Comparison of Observed and Simulated Spatial Patterns of Ice Microphysical Processes in 1

Tropical Oceanic Mesoscale Convective Systems 2

3

Hannah C. Barnes* and Robert A. Houze, Jr* #. 4

5

*Department of Atmospheric Sciences, University of Washington, Seattle, Washington, USA 6

#Pacific Northwest National Laboratory, Richland, Washington, USA 7

8

Corresponding author: H. C. Barnes, Department of Atmospheric Sciences, University of Washington, Box 9

351640, Seattle, WA 98195 (hannah@atmos.washington.edu) 10

11

Submitted to Journal of Geophysical Research - Atmospheres 12

13

Submitted March 2016 14

Revised May 2016 15

Revised June 2016 16

17

18

Key Points 19

Simulated ice processes in the midlevel inflow are consistent with radar observations and 20

theory 21

Of the processes considered, simulated riming and aggregation differ the most from 22

observations 23

Disparate simulated ice microphysical patterns factor into reflectivity differences 24

Abstract 25

To equitably compare the spatial pattern of ice microphysical processes produced by 26

three microphysical parameterizations with each other, observations, and theory, simulations of 27

tropical oceanic mesoscale convective systems (MCSs) in the Weather Research and Forecasting 28

(WRF) model were forced to develop the same mesoscale circulations as observations by 29

assimilating radial velocity data from a Doppler radar.. The same general layering of 30

microphysical processes was found in observations and simulations with deposition anywhere 31

above the 0°C level, aggregation at and above the 0°C level, melting at and below the 0°C level, 32

and riming near the 0°C level. Thus, this study is consistent with the layered ice microphysical 33

pattern portrayed in previous conceptual models and indicated by dual-polarization radar data. 34

Spatial variability of riming in the simulations suggest that riming in the midlevel inflow is 35

related to convective-scale vertical velocity perturbations. Finally, this study sheds light on 36

limitations of current generally available bulk microphysical parameterizations. In each 37

parameterization, the layers in which aggregation and riming took place were generally too thick 38

and the frequency of riming was generally too high compared to the observations and theory. 39

Additionally, none of the parameterizations produced similar details in every microphysical 40

spatial pattern. Discrepancies in the patterns of microphysical processes between 41

parameterizations likely factor into creating substantial differences in model reflectivity patterns. 42

It is concluded that improved parameterizations of ice-phase microphysics will be essential to 43

obtain reliable, consistent model simulations of tropical oceanic MCSs. 44

Index terms and Keywords 45

AGU: 3329 Mesoscale Meteorology, 3371 Tropical Convection, 3374 Tropical Meteorology, 46

3315 Data Assimilation, 3314 Convective Processes 47

Author: Mesoscale Convective Systems, Cloud Microphysics, Stratiform Precipitation, Dual-48

Polarimetric Radar 49

Text 50

1. Introduction 51

The spatial distribution of microphysical processes within a convective cloud 52

system indicates how the microscale interactions among hydrometeors are affecting the 53

entire dynamical and thermodynamical system. Studies such as Chen and Cotton [1988] 54

have demonstrated that accurate mesoscale simulations require accurate representations 55

of microphysical processes, latent heating, and radiative transfer and their interactions. 56

Latent heat absorbed or emitted in cloud microphysical processes modifies buoyancy, 57

which, in turn, contributes to the development and maintenance of vertical air motion 58

[e.g. Szeto et al., 1988; Tao et al., 1995; Adams-Selin et al., 2013]. Ice-phase 59

microphysical processes are essential to the development of both the convective and 60

stratiform components of mesoscale convective systems [e.g. Chen and Cotton, 1988; 61

Tao et al., 1991; Zipser, 2003]. Once stratiform precipitation is formed, ice microphysical 62

processes modify radiative heating, which can increase instability, cause turbulence, and 63

extend the lifetime of stratiform precipitation and its associated anvil cloud [e.g. Webster 64

and Stephens, 1980; Chen and Cotton, 1988; Churchill and Houze, 1991; Tao et al., 65

1996]. 66

The impact of microphysical processes is not limited to the convective cloud 67

systems in which they occur. The modification of the diabatic heating structure by 68

microphysical processes influences dynamics at the global-scale. This evolving diabatic 69

heating profile alters the global circulation through teleconnections [e.g. Hartmann et al., 70

1984; Schumacher et al., 2004]. For example, Barnes and Houze [2015] demonstrated 71

that the latent heating profile systematically changes during the Madden-Julian 72

Oscillation (MJO). This evolving latent heating profile is likely one reason why studies 73

such as Vitart and Molteni [2010] have demonstrated that extratropical weather patterns 74

are correlated to the MJO. 75

Carefully collected and processed dual-polarization radar data provides some of 76

the most comprehensive data on microphysical processes. One approach is to use a 77

particle identification algorithm (PID). Traditionally, PIDs have been used to identify the 78

dominant hydrometeor type within convection [e.g. Hendry and Antar, 1984; 79

Vivekanadan et al., 1999; Straka et al., 2000; Thompson et al., 2014; Grazioli et al., 80

2015; Kouketsu et al., 2015]. Because the PID classification is based on physical 81

characteristics of particles, not their exact size or density, they do not indicate 82

hydrometeor mixing ratios. However, the particle type indicated by the PID is a good 83

indicator of the microphysical processes that have produced the particles (i.e. deposition, 84

aggregation/accretion, or riming). This study capitalizes on this capability of the PID 85

methodology. 86

PID classifications have limitations that must be considered when the data is 87

analyzed. Data provided by PIDs is restricted since the radar statistically samples large 88

volumes that contain many particles and only provides data on the most likely dominant 89

microphysical process. Additionally, the PID data is somewhat uncertain due to 90

overlapping classification boundaries. Nonetheless, the PID provides the most 91

comprehensive high-resolution three-dimensional mapping of the spatial distribution of 92

microphysical processes in convection. Ground instruments and aircraft probes provide in 93

situ observations [e.g. Baumgardner et al., 2011], but their spatial coverage is limited. 94

Additionally, previous aircraft probes biased results by shattering ice crystals [Korolev et 95

al., 2011]. Satellites can retrieve details about the microphysical structure over large 96

areas and long temporal periods [e.g. Comstock et al., 2007], but the resolution is coarse. 97

PIDs, therefore, remain the best option for comprehensive high resolution mapping of 98

microphysical processes in convection. 99

The microphysical processes at a given location are closely linked to the evolution 100

of the airflow. Mesoscale convective systems (MCSs), which are broadly defined as 101

cloud systems whose contiguous precipitation span at least 100 km in any direction 102

[Houze, 2004; 2014], are ideal for investigating the spatial microphysical patterns in 103

convection because they have a well-known, repeatable kinematic structure [Kingsmill 104

and Houze, 1999a]. MCSs are generally composed of two parts: a convective and 105

stratiform region. The convective region is composed of intensely precipitating cores of 106

relatively small horizontal scale. In the convective region, the air steeply rises as a layer 107

originating in the lower troposphere. The stratiform region is a more expansive region of 108

weaker precipitation characterized by a current of air that starts at midlevels within the 109

anvil and extends towards the center of the storm as it gently subsides. This current of air 110

is referred to as the midlevel inflow. Vertical velocities in the stratiform zone are weak, 111

often an order of magnitude less than the horizontal wind speeds, and are characterized 112

by broad regions of weak ascent above the midlevel inflow and weak descent below the 113

midlevel inflow. Differences in the large-scale environment cause the horizontal 114

morphology of MCSs to vary [Tollerud and Esbensen, 1985]. Often in environments with 115

strong lower tropospheric vertical wind shear MCSs take the form of a leading 116

convective line with a trailing stratiform region. These MCSs are called squall lines. 117

When MCSs develop in environments with weak low-level vertical wind shear they tend 118

to be more amorphous with the convective regions lying alongside the stratiform region 119

in various patterns. Despite these morphological differences all MCSs have the same 120

midlevel inflow structure [Kingsmill and Houze, 1999a]. Because of this repeatable 121

kinematic structure, it is possible to composite the pattern of microphysical processes 122

relative to the mesoscale midlevel inflow dominating the stratiform region and the mean 123

steeply rising current in the convective region. 124

Barnes and Houze [2014] took this compositing approach and applied a PID 125

developed at the National Center for Atmospheric Research (NCAR) to data obtained 126

using the NCAR S-PolKa radar during the Dynamics of the Madden-Julian Oscillation 127

field campaign / ARM MJO Investigation Experiment (DYNAMO/AMIE) [Yoneyama et 128

al., 2013]. Compositing PID data from 36 MCSs with a midlevel inflow, they found that 129

frozen hydrometeors had a systematic layered pattern with small ice crystals near echo 130

top, dry aggregates at midlevels above the 0°C level, and melting particles near the 0 °C 131

level. Additionally, graupel/rimed aggregates were found occasionally in shallow pockets 132

just above the 0°C level. Their results were insensitive to the morphology of the MCS. 133

Previous studies have found a similar layered frozen hydrometeor structure and identified 134

graupel in stratiform precipitation in individual case studies [e.g. Leary and Houze, 1979; 135

Houze and Churchill, 1987; Takahashi and Kuhara, 1993; Hogan et al., 2002; Park et 136

al., 2009; Evaristo et al., 2010; Jung et al., 2012; Bechini et al., 2013; Martini et al., 137

2015]. However the PID-based composite analysis conducted by Barnes and Houze 138

[2014] was the first study to demonstrate that these hydrometeor patterns are statistically 139

robust features of tropical, oceanic MCSs and have a systematic relationship to the 140

characteristic mesoscale midlevel inflow. 141

If the PID results from Barnes and Houze [2014] are interpreted in terms of ice 142

microphysical processes, their results indicate that the dominant ice microphysical 143

processes in MCSs over the tropical ocean systematically transition downward from 144

primarily deposition at high levels to a layer of considerable aggregation and some 145

deposition before melting, with riming occasionally occurring in shallow pockets just 146

above the melting level. This layered pattern is consistent with the conclusions and 147

conceptual diagrams of Leary and Houze [1979], Houze [1981; 1989], Houze and 148

Churchill [1987], and Braun and Houze [1994]. Given that Barnes and Houze [2014] has 149

demonstrated this ice microphysical pattern is statistically robust, the question arises as to 150

whether simulated microphysical processes exhibit a similar pattern. 151

In recent decades the number of microphysical parameterizations available has 152

substantially increased, which has led to a large number of studies focused on comparing 153

the performance of these schemes in a variety of storm types, atmospheric environments, 154

and model frameworks, including tropical MCSs [e.g. Blossey et al., 2007; Wang et al., 155

2009; Varble et al., 2011; Van Weverberg et al., 2013; Hagos et al., 2014; Roh and 156

Satoh, 2014]. One aspect of microphysical parameterizations that has been largely 157

neglected in prior intercomparison studies is the spatial distribution of microphysical 158

processes. The spatial distribution of ice microphysical processes is particularly 159

important since Varble et al. [2011] demonstrated that significant differences exist in the 160

mixed and ice phase regions of observations and simulations and Wang et al. [2009] 161

suggested that inconsistencies among observations and simulations result from the 162

treatment of frozen hydrometeors in parameterizations. Caniaux et al. [1994] is one of 163

the only studies that explicitly shows the spatial pattern of microphysical processes 164

within simulated convection. However, their study was conducted using a two-165

dimensional anelastic model. Somewhat surprisingly, it is relatively unknown how these 166

processes arrange themselves in three-dimensional full-physics simulations. Several 167

studies including Donner et al. [2001] have shown the spatial pattern of latent heating 168

attributed to specific microphysical processes. While these studies provide an indication 169

of the spatial pattern of microphysical processes that change the phase of water, this 170

technique provides no insight into processes that do not impact latent heating, such as 171

aggregation/accretion. A common way to analyze the simulated spatial pattern of 172

microphysical processes is to apply a dual-polarization radar simulator to the model 173

output and compare the simulator’s output with observations [e.g. Jung et al., 2012; 174

Putnam et al., 2014; Brown et al., 2015; Jung et al., 2016]. While these studies provide 175

insight into the simulated microphysical structure, it is unclear whether observed 176

differences are attributable to the model, the simulator, or a combination of both. 177

The current study assimilates radial velocity data from DYNAMO/AMIE into the 178

Weather Research and Forecasting (WRF) model and outputs fields that show the spatial 179

pattern of deposition, aggregation, riming, and melting. This technique constrains 180

naturally evolving MCSs in a full-physics three-dimensional model to have the same 181

kinematic structure as observed MCSs, while allowing full interaction with the 182

microphysical processes. This method enables the spatial pattern of simulated and 183

observed microphysical processes to be directly and equitably compared with minimal 184

contamination from dynamical differences. Using these model results, this study will 185

explicitly investigate if three different microphysical parameterization schemes produce 186

large-scale spatial patterns of deposition, aggregation/accretion, riming, and melting that 187

are: 188

1. Consistent with the systematic ice microphysical pattern observed in the 189

PID data from DYNAMO/AMIE, 190

2. Consistent with our theoretical understanding of microphysics and their 191

interaction with the dynamical structure of convection, and 192

3. Consistent with each other. 193

2. Methodology 194

2.1 Microphysical Interpretation of Particle Identification (PID) Algorithm 195

By emitting and receiving horizontally and vertically polarized pulses, dual-196

polarization radars obtain moments of the particle distribution in the radar sampling 197

volume that indicate the physical characteristics of the particles. The variables obtained 198

in this way include differential reflectivity (ZDR), linear depolarization ratio (LDR), 199

correlation coefficient (ρhv), and specific differential phase (KDP). (See Bringi and 200

Chandrasekar [2001] for a comprehensive description of dual-polarimetric radar 201

variables.) Vivekanandan et al. [1999] developed a technique that classifies regions 202

within convection by combining dual-polarization radar variables and rawinsonde 203

temperature profiles. Included in this frequently used algorithm, which is referred to as a 204

PID, are five frozen hydrometeor categories. Barnes and Houze [2014] referred to these 205

frozen hydrometeor categories as horizontally oriented ice crystals, small ice crystals, dry 206

aggregates, wet aggregates, and graupel/rimed aggregates. (For additional details about 207

the PID used during DYNAMO/AMIE see Rowe and Houze [2014] and Barnes and 208

Houze [2014].) While these categories are named in terms of particle type, the names 209

imply the microphysical processes producing the particles. 210

Table 1 shows the dual-polarimetric and temperature thresholds used to define 211

each of these categories. One of the defining characteristics of horizontally oriented ice 212

crystals is large ZDR values, which indicate that the major axis of these particles have a 213

strong horizontal component to their orientation. However, both horizontally oriented ice 214

crystals and small ice crystals are associated with the smallest horizontal reflectivity and 215

have large negative values of LDR, which suggests that both particles are the smallest 216

frozen hydrometeors identified by the radar and have an undefinable shape. The most 217

likely way that a frozen particle can grow large enough to be detected by the radar and 218

not have the characteristics of an aggregate of particles is through deposition. Barnes and 219

Houze [2014] found that horizontally oriented ice crystals during DYNAMO/AMIE were 220

often observed at temperatures between -10°C and -20°C and in regions of vertical wind 221

shear. Mason [1971] and Hobbs [1974] suggest that vertical wind shear at these 222

temperatures promotes enhanced depositional growth. Both horizontally oriented ice 223

crystals and small ice crystals could have been advected from another part of the storm, 224

but even in that case any subsequent growth of these advected particles would most likely 225

result from deposition. Compared to the horizontally oriented ice crystals and small ice 226

crystals, dry aggregates have higher reflectivity and their LDR values are less negative. 227

