Effects of Cloud Liquid‐Phase Microphysical Processesin Mixed‐Phase Cumuli Over the Tibetan PlateauXiaoqi Xu1 , Chunsong Lu1 , Yangang Liu2 , Wenhua Gao3, Yuan Wang1 ,Yueming Cheng1, Shi Luo1, and Kwinten Van Weverberg4

1Key Laboratory for Aerosol‐Cloud‐Precipitation of China Meteorological Administration/Collaborative InnovationCenter on Forecast and Evaluation of Meteorological Disasters (CIC‐FEMD), Nanjing University of Information Scienceand Technology, Nanjing, China, 2Environmental and Climate Sciences Department, Brookhaven National Laboratory,Upton, NY, USA, 3State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences, Beijing, China,4Met Office, Exeter, UK

Abstract Numerical simulations often overpredict precipitation over the Tibetan Plateau (TP). Toexamine the factors causing precipitation overprediction, different parameterizations of liquid‐phasemicrophysical processes (accretion, autoconversion, and entrainment mixing) are implemented into theMorrison microphysics scheme to simulate a TP precipitation event in summer with the Weather Researchand Forecasting (WRF) model. The general spatial distribution and temporal trend of precipitation arecaptured by all simulations, but the precipitation rate is overpredicted. The results from sensitivityexperiments suggest that compared to other examined liquid‐phase processes, the accretion process is moreimportant in precipitation simulation over the TP region. Further investigation with the Heidke skillscores reveals that accretion parameterization that takes into account the raindrop size produces the mostaccurate results in terms of the total surface precipitation. This parameterization suppresses spuriousaccretion and does not produce liquid‐phase precipitation until cloud droplets are big enough. It is alsoconfirmed that increasing the model resolution can reduce precipitation overprediction. Results from thecase study are confirmed by the use of a 1‐month simulation.

1. Introduction

The Tibetan Plateau (TP) possesses an area larger than 2.5 × 106 km2 and an average elevation of more than4 km above the sea level, which is the largest and highest plateau of the world. Through strong thermal anddynamic forcing, the vibrant exchanges of heat andmoisture over the TP have significant impacts on climateand environmental changes, not only in China but also throughout East Asia and even the entire NorthernHemisphere (Ding et al., 2001; Flohn, 1957; Fu et al., 2006; Hahn & Manabe, 1975; Li et al., 2020; Li &Zhang, 2017; Molnar et al., 2010; Qie et al., 2005; Wang et al., 2008; Wu & Chen, 1985; Yanai et al., 1992;Yang et al., 2014; Ye, 1981; Yeh, 1950). The upward transport of latent heat and sensible heat are importantheat sources in the upper troposphere, which have vital effects on the maintenance of the Asian summermonsoon and associated precipitation events (Fan et al., 2013; Hsu & Liu, 2003; Jiang et al., 2004; Liet al., 2013; Luo & Yanai, 1984; Nitta, 1983; Ueda et al., 2003; Ueda & Yasunari, 1998; Yanai & Li, 1994).In the summer monsoon season, the developments of deep convection over the TP are frequently relatedtomesoscale vortices (Li, Wang, Song, et al., 2008; Shen et al., 1986; Wang et al., 1993). Summer precipitationon the TP has a significant diurnal cycle (Chen, Hu, et al., 2017; Fujinami & Yasunari, 2001; Kurosaki &Kimura, 2002) and can be characterized by weak and frequent convection (Chen, Hu, et al., 2017; Gaoet al., 2016; Li, Wang, Song, et al., 2008; Li & Zhang, 2016), which is significantly affected by the TP terrain(Chen, Wu et al., 2017; Porcù et al., 2014; Wu & Liu, 2017).

Many studiesfind simulated precipitation to be overpredicted (Gao et al., 2016;Maussion et al., 2011; Xu et al.,2012). There are several reasons for this overprediction, includingmodel resolution (Sato et al., 2008; Xu et al.,2012) and the initial and boundary conditions (Gerken et al., 2013). Moreover, the assumptions in the para-meterization of microphysics could be another culprit for the overprediction of precipitation over the TPregion. Maussion et al. (2011) found a significant sensitivity to changing the microphysics parameterizationfor convective precipitation over the TP. They compared different physical schemes with different statistical

©2020. American Geophysical Union.All Rights Reserved.

RESEARCH ARTICLE10.1029/2020JD033371

Special Section:The Land‐Air Coupling OverTibetan Plateau and Its GlobalClimate Effects

Key Points:• The accretion process plays more

important roles than otherexamined liquid‐phase processes

• Considering raindrop size inaccretion process suppressesliquid‐phase rain and mitigates theoverprediction of precipitationover the TP

• Increasing the model resolutioncan reduce precipitationoverprediction

Supporting Information:• Supporting Information S1

Correspondence to:C. Lu and Y. Liu,luchunsong110@gmail.com;lyg@bnl.gov

Citation:Xu, X., Lu, C., Liu, Y., Gao, W., Wang,Y., Cheng, Y., et al. (2020). Effects ofcloud liquid‐phase microphysicalprocesses in mixed‐phase cumuli overthe Tibetan Plateau. Journal ofGeophysical Research: Atmospheres,125, e2020JD033371. https://doi.org/10.1029/2020JD033371

Received 27 JUN 2020Accepted 15 SEP 2020Accepted article online 23 SEP 2020

Author Contributions:Conceptualization: Xiaoqi XuInvestigation: Xiaoqi Xu, WenhuaGao, Yuan Wang, Yueming ChengMethodology: Xiaoqi Xu, Yangang LiuResources: Kwinten Van WeverbergValidation: Yuan WangWriting ‐ original draft: Xiaoqi XuWriting – review & editing: XiaoqiXu, Yangang Liu, Wenhua Gao,Yueming Cheng, Shi Luo, Kwinten VanWeverberg

XU ET AL. 1 of 23

evaluation scores and the results showed that the effects ofmicrophysics schemeswere larger than other phy-sical schemes including cumulus schemes, land surface schemes, and boundary layer schemes.

The high TP altitude results in low melting levels and enables a large amount of supercooled cloud water(Gao et al., 2016; Tang et al., 2019; Zhao et al., 2017). Zhao et al. (2017) observed precipitating cumuli overNaqu and confirmed the dominance of supercooled liquid water in the temperature range of −3.5°C to−2.5°C. Therefore, liquid precipitation processes could be one of the reasons for the precipitation overesti-mation in this area. For instance, Li et al. (2006) analyzed the raindrop size distributions at Maqu over theTP and argued that despite ice‐phase rain processes were dominant over the region, liquid‐phase processeswere still critical for precipitation. Gao et al. (2016) examined the effects of liquid‐phase processes by exclud-ing ice‐phase microphysics, doubling the condensation rate, halving the evaporation rate and increasing theinitial droplet radius, and found significant effects from all these sensitivity tests on the surface precipitation;they also suggested that liquid‐phase rain processes could be more important than ice‐phase processes overthe precipitation cores during weak convection over the TP.

However, it is still unclear how liquid‐phase processes influence precipitation over the TP, and which liquid‐phase process is the most important. Also unknown is whether the problem of precipitation overestimationover the TP can be mitigated by improving the liquid‐phase process parameterizations, which parameteriza-tion can best describe the most important process and why. To overcome these problems, this study com-pares simulations of a typical precipitation event over the TP with different parameterizations ofaccretion, autoconversion, and entrainment‐mixing processes (hereafter referred to as liquid‐phase pro-cesses for convenience) and reveals the physical mechanisms. This is done using a case study, as well as a1‐month simulation. Moreover, the impact of model resolution is investigated following Sato et al. (2008)and Xu et al. (2012).

In this study, we focus on three parameterized liquid‐phase processes related to precipitation, includingaccretion, autoconversion, and entrainment mixing. These three processes are likely to be intertwined. Itis expected that accretion and ice/mixed‐phase processes matter in the precipitation events over the TP.But also, autoconversion and entrainment‐mixing process may be important. The reason is that accretionis the collection process between liquid raindrops and cloud droplets, and the number concentration andsize of liquid droplets are affected by autoconversion and entrainment‐mixing process. During thecollision‐coalescence process, total mass conversion from cloud to raindrop populations are usually dividedinto two aspects inmicrophysics schemes: Autoconversion is expressed as collision‐coalescence of cloud dro-plets while accretion is defined as the collection of cloud droplets by raindrops (Fan et al., 2016; Jinget al., 2019; Liu et al., 2011; Liu & Daum, 2004; Wang, Niu, Lv, et al., 2019; White et al., 2017;Wood, 2005b; Wood et al., 2009). Wang et al. (2012) and Gettelman et al. (2013) highlighted that the initia-tion of liquid‐phase precipitation was strongly affected by autoconversion, whereas the amount ofliquid‐phase precipitation was usually ascribed to accretion. Unlike autoconversion and accretion, theentrainment‐mixing process does not affect raindrops directly but affects cloud microphysical properties(number concentration and droplet size). Based on different cloud microphysical properties, cloud‐relatedprocesses such as radiation and precipitation may be different when different entrainment‐mixing mechan-isms dominate (Chosson et al., 2007; Cooper et al., 2013; Grabowski, 2006; Lasher‐Trapp et al., 2005; Luet al., 2013; Slawinska et al., 2008). The mixing process may cause a reduction of only droplet size (homoge-neous mixing), only droplet number (extremely inhomogeneous mixing), or both after evaporation.Entrainment‐mixing process remains uncertain in cloud physics (Gao et al., 2020; Grabowski, 2006;Hoffmann & Feingold, 2019; Kollias & Albrecht, 2000; Lu et al., 2013).

The organization of this paper is as follows: Section 2 gives an introduction of the precipitation event andconfigurations for simulations. Section 3 discusses the effects of liquid‐phase processes on precipitation over-prediction and underlying physical mechanisms. Section 4 extends the case study to 1‐month simulations.The effects of resolutions on precipitation overprediction are discussed in section 5. Section 6 presents thesummary and conclusions.

2. Precipitation Event, Experiment Description, and Method2.1. Case Description and Observational Data

A large frontal system passed through the TP from 21 to 23 July 2014 and precipitation started at 0400 UTC(Coordinated Universal Time) 22 July, according to Gao et al. (2016). The motivation for using this case is

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 2 of 23

that WRF can capture the main precipitation pattern (Gao et al., 2016). Gao et al. (2016) and this study focuson different periods and different scientific questions. As mentioned in the introduction, Gao et al. (2016)studied the sensitivity of precipitation to doubling droplet concentration, halving raindrop evaporation,and increasing initial cloud droplet size. This study focuses on different parameterizations ofautoconversion, accretion, and entrainment mixing. The simulations are compared against theprecipitation data set that Ma et al. (2018) derived from sparse gauge observations and multiple satelliteprecipitation data sets, including Tropical Rainfall Measuring Mission (TRMM) MultisatellitePrecipitation Analysis (TMPA) 3B42RT and 3B42V7 (Huffman et al., 2007), Climate Prediction CenterMORPHing technique (CMORPH) (Joyce et al., 2004), and Precipitation Estimation from RemotelySensed Information using Artificial Neural Networks‐Climate Data Record (PERSIANN‐CDR) (Ashouriet al., 2015). This rain gauge data set developed by Ma et al. (2018) is more accurate than individualgauges for complex terrains, for example, the TP region. Compared with TRMM that is often used in thisregion (Fu et al., 2007; Maussion et al., 2011; Qie et al., 2014; Xu et al., 2012; Yin et al., 2008), this data setalso has higher temporal (1 hr) and spatial (0.1°) resolution.

