1
2
3
The Treatment of Convection in the Next
Generation Global Models: Challenges
and Opportunities
4
Samson Hagos*, Robert Houze*,# , Zhe Feng*, and Angela Rowe#
5
*
6
Pacific Northwest National Laboratory
#
7
University of Washington
8
9
10
Corresponding Author Address
11
Samson Hagos
12
Pacific Northwest National Laboratory
13
902 Battelle Boulevard
14
Richland WA 99352
15
Email: [email protected]
1
16
Abstract
17
Cloud-permitting global modeling is becoming a new reality, and rethinking of the
18
objectives, assumptions, and methods of convection parameterizations is already occurring.
19
Models must now represent sub-grid and resolved processes as part of the same continuum.
20
Scale separation, commonly used in the past, must be replaced by the stringent requirement of
21
"scale awareness." To this end, advances are needed in understanding of transitions in cloud
22
populations including boundary-layer evolution, deepening of initially shallow clouds,
23
formation of precipitation and cold pools, and modes of growth and aggregation of clouds to
24
form mesoscale units of convection, which have dynamical features larger than individual
25
clouds.
26
Such advances require appropriate observational data collection, processing, and
27
packaging strategies that include the development of merged datasets on convective and
28
microphysical processes along with the environmental context. In particular, concurrent ice-
29
phase microphysical processes and corresponding updraft and downdraft statistics are needed
30
but presently missing. These observational gaps need to be filled by increasingly advanced
31
remote-sensing techniques and research aircraft capable of penetrating intense convection at a
32
wide range of altitudes. This new information, provided on the scales most relevant to
33
parameterization, can be related to model output via advanced radar and satellite simulators that
34
convert model output to observable variables. Because no single observational method is self-
35
sufficient, field experiments that integrate as many instrument platforms as possible, including
36
emerging technologies, will remain a necessary avenue for model validation and hypothesis
37
testing. This holistic approach will accelerate parameterization development by allowing
2
38
validation of a hierarchy of modeling frameworks using the multi-variable data collected in
39
similar environmental conditions.
3
40
1.
Introduction
41
For decades, the development of convective cloud parameterizations in global models
42
has implicitly or explicitly relied on a perceived spatio-temporal scale separation, in which the
43
resolved large-scale environment is in a statistical equilibrium with the unresolved cloud
44
processes. Such an approach, however, neglects the continuum of scales of motion shown by
45
field programs and satellite remote sensing to be present in convection. As shown in the satellite
46
image in Figure 1, convective cloud populations are a mix of cumulus, cumulonimbus, and
47
mesoscale convective systems. It is well known that the larger forms of these convective clouds
48
(e.g., mesoscale convective systems, MCSs) can evolve upscale from the smaller elements.
49
Furthermore, understanding and accurately representing the processes involved in the
50
transitions from shallow non-precipitating clouds to precipitating shallow clouds to deep
51
convection to MCSs and planetary scale phenomena, such as the Madden-Julian Oscillation
52
(MJO), is important for accurate representation of the mean state of the climate, its natural and
53
forced variability, and for the prediction of drought and extreme precipitation events.
54
The need to understand how convective processes interact across a continuum of scales
55
and with the atmospheric circulation and climate intersects with the current state of global
56
modeling. Rapid expansion of computational resources is pushing global model resolution into
57
a "gray zone," where some aspects of convection are partially resolved and the traditional scale-
58
separation argument no longer applies. The next generation of global models, here defined as
59
those with grid spacing of 1–10 km, urgently requires novel strategies of combining
60
parameterization and explicit representation of convection processes that represent convective
61
cloud populations and variability consistent with observations.
4
62
The overview presented here is motivated by discussions at a U.S. Department of
63
Energy -supported workshop aimed at devising strategies for addressing these issues. The
64
complete workshop report from which much of the material discussed in this article is derived
65
will be available at http://asr.science.energy.gov/publications/program-docs. The premise of the
66
workshop was that a complete strategy for improving the treatment of convection in the next-
67
generation global models could be organized into three intersecting core themes:
68
•
Basic understanding of cloud processes
69
•
Parameterization
70
•
Collection, processing, and analysis of observational data.