Thus, dry aggregates are larger than horizontally oriented ice crystals and small ice 228

crystals and have a more uniform shape [e.g. Bader et al., 1987; Straka et al., 2000; 229

Andrić et al., 2013]. They are referred to as dry aggregates because they are generally 230

found well above the 0°C level, where they would not have any liquid attribute as would 231

melting aggregates. In a region of upward motion, such as occurs above the midlevel 232

inflow layer of an MCS, these dry aggregates would also be expected to be increasing in 233

mass by vapor deposition, as suggested by Houze [1989]. Some of the distinguishing 234

characteristics of wet aggregates include reduced correlation coefficient and high 235

differential reflectivity, which are indications of mixed phase particles and melting [e.g. 236

Zrnić et al., 1993; Straka et al., 2000; Brandes and Ikeda, 2004]. Previous studies have 237

demonstrated that the combination of high reflectivity and low differential reflectivity 238

assigned to the graupel/rimed aggregate category is predominantly associated with riming 239

[e.g. Aydin and Seliga, 1984; Straka et al., 2000]. Rimed particles, such as graupel, 240

would also be growing by vapor deposition in in-cloud regions of upward motion. This 241

interpretation of the graupel/rimed aggregates assumes that these particles are formed 242

locally, not advected from other regions of the storm. This is a valid assumption since the 243

authors have analyzed the dual-polarimetric variables associated with dozens of MCSs 244

with a midlevel inflow and graupel/rimed aggregates were always restricted to a narrow 245

layer just above the 0°C level. If these rimed particles had been advected from a more 246

convectively active portion of the storm, we would expect to see graupel/rimed 247

aggregates at higher altitudes. Using these interpretations the PID categories can be used 248

to map the spatial pattern of ice microphysical processes within a region of radar echo in 249

an MCS. Specifically, horizontally oriented ice crystals and small ice crystals indicate 250

primarily deposition, dry aggregates represent aggregation combined with deposition, 251

graupel/rimed aggregates indicate riming possibly mixed with deposition, and wet 252

aggregates occur where melting is occurring. 253

While the PID is a powerful technique for mapping the locations of different 254

microphysical processes, it has limitations. First, the PID identifies only the process 255

creating the hydrometeor with the largest radar return, even though multiple processes are 256

likely occurring in a given volume of air. As a result, PID results are biased towards 257

processes that create particles that are the largest, densest, or have the highest dielectric 258

constant. Consequently, the process identified may not be the most prevalent process. 259

This problem becomes more serious with distance from the radar due to the broadening 260

of the radar beam [Park et al., 2009]. The certainty of the PID is also questioned because 261

the theoretical associations between the dual-polarimetric variables and categories are 262

complex and involve overlapping boundaries (Table 1) and because few validation 263

studies exist. Vivekanadan et al. [1999] pointed out that the complex relationships used in 264

the PID are unavoidable and the soft boundaries used in the fuzzy-logic algorithm 265

provides one of the best ways to handle these relationships. 266

We do not attempt here to address all the uncertainties in PID algorithms. We are 267

applying the technique only to a limited type of convection, namely, MCSs observed in a 268

tropical oceanic environment, specifically in DYNAMO/AMIE. In the case of this 269

dataset, Barnes and Houze [2014] and Martini et al. [2015] have analyzed the 270

performance of the PID during DYNAMO/AMIE and concluded that it was accurate. We 271

are therefore confident that the DYNAMO/AMIE PID dataset describes the dominant ice 272

microphysical process with sufficient accuracy for the purposes of this study. 273

Our aim is to interpret the PID in terms of microphysical processes in a way that 274

can determine if the spatial pattern of microphysical processes in numerical simulations is 275

consistent with that seen in the DYNAMO/AMIE dataset. The PID cannot validate 276

numerical simulations when it is interpreted in terms of hydrometeors because the 277

defining characteristics of hydrometeors in the PID and numerical simulations are 278

fundamentally different. In numerical simulations, hydrometeors are defined based on the 279

particle’s size and density. The PID defines hydrometeors based on their physical 280

characteristics including their relative shape, their water phase, and the uniformity of the 281

particles. Models compute mixing ratios of particle types; PID methods cannot determine 282

mixing ratios. However, PID data can be interpreted in terms of the microphysical 283

processes producing the particles, and numerical simulations directly calculate the same 284

processes. Our comparisons therefore are based on processes, not mixing ratios. 285

2.2 Classification of Microphysical Processes in WRF 286

Microphysical parameterizations within WRF only routinely output the hydrometeor 287

mixing ratios and, if the scheme is double-moment, number concentrations. However, 288

dozens of variables describe the interactions among the mixing ratios of individual 289

hydrometeor types. By grouping these hydrometeor mass interaction variables in terms of 290

the microphysical process they describe, additional three-dimensional fields depicting the 291

mass production rate of microphysical processes can be output. This study considers four 292

ice microphysical processes: deposition, aggregation, riming, and melting. The following 293

definitions were used to group the hydrometeor interaction variables: 294

Deposition included any hydrometeor mass interaction variable that 295

described how a frozen hydrometeor collected water vapor. 296

Aggregation included any hydrometeor mass interaction variable that 297

described the collection of a frozen hydrometeor on to another frozen 298

hydrometeor. This included any variable that described the interaction 299

among snow, ice, graupel, and/or hail. Thus our definition of aggregation 300

includes all forms of ice accretion. Variables describing how aggregation 301

impacts the number concentration of these particles were not included. For 302

purposes of this study, the impact of this exclusion is relatively minor and 303

will be discussed below. 304

Riming included any hydrometeor mass interaction variable that described 305

how a frozen hydrometeor collected a liquid hydrometeor and remained 306

classified as frozen. It is possible that a thin layer of liquid developed 307

along the edge of the hydrometeor even though its core remained frozen. 308

Melting included any hydrometeor mass interaction variable that described 309

how a frozen hydrometeor became a liquid hydrometeor. 310

The current study tests three routinely available microphysical parameterization 311

schemes in WRF: the Milbrandt-Yau Double Moment Scheme (MY) [Milbrandt and 312

Yau, 2005a; b], the Morrison 2-Moment Scheme (MOR) [Morrison et al., 2009], and the 313

WRF Double-Moment 6-Class Scheme (WD) [Lim and Hong, 2010]. While MY is fully-314

double moment, the MOR and WD are only partially double-moment. MOR is double 315

moment in rain, ice, snow, and graupel. WD scheme is double moment in cloud water 316

and rain. Table 2 shows how the hydrometeor mass interaction variables in each 317

parameterization were grouped into microphysical processes. 318

2.3 Data Assimilation Simulations 319

One of the most challenging aspects of comparing observed and simulated 320

microphysical processes is their sensitivity to the dynamics. If the observed and 321

simulated dynamical structure differ it is impossible to discern if differences in the 322

microphysical processes are related to the processes themselves or dynamical differences. 323

Shipway and Hill [2012] addressed this complication by developing a one-dimensional 324

kinematic driver model that prescribed the flow and did not allow interaction among the 325

dynamics and microphysics. While their methodology provides a straight-forward 326

method to equitably compare parameterizations, it does not address the fact that the 327

microphysics evolve in concert with the dynamics of the convection. 328

Our study uses data assimilation to require simulations to develop a mesoscale 329

circulation similar to observations. By so doing, we enable the microphysical processes to 330

evolve freely but only in concert with a dynamically accurate simulation. The 331

microphysics can then be compared to dual polarization radar observations with minimal 332

contamination from erroneous dynamics. Previous studies suggest that this technique is a 333

valid way to investigate microphysical variability among parameterizations. Wheatley et 334

al. [2014] assimilated radar data using a WRF-based ensemble Kalman filter and 335

demonstrated that differences among microphysical schemes were the same whether or 336

not radar data was assimilated. Assimilating radial velocity data and mesonet 337

observations, Marquis et al. [2014] demonstrated that a WRF-based ensemble Kalman 338

Filter (EnKF) could simulate supercell thunderstorms whose kinematic structure was 339

consistent with observations. While these studies were investigating a type of convection 340

very different from tropical oceanic convection, their results suggest our approach is 341

reasonable. However, data assimilation in Marquis et al. [2014] only accounted for the 342

mesoscale circulation. While the mesoscale circulation is the dominant dynamical feature 343

of MCSs, convective-scale dynamical differences exist within our simulations and likely 344

influence some of our results. The impact of these differences are relatively minor and 345

will be discussed where applicable. 346

Radial velocity is the most accurate radar variable. It involves no assumptions 347

about the nature of the particles, and it is a direct measurement. All other radar variables 348

are derived. Additionally, all other radar variables measure characteristics of the particle 349

population and assimilating them could manipulate the simulated microphysical pattern. 350

This manipulation would obscure differences among observations and simulations. Thus, 351

radial velocity is the most appropriate radar variable to assimilate in this study since it 352

forces simulations to have similar mesoscale dynamical evolution while impacting the 353

microphysical processes as little as possible. 354

Radial velocity data used in this study was obtained by the NCAR S-PolKa radar 355

during the DYNAMO/AMIE field campaign, which was located on Addu Atoll in the 356

Maldives from November 2011 through January 2012 [Yoneyama et al., 2013]. The 357

NCAR S-PolKa radar is a dual-wavelength (10.7 and 0.8 cm), dual-polarimetric, Doppler 358

radar that has a beam width of 0.92° and maximum range of 150 km. In order to map the 359

horizontal and vertical distribution of convection, S-PolKa’s scan strategy during 360

DYNAMO/AMIE consisted of a set of horizontal surveillance scans (called plain 361

position indicator (PPI) scans) and vertical scans along specified azimuthal angles (called 362

range height indicator (RHI) scans) repeated every 15 minutes. (For more details about 363

the scan strategy used during DYNAMO/AMIE see Zuluaga and Houze [2013], Powell 364

and Houze [2013], and Rowe and Houze [2014].) This study only considers PPI scans 365

from the 10.7 cm (S-band) wavelength radar. Prior to assimilation the radial velocity data 366

was quality-controlled and thinned. Quality-control measures included removing data in 367

pixels that contained low signal-to-noise ratios, clutter, and/or high spectral width. 368

Additionally, data in pixels with reflectivity less than 3 dBZ were excluded. The quality-369

control of the radial velocity data was conducted with the assistance of a Matlab-based 370

quality-control package being developed by the NCAR Earth Observing Laboratory 371

(Scott Ellis, NCAR EOL, personal communication). Given that the NCAR S-PolKa radar 372

data had a much finer horizontal resolution (150 m) than our simulations (3 and 1 km), 373

the radial velocity data was thinned prior to assimilation onto a 2° x 1 km grid using the 374

super-observation technique discussed in Zhang et al. [2009]. 375

This study uses the Advanced Research version of the Weather Research and 376

Forecasting (WRF ARW) model version 3.5 [Skamarock et al., 2008]. Details about the 377

WRF architecture used in this study is provided in Table 3 and the nested domains with 3 378

and 1 km resolution, respectively, are shown in Figure 1. Each of the domains were two-379

way nested and had 39 unevenly spaced vertical levels between the surface and 26 km 380

with maximum resolution near the surface. Analysis was only completed in the inner 381

domain with 1 km resolution. The inner domain is located mostly east of the radar 382

because the scans to the west of the radar were obscured by ground clutter. The high 383

computation expense of data assimilation prevented finer horizontal resolutions from 384

being used. The model time step was 18 seconds and data was output every 15 minutes. 385

This model output frequency is appropriate given that the current study is only interested 386

in the spatial pattern of the microphysical processes, not the temporal evolution of the 387

microphysical pattern. 388

The WRF-based ensemble Kalman filter (EnKF) system used in this study is same 389

as Meng and Zhang [2008a; b] and Zhang et al. [2009]. Details about the EnKF 390

architecture are provided in Table 3. The EnKF was run using 50 members whose initial 391

and boundary conditions were perturbed from ERA-interim reanalysis using a domain-392

specific background error covariance and the “cv5” option of the WRF 3DVar package 393

[Barker et al., 2004]. The perturbed variables in the ensemble included perturbation 394

potential temperature, zonal and meridional wind, cloud water mixing ratio, vapor mixing 395

ratio, rain mixing ratio, base-state and perturbation geopotential, base-state and 396

perturbation dry air mass in the column, surface pressure, base-state and perturbation 397

pressure, and zonal and meridional wind at 10 km. Radial velocity was assimilated into 398

both domains every 15 min after WRF spun up for 6 h. The background error covariance 399

was inflated prior to each assimilation using the covariance relaxation method proposed 400

by Zhang et al. [2004, their Eq. 4] with a relaxation coefficient of 0.8. The successive 401

covariance localization (SCL) method [Zhang et al., 2009] was used with a fifth-order 402

correlation function [Gaspari and Cohn, 1999] and a horizontal radius of influence of 90 403

and 30 km for the outer and inner domains, respectively. The vertical radius of influence 404

was set to six model levels. The EnKF updated perturbation potential temperature, zonal 405

and meridional wind, cloud water mixing ratio, vapor mixing ratio, rain mixing ratio, 406

perturbation geopotential, perturbation dry air mass in the column, surface pressure, 407

perturbation pressure, and zonal and meridional wind at 10 km. 408

Two sets of simulations were conducted. One of a squall line on 23 December 409

2011 and one of a non-squall MCS on 16 October 2011. Details about these storms are 410

provided in Section 3. Each set of simulations contained three model runs whose only 411

difference was their microphysical parameterization. The same ensemble was used to 412

initialize all simulations on 23 December and 16 October, respectively. 413

2.4 Model Spatial Compositing Technique 414

Assimilating radial velocity results in each ensemble member developing a 415

mesoscale midlevel inflow. However, details about the midlevel inflow, including its 416

magnitude, slope, length, and location slightly vary among members. This variability is 417

expected and required since the EnKF is a best-linear estimator and assumes that the 418

ensemble represents the range of all possible outcomes. This variability is problematic 419

when the spatial pattern of microphysical processes around the midlevel inflow is 420

analyzed because it smears the spatial pattern and makes its association with the midlevel 421

inflow unclear. Thus, prior to analyzing the spatial pattern of microphysical processes, 422

we determined if a robust midlevel inflow existed in a given solution and spatially 423

composited all robust midlevel inflows so they were the same horizontal size. 424

The first step in creating the spatial composites was to identify robust midlevel 425

inflows. At a given time, a set of vertical cross sections was taken from each ensemble 426

member. Cross sections from the 23 December simulations were examined at 1930 UTC, 427

after radial velocity data had been assimilated seven times. Cross sections from the 16 428

October simulations vertical cross sections were analyzed at 1445 UTC, after radial 429

velocity data had been assimilated eleven times. Analysis was done after a different 430

number of assimilation periods in the two sets of simulations since the squall line on 23 431

December rapidly formed a midlevel inflow but the MCS on 16 October did not form a 432

midlevel inflow until late in its lifecycle (not shown). This analysis methodology has 433

been applied to different time steps and has been used to composite the same cross 434

section at multiple times. In all cases the composites are fundamentally the same (not 435

shown). 1930 UTC 23 December and 1445 UTC 16 October were specifically selected 436

since they provided the largest sample size. 437

The midlevel inflow observed by S-PolKa on 23 December descended 438

predominantly from west to east. On 16 October the midlevel inflow observed by S-439