2.2. Numerical Experiment Description

The summer TP precipitation event mentioned above is simulated with the WRF model Version 3.8.1. Here,WRF is set as a cloud‐resolving model and the horizontal grid spacing of its innermost domain (03) is 1 kmwith 276 × 276 × 45 grid points, covering most of the plateau center. The two outer domains (02 and 01) havespatial resolutions of 5 and 25 km, and their grid points are 176 × 176 × 45 and 200 × 200 × 45, respectively(Figure 1). The WRF runs are driven by the National Center for Environmental Prediction Final operationalglobal analysis data. The simulations span from 1200 UTC 21 July to 0000 UTC 24 July and the last 48 hrresults from Domains 02 and 03 with a 30‐min interval are analyzed.

The Morrison double‐moment microphysics scheme is used in the control run (Morrison &Grabowski, 2008). This scheme can predict the mass mixing ratios and number concentrations of five hydro-meteors: ice crystals (qi, Ni), graupel particles (qg, Ng), snow particles (qs, Ns), raindrops (qr, Nr), and clouddroplets (qc, Nc), different from the default version that has a constant Nc. The expressions fromKhairoutdinov and Kogan (2000) (referred to as the KK00 schemes) are used to parameterize autoconversionrate (Au; kg/kg/s) and accretion rate (Ac; kg/kg/s):

Figure 1. Geographic locations of the three one‐way nested domains over the Tibetan Plateau. The color bar representsthe height (m) above the sea level (A.S.L.).

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 3 of 23

Au ¼ 1350 × q2:47c Nc × 10−6 × ρa� �−1:79

; (1)

Ac ¼ 67 × qcqrð Þ1:15; (2)

where ρa is the air density.

In order to achieve our goal of exploring the effects of liquid‐phase processes, several different expressionsfor the liquid‐phase processes are implemented into the Morrison scheme. For autoconversion, three com-monly used schemes (referred to as Be68, Bh94, and LD04) are employed:

1. Be68:

Au ¼ 3:5 × 10−2q2c

0:12þ 1:0 × 10−12Nc

qc

: (3)

This is the version usually used in global climate models derived from Berry (1968), for example, Modelfor Interdisciplinary Research on Climate Version 5 (MIROC5; Jing & Suzuki, 2018; Michibata &Takemura, 2015).

2. Bh94:

Au ¼ 6:0 × 1028n−1:7 qc × 10−3� �4:7

Nc × 10−6� �−3:3

; (4)

where n represents the cloud droplet size distribution width (Beheng, 1994) and is set to 10 here.

3. LD04:

Au ¼ P0T; (5a)

P0 ¼ 1:1 × 10131þ 3ε2ð Þ 1þ 4ε2ð Þ 1þ 5ε2ð Þ

1þ ε2ð Þ 1þ 2ε2ð Þq3cNc

� �; (5b)

T ¼ 12x2c þ 2xc þ 2� �

1þ xcð Þe−2xc ; (5c)

xc ¼ 9:7 × 10−14N3=2c q−2

c : (5d)

Besides liquid water mixing ratio and droplet concentration, LD04 from Liu and Daum (2004) andLiu (2005) also takes into account the ratio of the standard deviation to the mean radius, that is., rela-tive dispersion (ε). Xie and Liu, 2011 and Xie et al. (2013) implemented this scheme into theWRF double‐moment microphysics schemes. P0 and T are the rate function and threshold function,respectively; xc is the normalized critical mass and can be written as a function of Nc and qc(Liu, 2005). Based on observational studies of Wang, Niu, Lu, et al. (2019) and Zhao et al. (2006), εis set to 0.4.

Different from most accretion schemes which only relate accretion rate to mass mixing ratios of raindropsand cloud droplets (e.g., Beheng, 1994; Khairoutdinov & Kogan, 2000; Kogan, 2013), a parameterizationsimultaneously considering liquid droplet sizes and number concentrations (CP2k; Cohard & Pinty, 2000)is adopted:

Ac ¼ π6ρWρaK1

NcN r

λ3c

A1

λ3cþ B1

λ3r

� �if Rr ≥ 50 μm; and (6a)

Ac ¼ π6ρWρaK2

NcN r

λ3c

A2

λ6cþ B2

λ6r

!if Rr < 50 μm; (6b)

where K1 and K2 are empirical parameters; R is the drop radius; λ, the slope parameter, is related to themixing ratio, number concentration, and dispersion parameter (Morrison et al., 2005); as shown inAppendix B, A1, A2, B1, and B2 are derived based on the dispersion parameters in the modified Gammasize distribution (Cohard & Pinty, 2000); the subscripts r and c represent raindrops and cloud droplets,

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 4 of 23

respectively; ρW is the water density. In addition to the KK00 and the CP2k schemes, another accretionscheme (Kogan, 2013), named as Ko13, is also examined:

Ac ¼ 8:53 × q1:05c qr0:98: (7)

A single parameter α is used to represent the effect of entrainment‐mixing process in this microphysicalscheme (Lu et al., 2013; Morrison & Grabowski, 2008):

Nc ¼ Nc0qcqc0

� �α

; (8)

where qc0 and Nc0 are the mixing ratio and number concentrations of cloud water before the evaporationprocess, respectively; qc and Nc represent the corresponding cloud properties when new saturation isachieved after evaporation. It is noteworthy that qc is determined by qc0, relative humidity, air pressure,and temperature. Here, the parameter α is set to 0 for homogeneous mixing (the control run) and 1 forextremely inhomogeneous mixing (the INHOMO run).

In total, seven simulations are conducted: the control run with the default schemes, and sensitivity testsincluding CP2k and Ko13 (two accretion schemes); Be68, Bh94, and LD04 (three autoconversion schemes);and INHOMO (one entrainment‐mixing scheme). The case names and corresponding formulations are sum-marized in Table 1.

2.3. Calculations of Microphysical/Radiative Properties

The liquid cloud water path (LCWP) is calculated by

LCWP ¼ ∫H

0 ρaqc zð Þdz; (9)

where qc(z) is the cloud water mixing ratio at each height (z). The equation for cloud optical depth (τ) is

τ ¼ 321ρw

∫H

0ρaqc zð Þre zð Þ dz; (10)

where re(z) is the effective radius of cloud droplets at each height (z); H is the height of cloud top; the

Table 1Summary of Case Names and Corresponding Formulations for the Seven Experiments

Liquid‐phase process Case name Formulations

— Control run Au ¼ 1350 × q2:47c Nc × 10−6−1:79ρ−1:47a

� Ac ¼ 67 × qcqrð Þ1:15ρ−2:3

a

Autoconversion Be68Au ¼ 3:5 × 10−2q2c

0:12þ 1:0 × 10−12Nc

qc

Bh94 Au = 6.0 × 1028n−1.7(qc × 10−3)4.7(Nc × 10−6)−3.3

LD04Au ¼ 1:1 × 1013

1þ 3ε2ð Þ 1þ 4ε2ð Þ 1þ 5ε2ð Þ1þ ε2ð Þ 1þ 2ε2ð Þ

q3cNc

� �

×12x2c þ 2xc þ 2� �

1þ xcð Þe−2xc

Accretion Ko13 Ac ¼ 8:53 × q1:05c qr0:98ρ−2:03

a

CP2kAc ¼ π

6ρWρaK1

NcNr

λ3c

A1

λ3cþ B1

λ3r

!; if Rr ≥ 50 μm, and

Ac ¼ π6ρWρaK2

NcNr

λ3c

A2

λ6cþ B2

λ6r

!; if Rr < 50 μm.

Entrainment mixing INHOMO Nc ¼ Nc0qcqc0

� α, α = 1.

Note. The meaning of each symbol and reference for each case can be found in the text.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 5 of 23

extinction efficiency is equal to 2 approximately (Grabowski, 2006). With Equations 9 and 10, the columnmean effective radius (re ) is given by

re ¼ 32LCWPρwτ

: (11)

Note that only the cloud data in the grid boxes with hydrometeor mixing ratios larger than 0.01 g/kg areincluded.

3. Case Study Analysis3.1. Control Run3.1.1. Precipitation From the Control Run and ObservationsThe 48‐hr accumulated precipitation from 0000 UTC 22 July to 0000 UTC 24 July 2014 from the control runover Domain 02 and Domain 03 are averaged to the resolution of 0.1° to compare with observations(Figure 2). The results from Domain 02 indicate that, although the simulated results are spatially

Figure 2. Spatial distributions of observed and simulated (the control run) 48‐hr accumulated precipitation (mm) overDomains (a, b) 02 and (c, d) 03 from 0000 UTC 22 July to 0000 UTC 24 July 2014.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 6 of 23

consistent with the observations, the maximum simulated precipitation isapproximately 200 mm, largely exceeding the maximum observed value80 mm; furthermore, the observed precipitation amount is less than50 mm in most regions. Similar biases were reported in Xu et al. (2012)and Gao et al. (2016). For Domain 03, the spatial precipitation bias shownin Figures 2c and 2d is largely due to the insufficient data volume (onlyabout 27 × 27 data points), when the simulated precipitation is averagedto 0.1°. Besides spatial distributions, the temporal evolutions ofarea‐averaged precipitation rate (mm/hr) from the observation and thecontrol run over the two inner domains are compared in Figures 3a and3b, respectively. The trends of precipitation rate in the control run fromboth domains correlate well with those of the observations (black solidlines), but the result of Domain 03 is clearly closer to the observation.Observation shows two peaks of precipitation rate corresponding to thetime of local afternoon (UTC+ 6 hr) for Domain 03. The first precipitationevent spans from 0400 UTC to 1800 UTC and attains maximum precipita-tion rate (~1.0 mm/hr) near 0900 UTC. Another precipitation peak isweaker than the former one and its maximum precipitation rate is onlyabout 0.4 mm/hr. Similar to the results from Gao et al. (2018), the simu-lated peaks are about 2 hr delayed, compared with those in theobservations.