71
Equally important is that these three topics need to be approached in a coordinated way,
72
with particular attention to the integration of each of the three core themes with the others, as
73
depicted in Figure 2. With this framework in mind, this overview attempts to address the
74
following questions:
75
•
What are the current challenges in each of the three core themes and their integration?
76
•
What can be done in the short term (~3 years) using existing resources, and what new
77
capabilities and/or long-term (~10 years) investments are required to address these
78
challenges?
79
This article is intended to serve as a road map for integrated research and development
80
activities aimed at accurate treatment of convection in the next-generation (1–10 km grid
81
spacing, non-hydrostatic dynamical core) global models. In the remainder of the article,
82
convection-related issues in current global models are briefly reviewed, the challenges in the
83
three core themes and their integration are identified, and strategies for meeting these challenges
5
84
are laid out. Finally, the article concludes with a summary of short- and long-term activities that
85
could improve the understanding and model representation of convective clouds.
86
2.
87
models
Issues associated with convection in current global
88
As a primary mechanism of heat transport between Earth’s surface and upper atmosphere,
89
convection is a critical component of the climate system and its variability. As a result,
90
limitations in our understanding and representation of convection in global models are
91
manifested in the biases in the simulated climate. The list of convection-related biases in
92
present-day global models is extensive and covers all spatio-temporal scales. Highlighted below
93
are a few of the most prominent biases.
94
a) Diurnal cycle
95
Convection often initiates as shallow cumulus clouds in response to solar forcing. The
96
shallow clouds mostly constitute a field of transient clouds, each bubbling up then dying out
97
quickly, while a few last longer and may grow larger to individual congestus or isolated
98
cumulonimbus clouds in late afternoon if humidity in the lower free troposphere is sufficiently
99
high. Many of these taller clouds decay near the area of the clouds' initiation, while others
100
organize into MCSs, propagate over long distances, and precipitate well into the next morning.
101
Thus, the diurnal cycle of precipitation varies with the scale of the convective entity.. Figure 3
102
shows the diurnal cycle of precipitation over the Southern Great Plains (SGP) of the U.S. from
103
observations and CMIP5 models and seasonal cycle of surface temperature. Globa models such
104
as those in CMIP5 do not explicitly account for MCSs, which produce over half of the warm
6
105
season precipitation (Fritsch et al. 1986; Nesbitt et al. 2006) and most often occur at night
106
(McAnelly and Cotton 1988; Carbone et al. 2002). As a result, nocturnal convective
107
precipitation is essentially absent in the CMIP5 models and many such models have peak
108
precipitation just before or after noon. The underestimated propagating nocturnal precipitation
109
has important implications for land-atmosphere interactions and the seasonal cycle of
110
temperature, probably contributing to the models' tendency to have a warm bias in summertime
111
near-surface temperatures on account of dry soil possibly due to low precipitation. This bias is
112
notable over the central part of the U.S. (Figure 3b).
113
b) Madden-Julian Oscillation
114
The MJO is a major component of tropical intra-seasonal variability with far-reaching
115
impacts on regional extremes such as tropical cyclone activity, atmospheric rivers, heat waves,
116
and floods (Zhang 2005, 2013). Despite extensive research over the last four decades,
117
fundamental understanding and accurate representation of MJO initiation, propagation, and
118
interaction with other regional processes remain as unmet challenges. This is mainly because of
119
the multiscale nature of the cloud processes involved and associated difficulties in
120
parameterizing these processes. Parameterizations are as yet unable to handle satellite-observed
121
multi-scale aspects of the convective population, which varies with MJO phase (Barnes and
122
Houze 2013; Yuan and Houze 2013). In particular, the cloud population is composed of
123
different proportions of shallow, deep, and mesoscale convection during each MJO phase. The
124
consequence of this problem is apparent in the comparison of MJO variance between CMIP
125
models and observations (Figure 4). While there have been improvements in CMIP5 over
126
CMIP3, the more recent models generally continue to underestimate the variance and have more
127
persistent precipitation over equatorial regions than is observed (Hung et al. 2013). Another
7
128
aspect of the MJO that challenges global models is the “MJO prediction barrier” over the
129
maritime continent (Neena et al. 2014; Kim et al. 2009), where problems in the representation
130
of convection in the MJO are compounded by interactions with a pronounced diurnal cycle
131
(Johnson and Priegnitz 1981; Williams and Houze 1987) and affected by the complex
132
topography of the islands (Hagos et al. 2016).