PolKa was primarily oriented from south to north. Based on this observed geometry, the 440

cross sections from the WRF simulations were oriented zonally on 23 December and 441

meridonally on 16 October. Given that the midlevel inflow region was broader during the 442

16 October non-squall MCS, 6 cross sections were taken from each ensemble member 443

during the 16 October simulations and 5 cross sections were taken from each ensemble 444

member during the 23 December simulations. Multiple cross sections were considered 445

from each member in order to increase the sample size within the composites. However, 446

each cross section was evaluated independently. Thus, the composites may contain no 447

cross section, one cross section, or multiple cross sections from a given ensemble 448

member. The number of cross sections from each ensemble member does not change the 449

results of this study. 450

The process of identifying robust midlevel inflows in each cross section and 451

determining the core of the robust inflows was as follows: 452

1. Convert the vertical coordinate in the cross sections from their native unevenly 453

spaced eta coordinates to uniform height coordinates. 454

2. Isolate convection. Any vertical column within the cross section that lacked 455

reflectivity greater than 5 dBZ above 5 km was ignored. 456

3. Remove convective cores. In the 23 December simulations, data within vertical 457

columns that had reflectivity greater than 30 dBZ above 8 km and vertical velocities 458

greater than 2 m s-1 at any altitude was ignored. In the 16 October simulations, data 459

within vertical columns that had reflectivity greater than 30 dBZ above 6 km and 460

vertical velocities greater than 2 m s-1 at any altitude was ignored. The threshold 461

changed due to convective intensity differences between the storms. 462

4. Identify the location of the potential midlevel inflow. Isolate locations within the cross 463

section that had horizontal wind speeds greater than 18 m s-1 during the 23 December 464

simulations or greater than 9 m s-1 during the 16 October simulations. This region was 465

called the potential midlevel inflow. The threshold changed due to wind speed 466

differences between the storms. If no region within the cross section satisfied this 467

criteria the cross section was excluded from the analysis. 468

5. At each model level, identify the location of the maximum wind speed within the 469

potential midlevel inflow. These locations were referred to as maximum points. 470

6. Ensure that the potential midlevel inflow is linear. Use the maximum points to 471

calculate the best fit linear line. Exclude the cross section if the correlation between 472

the maximum points and the best fit line was less than 0.9. It is appropriate to use a 473

best fit linear line since we previously converted the vertical coordinate of the cross 474

section from unevenly spaced eta values to uniform height intervals. 475

7. Ensure that the midlevel inflow was not sloped too steeply. Exclude the cross section 476

if the slope of the best fit linear line was less than -15 or greater than 0. This ensured 477

that the potential midlevel inflow gradually descended from left to right. It is 478

appropriate to use a best fit linear line since we previously converted the vertical 479

coordinate of the cross section from unevenly spaced eta values to uniform height 480

intervals. 481

8. Find the base of the core of the midlevel inflow. Find the maximum point with the 482

lowest altitude. This was called the convective end of the midlevel inflow core. 483

Ignore all data to the right of this point. 484

9. Ensure that the core of the midlevel inflow was coherent. If a maximum point was 485

more than 0.1° longitude away from the maximum point immediately to its right, 486

ignore that maximum point and all subsequent maximum points to its left. This length 487

scale was arbitrary defined. Results were insensitive to the length scale. Changing the 488

length scale only changed the number of cross sections in the composites. 489

10. Find the top of the core of the midlevel inflow. Find the maximum point that had the 490

highest altitude and the left most horizontal position. This was called the anvil end of 491

the midlevel inflow core. Any maximum point that was above and/or to the right of 492

this point was ignored. 493

11. Ensure that the core of the midlevel inflow was large enough. Exclude the cross 494

section if the distance between the convective and anvil end of the midlevel inflow 495

core was less than 0.2° longitude. This length scale was arbitrary defined. Results 496

were insensitive to this length scale. Changing this length scale only changed the 497

number of cross sections in the composites. 498

12. Ensure that only one coherent storm was contained within the midlevel inflow core. 499

Exclude the cross section if more than five horizontally adjacent model grid points 500

had reflectivity less than 5 dBZ at any vertical level or if there were vertical holes 501

where reflectivity was less than 5 dBZ. 502

If these criteria were satisfied the cross section was said to contain a robust midlevel 503

inflow and the midlevel inflow core as defined as the region between the maximum 504

points at convective and the anvil end. All subsequent analysis only used data within the 505

midlevel inflow core. 506

Once all midlevel inflow cores were located, they were arbitrarily scaled to the 507

same length and shifted to the same location. Results were insensitive to the assumed 508

length scale and shifted location. No vertical scaling was conducted since the height 509

profile was nearly the same in each simulation. 510

One of the primary objectives of this study is to compare the spatial pattern of ice 511

processes within the midlevel inflow region from WRF with PID observations. The PID 512

only indicates what process is dominant at a given location. No information about the 513

microphysical mass production rate is provided. Thus, the PID can only validate the 514

location of microphysical processes. To match the PID data, the spatial pattern of ice 515

microphysical processes generated by the three microphysical parameterizations in this 516

study was represented in terms of their composite frequency. At each grid point the 517

number of cores that had a nonzero microphysical mass production rate was normalized 518

by the total number of cores. The WRF composites were not generated using the 519

dominant simulated microphysical process since the definition of the dominant process 520

differs in the PID and WRF. In the PID, the dominant microphysical process is the 521

process that produces the hydrometeor that is the largest, densest, or has the highest 522

dielectric constant since these particles account for the largest radar return power. The 523

dominant microphysical process in WRF is defined based on the mass production rate of 524

the process, which cannot be observed using the PID. In order to ensure that composites 525

were statistically robust, the frequency was only reported if more than half of the cores at 526

that grid point had reflectivity greater than 5 dBZ. Thus, the frequency composites shown 527

in this study indicate where processes were statistically more and less likely to occur 528

relative to the midlevel inflow, which is directly comparable to the data provided by the 529

PID. 530

This study also generated composites of radar reflectivity, horizontal wind speed, 531

vertical velocity, and temperature. Similar to above, only grid points where more than 532

half of the cores had reflectivity greater than 5 dBZ were included. The reflectivity used 533

in this study was directly output by WRF and was produced by an S-band radar simulator 534

that had been adapted to fit the assumed hydrometeor size distributions in each scheme. 535

All scattering was assumed to be in the Rayleigh regime. WRF simulated reflectivity is 536

directly comparable to observed S-PolKa reflectivity since the S-PolKa radar is also an S-537

band radar. 538

3. Dual-Polarization Radar Observations of Mesoscale Convective Systems 539

Barnes and Houze [2014] identified 36 MCSs during DYNAMO/AMIE that 540

contained a midlevel inflow and created spatial composites that showed that the 541

DYNAMO/AMIE PID data had a systematic spatial pattern relative to the midlevel 542

inflow. Squall line and non-squall MCSs were composited separately in Barnes and 543

Houze [2014] and the organization of the PID data relative to the midlevel inflow was 544

unchanged by whether or not the MCS was organized into a squall line. The current study 545

simulates a squall line and non-squall MCS that were included in the composites in 546

Barnes and Houze [2014]. It is important that this study considers two different types of 547

MCSs in order to investigate if the morphological structure of convection impacts the 548

simulated spatial pattern of ice processes. 549

The squall line simulated in this study occurred on 23 December 2011 (Figure 2a-550

2c). This case was selected since it was one the most well-developed and strongest squall 551

lines observed by S-PolKa during DYNAMO/AMIE. At 1935 UTC a broken convective 552

line was oriented north-south and its leading edge was about 100 km to the east of the S-553

PolKa radar (Figure 2a). Stratiform precipitation extended from the convective line 554

westward for 100-200 km. Figure 2b and 2c shows a vertical cross section of reflectivity 555

and radial velocity data along the red line in Figure 2a. Reflectivity shows that this storm 556

had a well-defined convective zone characterized by vertically erect convective cores, a 557

transition region characterized by weaker reflectivity, and a stratiform region with a 558

distinct brightband. Radial velocity shows that the squall line had a strong descending 559

midlevel inflow whose radial velocities exceeded 20 m s-1 approximately 50-75 km away 560

from the S-PolKa radar. This midlevel inflow extended from beneath the anvil 561

(approximately 15 km from S-PolKa) into the stratiform region and provided the primary 562

forcing for the squall line. 563

Five vertical cross sections were taken through the 23 December squall line and 564

the PID data from these cross sections were spatially composited around the observed 565

midlevel inflow using a methodology directly analogous to the method described in 566

Section 2.4 (For a complete description of the spatial compositing technique applied to 567

the PID see Barnes and Houze [2014]). The shading in Figures 3a-3e show the frequency 568

in which horizontally oriented ice crystals, small ice crystals, dry aggregates, 569

graupel/rimed aggregates, and wet aggregates occurred at a given location relative to the 570

midlevel inflow. The dotted magenta line represents the approximate location of the 571

midlevel inflow. These PID composites indicate that the frozen hydrometeors had a 572

layered structure with small ice crystals at cloud top, dry aggregates at midlevels above 573

the melting level, and wet aggregates near the melting level. The isolated region of 574

horizontally oriented ice crystals suggests that deposition was enhanced just above the 575

midlevel inflow near the anvil end of the midlevel inflow, where the shear and 576

temperature conditions correspond to those identified by Mason [1971] and Hobbs [1974] 577

as favoring deposition. Graupel/rimed aggregates were located in a few isolated regions 578

along the upper boundary of the wet aggregate layer. The isolated nature of the 579

graupel/rimed aggregates is not a feature unique to this case. Whenever graupel/rimed 580

aggregates were observed within one of the 36 midlevel inflow regions investigated in 581

Barnes and Houze [2014] these hydrometeors always occurred in shallow, isolated 582

pockets. Additionally, Rowe and Houze [2014] analyzed all MCSs observed during 583

DYNAMO/AMIE and also found that graupel/rimed aggregates in stratiform regions 584

were confined to a narrow region above the 0°C level. Thus, the hydrometeor pattern 585

observed during the 23 December squall line is entirely consistent with the systematic 586

pattern identified in Barnes and Houze [2014] and is long thought to exist in oceanic 587

tropical convective systems [Leary and Houze, 1979b; Houze, 1989]. 588

Interpreting the PID in terms of processes suggests that ice processes on 23 589

December were layered with only depositional growth near echo top, aggregation at 590

midlevels above the 0°C level, and melting and shallow pockets of riming near the 0°C 591

level. At the levels where aggregation was occurring, the larger particles dominated the 592

radar signal, thus preventing deposition from being identified. However, because upward 593

motion was likely occurring above the midlevel inflow layer, the aggregates would have 594

been also accumulating mass by depositional growth. We therefore interpret the layer 595

where aggregation is occurring as a zone where both aggregation and vapor deposition 596

were active. This interpretation is important when comparing the PID with model output 597

since the parameterizations calculate aggregation and deposition separately and therefore 598

keep track of both processes. A similar interpretation applies to riming. If rimed particles 599

are present in a zone of upward motion, they too would be growing by vapor deposition 600

as well as riming. 601

The non-squall MCS simulated in this study persisted over the S-PolKa radar 602

domain for nearly 18 h on 16 October 2011. This MCS was selected since it was one the 603

largest MCSs observed by S-PolKa during DYNAMO/AMIE. Figures 2d-2f shows the 604

storm at 1450 UTC as a midlevel inflow developed during the later stages of this MCS. 605

Prior to this time this MCS lacked a midlevel inflow. Thus, our analysis could not have 606

been conducted when the 16 October MCS was in the earlier stages of its lifecycle. The 607

convective and stratiform regions were organized less systematically with Figure 2d 608

showing pockets of high reflectivity convective cores scattered within an expansive 609

region of weaker reflectivity stratiform precipitation. Figures 2e and 2f show a cross 610

section of reflectivity and radial velocity data through a portion of the stratiform 611

precipitation along the red line in Figure 2d. Five cross sections were taken within 612

southeast portion of this storm and PID data was spatially composited around the 613

midlevel inflow (Figure 3f-3j) in a manner analogous to the methodology described in 614

Section 2.4. (For a complete description of the spatial compositing technique applied to 615

the PID see Barnes and Houze [2014]). 616

While some hydrometeor/microphysical differences existed between the 23 617

December squall line and 16 October non-squall MCS, the composited DYNAMO/AMIE 618

PID data had the same fundamental spatial pattern relative to the midlevel inflow in both 619

cases. Similar to 23 December, the 16 October non-squall MCS had a well-defined 620

midlevel inflow (Figure 2c, 2f). However, the midlevel inflow on 16 October remained at 621

a constant altitude of 5 km and only had a maximum intensity of 15 m s-1. Additionally, 622

the 16 October non-squall MCS was shallower than the 23 December squall line (Figures 623

2b, 2e) and its precipitation was weaker. A weak brightband was apparent on 16 October, 624

but a portion of the precipitation did not reach the surface. Loops of reflectivity and radial 625

velocity (not shown) suggest that the midlevel inflow on 16 October weakened the MCS, 626

instead of forcing the MCS as observed on 23 December. A few differences in the 627

detailed spatial patterns of the ice processes also exist between these storms. Unlike the 628

23 December squall line, PID data during the later stages of the 16 October MCS often 629

lacked shallow pockets of graupel/rimed aggregates, had a shallower dry aggregate layer, 630

and more prevalent horizontally oriented ice crystals. This comparison suggests that the 631

stratiform region associated with the 16 October non-squall MCS lacked riming and had 632

less aggregation but had stronger deposition than the 23 December squall line. While 633

these differences may be attributable to weaker vertical motions on 16 October than on 634

23 December, observational data to verify this theory is unavailable. Differences in the 635

prevalence of horizontally oriented ice crystals between these MCSs have little impact on 636

the conclusions of the current study since both horizontally oriented ice crystals and 637

small ice crystals are predominantly associated with deposition. The combination of these 638

two hydrometeors indicates that deposition is the predominant microphysical process 639

near echo top in each MCS. Thus, while differences in the spatial microphysical pattern 640

existed between these MCSs, these differences were relatively minor. Overall, both 641

MCSs were characterized with a layered ice process pattern that transitioned from mostly 642

deposition in upper levels to aggregation to melting with increasing distance from echo 643

top in a manner that was consistent with Barnes and Houze [2014]. 644

4. Kinematic Structure of Simulated Mesoscale Convective Systems 645

Before the spatial pattern of simulated ice processes can be compared with the 646

PID, it must be proven that assimilation successfully results in the ensembles developing 647

a mesoscale circulation consistent with observations. Figure 4 shows cross sections of S-648

PolKa radial velocity and composite simulated horizontal wind speed for the 23 649

December and 16 October MCSs in shading. S-PolKa radial velocity and WRF horizontal 650

wind speeds are not directly comparable since radial velocity describes a portion of the 651

horizontal and vertical wind component. However, the vertical velocity component is 652

often an order of magnitude less than the horizontal velocity component within the 653

midlevel inflow region. Streamlines calculated within the plane of the WRF cross 654

sections are oriented nearly horizontal (not shown). Additionally, while S-PolKa radial 655

velocity underestimates horizontal winds, this underestimation is minimized since cross 656

sections of S-PolKa data were oriented along the dominant direction of the midlevel 657

inflow (see section 2.4). This underestimation by S-PolKa does not impede this study 658

since validating the absolute magnitude of the simulated midlevel inflow is not our 659

objective. We only require the simulated midlevel inflows to have a magnitude similar to 660

observations. Thus, it is appropriate to compare the WRF horizontal wind speeds to S-661