In general, the main features of the precipitation distribution and evolu-tion (the trends and the peaks) are reproduced by the control run.However, compared with the observations of 48 hr accumulated precipita-tion over the corresponding areas, the control run overpredicts precipita-tion by 51.5% in Domain 02 and by 20.8% in Domain 03.3.1.2. Microphysical Processes in the Control RunTo examine the origin of the precipitation biases discussed above, a morein‐depth analysis of the microphysics is performed for both Domains 02

and 03. For Domain 02, liquid‐phase precipitation in the southeastern corner is expected to be strongerdue to its lower altitude compared to the other regions. Therefore, Domain 02 is divided into two parts:Domain 02 except the southeastern corner and Domain 02 southeastern corner. For Domain 03, the two pre-cipitation peak periods (5 hr for each peak) are picked out separately to examine the effects of liquid‐phaseprocesses. The discussion in this section focuses on these four different parts.

The mean vertical profiles of all hydrometeors and their primary microphysical processes from the four dif-ferent parts are shown in Figure 4. For Domain 02, compared to the results in the other regions (Figures 4aand 4b), mixing ratios of ice‐phase particles (ice, graupel, and snow) and rates of ice‐phase microphysicalprocesses (RIM‐g, RIM‐s, andMELT) over the southeastern corner (Figures 4c and 4d) are similar or slightlysmaller. On the contrary, mixing ratios of liquid‐phase particles (cloud and rain) and rates of liquid‐phasemicrophysical processes (ACCR‐r and AUTO‐r) over the southeastern corner are larger than those overthe other regions of Domain 02. As mentioned above, due to the lower terrain, liquid droplets grow morefavorably over the southeastern corner. For Domain 03, considering the small precipitation rate over the sec-ond peak, most of the particles (rain, snow, ice, and graupel) are fewer and the microphysical processes(RIM‐s, RIM‐g, ACCR‐s, EVAP‐r, and MELT) are weaker during this peak period (Figures 4g and 4h) thanthose during the first peak (Figures 4e and 4f). On the other hand, although melting still dominates, the sec-ond peak has a larger accretion of cloud droplets by rain (ACCR‐r), compared to the first peak. By calculatingthe vertical accumulated conversions of liquid‐phase processes (ACCR‐r and AUTO‐r) and ice‐phase pro-cesses to raindrops, the contributions of ACCR‐r and AUTO‐r are 32.9%, 65.2%, 27.0%, and 35.4% for theother regions in Domain 02 except the southeastern corner, the southeastern corner in Domain 02, andtwo precipitation peaks in Domain 03, respectively. Therefore, compared to the other regions in Domain02 and the first precipitation peak in Domain 03, the liquid‐phase processes are more important over theDomain 02 southeastern corner and in the latter precipitation peak in Domain 03, respectively. However,this happens for different reasons. For Domain 02, although ice‐phase processes are always important

Figure 3. The temporal evolutions of area‐averaged precipitation rate(mm/hr) in the observations and the control run over Domains(a) 02 and (b) 03 from 0000 UTC 22 July to 0000 UTC 24 July 2014.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 7 of 23

Figure 4. Mean vertical profiles of hydrometeors (cloud, rain, ice, snow, and graupel) mixing ratios (g/kg) and mainmicrophysical process rates (kg/kg/s) in the control run averaged for 48 hr over two separate regions in Domain 02 andaveraged for 5 hr during two precipitation peaks in Domain 03: (a, b) Domain 02 except southeastern corner, (c, d)Domain 02 southeastern corner, (e, f) Domain 030700–1200 UTC 22 July 2014, and (g, h) Domain 030700–1200 UTC 23July 2014. The heights of 0°C isotherm are denoted by the purple dash‐dotted lines. Appendix A shows the meaningsof the symbols in the legends.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 8 of 23

Figure 5. Spatial distributions of observed and simulated (Be68, Bh94, LD04, CP2k, Ko13, and the INHOMO run)48‐hr accumulated precipitation (mm) over Domains (a–f) 02 and 03 (g–l) from 0000 UTC 22 July to 0000 UTC24 July 2014.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 9 of 23

over the entire domain, the low altitude in the southeastern corner isfavorable for liquid‐phase processes. In Domain 03, the second peak cor-responds to weak ice‐related processes.

3.2. Sensitivity of Precipitation Overprediction to DifferentLiquid‐Phase Processes

The effects of sensitivity simulations (Be68, Bh94, LD04, CP2k, Ko13, andINHOMO) on precipitation overprediction are examined in this section atgreat length. Surface precipitation, microphysical properties, and micro-physical processes are discussed.3.2.1. Evaluation of Surface PrecipitationSimilar to the control run, Figures 5 and 6 show the results for spatial dis-tribution and temporal evolution of precipitation from the sensitivitytests, respectively. CP2k has distinctly weaker precipitation, especiallyover the Domain 02 southeastern corner and in the Domain 03 secondprecipitation peak period, compared to the other sensitivity tests,although all other tests also reproduce similar spatial distributions andtemporal trends of precipitation. CP2k overestimates precipitation by30.2% in Domain 02, which is an improvement compared to the controlrun and other sensitivity tests (48.8–52.5%); in Domain 03, the overpredic-tion in the CP2k is only 10.9%, also much smaller than the 15.1% inINHOMO and 20.8–26.9% in the control run and other sensitivity tests.Some previous studies found large variations of precipitation by using dif-ferent autoconversion schemes (Li, Wang, & Zhang, 2008; Wanget al., 2013), while other studies showed that surface precipitation wasinsensitive to the choice of autoconversion schemes (Michibata &Takemura, 2015; Morrison & Grabowski, 2007). It seems that such sensi-tivity may vary by cloud regimes and even case by case. The limited effectsof different autoconversion schemes on the surface precipitation rate in

this study could be related to the dominance of melting and accretion in raindrop formation, and large pre-cipitation amount.

The ability of each sensitive experiment to simulate precipitation is further quantitatively evaluated by theHeidke skill score (HSS), which can not only judge well‐simulated events but also recognize the poor forecast(Barnston, 1992):

HHS ¼ 2 ad − bcð Þaþ cð Þ cþ dð Þ þ aþ bð Þ bþ dð Þ: (12)

Elements a–d are the numbers of “hits”, “false alarms”, “misses”, and “correct negatives”, respectively. Asshown in Table 2, a–d are determined by observation value po, simulation value ps, and threshold valuept. Here, pt is set to 2 mm covering most of the simulated and observed precipitation regions. A higherHSS represents better simulation skills.

Table 3 shows the values of elements a–d and corresponding HSS for allsimulations over Domains 02 and 03. In Domain 02, CP2k has the largestHSS with 0.508 while the HSS in all other cases slightly exceeds 0.4 and isclose to each other. The high value of d (“correct negatives”) in CP2kmainly over the southeastern corner results in a significantly higherHSS than that in the other cases. The close HSS from the other casesexcept CP2k indicates that the impacts of changing autoconversionschemes and entrainment‐mixing mechanisms on improving the overesti-mated precipitation are limited. However, an appropriate accretionscheme is a possible way to improve precipitation simulation. InDomain 03, the HSS of all simulations are much smaller compared to

Figure 6. The temporal evolutions of area‐averaged precipitation rate(mm/hr) in the observations and all simulations over Domains(a) 02 and (b) 03 from 0000 UTC 22 July to 0000 UTC 24 July 2014.

Table 2Contingency Table for the Elements to Calculate the Heidke Skill Score(HSS)

Observation po > pt Observation po ≤ pt

Simulation ps > pt a bSimulation ps ≤ pt c d

Note. The four elements a–d for HSS are determined by observation valuepo, simulation value ps, and threshold value pt.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 10 of 23

those in Domain 02, because the data points for evaluation are too few, as mentioned above. Therefore, aslight change in any of the four elements can result in a large difference in the HSS. Most of the othersimulations have the HSS smaller than 0.1, and the HSS of CP2k is the highest (0.152).3.2.2. Impact of Liquid‐Phase Parameterization on Cloud PropertiesThemicrophysical properties for all the simulations, including LCWP, re andNc over Domains 02 and 03 aresummarized in Table 4, respectively. The CP2k scheme has the largest effect on LCWP and re among all thesensitivity tests. Differences of the CP2k case with respect to the control run over Domain 02 (03) are +64.6%(+51.0%) in LCWP and +7.9% (+5.6%) in re, while the other accretion sensitivity test, using Ko13, is close tothe control run. The reasons will be analyzed in section 3.3.2. Cloud observational data from Clouds and theEarth's Radiant Energy System (CERES) project with 1‐hr time resolution and 1° spatial resolution are alsoprovided in Table 4. Among all the simulations, the CP2k case has the results of LCWP and re closest to theobservations.

The difference between all autoconversion sensitivity tests and the control run for LCWP and re are, respec-tively, from −13.7% to 10.6% and −1.9% to 2.3% over Domain 02, and from −11.1% to 13.9% and −1.2% to3.5% over Domain 03. The wide ranges are mainly caused by one order of magnitude difference of Au amongthe different sensitivity tests (Figures 7a and 7b). However, the magnitude of the difference is much smallerthan that in typical marine boundary layer clouds (Wood, 2005b), because of the thinner liquid‐phase layerand the involvement of ice‐/mixed‐phase processes over the TP. The sign of the difference between the

Table 4The Area‐Averaged Liquid Cloud Water Path LCWP (g/m2), Mean Effective Radius re (μm), and Number ConcentrationNc (/cm

3) of Cloud Droplets Over Domains 02 and 03 (d02/d03) of the Control Run, Three Autoconversion Cases (Be68,Bh94, and LD04), Two Accretion Cases (CP2k and Ko13), and One Entrainment‐Mixing Case (the INHOMO Run)

LCWP (g/m2) re (μm) Nc (/cm3)

Control 73.5/66.8 6.97/6.77 71.5/91.2AutoconversionBe68 63.4/59.4 6.84/6.74 71.3/91.6Bh94 81.3/76.1 7.13/7.01 72.3/91.9LD04 63.8/60.9 6.85/6.69 71.6/91.3AccretionCP2k 121.0/97.0 7.52/7.15 72.4/90.1Ko13 74.4/64.4 6.92/6.72 71.5/91.0Mixing mechanismINHOMO 72.9/66.7 7.03/6.87 68.9/86.3ObservationCERES 100.4/91.2 9.13/8.75 —

Note. The observation results from Clouds and the Earth's Radiant Energy System (CERES) are also shown.

Table 3The Values of Four Elements a–d and Corresponding Heidke Skill Score (HSS) for All Simulations Over Domains 02 and03 (d02/d03) of the Control Run, Three Autoconversion Cases (Be68, Bh94, and LD04), Two Accretion Cases (CP2k andKo13), and One Entrainment‐Mixing Case (the INHOMO Run)

a b c d HSS

Control 2,636/304 1,224/148 773/76 2,231/48 0.419/0.049AutoconversionBe68 2,645/309 1,261/142 764/71 2,194/54 0.411/0.097Bh94 2,533/306 1,148/138 876/74 2,307/58 0.411/0.110LD04 2,628/313 1,264/154 781/67 2,191/42 0.405/0.043AccretionCP2k 2,583/304 1,063/129 632/76 2,586/67 0.508/0.152K013 2,620/303 1,223/146 770/77 2,251/50 0.420/0.057Mixing mechanismINHOMO 2,656/308 1,124/141 753/72 2,214/55 0.420/0.100

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 11 of 23

schemes is consistent with previous studies, for example, Be68 and LD04 have larger Au than KK00(Figures 7a and 7b) (Lee & Baik, 2017).

