole China.

(2)Combining individual element observationwith multiple elements systematic observation. Futureprojects need pay attention to the mismatchbetween single observations and composite observationsin previous studies. It is difficult to develop afull underst and ing of cloud microphysical processesby discussing only a few variables, particularly giventhe complicated interactions among particles duringcloud-precipitation processes.

(3)Coupling instantaneous observation with longtermprocess observation. Previous observations havealso often failed to establish the links between instantaneousconditions and evolutionary states. The limitationsof observational methods and instruments haveprevented the collection of continuous data during thedevelopment of precipitation events, particularly duringthe early days of microphysical observations. Aseries of instantaneous conditions cannot completelydescribe the evolution of cloud microphysical characteristics.

(4)Complementing in-situ measurements withremote sensing observations. Careful combinations ofthese two data sources may significantly improve ourunderst and ing of cloud-precipitation microphysics.

(5)Extending the measurements from surface to upper air. The vertical distributions of microphysicalvariables are important information required bynumerical models. Thus, observations conducted at multiple vertical levels are necessary.