PolKa radial velocity. Figure 4 shows that each ensemble had a midlevel inflow whose 662

slope and magnitude was similar to the S-PolKa observations. 663

The shading in Figure 5 shows the composite vertical velocities on 23 December 664

and 16 October. The midlevel inflow region is characterized by a broad weak updraft 665

above the midlevel inflow and a broad weak downdraft below the midlevel inflow, 666

consistent with previous studies [Houze 2004; 2014]. Note that the color bars in Figure 5 667

have different scales. The scale used for the 23 December squall line is twice as large as 668

the scale used for the 16 October non-squall MCS. The 23 December ensembles were 669

expected to have consistently greater vertical velocities than the 16 October ensembles 670

since the 23 December case was an active squall line and the 16 October case was a late 671

stage MCS. The magnitude of these vertical velocities cannot be validated since in situ 672

vertical velocity data was unavailable during these storms. 673

The small-scale vertical velocity differences between the parameterizations in 674

Figure 5 indicates that convective-scale differences in the simulated vertical velocity 675

pattern exist despite data assimilation. Given that the airflow through these storms was 676

dominated by the midlevel inflow, this convective-scale vertical velocity variability is not 677

a serious problem. Thus, this study proceeds and compares the observed and simulated 678

spatial pattern of ice processes within the midlevel inflow region with confidence that 679

most of the microphysical differences exist due to differences in the processes and not 680

because of large-scale dynamical differences. The impact of convective-scale vertical 681

velocity variability on the results will be discussed below. 682

5. Microphysical Comparison 683

5.1 Deposition 684

Figure 6 shows the frequency of simulated deposition within the midlevel inflow 685

region on 23 December and 16 October in shading. In both sets of simulations, each of 686

the three microphysical parameterizations indicated that deposition was possible 687

anywhere above an altitude of 5 km, which is approximately the 0°C level. This spatial 688

pattern is consistent with Figure 3 and composites presented in Barnes and Houze [2014], 689

which show a layer of dry aggregates and graupel/rimed aggregates topped by a layer of 690

horizontally oriented ice crystals and small ice crystals, all of which are inferred to be 691

growing by deposition in an environment of in-cloud upward air motion. Additionally, 692

this simulated pattern is consistent with laboratory and field observations that indicate 693

that ice can grow via deposition between 0° and -70°C [e.g. Wallace and Hobbs, 2006; 694

Bailey and Hallett, 2009] and has been observed in previous modeling studies including 695

Caniaux et al. [1994] and Donner et al. [2001]. 696

The spatial pattern of simulated deposition varied in both sets of simulations. 697

Deposition did not occur always everywhere above an altitude of 5 km, each 698

parameterization had regions statistically more and less likely to have deposition. 699

Additionally, the location of the maximum and minimum frequency of deposition 700

differed among the parameterizations. This spatial variability was linked to variations in 701

the vertical air motion. Deposition was always collocated with regions of upward vertical 702

motion (yellow shading and orange lines in Figure 6). Because upward motion was 703

statistically more frequent above the midlevel inflow and near echo top, deposition was 704

more common near echo top. The frequency of deposition increased the least near echo 705

top in MOR on 23 December, because the upward vertical velocity was statistically 706

weaker (Figure 5). The relationship between deposition and upward motion agrees with 707

previous studies concluding that ice particles grow via deposition within the mesoscale 708

updraft zone [e.g. Houze, 1981; Gamache and Houze, 1983; Houze and Churchill, 1987; 709

Houze, 1989; Braun and Houze, 1994]. 710

Overall, deposition occurred more frequently in each of the three 711

parameterizations on 16 October than 23 December. This difference is consistent with the 712

PID composites. Recall that horizontally oriented ice crystals are indicative of regions of 713

enhanced deposition and these particles were more prevalent on 16 October than 23 714

December (Figures 3a and 3f). These simulations suggest that the differences in the 715

prevalence of horizontally oriented ice crystals between these MCSs may be related to 716

vertical velocity differences. Specifically, deposition may have been more enhanced on 717

16 October than 23 December since vertical motions on 16 October were more 718

horizontally uniform and had less small-scale variability within the -10°C to -20°C 719

temperature range (Figure 5) favorable for enhanced deposition [e.g. Mason, 1971, 720

Hobbs, 1974]. Vertical velocity is expected to differ in these cases. While the MCS on 16 721

October was in a later stages of its lifecycle and had midlevel inflow that remained nearly 722

parallel with the surface, the MCS on 23 December was an active squall line with a 723

steeply sloping midlevel inflow (Figure 2c and 2f) 724

5.2 Aggregation 725

The composite frequency of simulated aggregation on 23 December and 16 726

October is shown in shading in Figure 7. Aggregation had the most consistent spatial 727

pattern considered in this study with all parameterizations producing aggregation 728

everywhere above the 0°C level. While MOR had a slightly reduced aggregation 729

frequency just above the 0°C level, the frequency of aggregation in MY and WD was 730

almost uniformly one everywhere above the 0°C level. There was no significant 731

difference in the frequency of simulated aggregation between the 23 December and 16 732

October MCSs. Note that this study does not include the variables within MY and MOR 733

that change the number concentration of snow due to aggregation. While these number 734

concentration variables are important terms in the parameterizations, including them 735

would not have changed our results or conclusions since our current definition already 736

shows that aggregation always occurs wherever temperatures are below 0°C. 737

While the microphysical parameterizations were consistent with the PID 738

composites in indicating that aggregation occurred near the 0°C level, serious 739

discrepancies existed in terms of the maximum depth of the aggregation layer. Each 740

simulation has aggregation always reaching echo top and occurring at temperatures 741

below -20°C. Figure 3c and 3h indicates that aggregation never extended to echo top on 742

23 December or 16 October. The relatively shallow extent of aggregation is a robust 743

pattern in the PID data that was observed in each of the 36 midlevel inflows analyzed in 744

Barnes and Houze [2014]. Studies such as Houze and Churchill [1987] and Braun and 745

Houze [1994] indicate that aggregation is most common within 1-2 km of the melting 746

level in the stratiform portion of an MCS. Additionally, this simulated pattern disagrees 747

with laboratory and observational studies that have demonstrated that aggregation rarely 748

occurs below -20°C [e.g. Hobbs et al., 1974]. 749

One factor in this altitude discrepancy between the simulations and PID may be 750

related to the fact that simulated aggregation is defined in this study to be any process in 751

which frozen particles collect each other. This includes processes in which very small ice 752

crystals collide and coalesce, which is likely undetectable by the PID. However, based on 753

laboratory studies even this weak aggregation is not expected at very low temperatures 754

[e.g. Hobbs et al., 1974]. Thus, a systematic error in the representation of aggregation at 755

low temperatures likely exists in these three parameterizations. 756

This abundance of aggregation may have serious implications for simulated 757

convection by impacting radiative fluxes and causing ice particles to become too large 758

and fall too quickly. Van Weverberg et al. [2013] found that simulated tropical MCSs 759

were very sensitive to the fall speed of frozen hydrometeors and accurate simulations 760

required small, slowing falling particles. Roh and Satoh [2014] demonstrated that shallow 761

and midlevel convection was reduced and deep MCSs with stratiform precipitation was 762

increased when snow and ice were prevented from accreting onto graupel in simulations 763

that used a single-moment microphysical parameterization. Recall that these accretion 764

variables were included within the aggregation term used in this study. Thus, the 765

parameterization of aggregation in these schemes may be a factor limiting the ability to 766

accurately simulate stratiform precipitation. This deficiency in stratiform precipitation is 767

a problem that has been discussed in several modeling studies including Morrison et al. 768

[2009], Hagos et al. [2014], and others extending back to Fovell and Ogura [1988]. 769

Evidence of insufficient stratiform precipitation in our simulations will be presented in 770

section 6. 771

5.3 Riming 772

Figure 8 shows the frequency of simulated riming on 23 December and 16 773

October in shading. Each parameterization indicated that both cases were characterized 774

by frequent riming near the 0°C level that occasionally occurred at altitudes as great as 10 775

km. It is important to keep in mind that simulated riming in this study includes processes 776

that create water-coated frozen hydrometeors, which may be partially responsible for the 777

high frequency of riming near an altitude of 5 km (i.e. the approximate location of the 778

0°C level). 779

Despite these broad similarities between these two MCSs, riming was the process 780

analyzed in this study that displayed some of the largest differences among individual 781

parameterizations and between the MCSs. A striking difference between the schemes in 782

both cases was that MY sometimes produced riming well below an altitude of 5 km. 783

Figure 8a indicates that the 23 December MY simulations had near surface riming in 784

nearly 50% of the cross sections. This near surface riming is physically impossible since 785

the 0°C level is at 5 km. Frozen hydrometeors were never observed near the surface 786

during DYNAMO/AMIE, and indeed are never observed over oceanic environments 787

along the equator. While our definition of riming includes processes that could result in 788

water-coated frozen hydrometeors, the frozen core of these hydrometeors are still 789

expected to melt rapidly and should not be present well below the 0°C level. Thus, the 790

near surface riming in MY is erroneous. While Figures 8a and 8d suggest riming 791

simulated by MY occurred below the 0°C level, the rate of this riming was very low (not 792

shown). Thus, this behavior is a model artifact that does not substantially affect the 793

results of this study and likely does not significantly impact the structure of the simulated 794

MCS. 795

Riming in each scheme on 23 December extended to greater altitudes near the 796

convective end than the anvil end of the midlevel inflow (left compared to right side of 797

each panel top row Figure 8). This spatial riming pattern is consistent with simulations 798

conducted by Caniaux et al. [1994]. The schemes differed in how fast riming reduced 799

with height above the melting level and the amount of riming near the 0°C level at the 800

anvil end of the midlevel inflow. Riming in MY and WD was frequently deep near the 801

convective end and was concentrated into a narrow layer near the anvil end. Peak riming 802

in MOR was concentrated in the convective half of the midlevel inflow and was 803

relatively rare in the anvil half. 804

Riming occurred within a shallower layer and had a more consistent spatial 805

pattern on 16 October (bottom row of Figure 8). Each parameterization on 16 October 806

had riming concentrated to locations within a layer near the 0°C level. Unlike the 23 807

December squall line, simulations of the 16 October MCS indicated that the frequency of 808

riming was similar near the convective and anvil end of the midlevel inflow. These 809

differences between the cases are expected since the 16 October non-squall MCS was in a 810

later stage of its lifecycle and had less intense, more horizontally uniform vertical 811

motions than the 23 December active squall line (Figure 5). 812

Similar to the PID composites, each parameterization indicated that riming 813

sometimes occurred near the 0°C level within the midlevel inflow. Leary and Houze 814

[1979b] were the first to discuss the possible presence of rimed particles in the stratiform 815

region of MCSs. They inferred the presence of riming in this layer indirectly from drop-816

size distributions measured below the brightband. Since then observational and modeling 817

studies have identified rimed particles within the stratiform precipitation over the Indian 818

Ocean [Suzuki et al., 2006; Rowe and Houze, 2013], equatorial maritime continent 819

[Takahashi and Kuhara, 1992; Takahashi et al., 1995], West Africa [Evaristo et al., 820

2010; Bouniol et al., 2010], Oklahoma [Zrnić et al., 1993], Taiwan [Jung et al., 2012], 821

and Europe [Hogan et al., 2002]. It is a positive result that all three microphysical 822

schemes identify this riming layer. 823

Riming requires supercooled water, which is produced through upward motion or 824

turbulence. Barnes and Houze [2014] suggested that the riming within the midlevel 825

inflow region during DYNAMO/AMIE could have resulted from shear-induced 826

turbulence, small embedded convective cells, or residual upward motion left from a more 827

convective phase of the precipitation. Each of these potential sources of riming would 828

result in pockets of supercooled water, and associated riming, intermittent on space and 829

time scales smaller than the mesoscale midlevel inflow current. However, Barnes and 830

Houze [2014] could not confirm the association between convective-scale vertical 831

velocity perturbations and riming since they lacked vertical velocity observations. 832

Additionally, it remains possible, though unlikely, that the rimed particles observed in the 833

PID data could have been advected into the midlevel inflow region from a more 834

convectively active portion of the storm. The exact sources of the rimed particles is a 835

topic beyond the scope of the present study. 836

Even though assimilation has forced our simulations to have a similar mesoscale 837

vertical velocity pattern, convective-scale vertical velocity differences exist. In each 838

simulation parameterized riming often was not strongly correlated with the mesoscale 839

vertical velocity (orange contours in Figure 8). The fact that this behavior occurred in 840

every simulation analyzed in this study provides strong evidence that the riming is 841

associated with convective-scale vertical velocity perturbations superimposed on the 842

midlevel inflow. The current study cannot determine whether the convective-scale 843

velocity perturbations necessary for riming are shear or buoyancy induced. However, our 844

results confirm the inference of Barnes and Houze [2014] that riming within the midlevel 845

inflow region during DYNAMO/AMIE was most likely the result of convective-scale 846

vertical velocity perturbations 847

While riming is present in both in observations and simulations, there are 848

discrepancies between observed and simulated composites in terms of the detailed spatial 849

pattern of riming. The most systematic and significant of these discrepancies is the 850

frequency of riming and the depth of the riming layer. The bottom row of Figure 8 851

indicates that riming could occur at any given location along the midlevel inflow in the 852

16 October case with a frequency of at least 0.7 in each parameterization. This high 853

frequency strongly contrasts the observational PID. During the later stages of the 16 854

October MCS, PID observations indicated that riming was absent from the stratiform 855

portion of the MCS (Figure 3i). Riming was observed during the 23 December squall 856

line, but the riming frequency peaked at 0.6 (Figure 3d). PID composites presented in 857

Barnes and Houze [2014] (their Figure 10), which were generated from cross sections 858

using PID data from 36 individual MCSs, indicate that the frequency of riming at a given 859

location within the midlevel inflow region during DYNAMO/AMIE was only 0.2. While 860

the PID may miss regions of weak riming, the vertical motions and turbulence were 861

likely weak at this late stage of the storm’s lifetime and would therefore not be expected 862

to support frequent riming. Thus, the microphysical parameterizations appear to be 863

creating too much riming near the 0°C level. 864

Nearly 70% of the cores in each scheme on 23 December had riming at an altitude 865

of 8 km (top row Figure 8). Figure 3d indicates that riming was detected only in very 866

shallow pockets within the midlevel inflow during the 23 December squall line. These 867

shallow pockets of riming are not a feature unique to the 23 December squall line. Rowe 868

and Houze [2014] and Barnes and Houze [2014] found that all instances of graupel/rimed 869

aggregates (i.e. riming) within the stratiform portion of MCSs were concentrated into a 870

shallow region near the 0°C level. While the PID may be unable identify instances of 871

very weak riming, previous studies suggest that riming should not extend over deep 872

layers within the midlevel inflow region of an oceanic MCS (e.g. Leary and Houze, 873

1979b). Additionally, riming is expected to be shallow in these storms since these MCSs 874

occurred over the near equatorial ocean where vertical motions are weaker [e.g. LeMone 875

and Zipser, 1980; Zipser and LeMone, 1980; Mohr and Zipser, 1996; Nesbitt et al., 876

2000]. Even in convective cores, DYNAMO/AMIE PID data indicated that rimed 877

particles only reached a few kilometers above the 0°C level [Rowe and Houze, 2014; 878