For the entrainment‐mixing process, the INHOMO experiment has the largest effect on Nc among all thesensitivity tests. Even so, there is only a modest reduction of only 2.6 (4.9)/cm3 compared to the controlrun, resulting in 0.9% (1.5%) larger re over Domain 02 (03). It seems that the higher resolution of Domain03 results in the stronger effect of INHOMO in Domain 03 than that in Domain 02, because the scales rele-vant to the entrainment‐mixing process are often small. Such a variation of re over Domain 03 is comparablewith that in all the autoconversion schemes and Ko13. The variation of LCWP is smaller than that in theliquid conversion process. Sensitivity tests of all autoconversion and accretion schemes in Table 4 are alsoconducted assuming different mixing mechanisms (see supporting information for details). The effects arealso small, similar to several previous studies using double‐moment microphysics schemes (Grabowski &Morrison, 2011; Hill et al., 2009; Slawinska et al., 2012). One important reason is that the relative humidityin entrained air is high (Hoffmann & Feingold, 2019; Slawinska et al., 2012).

Figure 7. The temporal evolutions of area‐averaged autoconversion rate (kg/kg/s), accretion rate (kg/kg/s), and the ratio of accretion rate to autoconversion rate(Ac/Au) for all simulations over (a, c, and e) Domains 02 and (b, d, and f) 03, respectively.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 12 of 23

3.3. Reasons for Improvements of Precipitation Overprediction in CP2k

Asmentioned before, CP2k reduces the precipitation overpredictionmore significantly than other sensitivitytests. The reasons are discussed in this section.3.3.1. Detailed Microphysical Processes in CP2kThe Ac rate in CP2k is one to 2 orders of magnitude smaller than that in the other tests (Figures 7c and 7d).The reason whyAc in the other tests is so close is that the KK00 and the Ko13 schemes have similar functionsof rain and cloud water content (Equations 2 and 7) and similar variation trends in Figure 9 (which is ana-lyzed in section 3.3.2 in detail). In each case, larger Au enhances Ac by providing more raindrops. Therefore,the trend of Ac corresponds well to that of Au. The relationship between Ac and Au between different cases isdifferent from that in each case. Compared with the control run,Ac in CP2k is smaller, and thus, fewer clouddroplets are collected by raindrops; these surviving cloud droplets are then available for autoconversion,

Figure 8. The vertical differences of the dominant microphysical process rates between CP2k and the control run(CP2k‐Control) averaged from two separate regions in Domain 02 and two precipitation peaks in Domain 03: (a) Domain02 except southeastern corner, (b) Domain 02 southeastern corner, (c) Domain 030700–1200 UTC 22 July 2014, and(d) Domain 030700–1200 UTC 23 July 2014. The heights of 0°C isotherm are denoted by the purple dash‐dotted lines.Appendix A shows the meanings of the symbols in the legends.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 13 of 23

which leads to the largerAu in CP2k (e.g., Gettelman et al., 2013; Posselt &Lohmann, 2008). As shown in Figures 7a and 7b, the difference of Au

between CP2k and the control run has similar magnitude as applying dif-ferent autoconversion schemes. Therefore, CP2k has the lowest ratio Ac/Au with the mean value of 2.9 (2.8) over Domain 02 (03), mainly becauseof small Ac. Bh94 has the largest ratio Ac/Au with a mean value of 151.2(144.6) over Domain 02 (03). Indeed, Bh94 exhibits the smallest Au

(Figures 7e and 7f) of all schemes tested here. Ac/Au of all schemes is inthe range of 0.1–296.3, consistent with previous studies (Gettelmanet al., 2013; Jiang et al., 2010; Lee & Baik, 2017; Michibata &Takemura, 2015; Seifert & Onishi, 2016). In Domain 03, Ac/Au larger than1 usually corresponds to larger precipitation intensity; Ac/Au smaller than1 usually corresponds to smaller precipitation at the start or the end ofprecipitation events. This correspondence is consistent with the argu-ments as to accretion‐dominated and autoconversion‐dominated regimes(Jiang et al., 2010; Michibata & Takemura, 2015; Wood et al., 2009). In

Domain 02, most of Ac/Au is larger than 1; some Ac/Au values are smaller than 1 but still with strong preci-pitation likely caused by the influence of ice‐/mixed‐phase processes.

Combining two main liquid‐phase conversion processes from cloud droplets to raindrops (autoconversionand accretion), the loss of cloud water is less in CP2k than that in the control run, which leads to over50.0% larger LCWP in CP2k than that of the control run, as shown in Table 4. The vertical differences ofthe dominant microphysical process rates between CP2k and the control run (CP2k‐Control) over differentregions and during different periods are shown in Figure 8. Similar to Figure 7, CP2k has much largerAu andsmaller Ac than the control run. Despite the larger Au, the larger LCWP of CP2k indicates more cloud dro-plets survive in the atmosphere above the 0°C isotherm, which are beneficial for the riming processes(RIM‐s + RIM‐g). Figure S1 in the supporting information shows the microphysical processes conversionrates in CP2k with riming minus those without riming. Riming suppresses the liquid‐phase rain formationprocesses through reducing Ac but enhances ice‐/mixed‐phase rain formation processes through increasingmelting rate. The sensitivity of warm/cold rain formation to riming ultimately trickles down to uncertaintiesin the simulation of surface precipitation.

Note that the melting level in CP2k is lower than that in the control run over Domain 03. Given the largerwater content, CP2k also has a larger optical depth τ (14.3) than the control run (11.1), which leads to morereflection of solar radiation to the upper atmosphere and less downward short‐wave radiation at the surface(219.6 and 226.5 W/m2 in CP2k and the control run, respectively). Qualitatively, such a difference in radia-tion could be one reason responsible for a lower temperature in CP2k in the low atmosphere than in the con-trol run, other things being equal (e.g., latent heating release). Therefore, the melting level is lower in CP2kand then the melting rate is smaller near 6–6.5 km.

For the first precipitation peak period in Domain 03, despite Ac in CP2k is smaller than that in the controlrun, more riming leads to more ice‐phase particles, and thus more melting below the melting level. Theweaker accretion but stronger melting in CP2k offset each other. As a result, the precipitation from CP2kis very close to the control run in this period (Figure 6b). A similar explanation can also apply to Domain02 except for the southeastern corner (Figures 2b and 5d). However, for the southeastern corner inDomain 02 and during the second peak period in Domain 03, the liquid‐phase processes become relativelymore important as discussed before. Therefore, the melting process cannot compensate for the suppressionof accretion in CP2k, which appears to alleviate the overestimation of precipitation and the surface precipi-tation in CP2k is closer to the observations than that in the control run.3.3.2. Theoretical Analysis of the CP2k ParameterizationThe different equations for accretion process (Equations 2, 6a, and 7) should account for the large differencesin precipitation and related properties between CP2k and other cases. As mentioned in Wood (2005b), thereare two main methods to establish the autoconversion and accretion schemes. First, such as the KK00scheme, the stochastic collection equation is integrated from a wide range of drop size distribution and thenthe equation is simply fitted by a power law function. Second, such as the autoconversion scheme in LD04

Figure 9. The relationship between accretion rate and raindrop radius forthe three accretion schemes (the control run, CP2k, and Ko13) underthe condition of raindrops number concentration Nr = 4,000/m3, cloudmixing ratio qc = 1 g/kg, and the cloud droplet radius Rc = 10 μm.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 14 of 23

and accretion scheme in CP2k, the collection kernel is simplified assum-ing drop size distributions, and then the autoconversion or accretion pro-cess is parameterized. Autoconversion schemes are commonly developedby using one of these basic methods and there are large differencesbetween different autoconversion schemes. However, the accretionschemes used in most of the microphysical schemes and compared in pre-vious studies are based on the first method (Hill et al., 2015; Wood, 2005b).The CP2k accretion parameterization derived by the second methodappears superior to other parameterizations, but this scheme is onlyapplied in a few microphysics schemes (e.g., WDM6 scheme in WRF;Lim & Hong, 2010) and seldom discussed in the comparison of differentaccretion schemes.

The Ac calculated as a function of raindrop radius under the conditionsof qc = 1 g/kg, Rc = 10 μm, and Nr = 4,000/m3 for the three accretionschemes (the control run, CP2k, and Ko13) is compared in Figure 9. Forthe first method with the power law function, that is, the KK00 schemein the control run and the Ko13 scheme, Ac only depends on the rainwater mixing ratio if cloud water is adequate and is linearly related toraindrop radius in the logarithmic space. However, due to the piecewisefunction bounded by 50 μm of raindrop radius, CP2k has an inflectionpoint at 50 μm. The Ac is very small in CP2k, if the raindrop radius isless than 50 μm. Figure 9 shows that, compared to the other twoschemes, the Ac in the CP2k scheme is always smaller when the rain-drop radius is smaller than 2,000 μm. When the raindrop radius is smal-ler than 50 μm, the maximum difference between CP2k and the othertwo schemes can reach more than 2 orders of magnitude. Therefore,the differences between different accretion schemes are largely attribu-ted to the probability density distributions (PDFs) of raindrop radius.

Figure 10 shows the PDFs of raindrop radius. The peaks of PDFs in thecontrol run, Ko13, and CP2k are ~30, ~30, and ~25 μm, respectively,and the cumulative PDFs show that the proportions of raindrops radiussmaller than 50 μm are 58.8%, 53.8%, and 46.0%, respectively. These largepercentages of small raindrops cause the Ac and precipitation of CP2k tobe quite different from those in the other schemes (Figure 9). A large pro-portion of liquid droplets with radii in the range of 25–50 μm is also con-firmed in the previous aircraft observations and bin model simulations(Morrison & Grabowski, 2007; Wood, 2005a). Furthermore, there is apositive feedback mechanism, because the larger the Ac, the larger theqr. The larger qr further leads to a larger Ac in the case of sufficient cloudwater. The overestimation of the Ac in KK00 or Ko13 hence feeds back on

itself. This is the reason why the differences of precipitation and Ac between CP2k and the other cases are sosignificant over the Domain 02 southeastern corner and during the Domain 03 second peak period.