Barnes and Houze, 2014]. Thus, we infer that simulated riming in these three 879

parameterizations extends over too deep of a layer. This excessive riming may create too 880

much graupel and cause too much latent heat release, which may contribute to the 881

tendency for simulated oceanic convection in previous studies to be too strong and 882

contain too much graupel [e.g. Wiedner et al., 2004; McFarquhar et al., 2006; Blossey et 883

al., 2007; Lang et al., 2007; Li et al., 2008; Lang et al., 2011; Hagos et al., 2014; Varble 884

et al., 2014]. 885

5.4 Melting 886

The frequency of simulated melting on 23 December and 16 October is shown in 887

Figure 9. In each set of simulations, these parameterizations indicated that melting 888

occurred near the 0°C (bottom orange line in Figure 9). The depth of melting was 889

smallest in MOR with melting frequently isolated to one model level. Melting occurred 890

over a somewhat deeper layer in WD. The spatial pattern of melting in MY was distinctly 891

different. Melting in MY almost always reached the surface on 23 December and often 892

extended down to an altitude of 2 km on 16 October. Again, we see that MY allowed ice 893

well below the melting layer, which is a behavior never actually observed in the oceanic 894

tropics. However, the near-surface melting in MY occurred at a very low rates (not 895

shown), again indicating that it is a model artifact that does not affect the basic 896

conclusions of this paper and likely does not significantly impact the structure of the 897

simulated MCS. 898

Excluding MY, the simulated spatial pattern of melting was the most consistent 899

with the PID composites and previous studies of any process analyzed in this study. PID 900

data close to the S-PolKa radar, where the impact of beam broadening was minimal, 901

indicated that melting occurred in a very narrow band. Previous studies have suggested 902

that melting in the stratiform region occurs within 1 km of the 0°C level [e.g., Leary and 903

Houze, 1979b; Houze, 1981; Tao and Simpson, 1989; Caniaux et al., 1994; Donner et al., 904

2001]. Thus, while MOR and WD were both consistent with the PID and previous 905

studies, the shallower layer of melting in MOR was slightly more accurate than the 906

somewhat deeper layer in the WD simulation. 907

6. Impact of Microphysical Differences 908

From the foregoing discussions, it is evident that no single scheme accurately 909

portrays the precise spatial distribution of microphysical processes in either of the two 910

MCSs considered in this study. All the schemes reproduce the right basic layering of 911

microphysical processes, but disparities in the detailed spatial patterns are prevalent. Do 912

these spatial dissimilarities imply substantial differences in the reflectivity (i.e. 913

precipitation structure) between the parameterizations? 914

The right three columns of Figure 10 shows the composite reflectivity cross 915

section for 23 December and 16 October. For comparison, a composite S-PolKa 916

reflectivity cross section is shown in the left column. Figure 11 shows the corresponding 917

horizontal maps of reflectivity at an altitude of 5 km. While each parameterization 918

correctly simulated a squall line with a convective and stratiform end on 23 December 919

and an expansive region of weak precipitation on 16 October, each parameterization 920

exhibited different patterns of reflectivity and none matched S-PolKa observations. 921

MY and WD were similar on 23 December with both having an anvil in Figures 922

10b and 10d. However, S-PolKa data does not indicate an anvil (Figure 10a). The 923

reflectivity near the convective end of the midlevel inflow was stronger in MY (Figure 924

10b) and the brightband was more distinct in WD (Figure 10d). Figures 11b and 11d 925

shows that WD produced a better-defined leading convective line than MY. However, 926

these figures also demonstrate that both simulations produced too little stratiform 927

precipitation, a trend that has been consistently reported since Fovell and Ogura [1988]. 928

MOR lacked an anvil and had the most distinct convective, transition, and stratiform 929

regions, making it the most consistent with S-PolKa observations (Figure 10c and 10a). 930

Additionally, the horizontal distribution of reflectivity in MOR was most similar to S-931

PolKa on 23 December (Figure 11c and 11a). 932

The parameterizations had even more difficulty representing the reflectivity 933

structure on 16 October. As in the 23 December simulations, the MOR simulation on 16 934

October was most similar to S-PolKa observations, with a well-defined bright band 935

(Figure 10g and 10e) and isolated convective pockets whose locations were similar to 936

observations (Figure 11g and 11e). However, reflectivity in MOR on 16 October was too 937

high, especially in the anvil end of the composites (right side of Figure 10g) and near the 938

surface in the convective end (left side of Figure 10g). Additionally, the horizontal map 939

of reflectivity in Figure 11g indicates reflectivity in the convective pockets on 16 October 940

tended to be too high in MOR. MY lacked a brightband on 16 October and reflectivity in 941

the anvil end of the composites was often too high (Figure 10f). The horizontal map of 942

reflectivity at the 5 km level in Figure 11f indicates that reflectivity simulated by MY 943

was often too low, reflecting the fact that the brightband in MY was too weak. While the 944

horizontal distribution of reflectivity from WD on 16 October was similar to S-PolKa 945

(Figure 11h and 11e), its reflectivity cross section was drastically different than S-PolKa 946

and the other schemes. Figure 10h shows that WD had a strong brightband across the 947

entire midlevel inflow region and reflectivity was not observed below 3 km. Thus, WD 948

was extremely deficient in stratiform precipitation on 16 October. This result is consistent 949

with Varble et al. [2011] and Hagos et al. [2014], who found that microphysical 950

parameterizations in WRF tended to accurately simulate the areal coverage of stratiform 951

precipitation in tropical MCSs but underestimated their stratiform rainfall rates. The 952

increased discrepancy among the parameterizations and observations on 16 October 953

compared to 23 December is not surprising since the forcing on 16 October was weaker 954

and the MCS was characterized by more stratiform precipitation. Studies such as 955

Morrison et al. [2009] and Hagos et al. [2014] and others extending back to Fovell and 956

Ogura [1988], have consistently demonstrated that models have difficulty simulating the 957

stratiform precipitation of MCSs. 958

As a unit Figures 10 and 11 indicate that large reflectivity discrepancies were 959

witnessed among the parameterizations even though data assimilation forced each 960

ensemble to have similar mesoscale circulations and their spatial distributions of 961

microphysical processes were similar to first order. It should be noted that the spatial 962

pattern of the ice-phase processes were not the only differences among the 963

parameterizations. The schemes were characterized by different assumed drop size 964

distributions and fall speeds. Additionally, MY was the only scheme that simulated hail 965

in addition to low-density graupel. While these assumed parameter differences contribute 966

to variations in the simulated spatial distribution of precipitation in both the horizontal 967

and vertical dimensions, the details of the spatial distribution of the ice-phase 968

microphysical process must also factor into obtaining the correct precipitation 969

distribution. Detailed differences in both the liquid and ice microphysics between 970

parameterization schemes result in different interactions with the convective-scale 971

dynamics, which in turn affect the mesoscale evolution of the storms. Differences in 972

dynamical/microphysical evolution manifest themselves in the magnitude and location of 973

heating and cooling processes in the storms [Tao and Simpson, 1989]. Szeto et al. [1988] 974

and Tao et al. [1995] have suggested that squall line simulations are especially sensitive 975

to simulated melting patterns. Van Weverberg et al. [2014] systematically changed MOR 976

so that its parameterization of microphysical processes exactly matched MY. In the 977

process they showed that differences in the parameterization of microphysical processes 978

accounted for structural differences between the schemes. Thus, while the spatial pattern 979

of ice processes is not the only reason why convection in WRF varies among 980

parameterizations, it is likely an important factor. Simulations will likely become more 981

accurate as the spatial representation of these processes are improved [Grasso et al., 982

2014]. 983

7. Conclusions 984

Barnes and Houze [2014] used the NCAR particle identification (PID) algorithm 985

to demonstrate that ice hydrometeors in the midlevel inflow region of mesoscale 986

convective systems (MCSs) during DYNAMO/AMIE had a systematic spatial pattern 987

with sequential layers of horizontally oriented ice crystals/small ice crystals, dry 988

aggregates, and wet aggregates descending from echo top and shallow pockets of 989

graupel/rimed aggregates just above the melting level. This overall pattern was observed 990

in all types of MCSs. While PIDs have been traditionally interpreted as an indication of 991

the dominant hydrometeor, the frozen categories identified by the PID can be interpreted 992

as ice-phase microphysical processes. In particular, the categories identified as 993

horizontally oriented ice crystals and small ice crystals represents populations of particles 994

growing mostly by vapor deposition. Dry aggregates represent zones with particles that 995

are redistributing their mass through aggregation while still increasing in mass via vapor 996

deposition. The category referred to as graupel/rimed aggregates denotes regions where 997

the ice particles are predominantly increasing in mass as supercooled water freezes to 998

their surfaces in a process referred to as riming. These particles are also likely 999

simultaneously increasing in mass due to deposition. The category called wet aggregates 1000

indicates where ice particles are melting. The results of Barnes and Houze [2014] 1001

therefore indicate that these basic ice-phase microphysical process are distributed in the 1002

midlevel inflow region of tropical, oceanic MCSs in a systematic layered pattern. 1003

The current study investigates whether the spatial pattern of ice processes 1004

simulated by three routinely available microphysical parameterizations in the Weather 1005

Research and Forecasting (WRF) model are consistent with the patterns obtained from 1006

radar data in Barnes and Houze [2014], our theoretical understanding of microphysical 1007

processes, and each other. The microphysical parameterizations included in the current 1008

study were the Milbrandt-Yau Double-Moment scheme (MY), the Morrison 2-Moment 1009

scheme (MOR), and the WRF Double-Moment 6-Class scheme (WD). Radial velocity 1010

data were assimilated into WRF so that all simulations would develop a mesoscale 1011

circulation similar to observations and ensure that microphysical differences were not 1012

fundamentally caused by large-scale dynamical differences. Cross sections through each 1013

storm were selected based on the robustness of their midlevel inflow and were spatially 1014

composited so that slight variations among the individual cross sections did not smear the 1015

spatial microphysical pattern. Microphysical processes were obtained by summing the 1016

variables within each scheme that calculated the rate at which the mixing ratio of 1017

individual particles interact according to the process involved. Two sets of simulations 1018

were done to test if the simulated spatial pattern was dependent on the morphology of the 1019

MCS. 1020

The three microphysical parameterizations had a broad spatial ice microphysical 1021

pattern that was consistent with the PID, each other, and our theoretical understanding of 1022

ice microphysics. These consistencies included: 1023

The parameterizations had the same process at a given location within the midlevel 1024

inflow as the PID. All were characterized by deposition anywhere above the 0°C 1025

level, aggregation at and above 0°C level, melting at and below the 0°C level, and 1026

riming near the 0°C level. 1027

The spatial pattern of the ice microphysical processes was not fundamentally 1028

impacted by whether or not the MCS was organized into a squall line. 1029

Each parameterization had deposition located above the 0°C level wherever upward 1030

motion existed. The PID indicated ice particles (horizontally oriented ice crystals, 1031

small ice crystals, dry aggregates, or graupel/rimed aggregates) existed everywhere 1032

above the 0°C level. 1033

The small-scale spatial variability of simulated riming is likely caused by differences 1034

in the convective-scale vertical velocity motions in the simulations, which supports 1035

the theory put forth by Barnes and Houze [2014] that the riming during 1036

DYNAMO/AMIE occurred in pockets within the midlevel inflow region due to 1037

vertical motion from small-scale shear induced turbulence or embedded convective 1038

cells or left as residual from a more convective phase of the storm. 1039

Except for MY, melting was the microphysical process with the greatest 1040

consistency. Melting in the PID, MOR, and WD was concentrated in a narrow band at the 1041

0°C level, which agrees with previous observational and modeling studies [e.g., Leary 1042

and Houze, 1979b; Houze, 1981; Tao and Simpson, 1989; Caniaux et al., 1994; Donner 1043

et al., 2001]. Details within these general microphysical patterns varied considerably 1044

among the parameterizations, differing from each other, the PID, and previous studies. 1045

The main differences were: 1046

None of the schemes produced similar details in every microphysical pattern 1047

considered in this study. While the spatial pattern of deposition and aggregation were 1048

most similar in MY and WD, the spatial pattern of riming and melting were most 1049

similar in MOR and WD. 1050

Aggregation simulated by each parameterization existed everywhere above the 0°C 1051

level, regardless of the life stage of the MCS. PID data indicated that aggregation 1052

never occurred near echo top and extended only a short distance above the 0°C level, 1053

especially in MCSs at a later stage of development. Simulated aggregation was 1054

notably inconsistent with previous observational and laboratory studies, which show 1055

that aggregation is nearly impossible below -20°C [e.g. Hobbs et al., 1974]. 1056

Aggregation simulated at these very low temperatures could create biases in the 1057

radiative flux and be inimical to the development of stratiform precipitation [e.g. 1058

Varble et al., 2011; Van Weverberg et al., 2013; Hagos et al., 2014; Roh and Satoh, 1059

2014]. 1060

The PID showed riming in the midlevel inflow region was relatively rare, especially 1061

during the later developmental stages of MCSs, and only occurred in shallow pockets 1062

just above the 0°C level. In general, riming was very common in all simulations, 1063

especially near the 0°C level, and sometimes reached altitudes as high as 10 km. Too 1064

much riming can be expected to have adverse dynamical feedbacks since too much 1065

latent heat is released. However, it is important to note that simulated riming is 1066

defined in this study to include processes that create water-coated frozen 1067

hydrometeors, which may partially account for the high riming frequency near the 1068

0°C level in these simulations. 1069

Riming and melting in MY were sometimes located well below the 0°C level, which 1070

is incorrect; frozen hydrometeors never occur near the surface over equatorial oceans. 1071

However, the rate at which these near surface processes occurred in the model was 1072

relatively slight and likely do not significantly impact the structure of the MCS. These 1073

erroneous near surface rates are probably an easily correctable problem. 1074

Houze [1989] presented a conceptual model that suggested that ice microphysical 1075

processes above the 0°C level within the stratiform portion of an MCS have a layered 1076

pattern with deposition aloft and aggregation combined with deposition and riming 1077

between 0 and –12°C. This pattern is generally similar to the PID composites in Figure 3 1078

and Barnes and Houze [2014]. Leary and Houze [1979b] concluded that the riming was 1079

occurring in a rather shallow layer just above the melting layer. The WRF results 1080

presented in this study generally support the layered pattern presented in the PID and 1081

these previous studies. 1082

While rimed particles have been identified within stratiform precipitation in 1083

convective systems in a variety of locations around the globe, the source of these 1084

particles has been debated [Takahashi and Kuhara, 1992; Zrnić et al., 1993; Takahashi et 1085

al., 1995; Hogan et al., 2002; Suzuki et al., 2006; Evaristo et al., 2010; Bouniol et al., 1086

2010, Jung et al., 2012; Rowe and Houze, 2014; Barnes and Houze, 2014]. While the 1087

detailed riming pattern varied among each parameterization used in this study, there was 1088

a key similarity among the parameterizations. Riming often occurred outside of the 1089

region of large-scale upward motion associated with the midlevel inflow. While the 1090

assimilation of radial velocity had produced a similar midlevel inflow in each simulation, 1091

convective-scale vertical velocity differences existed among the simulations. Thus, the 1092

spatial riming patterns presented in this study provides strong evidence that riming within 1093

the midlevel inflow region of these tropical oceanic MCSs was associated with 1094

convective-scale velocity perturbations. Whether the vertical velocity perturbations are 1095

shear or buoyancy induced remains to be determined. 1096

This study has shown that even while data assimilation produces simulations that 1097

evolve mesoscale circulations similar to those observed, changing the microphysical 1098

parameterization creates substantial differences in the details of the horizontal and 1099

vertical reflectivity patterns. Given that microphysical processes influence the structure 1100

of latent and radiative heating, improving microphysical parameterizations will improve 1101

the accuracy of convection at the convective- and global- scale. Until the representation 1102

of microphysical processes is improved and made more consistent, simulations will 1103

continue to struggle to accurately represent convection and results will depend on the 1104

parameterization. These new results add to previous findings that demonstrate that the 1105

simulated structure of MCSs is sensitive to ice processes [Chen and Cotton, 1988; Szeto 1106

et al., 1988; Tao et al., 1991; Tao et al., 1995] and motivate continued ongoing efforts to 1107

improve microphysical schemes in cloud resolving models. 1108

Acknowledgements 1109

This research was supported by National Science Foundation grants AGS-1059611 and 1110

AGS-1355567. S-PolKa radar data used in this study is maintained by the NCAR Earth 1111

Observing Laboratory and can be freely obtained from 1112

http://data.eol.ucar.edu/codiac/dss/id=347.017. The authors would like to thank Professor Fuqing 1113

Zhang, Christopher Melhauser, and Yue (Michael) Ying for providing access, training, and 1114

support for the Pennsylvania State University Ensemble Kalman Filter version of WRF, NCAR 1115

EOL and Scott Ellis for assistance preparing the S-PolKa radar data for assimilation, David 1116

Warren and Stacy Brodzik for providing technological support, Beth Tully for improving the 1117

graphics, and two anonymous reviewers who provided helpful comments that improved this 1118

study. 1119

References 1120

Adams-Selin, R. D., S. C. van den Heever, and R. H. Johnson (2013), Impact of graupel 1121

parameterization schemes on idealized bow echo simulations, Mon. Wea. Rev., 141(4), 1241-1122

1262. 1123

Andrić, J., M. R. Kumjian, D. S. Zrnić, J. M. Straka, and V. M. Melnikov (2013), Polarimetric 1124

signatures above the melting layer in winter storms: An observational and model study, J. Appl. 1125

Meteor. Climatol., 52(3), 682-700. 1126

Aydin, K., and T. A. Seliga (1984), Radar polarimetric backscattering properties of conical 1127

graupel, J. Atmos. Sci., 41(11), 1887-1892. 1128

Bader, M. J., S. A. Clough, and G. P. Cox (1987), Aircraft and dual-polarization radar 1129

observations of hydrometeors in light stratiform precipitation, Quart. J. Roy. Meteor. Soc., 1130

113(476), 491-515. 1131

Bailey, M. P., and J. Hallett (2009), A comprehensive habit diagram for atmospheric ice crystals: 1132

Confirmation from the laboratory, AIRS II, and other field studies, J. Atmos. Sci., 66(9), 2888-1133

2899. 1134

Barker, D. M., W. Huang, Y. R. Guo, A. J. Bourgeois, and Q. N. Xiao (2004), A three-1135

dimensional variational data assimilation system for MM5: Implementation and initial results, 1136

Mon. Wea. Rev., 132(4), 897-914. 1137

Barnes, H. C. and R. A. Houze Jr. (2014), Precipitation hydrometeor type relative to the 1138

mesoscale airflow in mature oceanic deep convection of the Madden-Julian Oscillation, J. 1139

Geophys. Res. Atmos., 119(24), 13990-14014. 1140

Barnes, H. C., M. D. Zuluaga, and R. A. Houze Jr. (2015), Latent heating characteristics of the 1141

MJO computed from TRMM observations, J. Geophys. Res. Atmos., 120(4), 1322-1334. 1142

Baumgardner, D., J. L. Brenguier, A. Bucholtz, H. Coe, P. DeMott, T. J. Garrett, J. F. Gayet, M. 1143

Hermann, A. Heymsfield, and A. Korolev (2011), Airborne instruments to measure atmospheric 1144

aerosol particles, clouds and radiation: A cook’s tour of mature and emerging technology, Atmos. 1145

Res., 102(1-2), 10-29. 1146

Bechini, R., L. Baldini, and V. Chandrasekar (2013), Polarimetric radar observations in the ice 1147

region of precipitating clouds at c-band and x-band radar frequencies, J. Appl. Meteor. Climatol., 1148

52(5), 1147-1169. 1149

Blossey, P. N., C. S. Bretherton, J. Cetrone, and M. Kharoutdinov (2007), Cloud-resolving 1150

model simulations of KWAJEX: Model sensitivities and comparisons with satellite and radar 1151

observations, J. Atmos. Sci., 64(5), 1488-1508. 1152

Bouniol, D., J. Dalanoe, C. Duroure, A. Protat, V. Giraud, and G. Penide (2010), Microphysical 1153

characterization of West African MCS anvils, Quart. J. Roy. Meteor. Soc., 136, 323-344. 1154

Brandes, E. A., and K. Ikeda (2004), Freezing-level estimation with polarimetric radar, J. Appl. 1155

Meteor., 43(11), 1541-1553. 1156

Braun, S. A., and R. A. Houze Jr. (1994), The transition zone and secondary maximum of radar 1157

reflectivity behind a midlatitude squall line – results retrieved from Doppler radar data, J. Atmos. 1158

Sci., 51(19), 2733-2755. 1159

Bretherton, C. S. and S. Park (2009), A new moist turbulence parameterization in the community 1160

atmosphere model, J. Climate, 22(12), 3422-3448. 1161

Bringi, V. N. and V. Chandrasekar (2001), Polarimetric Doppler Weather Radar Principles and 1162

Applications, 636 pp., Cambridge Univ. Press, New York. 1163

Brown, B. R., M. M. Bell, and A. J. Frambach (2016), Validation of simulated hurricane drop 1164

size distributions using polarimetric radar, Geophys. Res. Let., 43(2), 910-917. 1165

Caniaux, G., J. L. Redelsperger, and J. P. Lafore (1994), A numerical study of the stratiform 1166

region of a fast-moving squall line. Part 1: General description and water and heat budgets, J. 1167

Atmos. Sci., 51(14), 2046-2074. 1168

Chen, S. and W. R. Cotton (1988), The sensitivity of a simulated extratropical mesoscale 1169

convective system to longwave radiation and ice-phase microphysics, J. Atmos. Sci., 45(24), 1170

3897-3910. 1171

Churchill, D. D., and R. A. Houze Jr. (1991), Effects of radiation and turbulence on the diabatic 1172

heating and water-budget of the stratiform region of a tropical cloud cluster, J. Atmos. Sci., 1173

48(7), 903-922. 1174

Comstock, J. M., R. d’Entremont, D. DeSlover, G. G. Mace, S. Y. Matrosov, S. A. McFarlane, P. 1175

Minnis, D. Mitchell, K. Sassen, M. D. Shupe, D. D. Turner, and Z. Wang (2007), An 1176

intercomparison of microphysical retrieval algorithms for upper-tropospheric ice clouds, Bult. 1177

Amer. Meteor. Soc., 88(2), 191-204. 1178

Donner, L. J., C. J. Seman, R. S. Hemler, and S. M. Fan (2001), A cumulus parameterization 1179

including mass fluxes, convective vertical velocities, and mesoscale effects: Thermodynamics 1180

and hydrological aspects in a general circulation model, J. Climate, 14(16), 3444-3463. 1181

Dudhia, J. (1989), Numerical study of convection observed during the winter monsoon 1182

experiment using a mesoscale two-dimensional model, J. Atmos. Sci., 46(20), 3077-3107. 1183

Evaristo, R., G. Scialom, N. Vitart, and Y. Lemaitre (2010), Polarimetric signatures and 1184

hydrometeor classification of West African squall lines, Quart. J. Roy. Meteor. Soc., 136, 272-1185

288, doi: 10.1002/qj.561. 1186

Fovell, R. G., and Y. Ogura (1988), Numerical simulation of a midlatitude squall line in two 1187

dimensions, J. Atmos. Sci., 45(24), 3846–3879. 1188

Gamache, J. F., and R. A. Houze Jr. (1983), Water-budget of a mesoscale convective system in 1189

the tropics, J. Atmos. Sci., 40(7), 1835-1850. 1190

Gaspari G. and S. E. Cohn (1999), Construction of correlation functions in two and three 1191

dimensions, Quart. J. Roy. Meteor. Soc., 125(554), 723-757. 1192

Grasso, L., D. T. Lindsey, K-S. S. Lim, A. Clark, D. Bikos, and S. R. Dembek (2014), 1193

Evaluation of and suggested improvements to the WSM6 microphysics in WRF-ARW using 1194

synthetic and observed GOES-16 imagery, Mon. Wea. Rev., 142(10), 3635-3650. 1195

Grazioli, J., D. Tuia, and A. Berne (2015), Hydrometeor classification from polarimetric radar 1196

measurements: A cluster approach, Atmos. Meas. Tech., 8(1), 149-170. 1197

Hagos, S. Z. Feng, C. D. Burleyson, K. S. S. Lim, C. N. Long, D. Wu, and G. Thompson (2014), 1198

Evaluation of convection-permitting model simulations of cloud populations associated with the 1199

Madden-Julian Oscillation using data collected during the AMIE/DYNAMO field campaign, J. 1200

Geophys. Res. Atmos., 119(21), 12052-12068. 1201

Hartmann, D. L., H. H. Hendon, and R. A. Houze, Jr. (1984), Some implications of the 1202

mesoscale circulation in tropical cloud clusters for large-scale dynamics and climate, J. Atmos. 1203

Sci., 41(1), 113-121. 1204

Hendry, A. and Y. M. M. Antar (1984), Precipitation particle identification with centimeter 1205

wavelength dual-polarization radars, Radio Sci., 19(1), 115-122. 1206

Hobbs, P. V., S. Chang., and J. D. Locatelli (1974), The dimensions and aggregation of ice 1207

crystals in natural cloud, J. Geophys. Res. Oceans Atmos., 79, 2199-2206. 1208

Hobbs, P. V. (1974), Ice Physics, 837 pp., Oxford Press, Bristol. 1209

Hogan, R. J., P. R. Field, and A. J. Illingworth (2002), Properties of embedded convection in 1210

warm-frontal mixed-phase cloud from aircraft and polarimetric radar, Quart. J. Roy. Meteor. 1211

Soc., 128(580), 451-476. 1212

Houze, R. A. Jr. (1981), Structure of atmospheric precipitation systems – A global survey, Radio 1213

Sci., 16(5), 671-689. 1214

Houze, R. A. Jr. (1989), Observed structure of mesoscale convective systems and implications 1215

for large-scale heating, Quart. J. Roy. Meteor. Soc., 115(487), 425-461. 1216

Houze, R. A. Jr., and D. D. Churchill (1987), Mesoscale organization and cloud microphysics in 1217

a Bay of Bengal depression, J. Atmos. Sci., 44(14), 1845-1867. 1218

Houze, R. A. Jr (2004), Mesoscale convective systems, Rev. Geophys., 42(4), 1219

doi:10.029/2004RG0000150, 43pp. 1220

Houze, R. A. Jr. (2014), Cloud Dynamics, 2nd Ed., 432 pp., Elsevier/Academic Press, Oxford. 1221

Jung, S-A., D-I., Lee, B. J-D. Jou, and H. Uyeda (2012), Microphysical properties of maritime 1222

squall line observed on Jun 2, 2008 in Taiwan, J. Meteor. Soc. Japan, 90(5), 833-850. 1223

Kingsmill, D. E., and R. A. Houze Jr. (1999a), Kinematic characteristics of air flowing into and 1224

out of precipitating convection over the west pacific warm pool: An airborne Doppler radar 1225

survey, Quart. J. Roy. Meteor. Soc., 125(556), 1165-1207. 1226

Korolev, A. V., E. F. Emery, J. W. Strapp, S. G. Cober, G. A. Isaac, M. Wasey, and D. Marcotte 1227

(2011) Small ice particles in tropospheric clouds: Fact or artifact? Airborne icing instrumentation 1228

evaluation experiment, Bult. Amer. Meteor. Soc., 92(8), 967-973. 1229

Kouketsu, T., H. Uyeda, T. Ohigashi, M. Oue, H. Takeuchi, T. Shinoda, K. Tsuboki, M. Kubo, 1230

and K. Muramoto (2015), A hydrometeor classification method for X-band polarimetric radar: 1231

Construction and validation focusing on solid hydrometeors under moist environments, J. Atmos. 1232

Oceanic Technol., 32(11), 2052-2074. 1233

Lang, A., W. K. Tao, R. Cifelli, W. Olsen, J. Halverson, S. Rutledge, and J. Simpson (2007), 1234

Improving simulations of convective systems from TRMM LBA: Easterly and westerly regimes, 1235

J. Atmos. Sci., 64(4), 1141-1164. 1236

Lang, S. E., W. K. Tao, X. P. Zeng, and Y. P. Li (2011), Reducing the biases in simulated radar 1237

reflectivities from a bulk microphysical scheme: Tropical convective systems, J. Atmos. Sci., 1238

68(10), 2306-2320. 1239

Leary, C. A., and R. A. Houze Jr. (1979b), Melting and evaporation of hydrometeors in 1240

precipitation from the anvil clouds of deep tropical convection, J. Atmos. Sci., 36(4), 669-679. 1241

Li, Y., E. J. Zipser, S. K. Kreuger, and M. A. Zulauf (2008), Cloud-resolving modeling of deep 1242

convection during KWAJEX. Part 1: Comparison to TRMM satellite and ground-based radar 1243

observations, Mon. Wea. Rev., 136(7), 2699-2712. 1244

Lim, K-S. S. and S-Y. Hong (2010), Development of an effective double-moment cloud 1245

microphysics scheme with prognostic cloud condensation nuclei (CNN) for weather and climate 1246

models. Mon. Wea. Rev., 138(5), 1587-1612. 1247

Mason, B. J. (1971), The Physics of Clouds, 2nd ed., 671pp., Clarendon Press, Oxford. 1248

Marquis, J., Y. Richardson, P. Markowski, D. Dowell, J. Wurman, K. Kosiba, P. Robinson, and 1249

G. Romine (2014), An investigation of the Goshen County, Wyoming, tornadic supercell of 5 1250

June 2009 using EnKF assimilation of mobile mesonet and radar observations collected during 1251

VORTEX2. Part 1: Experiment design and verification of the EnKF analyses, Mon. Wea. Rev., 1252

142(2), 530-554. 1253

Martini, A., N. Viltard, S. M. Ellis, and E. Fontaine (2015), Ice microphysics retrieval in the 1254

convective systems of the Indian Ocean during the CINDY-DYNAMO campaign, Atmos. Res., 1255

163, 13-23. 1256

McFarquhar, G. M., H. N. Zhang, G. Heymsfield, R. Hood, J. Dudhia, J. B. Halverson, and F. 1257

Marks (2006), Factors affecting the evolution of Hurricane Erin (2001) and the distributions of 1258

hydrometeors: Role of microphysical processes, J. Atmos. Sci., 63(1), 127-150. 1259

Meng, Z. Y., and F. Q. Zhang (2008a), Tests of an ensemble Kalman filter for mesoscale and 1260

regional-scale data assimilation. Part III: Comparison with 3DVAR in a real-data case study, 1261