Previous studies have shown that liquid‐phase precipitation does not exceed 2 mm/day until the cloud effec-tive radius reaches about 14 μm (Rosenfeld et al., 2019). To make an in‐depth study and understand the dif-ferences in precipitation behavior between CP2k and the other experiments, Figure 11 shows therelationship between liquid‐phase precipitation rate and cloud droplet effective radius for the three accretioncases. The total precipitation can be regarded as the sum of liquid‐phase processes (autoconversion + accre-tion) and ice‐/mixed‐phase processes (melting from snow + graupel); therefore, the liquid‐phase precipita-tion rate is estimated according to the total precipitation rate and the ratio of liquid‐phase processes to ice‐/mixed‐phase processes. In the control run and Ko13, the liquid‐phase precipitation rate is larger than 2 mm/day when the cloud effective radius reaches 9 μm, which is much smaller than the critical value suggested byobservational studies (Rosenfeld et al., 2019). Differently, the critical value of cloud effective radius in CP2k

Figure 10. Probability distribution functions (PDFs) of raindrop radiusused in the accretion parameterizations and corresponding cumulativePDFs for (a) the control run, (b) CP2k, and (c) Ko13. The purple linedenotes the radius of raindrop equal to 50 μm.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 15 of 23

is about 15 μm. Moreover, the contribution from autoconversion is closeto 0 in the control run and Ko13, which could be caused by the consump-tion of cloud droplets by the strong accretion processes after cloud dro-plets reaching 9 μm. Since observations suggest that 14 μm is needed toinitiate liquid‐phase precipitation (Rosenfeld et al., 2019), 9 μm is smallerand seems not reasonable. On the contrary, both autoconversion andaccretion rates significantly increase after cloud effective radius reaching15 μm in CP2k. Although Au in CP2k is larger than that in other schemes,Ac ultimately determines the liquid‐phase precipitation rate, which hasbeen discussed in many previous studies (e.g., Gettelman et al., 2015;Jiang et al., 2010; Michibata & Takemura, 2015; Wood et al., 2009). Theliquid‐phase precipitation is suppressed by weak Ac. Furthermore, largeAu in CP2k can increase qr but decrease qc, which may enhance or sup-press Ac (Posselt & Lohmann, 2008). In other schemes, the accretion pro-cess is triggered to a considerable amount with small liquid drops due tothe overestimation of Ac when confined to small drops. Therefore, themitigation of precipitation overestimation in CP2k appears to be rightand has physical reasons.

4. Long‐Term Analysis

While the analysis of the single case study has allowed for an in‐depthanalysis, it remains to be verified whether this case study is representa-tive of the general behavior of the model. As pointed out by Whiteet al. (2017), it is hard to be conclusive that one scheme is better thanothers based on a few cases. Hence, 1‐month simulations are performedfrom 0000 UTC 21 July to 0000 UTC 21 August 2014 with the samedomains in Figure 1, using the three accretion schemes (the controlrun, CP2k, and Ko13). Only the results starting from 0000 UTC 22 Julyare analyzed. The horizontal resolutions for Domains 01, 02, and 03 are30, 10, and 3.3 km, respectively; except for the resolutions and simulationtime, all other settings are the same as those in the 2‐day simulations insection 3.

Figure 12 shows the temporal evolution of the area‐averaged daily preci-pitation rate in Domains 02 and 03 from the three accretion simulationsand the observations. Compared with the observed precipitation, the con-trol run significantly overestimates precipitation for most days, especiallyin Domain 02. The average precipitation rate in the observation, the con-trol run, Ko13, and CP2k are, respectively, 1.6, 2.5, 2.5, and 2.2 mm/dayover Domain 02, and 4.5, 5.8, 5.9, and 5.2 mm/day over Domain 03. Theresults of Ko13 are very close to those in the control run, while CP2k sig-nificantly reduces precipitation overprediction with p values of Student's ttest less than 0.01 for both Domain 02 and Domain 03. Table 5 shows thatCP2k has higher HSS than the control run and Ko13 over both Domains02 and 03. Therefore, the effects of CP2k on reducing precipitation over-prediction are not limited to one specific case but appear to be a plausibleway to improve precipitation overprediction, at least of the TP.

The 1‐month simulations provide the opportunity to investigate theresponse of the PDF of the surface precipitation to the changes ofaccretion schemes over the TP region. The PDFs are based on thehourly precipitation rate from 0000 UTC 21 July to 0000 UTC 21August 2014 with the three accretion schemes (the control run,CP2k, and Ko13). As expected, Figure 13 shows that CP2k has more

Figure 11. Relationship between liquid‐phase precipitation rate and clouddroplet effective radius for the three accretion schemes (the control run,CP2k, and Ko13) over Domain 03 from 0000 UTC 22 July to 0000 UTC24 July 2014.

Figure 12. Time series of area‐averaged daily precipitation rate (mm/day)from 0000 UTC 22 July to 0000 UTC 20 August 2014 over (a) Domain 02and (b) Domain 03 in the observations and three accretion cases(the control run, CP2k, and Ko13).

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 16 of 23

weak precipitation (<~0.2 mm/hr) over both Domains 02 and 03 thanthe control run and Ko13, corresponding to the smaller total surfaceprecipitation in CP2k. The results indicate that the PDF of the surfaceprecipitation is subject to the changes in the microphysical schemesover the TP region. Furthermore, considering the significant climateeffects of the TP region, it is interesting to see whether microphysicalschemes have significant effects on the historical trend of precipitationintensity distribution with simulations over years in the future researchusing the method in Wang et al. (2016).

5. Sensitivity to Horizontal Resolution

Different resolutions in the two simulations (case study from 22 to 23 July2014 and long‐term 1‐month simulation from 22 July to 21 August 2014)

provide a good opportunity to examine the effects of resolution on precipitation overprediction. Thearea‐averaged precipitation rate during the first 2 days (from 22 to 23 July) from the 1‐month simulationis 0.36 mm/hr in Domain 03 with the resolution of 3.3 km (Figure 12b), which is larger than the0.29 mm/hr with the resolution of 1 km and the 0.24 mm/hr from the observations (Figure 6b).Similarly, the first 2‐day area‐averaged precipitation rate is 0.41 mm/hr in Domain 02 with the resolutionof 10 km (Figure 12a), larger than 0.32 mm/hr with a resolution of 5 km and 0.21 mm/hr from the observa-

tions (Figure 6a). Furthermore, the comparison with the precipitation inDomains 02 and 03 also provides some hints on the effects of resolutions.For the one case study, Domain 02 with the resolution of 5 km(Figure 6a) overpredicts 51.5% of precipitation compared with observa-tions, and the number for Domain 03 with the resolution of 1 km is only20.8% (Figure 6b). For the 1‐month study, Domains 02 and 03 overpredictprecipitation by 57.7% and 27.8%, respectively (Figure 12). These resultsconfirm that the model grid size plays an important role in the overpre-diction of precipitation over the TP.

Sato et al. (2008) investigated diurnal variation of precipitation over theTP with different model resolutions and showed that the results fromhigher‐resolution simulations were more consistent with observations.They claimed that higher resolution (<7 km) may resolve the convectioninitially occurred by the surface heating and consequently conduct aproper simulation of the precipitation. Xu et al. (2012) found that theWRF simulations at a resolution of 3 km could reproduce the timing ofprecipitation events but the intensities were doubled. The first 2‐day com-parison in our study shows that, when the horizontal resolution increasesfrom 3.3 to 1 km, the simulation of precipitation intensity can be effec-tively improved without affecting the trends of precipitation. This indi-cates that 1 km or even higher resolution is needed to accuratelysimulate the precipitation over the TP, possibly because such high resolu-tions can better resolve the orography of this region.

6. Summary and Conclusions

In order to understand the effects of liquid‐phase microphysical processeson the overprediction of precipitation over the TP, a typical summer pla-teau precipitation event is simulated using theWRFv3.8.1 model with sev-eral sensitivity experiments. Sensitivity tests are conducted by successivelyintroducing three autoconversion schemes (Be68, Bh94, and LD04), twoaccretion schemes (CP2k and Ko13), and one entrainment‐mixing scheme(INHOMO) into the Morrison double‐moment scheme.

Table 5The Values of Four Elements a–d and Heidke Skill Score (HSS) for Three1‐Month Simulations Over Domain 02 and Domain 03 of the ControlRun and CP2k and Ko13 (Different Accretion Schemes)

a b c d HSS

Domain 02Control 3,780 5,052 1924 24,584 0.403CP2k 3,749 4,369 1955 25,267 0.435Ko13 3,764 4,825 1940 24,811 0.413Domain 03Control 1,188 2,856 93 2,538 0.220CP2k 1,163 2,355 118 3,084 0.262Ko13 1,181 2,908 100 2,531 0.211

Figure 13. Probability distribution functions (PDFs) of precipitation rate(mm/hr) from 0000 UTC 21 July to 0000 UTC 21 August 2014 over (a)Domain 02 and (b) Domain 03 in the three accretion cases (the control run,CP2k, and Ko13).

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 17 of 23

The control run can capture the primary spatial distribution and temporal evolution of the precipitationevent, but the amount of precipitation is significantly overestimated. The precipitation overprediction issignificantly mitigated with the accretion scheme from Cohard and Pinty (2000), which is confirmed byHSSs. Furthermore, to understand the reason for the large difference between CP2k and the other cases,each simulation is further divided into two parts according to the relative importance between theliquid‐phase and ice‐phase processes. When ice‐/mixed‐phase processes are relatively weak, simulationshave significant differences and the mitigation of precipitation overestimation in the CP2k experiment ismore pronounced. When the ice‐/mixed‐phase processes dominate, CP2k is not much different fromother simulations. There are several reasons. On one hand, the accretion rate in CP2k is smaller thanthat in the other cases, which suppresses liquid‐phase precipitation. On the other hand, ice‐phase pre-cipitation due to melting from snow and graupel is enhanced because weaker accretion means morecloud droplets remain in the atmosphere and are available for the riming process. For the period whenice/mixed‐phase processes dominate, the combination of the weak accretion but strong melting in CP2koffsets each other and leads to similar precipitation in CP2k as in the other cases. For the period whenice/mixed‐phase processes are relatively weak, the suppression of liquid‐phase precipitation arises fromthe weak accretion process cannot be compensated by the enhanced precipitation from the strongriming and melting processes. Therefore, the precipitation rate is smaller in CP2k than that in theother cases.

To understand the physical mechanisms for the improvement in CP2k, the equations for parameterizingthe accretion process in the control run, CP2k, and Ko13 are compared directly. When the raindropradius is smaller than 2000 μm, the accretion rate in CP2k is always smaller than that in the other twoschemes. Furthermore, when raindrop radius decreases, the difference increases. Especially, for the rain-drop radius smaller than 50 μm, the difference can reach more than two orders of magnitude. The largeproportion of raindrops having radius less than 50 μm (~50%) results in the suppression of accretion andliquid‐phase precipitation in CP2k than in the other two schemes. Further insight into the reasons for theunique behavior in CP2k is provided through the relationship between liquid‐phase precipitation rate andcloud droplet size. When liquid‐phase precipitation exceeds 2 mm/day, the cloud effective radius in thecontrol run and Ko13 is only about 9 μm, which is much smaller than the critical value of 14 μm in obser-vations (Rosenfeld et al., 2019). However, liquid‐phase precipitation does not initiate until the cloud effec-tive radius reaches about 15 μm in CP2k, which is more consistent with observations than that in theother cases.

The results in terms of precipitation sensitivity are confirmed in a long‐term 1‐month simulation as well. Thetime series of daily precipitation rate indicates that the reduction in precipitation bias using CP2k is gener-ally valid. CP2k also has the highest HSS. Hence, it is assumed that the CP2k scheme generally producesmore accurate simulations of precipitation, at least over the TP. More studies are needed to understandwhether these findings are applicable to regions beyond the TP as well. Theoretically, accretion is signifi-cantly affected by cloud droplet and raindrop sizes, and these sizes are related to number concentrationsand liquid water mixing ratios of cloud droplets and raindrops. Therefore, it may be more convincing tosimultaneously consider number concentrations and liquid water mixing ratios in the future developmentof the accretion parameterizations, similar to CP2k.

We also confirm that higher‐resolution simulations reduce precipitation overestimation compared tolower‐resolution simulations, as pointed out by previous studies (e.g., Sato et al., 2008; Xu et al.,2012). For the same simulation domains with different resolutions, the results in the high‐resolutionsimulations are much closer to the observations.

It is noteworthy that small impacts of changing entrainment‐mixing mechanisms and autoconversionschemes could be related to the cloud type in the TP area. More studies are needed to further examine theimpacts of these processes and accretion on cloud and precipitation under different conditions, with the ratioof accretion rate to autoconversion rate (Gettelman et al., 2013; Jiang et al., 2010; Lee & Baik, 2017;Michibata & Takemura, 2015; Seifert & Onishi, 2016; Wood, 2005b) and the entrainment‐mixing parameter-izations (Lu et al., 2013). Furthermore, although our research shows the importance of the accretion process,it cannot be ignored that the ice‐phase processes are also important in this region. Therefore, it is necessaryto study the parameterizations of different ice‐phase processes in the TP area in the future.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 18 of 23

Appendix A: Symbol ListNc: number concentration of cloud dropletsqc: mixing ratio of cloud dropletNr: number concentration of raindropsqr: mixing ratio of raindropsNi: number concentration of ice crystalsqi: mixing ratio of ice crystalsNs: number concentration of snow particlesqs: mixing ratio of snow particlesNg: number concentration of graupel particlesqg: mixing ratio of graupel particlesAu : conversion rate of accretion processAc : conversion rate of autoconversion processρa : air densityε : dispersionρW : water densityλ : slope parameterNc0: number concentration of cloud water droplets before evaporation processqc0: mixing ratio of cloud water droplets before evaporation processpt: the threshold value of precipitation in the Heidke skill scoreps: value of precipitation from simulations in the Heidke skill scorepo: value of precipitation from observation in the Heidke skill scoreτ: cloud optical depthre : averaged effective radius of cloud water dropletsLCWP: liquid cloud water pathEVAP‐r: evaporation of raindropsACCR‐r: accretion of cloud liquid water by rainAUTO‐r: autoconversion from cloud droplets to raindropsMELT: melting from snow or graupel particles to raindropsAUTO‐s: autoconversion of cloud ice to snowACCR‐s: accretion of cloud ice by snowRIM‐s: accretion of cloud droplets by snow particleRIM‐g: accretion of cloud droplets by graupel particle

Appendix B: Four Parameters in Equation 6a and 6bIn Equation 6, A1, A2, B1, and B2 are the functions related to two dispersion parameters of the gamma sizedistribution given by

A1 ¼ Γ υc þ 6=αcð ÞΓ υcð Þ ; (B1a)

B1 ¼ Γ υc þ 3=αcð ÞΓ υcð Þ

Γ υr þ 3=αrð ÞΓ υrð Þ ; (B1b)

A2 ¼ Γ υc þ 9=αcð ÞΓ υcð Þ ; (B1c)

B2 ¼ Γ υc þ 3=αcð ÞΓ υcð Þ

Γ υr þ 6=αrð ÞΓ υrð Þ ; (B1d)

where υ and α are the two dispersion parameters in normalized form of cloud‐raindrop size distributions

n Dð Þ ¼ Nα

Γ υð ÞλαυDαυ − 1exp − λDð Þα½ �; λ is the slope parameter; D and N represent diameter and total num-

ber concentration, respectively. Subscripts c and r in Equations B1a–B1d represent cloud droplets andraindrops, respectively.

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 19 of 23

Data Availability Statement

The National Center for Environmental Prediction Final operational global analysis data were obtainedonline (https://rda.ucar.edu/datasets/ds083.2/index.html). Data from Clouds and the Earth's RadiantEnergy System (CERES) project can be obtained online (https://ceres-tool.larc.nasa.gov/ord-tool/jsp/SYN1degEd41Selection.jsp).

ReferencesAshouri, H., Hsu, K.‐L., Sorooshian, S., Braithwaite, D. K., Knapp, K. R., Cecil, L. D., et al. (2015). PERSIANN‐CDR: Daily precipitation

climate data record frommultisatellite observations for hydrological and climate studies. Bulletin of the American Meteorological Society,96(1), 69–83.

Barnston, A. G. (1992). Correspondence among the correlation, RMSE, and Heidke forecast verification measures: Refinement of theHeidke score. Weather and Forecasting, 7(4), 699–709.

Beheng, K. (1994). A parameterization of warm cloudmicrophysical conversion processes.Atmospheric Research, 33(1–4), 193–206. https://doi.org/10.1016/0169-8095(94)90020-5

Berry, E. X. (1968). Modification of the warm rain process. In Proc. First Conf. on Weather Modification, AMS, 1968.Chen, B., Hu, Z., Liu, L., & Zhang, G. (2017). Raindrop size distribution measurements at 4,500 m on the Tibetan Plateau during TIPEX‐III.

Journal of Geophysical Research: Atmospheres, 122, 11,092–11,106. https://doi.org/10.1002/2017JD027233Chen, J., Wu, X., Yin, Y., Huang, Q., & Xiao, H. (2017). Characteristics of cloud systems over the Tibetan Plateau and East China during

boreal summer. Journal of Climate, 30(9), 3117–3137. https://doi.org/10.1175/JCLI-D-16-0169.1Chosson, F., Brenguier, J.‐L., & Schüller, L. (2007). Entrainment‐mixing and radiative transfer simulation in boundary layer clouds.

Journal of the Atmospheric Sciences, 64(7), 2670–2682. https://doi.org/10.1175/JAS3975.1Cohard, J. M., & Pinty, J. P. (2000). A comprehensive two‐moment warm microphysical bulk scheme. I: Description and tests. Quarterly

Journal of the Royal Meteorological Society, 126(566), 1815–1842.Cooper, W. A., Lasher‐Trapp, S. G., & Blyth, A. M. (2013). The influence of entrainment and mixing on the initial formation of rain in a

warm cumulus cloud. Journal of the Atmospheric Sciences, 70(6), 1727–1743. https://doi.org/10.1175/jas-d-12-0128.1Ding, Y., Zhang, Y., Ma, Q., & Hu, G. (2001). Analysis of the large‐scale circulation features and synoptic systems in East

Asia during the intensive observation period of GAME/HUBEX. Journal of the Meteorological Society of Japan. Series II, 79(1B),277–300.

Fan, J., Leung, L. R., Rosenfeld, D., Chen, Q., Li, Z., Zhang, J., & Yan, H. (2013). Microphysical effects determine macrophysical responsefor aerosol impacts on deep convective clouds. Proceedings of the National Academy of Sciences of the United States of America, 110(48),E4581–E4590. https://doi.org/10.1073/pnas.1316830110

Fan, J., Wang, Y., Rosenfeld, D., & Liu, X. (2016). Review of aerosol–cloud interactions: Mechanisms, significance, and challenges. Journalof the Atmospheric Sciences, 73(11), 4221–4252. https://doi.org/10.1175/jas-d-16-0037.1

Flohn, H. (1957). Large‐scale aspects of the “summermonsoon” in South and East Asia. Journal of the Meteorological Society of Japan. SeriesII, 35, 180–186.

Fu, Y., Li, H., & Zi, Y. (2007). Case study of precipitation cloud structure viewed by TRMM satellite in a valley of the Tibetan Plateau.Plateau Meteorology, 26(1), 98–106.

Fu, Y., Liu, G., Wu, G., Yu, R., Xu, Y., Wang, Y., et al. (2006). Tower mast of precipitation over the central Tibetan Plateau summer.Geophysical Research Letters, 33, L05802. https://doi.org/10.1029/2005GL024713

Fujinami, H., & Yasunari, T. (2001). The seasonal and intraseasonal variability of diurnal cloud activity over the Tibetan Plateau. Journal ofthe Meteorological Society of Japan. Series II, 79(6), 1207–1227.

Gao, S., Lu, C., Liu, Y., Mei, F., Wang, J., Zhu, L., & Yan, S. (2020). Contrasting scale dependence of entrainment‐mixing mechanisms instratocumulus clouds. Geophysical Research Letters, 47, e2020GL086970. https://doi.org/10.1029/2020GL086970

Gao, W., Liu, L., Li, J., & Lu, C. (2018). The microphysical properties of convective precipitation over the Tibetan Plateau by a subkilometerresolution cloud‐resolving simulation. Journal of Geophysical Research: Atmospheres, 123, 3212–3227. https://doi.org/10.1002/2017JD027812

Gao, W., Sui, C. H., Fan, J., Hu, Z., & Zhong, L. (2016). A study of cloud microphysics and precipitation over the Tibetan Plateau by radarobservations and cloud‐resolving model simulations. Journal of Geophysical Research: Atmospheres, 121, 13,735–13,752. https://doi.org/10.1002/2015JD024196

Gerken, T., Babel, W., Sun, F., Herzog, M., Ma, Y., Foken, T., & Graf, H.‐F. (2013). Uncertainty in atmospheric profiles and its impact onmodeled convection development at Nam Co Lake, Tibetan Plateau. Journal of Geophysical Research: Atmospheres, 118, 12,317–12,331.https://doi.org/10.1002/2013JD020647

Gettelman, A., Morrison, H., Santos, S., Bogenschutz, P., & Caldwell, P. M. (2015). Advanced two‐moment bulk microphysics for globalmodels. Part II: Global model solutions and aerosol–cloud interactions. Journal of Climate, 28(3), 1288–1307. https://doi.org/10.1175/jcli-d-14-00103.1

Gettelman, A., Morrison, H., Terai, C. R., & Wood, R. (2013). Microphysical process rates and global aerosol–cloud interactions.Atmospheric Chemistry and Physics, 13(19), 9855–9867. https://doi.org/10.5194/acp-13-9855-2013

Grabowski, W. W. (2006). Indirect impact of atmospheric aerosols in idealized simulations of convective–radiative quasi equilibrium.Journal of Climate, 19(18), 4664–4682. https://doi.org/10.1175/JCLI3857.1

Grabowski, W. W., & Morrison, H. (2011). Indirect impact of atmospheric aerosols in idealized simulations of convective–Radiativequasi equilibrium. Part II: Double‐moment microphysics. Journal of Climate, 24(7), 1897–1912. https://doi.org/10.1175/2010JCLI3647.1