Mon. Wea. Rev., 136(2), 522-540. 1262

Meng, Z. Y., and F. Q. Zhang (2008b), Tests of an ensemble Kalman filter for mesoscale and 1263

regional-scale data assimilation. Part IV: Comparison with 3DVAR in a month-long experiment, 1264

Mon. Wea. Rev., 136(10), 3671-3682. 1265

Milbrandt, J. A., and M. K. Yau (2005a), A multimoment bulk microphysics parameterization. 1266

Part I: Analysis of the role of the spectral shape parameter, J. Atmos. Sci., 62(9), 3051-3064. 1267

Milbrandt, J. A., and M. K. Yau (2005b), A multimoment bulk microphysics parameterization. 1268

Part II: A proposed three-moment closure and scheme description, J. Atmos. Sci., 62(9), 3065-1269

3081. 1270

Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, S. A. Clough (1997), Radiative transfer 1271

for inhomogenous atmospheres: RRTM, A validated correlated-k model for the longwave, J. 1272

Geophys. Res. Atmos., 120(D14), 16663-16682. 1273

Mohr, K. I., and E. J. Zipser (1996), Mesoscale convective systems defined by the 85-GHz ice 1274

scattering signature: Size and intensity comparison over tropical oceans and continents, Mon. 1275

Wea. Rev., 124(11), 2417-2437. 1276

Morrison, H., G. Thompson, and V. Tatarskii (2009), Impact of Cloud microphysics on the 1277

development of trailing stratiform precipitation in a simulated squall line: Comparison of one- 1278

and two-moment schemes, Mon. Wea. Rev., 137(3), 991-1007. 1279

Nesbitt, S. W., E. J. Zipser, and D. J. Cecil (2000), A census of precipitation features in the 1280

tropics using TRMM: Radar, ice scattering, and lightning observations, J. Climate, 13(23), 4087-1281

4106. 1282

Park, H., A. V. Ryzhkov, D. S. Zrnić, and K. E. Kim (2009), The hydrometeor classification 1283

algorithm for the polarimetric WSR-88D: Description and application to an MCS, Wea. 1284

Forecasting, 24(1), 87-103. 1285

Powell, S. W. and R. A. Houze Jr. (2013), The cloud population and onset of the Madden-Julian 1286

Oscillation over the Indian Ocean during DYNAMO-AMIE, J. Geophys. Res. Atmos., 118(21), 1287

11979-11995. 1288

Putnam, B. J., M. Xue, Y. S. Jung, N. Snook, and G. F. Zhang (2014), The analysis and 1289

prediction of microphysical states and polarimetric radar variables in a mesoscale convective 1290

system using double-moment microphysics, multinetwork radar data, and the ensemble Kalman 1291

filter, Mon. Wea. Rev., 142(1), 141-162. 1292

Roh, W. and M. Satoh (2014), Evaluation of precipitating hydrometeor parameterizations in a 1293

single-moment bulk microphysics scheme for deep convective systems over the Tropical Central 1294

Pacific, J. Atmos. Sci., 71(7), 2654-2673. 1295

Rowe, A. K., and R. A. Houze Jr. (2014), Microphysical characteristics of MJO convection over 1296

the Indian Ocean during DYNAMO, J. Geophys. Res. Atmos., 119(5), 2543-2554. 1297

Schumacher, C., R. A. Houze Jr., and I. Kraucunas (2004), The tropical dynamical response to 1298

latent heating estimates derived from the TRMM precipitation radar, J. Atmos. Sci., 61(12), 1299

1341-1358. 1300

Shipway. B. J., and A. A. Hill (2012), Diagnosis of systematic differences between multiple 1301

parametrizations of warm rain microphysics using a kinematic framework, Quart. J. Roy. 1302

Meteor. Soc., 138(669), 2196-2211. 1303

Skamarcok, W. C., J. B. Klemp, J. Dudhia, D. O. gill, D. M. Barker, M. G. Duda, A-Y. Huang, 1304

W. Wang, and J. G. Powers (2008), A description of advanced research WRF version 3, NCAR 1305

Tech. Note MCAR/TN-4751STR, 125 pp, [Available online at 1306

www2.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf] 1307

Straka, J. M., D. S. Zrnić, and A. V. Ryzhkov (2000), Bulk hydrometeor classification and 1308

quantification using polarimetric radar data: Synthesis of relations, J. Apl. Meteor., 39(8), 1341-1309

1372. 1310

Straka, J. M., D. S. Zrnić, and A. V. Ryzhkov (2000), Bulk hydrometeor classification and 1311

quantification using polarimetric radar data: Synthesis of relations, J. Appl. Meteor., 39(8), 1341-1312

1372. 1313

Suzuki, K., Shigenaga, Y., Kawano, T., and K. Yoneyama (2006), Videosonde observations 1314

during the R/V Mirai MR04-08 Cruise, J. Mar. Meteorol. (Umi to Sora), 82(2), 1-10. 1315

Szeto K.K., C. A. Lin, and R. E. Stewart (1988), Mesoscale circulations forced by melting snow. 1316

Part 1: Basic simulations and dynamics, J. Atoms Sci., 45(11), 1629-1641. 1317

Takahashi T., and K. Kuhara (1993), Precipitation mechanisms of cumulonimbus clouds at 1318

Pohnpei, Micronesia, J. Meteor. Soc. Japan, 71(1), 21-31. 1319

Takahashi, T., Suzuki, K., Orita, M., Tokuno, M., and R. Delamar (1995), Videosonde 1320

observations of precipitation processes in equatorial cloud clusters, J. Meteor. Soc. Japan, 1321

73(2B), 509-534. 1322

Tao W. K., and J. Simpson (1989), Modeling study of a tropical squall-type convective line, J. 1323

Atmos. Sci., 46(2), 177-202. 1324

Tao W. K., J. R. Scala, B. Ferrier, and J. Simpson (1995), The effect of melting processes on the 1325

development of a tropical ad midlatitude squall line, J. Atmos. Sci., 52(11), 1934-1948. 1326

Tao, W. K., J. Simpson, and S. T. Soong (1991), Numerical-simulation of a subtropical squall 1327

line over the Taiwan Strait, Mon. Wea. Rev., 119(11), 2699-2723. 1328

Tao, W. K., S. Lang, J. Simpson, C. H. Sui, B. Ferrier, and M. D. Chou (1996), Mechanisms of 1329

cloud-radiation interaction in the tropics and midlatitudes, J. Atmos. Sci., 53(18), 2624-2651. 1330

Tewari, M., F. Chen, W. Wang, J, Dudhai, M. A. LeMone, K. Mitchell, M. Ek, G. Gayno, J. 1331

Wegiel, and R. H. Cuenca (2004), Implementation and verification of the united NOAH land 1332

surface model in the WRF model, 20th conference on weather analysis and forecasting / 16th 1333

conference on numerical weather prediction. pp. 11-15. 1334

Thompson, E. J., S. A. Rutledge, B. Dolan, and V. Chandrasekhar (2014), A dual-polarization 1335

radar hydrometeor classification algorithm for winter precipitation, J. Atmos. Oceanic Technol., 1336

31(7), 1457-1481. 1337

Tollerud, E. I., and S. K. Esbensen (1985), A composite life-cycle of nonsquall mesoscale 1338

convective systems over the tropical ocean. Part 1: Kinematic fields, J. Atmos. Sci., 42(8), 823-1339

837. 1340

Van Weverberg, K., A. M. Vogelmann, W. Lin, E. P. Luke, A. Cialella, P Minnis, M. Khaiyer, 1341

E. R. Boer, and M. P. Jensen (2013), The role of cloud microphysics parametrization in the 1342

simulation of mesoscale convective system clouds and precipitation in the Tropical Western 1343

Pacific, J. Atmos. Sci., 70(4), 1104-1128. 1344

Van Weverberg, K., E. Goudenhoofdt, U. Blahak, E. Brisson, M. Demuzere, P. Marbaix, and J. 1345

P. can Ypersle (2014), Comparison of one-moment and two-moment bulk microphysics for high-1346

resolution climate simulations of intense precipitation, Atmos. Res., 147,145-161. 1347

Varble, A., E. J. Zipser, A. M. Fridlind, P. Zhu, A. S. Ackerman, J. P. Chaboureau, S. Collis, J. 1348

W. Fan, A. Hill, and B. Shipway (2014), Evaluation of cloud-resolving and limited area model 1349

intercomparison simulations using TWP-ICE observations: Part 1. Deep convective updraft 1350

properties, J. Geophys. Res. Atmos., 119(24), 13891-13918. 1351

Vitart, F., and F. Molteni (2010), Simulation of the Madden-Julian Oscillation and its 1352

teleconnections in the ECMWF forecast system, Q. J. R. Meteorol. Soc., 136(649), 842-855. 1353

Vivekanandan, J., D. S. Zrnić, S. M. Ellis, R. Oye, A. V. Ryzhkov, and J. Straka (1999), Cloud 1354

microphysics retrieval using S-band dual-polarization radar measurements, Bult. Amer. Meteor. 1355

Soc., 80(3), 381-388. 1356

Wallace, J. M., and P. V. Hobbs (2006), Atmospheric science: An introduction survey, Academic 1357

Press, 483 pp. 1358

Wang, Y., C. N. Long, L. R. Leung, J. Dudhia, S. A. McFarlane, J. H. Mather, S. J. Ghan, and X. 1359

Liu (2009), Evaluating regional cloud-permitting simulations of the WRF model for the Tropical 1360

Warm Pool International Cloud Experiment (TWP-ICE), Darwin, 2006, J. Geophys. Res., 114, 1361

D21203, doi:10.1029/2009JD012729. 1362

Webster, P. J., and G. L. Stephens (1980), Tropical upper-tropospheric extended clouds: 1363

Inferences from Winter MONEX, J. Atmos. Sci., 37(7), 1521-1541. 1364

Wheatley, D. M., N. Yussouf, and D. J. Stensrud (2014), Ensemble Kalman filter analyses and 1365

forecasts of a severe mesoscale convective system using different choice of microphysics 1366

schemes, Mon. We. Rev., 142(9), 3243-3263. 1367

Wiedner, M., C. Prigent, J. R. Pardo, O. Nuissier, J. P. Chaboureau, J. P. Pinty, and P. Mascart 1368

(2004), Modeling of passive microwave responses in convective situations using output from 1369

mesoscale models: Comparison with TRMM/TMI satellite observations, J. Geophys. Res. 1370

Atmos., 109(D6), 1-13. 1371

Yoneyama, K., Zhang, C. D., and C. N. Long (2013), Tracking pulses of the Madden-Julian 1372

Oscillation, Bull. Amer. Meteor. Soc., 94(12), 1871-1891. 1373

Zipser, E. J (2003), Some views on “Hot Towers” after 50 years of tropical field programs and 1374

two years of TRMM data, Meteor. Monogr., 29(51), 49-58. 1375

Zhang, F. Q., C. Synder, and J. Z. Sun (2004), Impacts of initial estimate and observation 1376

availability on convective-scale data assimilation with an ensemble Kalman filter, Mon. Wea. 1377

Rev., 132(5), 1238-1253. 1378

Zhang, F., Y. H. Weng, J. A. Sippel, Z. Y. Meng, and C. H. Bishop (2009) Cloud-resolving 1379

hurricane initialization and prediction through assimilation of Doppler radar observation with an 1380

ensemble Kalman filter, Mon. Wea. Rev., 137(7), 2105-2125. 1381

Zrnić, D. S., N. Balakrishnan, C. L. Ziegler, V. N. Bringi, and K. Aydin (1993), Polarimetric 1382

signatures in the stratiform region of a mesoscale convective system, J. Appl. Meteor., 32(4), 1383

678-693. 1384

Zuluaga, M. D., and R. A. Houze, Jr. (2013), Evolution of the population of precipitating 1385

convective systems over the equatorial Indian Ocean in active phases of the Madden-Julian 1386

Oscillation. J. Atmos. Sci., 70(9), 2713-2725. 1387

1388

Tables 1389 1390

Table 1: Approximate range of values used in the NCAR S-PolKa PID during DYNAMO/AMIE. 1391

1392

Hydrometeor

Name

Microphysical

Process

ZH

(dBZ)

ZDR

(dB)

LDR

(dB)

KDP (°

km-1)

ρHV T (°C)

Horizontally

Oriented Ice

Crystals

Deposition 0-15 1-6 -31 - -

23.4

0.6-0.8 0.97-

0.98

-50 - 1

Small Ice

Crystals

Deposition 0 - 15 0 - 0.7 -31 - -

23.4

0 – 0.1 0.97 –

0.98

-50 - 1

Dry Aggregates Aggregation

and

Deposition

15 - 33 0 - 1.1 -26 - -

17.2

0 –

0.168

0.97 - -

0.98

-50 - 1

Graupel/Rimed

Aggregates

Riming and

Deposition

30 - 50 -0.1 -

0.76

-25 - -

20.17

0.08 -

1.65

0.89 –

0.96

-50 - 7

Wet Aggregates Melting 7 - 45 0.5 - 3 -26 - -

17.2

0.1 - 1 0.75 –

0.98

-4 – 12

1393

1394

Table 2: The definition of the ice microphysical processes adopted in this study and the variables 1395

from each parameterization used to define each process. 1396

1397

Parameterizations Ice Microphysical Processes

Aggregation Riming Melting Deposition

Frozen

hydrometeors

collecting other

frozen

hydrometeors

Frozen

hydrometeors

collecting liquid

hydrometeors

Frozen

hydrometeors

melting into

liquid

hydrometeors

Frozen

hydrometeors

collecting water

vapor

Milbrandt – Yau

(MY)

QCLis, QCLig,

QCLsh, QCNis2,

QCLih

QCLcs, QCLcg,

QCLch, QCLrg,

QCLrs, QCLri,

QCLrh, QCNsg,

QCNgh

QMLir, QMLsr,

QMLgr, QMLhr

QVDvi, QVDvs,

QVDvh, QVDvg

Morrison (MOR) prai, prci psacws, pgracs,

psacwi, psacwg,

pgsacw, psacr,

pracg, pracis,

praci, piacrs

psmlt, pgmlt prd, prdg, prds

WDM6 (WD) Psaci, Pgaci,

Psaut, Pgacs,

Pgaut

Psacw, Pgacw,

Paacw, Piacr,

Psacr, Pgacr,

Pracs

Psmlt, Pgmlt Pidep, Psdep,

Pgdep

1398

1399

Table 3: WRF and EnKF setup for simulations. 1400

1401

Parameters December

Simulation

October

Simulation

References

WRF Setup

Simulation Time 23 December

2011; 1200-2000

UTC

16 October 2011;

0600 -1800 UTC

Time Step 18 seconds

Vertical Levels 39, top at 26 km

Domain Horizontal Resolution 3 km, 1 km

Nesting Two-way

Planetary Boundary Layer

Parameterization

University of Washington (TKE)

Boundary Layer Scheme

Bretherton and

Park [2009]

Longwave Radiation

Parameterization

Rapid Radiative Transfer Model

(RRTM)

Mlawer et al.

[1997]

Shortwave Radiation

Parameterization

Dudhia Shortwave Scheme Dudhia [1989]

Surface Layer

Parameterization

MM5 Similarity Scheme

Land Surface Parameterization United Noah Land Surface Model Tewari et al.

[2004]

Microphysics

Parameterizations

Milbrandt - Yau Double - Moment

Scheme

Milbrandt and Yau

[2005a; b]

Morrison 2 - Moment Scheme Morrison et al.

[2009]

WRF Double - Moment 6 - Class

Scheme

Lim and Hong

[2010]

EnKF Setup

First Assimilation Time 1800 UTC 1200 UTC

Assimilation Interval 15 Minutes

Assimilated Data S-PolKa Radial Velocity

Number of Members 50

Ensemble Initiation and

Boundary Conditions

ERA-interim perturbed using WRFDA

cv option 5

Barker et al.

[2005]

HROI 90 km, 30 km, Gaspari and Cohn 5th

Order

Gaspari and Cohn

[1999]

VROI 6 Model Levels

Relaxation Coefficient 0.8

Perturbed Variables perturbation potential temperature,

zonal and meridional wind, cloud water

mixing ratio, vapor mixing ratio, rain

mixing ratio, base-state and

perturbation geopotential, base-state

and perturbation dry air mass in the

column, surface pressure, base-state and

perturbation pressure, and zonal and

meridional wind at 10 km

1402

1403

Figure Captions 1404

Figure 1: Outer (blue square) and inner (red square) domains used in WRF simulations. The 1405

resolution of each domain is listed in parenthesis. The S-PolKa radar is located at the black dot 1406

and its domain is outlined by the black circle. 1407

1408

Figure 2: (a) Horizontal map of column maximum S-PolKa reflectivity at 1935 UTC on 23 1409

December 2011. (b) Vertical cross section of S-PolKa reflectivity taken along the red line in (a). 1410

(c) Vertical cross section of S-PolKa radial velocity taken along the red line in (a). (d - f) Same 1411

as (a – c ) except at 1450 UTC on 16 October 2011. 1412

1413

Figure 3: (a) Composite showing the location of horizontally oriented ice crystals in the midlevel 1414

inflow region of the 23 December squall line. Shading represents the frequency of horizontally 1415

oriented ice crystals at that location relative to the midlevel inflow. The position of the midlevel 1416

inflow is approximated by the dashed diagonal magenta line. The boundaries of the midlevel 1417

inflow are marked by black vertical lines. The height axis is normalized so that the melting level 1418

is located at 0. (b) Same as (a) except for small ice crystals. (c) Same as (a) except for dry 1419

aggregates. (d) Same as (a) except for graupel/rimed aggregates. (e) Same as (a) except for wet 1420

aggregates. (f-j) Same as (a-e) except for the 16 October non-squall MCS. 1421

1422

Figure 4: (a) Cross section of the average S-PolKa radial velocity at 1930 UTC on 23 December 1423

2011. (b) Cross section of the composite average horizontal wind speeds simulated by the 1424

Milbrandt-Yau (MY) scheme at 1930 UTC on 23 December 2011 in shading. The composite 1425

average reflectivity is shown in gold contours with reflectivity values contoured every 5 dBZ 1426

from 5 to 45 dBZ. The boundaries of the midlevel inflow core are marked by black vertical lines. 1427

The position of the midlevel inflow is approximated by the dashed diagonal magenta line. The 1428

magenta number along the right side indicates how many cores are included in these composites. 1429

(c) Same as (b) except for the Morrison (MOR) scheme. (d) Same as (b) except for the WDM6 1430

(WD) scheme. (e – h) Same as (a – d) except taken at 1445 UTC on 16 October 2011. 1431

1432

Figure 5: (a) Cross section of the composite average vertical velocity simulated by the 1433

Milbrandt-Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 1434

2011 in shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, 1435

middle, and top green lines, respectively. The composite average reflectivity is shown in gold 1436

contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The boundaries of the 1437

midlevel inflow core are marked by black vertical lines. The position of the midlevel inflow is 1438

approximated by the dashed diagonal magenta line. The magenta number along the right side 1439

indicates how many cores are included in these composites. (b) Same as (a) except for the 1440

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1441

– c) except at 1445 UTC on 16 October 2011. 1442

1443

Figure 6: (a) Cross section of the composite frequency of deposition simulated by the Milbrandt-1444

Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in 1445

shading. The composite average upward motion in m s-1 is shown in orange contours with 1446

upward motion contoured is every 0.1 m s-1 from 0.1 m s-1 to 0.5 m s-1. The composite average 1447

reflectivity is shown in gold contours with reflectivity values contoured every 5 dBZ from 5 to 1448

45 dBZ. The boundaries of the midlevel inflow core are marked by black vertical lines. The 1449

position of the midlevel inflow is approximated by the dashed diagonal magenta line. The 1450

magenta number along the right side indicates how many cores are included in these composites. 1451

(b) Same as (a) except for the Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 1452

(WD) scheme. (d – f) same as (a – c) except at 1445 UTC on 16 October 2011. 1453

1454

Figure 7: (a) Cross section of the composite frequency of aggregation simulated by the 1455

Milbrandt-Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 1456

2011 in shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, 1457

middle, and top orange lines, respectively. The composite average reflectivity is shown in gold 1458

contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The edges of the core 1459

of the midlevel inflow are marked by black vertical lines. The position of the midlevel inflow is 1460

approximated by the dashed diagonal magenta line. The magenta number along the right side 1461

indicates how many cores are included in these composites. (b) Same as (a) except for the 1462

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1463

– c) except at 1445 UTC on 16 October 2011. 1464

1465

Figure 8: (a) Cross section of the composite frequency of riming simulated by the Milbrandt-Yau 1466

(MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in shading. 1467

The composite average upward motion in m s-1 is shown in orange contours with upward motion 1468

contoured is every 0.1 m s-1 from 0.1 m s-1 to 0.5 m s-1. The composite average reflectivity is 1469

shown in gold contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The 1470

boundaries of the midlevel inflow core are marked by black vertical lines. The position of the 1471

midlevel inflow is approximated by the dashed diagonal magenta line. The magenta number 1472

along the right side indicates how many cores are included in these composites. (b) Same as (a) 1473

except for the Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d 1474

– f) same as (a – c) except at 1445 UTC on 16 October 2011. 1475

1476

Figure 9: (a) Cross section of the composite frequency of melting simulated by the Milbrandt-1477

Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in 1478

shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, middle, 1479

and top orange lines, respectively. The composite average reflectivity is shown in gold contours 1480

with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The edges of the core of the 1481

midlevel inflow are marked by black vertical lines. The position of the midlevel inflow is 1482

approximated by the dashed diagonal magenta line. The magenta number along the right side 1483

indicates how many cores are included in these composites. (b) Same as (a) except for the 1484

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1485

– c) except at 1445 UTC on 16 October 2011. 1486

1487

Figure 10: (a) Cross section of the average S-PolKa radar reflectivity at 1930 UTC on 23 1488

December 2011. (b) Cross section of the composite average horizontal radar reflectivity 1489

simulated by the Milbrandt-Yau (MY) scheme at 1930 UTC on 23 December 2011 in shading. 1490

The boundaries of the midlevel inflow core are marked by black vertical lines. The position of 1491

the midlevel inflow is approximated by the dashed diagonal magenta line. The magenta number 1492

along the right side indicates how many cores are included in these composites. (c) Same as (b) 1493

except for the Morrison (MOR) scheme. (d) Same as (b) except for the WDM6 (WD) scheme. (e 1494

– h) Same as (a – d) except taken at 1445 UTC on 16 October 2011. 1495

1496

Figure 11: (a) Horizontal map of S-PolKa reflectivity at an altitude of 5 km at 1930 UTC on 23 1497

December 2011. The red dashed lines indicate where the cross sections shown in Figure 3a and 1498

9a are taken. (b) Horizontal map of reflectivity simulated by the Milbrandt-Yau (MY) scheme at 1499

an altitude of 5 km at 1930 UTC on 23 December 2011. The dashed red lines show where the 1500

cross sections analyzed in this study were taken. (c) Same as (b) except for the Morrison (MOR) 1501

scheme. (d) Same as (b) except for the WDM6 (WD) scheme. (e – h) Same as (a – d) except at 1502

1445 UTC at 1445 UTC on 16 October 2011. 1503

1504

Figures 1505

1506

Figure 1: Outer (blue square) and inner (red square) domains used in WRF simulations. The 1507

resolution of each domain is listed in parenthesis. The S-PolKa radar is located at the black dot 1508

and its domain is outlined by the black circle. 1509

1510

1511

Figure 2: (a) Horizontal map of column maximum S-PolKa reflectivity at 1935 UTC on 23 1512

December 2011. (b) Vertical cross section of S-PolKa reflectivity taken along the red line in (a). 1513

(c) Vertical cross section of S-PolKa radial velocity taken along the red line in (a). (d) Vertical 1514

cross section of S-PolKa PID data taken along the red line in (a). (e – h) Same as (a – d) except 1515

at 1450 UTC on 16 October 2011. 1516

1517

1518

Figure 3: (a) Composite showing the location of horizontally oriented ice crystals in the midlevel 1519

inflow region of the 23 December squall line. Shading represents the frequency of horizontally 1520

oriented ice crystals at that location relative to the midlevel inflow. The position of the midlevel 1521

inflow is approximated by the dashed diagonal magenta line. The boundaries of the midlevel 1522

inflow are marked by black vertical lines. The height axis is normalized so that the melting level 1523

is located at 0. (b) Same as (a) except for small ice crystals. (c) Same as (a) except for dry 1524

aggregates. (d) Same as (a) except for graupel/rimed aggregates. (e) Same as (a) except for wet 1525

aggregates. (f-j) Same as (a-e) except for the 16 October non-squall MCS. 1526

1527

1528

Figure 4: (a) Cross section of the average S-PolKa radial velocity at 1930 UTC on 23 December 1529

2011. (b) Cross section of the composite average horizontal wind speeds simulated by the 1530

Milbrandt-Yau (MY) scheme at 1930 UTC on 23 December 2011 in shading. The composite 1531

average reflectivity is shown in gold contours with reflectivity values contoured every 5 dBZ 1532

from 5 to 45 dBZ. The boundaries of the midlevel inflow core are marked by black vertical lines. 1533

The position of the midlevel inflow is approximated by the dashed diagonal magenta line. The 1534

magenta number along the right side indicates how many cores are included in these composites. 1535

(c) Same as (b) except for the Morrison (MOR) scheme. (d) Same as (b) except for the WDM6 1536

(WD) scheme. (e – h) Same as (a – d) except taken at 1445 UTC on 16 October 2011. 1537

1538

1539

Figure 5: (a) Cross section of the composite average vertical velocity simulated by the 1540

Milbrandt-Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 1541

2011 in shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, 1542

middle, and top green lines, respectively. The composite average reflectivity is shown in gold 1543

contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The boundaries of the 1544

midlevel inflow core are marked by black vertical lines. The position of the midlevel inflow is 1545

approximated by the dashed diagonal magenta line. The magenta number along the right side 1546

indicates how many cores are included in these composites. (b) Same as (a) except for the 1547

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1548

– c) except at 1445 UTC on 16 October 2011. 1549

1550

1551

Figure 6: (a) Cross section of the composite frequency of deposition simulated by the Milbrandt-1552

Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in 1553

shading. The composite average upward motion in m s-1 is shown in orange contours with 1554

upward motion contoured is every 0.1 m s-1 from 0.1 m s-1 to 0.5 m s-1. The composite average 1555

reflectivity is shown in gold contours with reflectivity values contoured every 5 dBZ from 5 to 1556

45 dBZ. The boundaries of the midlevel inflow core are marked by black vertical lines. The 1557

position of the midlevel inflow is approximated by the dashed diagonal magenta line. The 1558

magenta number along the right side indicates how many cores are included in these composites. 1559

(b) Same as (a) except for the Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 1560

(WD) scheme. (d – f) same as (a – c) except at 1445 UTC on 16 October 2011. 1561

1562

1563

Figure 7: (a) Cross section of the composite frequency of aggregation simulated by the 1564

Milbrandt-Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 1565

2011 in shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, 1566

middle, and top orange lines, respectively. The composite average reflectivity is shown in gold 1567

contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The edges of the core 1568

of the midlevel inflow are marked by black vertical lines. The position of the midlevel inflow is 1569

approximated by the dashed diagonal magenta line. The magenta number along the right side 1570

indicates how many cores are included in these composites. (b) Same as (a) except for the 1571

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1572

– c) except at 1445 UTC on 16 October 2011. 1573

1574

1575

Figure 8: (a) Cross section of the composite frequency of riming simulated by the Milbrandt-Yau 1576

(MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in shading. 1577

The composite average upward motion in m s-1 is shown in orange contours with upward motion 1578

contoured is every 0.1 m s-1 from 0.1 m s-1 to 0.5 m s-1. The composite average reflectivity is 1579

shown in gold contours with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The 1580

boundaries of the midlevel inflow core are marked by black vertical lines. The position of the 1581

midlevel inflow is approximated by the dashed diagonal magenta line. The magenta number 1582

along the right side indicates how many cores are included in these composites. (b) Same as (a) 1583

except for the Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d 1584

– f) same as (a – c) except at 1445 UTC on 16 October 2011. 1585

1586

1587

Figure 9: (a) Cross section of the composite frequency of melting simulated by the Milbrandt-1588

Yau (MY) scheme within the midlevel inflow core at 1930 UTC on 23 December 2011 in 1589

shading. The composite average 0°C, -20°C, and -40°C contours are shown in bottom, middle, 1590

and top orange lines, respectively. The composite average reflectivity is shown in gold contours 1591

with reflectivity values contoured every 5 dBZ from 5 to 45 dBZ. The edges of the core of the 1592

midlevel inflow are marked by black vertical lines. The position of the midlevel inflow is 1593

approximated by the dashed diagonal magenta line. The magenta number along the right side 1594

indicates how many cores are included in these composites. (b) Same as (a) except for the 1595

Morrison (MOR) scheme. (c) Same as (a) except for the WDM6 (WD) scheme. (d – f) same as (a 1596

– c) except at 1445 UTC on 16 October 2011. 1597

1598

1599

Figure 10: (a) Cross section of the average S-PolKa radar reflectivity at 1930 UTC on 23 1600

December 2011. (b) Cross section of the composite average horizontal radar reflectivity 1601

simulated by the Milbrandt-Yau (MY) scheme at 1930 UTC on 23 December 2011 in shading. 1602

The boundaries of the midlevel inflow core are marked by black vertical lines. The position of 1603

the midlevel inflow is approximated by the dashed diagonal magenta line. The magenta number 1604

along the right side indicates how many cores are included in these composites. (c) Same as (b) 1605

except for the Morrison (MOR) scheme. (d) Same as (b) except for the WDM6 (WD) scheme. (e 1606

– h) Same as (a – d) except taken at 1445 UTC on 16 October 2011. 1607

1608

1609

Figure 11: (a) Horizontal map of S-PolKa reflectivity at an altitude of 5 km at 1930 UTC on 23 1610

December 2011. The red dashed lines indicate where the cross sections shown in Figure 3a and 1611

9a are taken. (b) Horizontal map of reflectivity simulated by the Milbrandt-Yau (MY) scheme at 1612

an altitude of 5 km at 1930 UTC on 23 December 2011. The dashed red lines show where the 1613

cross sections analyzed in this study were taken. (c) Same as (b) except for the Morrison (MOR) 1614

scheme. (d) Same as (b) except for the WDM6 (WD) scheme. (e – h) Same as (a – d) except at 1615

1445 UTC at 1445 UTC on 16 October 2011. 1616

1617