Hahn, D. G., & Manabe, S. (1975). The role of mountains in the south Asian monsoon circulation. Journal of the Atmospheric Sciences,32(8), 1515–1541. https://doi.org/10.1175/1520-0469(1975)032<1515:TROMIT>2.0.CO;2

Hill, A. A., Feingold, G., & Jiang, H. (2009). The influence of entrainment and mixing assumption on aerosol–cloud interactions in marinestratocumulus. Journal of the Atmospheric Sciences, 66(5), 1450–1464. https://doi.org/10.1175/2008JAS2909.1

Hill, A. A., Shipway, B. J., & Boutle, I. A. (2015). How sensitive are aerosol‐precipitation interactions to the warm rain representation?Journal of Advances in Modeling Earth Systems, 7, 987–1004. https://doi.org/10.1002/2014MS000422

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 20 of 23

AcknowledgmentsThe authors thank Prof. Amy Solomonfor providing the microphysics scheme,Profs. Yinzhao Ma and Yang Hong forproviding the precipitation data, andthe data are available (https://data.mendeley.com/datasets/ynxtgtccj3/1.DOI: 10.17632/ynxtgtccj3.1). Thisresearch is supported by the NationalKey Research and DevelopmentProgram of China (2018YFC1505702),the Second Tibetan Plateau ScientificExpedition and Research (STEP) pro-gram (2019QZKK0105), the NaturalScience Foundation of Jiangsu Province(BK20160041), the National NaturalScience Foundation of China (41822504and 91537108), the Qinglan Project ofJiangsu Province of China (R2018Q05),the Six Talent Peak Project in Jiangsu(2015‐JY‐011), and the National Centerof Meteorology, Abu Dhabi, UAE,under the UAE Research Program forRain Enhancement Science. Liu is sup-ported by the U.S. Department ofEnergy Office of Science Biological andEnvironmental Research as part of theAtmospheric Systems Research (ASR)Program. Brookhaven NationalLaboratory is operated by Battelle forthe U.S. Department of Energy underContract DE‐SC0012704. The numeri-cal calculations in this paper have beendone on the supercomputing system inthe Supercomputing Center of NanjingUniversity of Information Science andTechnology.

Hoffmann, F., & Feingold, G. (2019). Entrainment and mixing in stratocumulus: Effects of a new explicit subgrid‐scale scheme forlarge‐eddy simulations with particle‐based microphysics. Journal of the Atmospheric Sciences, 76(7), 1955–1973. https://doi.org/10.1175/jas-d-18-0318.1

Hsu, H. H., & Liu, X. (2003). Relationship between the Tibetan Plateau heating and East Asian summer monsoon rainfall. GeophysicalResearch Letters, 30(20), 2003GL017909. https://doi.org/10.1029/2003GL017909

Huffman, G. J., Bolvin, D. T., Nelkin, E. J., Wolff, D. B., Adler, R. F., Gu, G., et al. (2007). The TRMM multisatellite precipitation analysis(TMPA): Quasi‐global, multiyear, combined‐sensor precipitation estimates at fine scales. Journal of Hydrometeorology, 8(1), 38–55.

Jiang, H., Feingold, G., & Sorooshian, A. (2010). Effect of aerosol on the susceptibility and efficiency of precipitation in warm trade cumulusclouds. Journal of the Atmospheric Sciences, 67(11), 3525–3540. https://doi.org/10.1175/2010JAS3484.1

Jiang, J. H., Wang, B., Goya, K., Hocke, K., Eckermann, S. D., Ma, J., et al. (2004). Geographical distribution and interseasonal variability oftropical deep convection: UARS MLS observations and analyses. Journal of Geophysical Research, 109, D03111. https://doi.org/10.1029/2003JD003756

Jing, X., & Suzuki, K. (2018). The impact of process‐based warm rain constraints on the aerosol indirect effect. Geophysical Research Letters,45, 10,729–10,737. https://doi.org/10.1029/2018GL079956

Jing, X., Suzuki, K., &Michibata, T. (2019). The key role of warm rain parameterization in determining the aerosol indirect effect in a globalclimate model. Journal of Climate, 32(14), 4409–4430. https://doi.org/10.1175/JCLI-D-18-0789.1

Joyce, R. J., Janowiak, J. E., Arkin, P. A., & Xie, P. (2004). CMORPH: A method that produces global precipitation estimates from passivemicrowave and infrared data at high spatial and temporal resolution. Journal of Hydrometeorology, 5(3), 487–503. https://doi.org/10.1175/1525-7541(2004)005<0487:CAMTPG>2.0.CO;2

Khairoutdinov, M., & Kogan, Y. (2000). A new cloud physics parameterization in a large‐eddy simulation model of marine stratocumulus.Monthly Weather Review, 128(1), 229–243. https://doi.org/10.1175/1520-0493(2000)128<0229:ANCPPI>2.0.CO;2

Kogan, Y. (2013). A cumulus cloud microphysics parameterization for cloud‐resolving models. Journal of the Atmospheric Sciences, 70(5),1423–1436.

Kollias, P., & Albrecht, B. (2000). The turbulence structure in a continental stratocumulus cloud from millimeter‐wavelength radarobservations. Journal of the Atmospheric Sciences, 57(15), 2417–2434. https://doi.org/10.1175/1520-0469(2000)057<2417:ttsiac>2.0.co;2

Kurosaki, Y., & Kimura, F. (2002). Relationship between topography and daytime cloud activity around Tibetan Plateau. Journal of theMeteorological Society of Japan. Series II, 80(6), 1339–1355.

Lasher‐Trapp, S. G., Cooper, W. A., & Blyth, A. M. (2005). Broadening of droplet size distributions from entrainment and mixing in acumulus cloud. Quarterly Journal of the Royal Meteorological Society, 131(605), 195–220. https://doi.org/10.1256/qj.03.199

Lee, H., & Baik, J.‐J. (2017). A physically based autoconversion parameterization. Journal of the Atmospheric Sciences, 74(5), 1599–1616.https://doi.org/10.1175/jas-d-16-0207.1

Li, G., Wang, Y., & Zhang, R. (2008). Implementation of a two‐moment bulk microphysics scheme to the WRF model to investigateaerosol‐cloud interaction. Journal of Geophysical Research, 113, D15211. https://doi.org/10.1029/2007JD009361

Li, J., Chen, J., Lu, C., & Wu, X. (2020). Impacts of TIPEX‐III rawinsondes on the dynamics and thermodynamics over the Eastern TibetanPlateau in the boreal summer. Journal of Geophysical Research: Atmospheres, 125, e2020JD032635. https://doi.org/10.1029/2020JD032635

Li, J., You, L.‐L., Hu, Z.‐J., Tu, D.‐B., & Li, L.‐L. (2006). Analysis on raindrop‐size distribution characteristics of Maqu region in upper reachof Yellow River. Plateau Meteorology, 25(5), 942–949.

Li, R., Min, Q., Wu, X., & Fu, Y. (2013). Retrieving latent heating vertical structure from cloud and precipitation profiles—Part II: Deepconvective and stratiform rain processes. Journal of Quantitative Spectroscopy & Radiative Transfer, 122, 47–63. https://doi.org/10.1016/j.jqsrt.2012.11.029

Li, Y., Wang, Y., Song, Y., Hu, L., Gao, S., & Rong, F. (2008). Characteristics of summer convective systems initiated over the TibetanPlateau. Part I: Origin, track, development, and precipitation. Journal of Applied Meteorology and Climatology, 47(10), 2679–2695.https://doi.org/10.1175/2008jamc1695.1

Li, Y., & Zhang, M. (2016). Cumulus over the Tibetan plateau in the summer based on CloudSat‐CALIPSO data. Journal of Climate, 29(3),1219–1230. https://doi.org/10.1175/JCLI-d-15-0492.1

Li, Y., & Zhang, M. (2017). The role of shallow convection over the Tibetan Plateau. Journal of Climate, 30(15), 5791–5803.Lim, K.‐S. S., & Hong, S.‐Y. (2010). Development of an effective double‐moment cloud microphysics scheme with prognostic Cloud

Condensation Nuclei (CCN) for weather and climate models. Monthly Weather Review, 138(5), 1587–1612. https://doi.org/10.1175/2009MWR2968.1

Liu, X., Xie, S., Boyle, J., Klein, S. A., Shi, X., Wang, Z., et al. (2011). Testing cloud microphysics parameterizations in NCAR CAM5 withISDAC and M‐PACE observations. Journal of Geophysical Research, 116, D00T11. https://doi.org/10.1029/2011JD015889

Liu, Y. (2005). Size truncation effect, threshold behavior, and a new type of autoconversion parameterization. Geophysical Research Letters,32, L11811. https://doi.org/10.1029/2005GL022636

Liu, Y., & Daum, P. H. (2004). Parameterization of the autoconversion process. Part I: Analytical formulation of the Kessler‐type para-meterizations. Journal of the Atmospheric Sciences, 61(13), 1539–1548.

Lu, C., Liu, Y., Niu, S., Krueger, S., & Wagner, T. (2013). Exploring parameterization for turbulent entrainment‐mixing processes in clouds.Journal of Geophysical Research: Atmospheres, 118, 185–194. https://doi.org/10.1029/2012JD018464

Luo, H., & Yanai, M. (1984). The large‐scale circulation and heat sources over the Tibetan Plateau and surrounding areas during the earlysummer of 1979. Part II: Heat and moisture budgets. Monthly Weather Review, 112(5), 966–989.

Ma, Y., Hong, Y., Chen, Y., Yang, Y., Tang, G., Yao, Y., et al. (2018). Performance of optimally merged multisatellite precipitation productsusing the dynamic Bayesian model averaging scheme over the Tibetan plateau. Journal of Geophysical Research: Atmospheres, 123,814–834. https://doi.org/10.1002/2017JD026648

Maussion, F., Scherer, D., Finkelnburg, R., Richters, J., Yang, W., & Yao, T. (2011). WRF simulation of a precipitation event over theTibetan Plateau, China—An assessment using remote sensing and ground observations. Hydrology and Earth System Sciences, 15(6),1795–1817. https://doi.org/10.5194/hess-15-1795-2011

Michibata, T., & Takemura, T. (2015). Evaluation of autoconversion schemes in a single model framework with satellite observations.Journal of Geophysical Research: Atmospheres, 120, 9570–9590. https://doi.org/10.1002/2015JD023818

Molnar, P., Boos, W. R., & Battisti, D. S. (2010). Orographic controls on climate and paleoclimate of Asia: Thermal and mechanical roles forthe Tibetan Plateau. Annual Review of Earth and Planetary Sciences, 38, 77–102.

Morrison, H., Curry, J. A., & Khvorostyanov, V. I. (2005). A new double‐moment microphysics parameterization for application in cloudand climate models. Part I: Description. Journal of the Atmospheric Sciences, 62(6), 1665–1677. https://doi.org/10.1175/JAS3446.1

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 21 of 23

Morrison, H., & Grabowski, W. W. (2007). Comparison of bulk and bin warm‐rain microphysics models using a kinematic framework.Journal of the Atmospheric Sciences, 64(8), 2839–2861. https://doi.org/10.1175/JAS3980

Morrison, H., & Grabowski, W. W. (2008). Modeling supersaturation and subgrid‐scale mixing with two‐moment bulk warmmicrophysics.Journal of the Atmospheric Sciences, 65(3), 792–812. https://doi.org/10.1175/2007JAS2374.1

Nitta, T. (1983). Observational study of heat sources over the eastern Tibetan Plateau during the summer monsoon. Journal of theMeteorological Society of Japan. Series II, 61(4), 590–605.

Porcù, F., D'Adderio, L. P., Prodi, F., & Caracciolo, C. (2014). Rain drop size distribution over the Tibetan Plateau. Atmospheric Research,150, 21–30.

Posselt, R., & Lohmann, U. (2008). Introduction of prognostic rain in ECHAM5: Design and single columnmodel simulations. AtmosphericChemistry and Physics, 8(11), 2949–2963. https://doi.org/10.5194/acp-8-2949-2008

Qie, X., Kong, X., Zhang, G., Zhang, T., Yuan, T., Zhou, Y., et al. (2005). The possible charge structure of thunderstorm and lightningdischarges in northeastern verge of Qinghai–Tibetan Plateau. Atmospheric Research, 76(1–4), 231–246. https://doi.org/10.1016/j.atmosres.2004.11.034

Qie, X., Wu, X., Yuan, T., Bian, J., & Lu, D. (2014). Comprehensive pattern of deep convective systems over the Tibetan Plateau–SouthAsian monsoon region based on TRMM data. Journal of Climate, 27(17), 6612–6626. https://doi.org/10.1175/JCLI-D-14-00076.1

Rosenfeld, D., Zhu, Y., Wang, M., Zheng, Y., Goren, T., & Yu, S. (2019). Aerosol‐driven droplet concentrations dominate coverage andwater of oceanic low‐level clouds. Science, 363(6427), eaav0566.

Sato, T., Yoshikane, T., Satoh, M., Miura, H., & Fujinami, H. (2008). Resolution dependency of the diurnal cycle of convective clouds overthe Tibetan Plateau in a mesoscale model. Journal of the Meteorological Society of Japan. Series II, 86, 17–31.

Seifert, A., & Onishi, R. (2016). Turbulence effects on warm‐rain formation in precipitating shallow convection revisited. AtmosphericChemistry and Physics, 16(18), 12,127–12,141. https://doi.org/10.5194/acp-16-12127-2016

Shen, R., Reiter, E. R., & Bresch, J. F. (1986). Some aspects of the effects of sensible heating on the development of summer weather systemsover the Tibetan Plateau. Journal of the Atmospheric Sciences, 43(20), 2241–2260. https://doi.org/10.1175/1520-0469(1986)043<2241:SAOTEO>2.0.CO;2

Slawinska, J., Grabowski, W. W., Pawlowska, H., & Morrison, H. (2012). Droplet activation and mixing in large‐eddy simulation of ashallow cumulus field. Journal of the Atmospheric Sciences, 69(2), 444–462. https://doi.org/10.1175/JAS-D-11-054.1

Slawinska, J., Grabowski, W. W., Pawlowska, H., & Wyszogrodzki, A. A. (2008). Optical properties of shallow convective cloudsdiagnosed from a bulk‐microphysics large‐eddy simulation. Journal of Climate, 21(7), 1639–1647. https://doi.org/10.1175/2007JCLI1820.1

Tang, J., Guo, X., & Chang, Y. (2019). A numerical investigation on microphysical properties of clouds and precipitation over the TibetanPlateau in summer 2014. Journal of Meteorological Research, 33(3), 463–477. https://doi.org/10.1007/s13351-019-8614-z

Ueda, H., Kamahori, H., & Yamazaki, N. (2003). Seasonal contrasting features of heat and moisture budgets between the eastern andwestern Tibetan Plateau during the GAME IOP. Journal of Climate, 16(14), 2309–2324. https://doi.org/10.1175/2757.1

Ueda, H., & Yasunari, T. (1998). Role of warming over the Tibetan Plateau in early onset of the summer monsoon over the Bay of Bengaland the South China Sea. Journal of the Meteorological Society of Japan. Series II, 76(1), 1–12.

Wang, B., Bao, Q., Hoskins, B., Wu, G., & Liu, Y. (2008). Tibetan Plateau warming and precipitation changes in East Asia. GeophysicalResearch Letters, 35, L14702. https://doi.org/10.1029/2008GL034330

Wang, M., Ghan, S., Liu, X., L'Ecuyer, T. S., Zhang, K., Morrison, H., et al. (2012). Constraining cloud lifetime effects of aerosols usingA‐Train satellite observations. Geophysical Research Letters, 39, L15709. https://doi.org/10.1029/2012GL052204

Wang, W., Kuo, Y.‐H., & Warner, T. T. (1993). A diabatically driven mesoscale vortex in the lee of the Tibetan Plateau. Monthly WeatherReview, 121(9), 2542–2561. https://doi.org/10.1175/1520-0493(1993)121<2542:ADDMVI>2.0.CO;2

Wang, Y., Fan, J., Zhang, R., Leung, L. R., & Franklin, C. (2013). Improving bulk microphysics parameterizations in simulations of aerosoleffects. Journal of Geophysical Research: Atmospheres, 118, 5361–5379. https://doi.org/10.1002/jgrd.50432

Wang, Y., Ma, P., Jiang, J. H., Su, H., & Rasch, P. J. (2016). Toward reconciling the influence of atmospheric aerosols and greenhouse gaseson light precipitation changes in Eastern China. Journal of Geophysical Research: Atmospheres, 121, 5878–5887. https://doi.org/10.1002/2016JD024845

Wang, Y., Niu, S., Lu, C., Liu, Y., Chen, J., & Yang, W. (2019). An observational study on cloud spectral width in North China. Atmosphere,10(3), 109. https://doi.org/10.3390/atmos10030109

Wang, Y., Niu, S., Lv, J., Lu, C., Xu, X., Wang, Y., et al. (2019). A newmethod for distinguishing Unactivated particles in cloud condensationnuclei measurements: Implications for aerosol indirect effect evaluation. Geophysical Research Letters, 46, 14,185–14,194. https://doi.org/10.1029/2019GL085379

White, B., Gryspeerdt, E., Stier, P., Morrison, H., Thompson, G., & Kipling, Z. (2017). Uncertainty from the choice of microphysics schemein convection‐permitting models significantly exceeds aerosol effects. Atmospheric Chemistry and Physics, 17(19), 12,145–12,175. https://doi.org/10.5194/acp-17-12145-2017

Wood, R. (2005a). Drizzle in stratiform boundary layer clouds. Part I: Vertical and horizontal structure. Journal of the Atmospheric Sciences,62(9), 3011–3033.

Wood, R. (2005b). Drizzle in stratiform boundary layer clouds. Part II: Microphysical aspects. Journal of the Atmospheric Sciences, 62(9),3034–3050.

Wood, R., Kubar, T. L., & Hartmann, D. L. (2009). Understanding the importance of microphysics and macrophysics for warm rain inmarine low clouds. Part II: Heuristic models of rain formation. Journal of the Atmospheric Sciences, 66(10), 2973–2990. https://doi.org/10.1175/2009JAS3072.1

Wu, G. X., & Chen, S. J. (1985). The effect of mechanical forcing on the formation of a mesoscale vortex. Quarterly Journal of the RoyalMeteorological Society, 111(470), 1049–1070. https://doi.org/10.1002/qj.49711147009

Wu, Y., & Liu, L. (2017). Statistical characteristics of raindrop size distribution in the Tibetan Plateau and southern China. Advances inAtmospheric Sciences, 34(6), 727–736. https://doi.org/10.1007/s00376-016-5235-7

Xie, X., & Liu, X. (2011). Effects of spectral dispersion on clouds and precipitation in mesoscale convective systems. Journal of GeophysicalResearch, 116, D06202. https://doi.org/10.1029/2010JD014598

Xie, X., Liu, X., Peng, Y., Wang, Y., Yue, Z., & Li, X. (2013). Numerical simulation of clouds and precipitation depending on differentrelationships between aerosol and cloud droplet spectral dispersion. Tellus Series B‐Chemical and Physical Meteorology, 65(1), 19054.https://doi.org/10.3402/tellusb.v65i0.19054

Xu, J., Zhang, B., Wang, M., & Wang, H. (2012). Diurnal variation of summer precipitation over the Tibetan Plateau: A cloud‐resolvingsimulation. Annales Geophysicae, 30(11), 1575–1586. https://doi.org/10.5194/angeo-30-1575-2012

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 22 of 23

Yanai, M., & Li, C. (1994). Mechanism of heating and the boundary layer over the Tibetan Plateau. Monthly Weather Review, 122(2),305–323.

Yanai, M., Li, C., & Song, Z. (1992). Seasonal heating of the Tibetan Plateau and its effects on the evolution of the Asian summer monsoon.Journal of the Meteorological Society of Japan. Series II, 70(1B), 319–351.

Yang, K., Wu, H., Qin, J., Lin, C., Tang, W., & Chen, Y. (2014). Recent climate changes over the Tibetan Plateau and their impacts on energyand water cycle: A review. Global and Planetary Change, 112, 79–91.

Ye, D. (1981). Some characteristics of the summer circulation over the Qinghai‐Xizang (Tibet) Plateau and its neighborhood. Bulletin of theAmerican Meteorological Society, 62(1), 14–19. https://doi.org/10.1175/1520-0477(1981)062<0014:SCOTSC>2.0.CO;2

Yeh, T. C. (1950). The circulation of the high troposphere over China in the winter of 1945–46. Tellus, 2(3), 173–183.Yin, Z.‐Y., Zhang, X., Liu, X., Colella, M., & Chen, X. (2008). An assessment of the biases of satellite rainfall estimates over the Tibetan

Plateau and correction methods based on topographic analysis. Journal of Hydrometeorology, 9(3), 301–326.Zhao, C., Liu, L., Wang, Q., Qiu, Y., Wang, Y., & Wu, X. (2017). MMCR‐based characteristic properties of non‐precipitating cloud liquid

droplets at Naqu site over Tibetan Plateau in July 2014.Atmospheric Research, 190, 68–76. https://doi.org/10.1016/j.atmosres.2017.02.002Zhao, C., Tie, X., Brasseur, G., Noone, K. J., Nakajima, T., Zhang, Q., et al. (2006). Aircraft measurements of cloud droplet spectral dis-

persion and implications for indirect aerosol radiative forcing. Geophysical Research Letters, 33, L16809. https://doi.org/10.1029/2006GL026653

10.1029/2020JD033371Journal of Geophysical Research: Atmospheres

XU ET AL. 23 of 23