133
c) South Asian Summer Monsoon
134
In general, precipitation issues in CMIP5 models are of two types: spread, which
135
pertains to large differences among the models that limits the confidence level in their
136
projections, and bias, which represents consistent deviations of the model results from
137
observations. These two forms of uncertainty are especially apparent in monsoon environments.
138
Consider, for example, the seasonal cycle of precipitation associated with the South Asian
139
monsoon. The solid black curve in Figure 5a shows the monthly mean precipitation averaged
140
over all of India; Figure 5b shows the same data normalized by the annual mean precipitation.
141
The CMIP5 ensemble has a very large spread in seasonal cycle amplitude, and when the
142
precipitation is normalized, it is also apparent that the onset of the monsoon is delayed in most
143
of the models.
144
3.
145
generation global models
Strategy of representing convection in the next
146
Convection-related model biases in key features of climate variability are reviewed in
147
the last section. In this section, specific challenges in each of the three core themes depicted in
8
148
Figure 2 that contribute to those biases are identified and strategies for addressing them in an
149
integrated way are proposed.
150
a) Improving the basic understanding of convective cloud processes
151
In order to highlight the gaps in our understanding of convection, we consider the full
152
lifecycle of convection as a starting point, which can be treated as a series of three transitional
153
processes:
154
1)
Boundary layer variability and the development of precipitating shallow cumulus clouds
155
2)
Transition to deep convection
156
3)
Upscale growth from deep convective cells to form MCSs
157
These transitions occur under certain environmental conditions and involve feedback
158
mechanisms. The overarching challenge is determining what environmental conditions and
159
feedback processes control these transitions? Each of the above-listed transitions and the
160
specific scientific questions related to them are individually discussed below.
161
162
i. Boundary layer dynamics and the development of precipitating shallow cumulus
clouds.
163
The boundary layer contains internal instabilities that produce rolls, hexagonal cells, and
164
other features of enhanced convergence that organize clouds into patterns and make some of the
165
clouds more robust. Such dynamical transitions can occur without external forcing; however, in
166
some situations vertical wind shear plays a crucial organizing role. Highly sensitive radars
167
deployed during the AMIE field campaign have led to some advances in understanding these
168
processes over a tropical oceanic environment. Rowe and Houze (2015) have shown how the
169
boundary layer develops rolls (due to internal instability), which favor some clouds to grow
9
170
deeper and precipitate. The resulting deep convection replaces boundary-layer air with
171
downdrafts on the scale of the precipitation, creating cold pools, which entirely change the
172
character of the boundary layer and dominate the formation of subsequent convection. Where
173
cold pools intersect, enhanced boundary-layer convergence often results in stronger secondary
174
convection (Feng et al. 2015), leading to a chain reaction as these intersecting cold pools trigger
175
ever-larger convective systems (Feng et al. 2015) and often leads to a cloud population
176
containing MCSs (Rowe and Houze 2015).
177
The studies of Rowe and Houze (2015) and Feng et al. (2015) were of oceanic convection;
178
similar studies are needed of boundary-layer evolution associated with cloud-field transitions
179
over continental regions. In addition to boundary-layer instabilities leading to rolls and other
180
patterns of convective initiation, the surface conditions, including land-use heterogeneity,
181
nearby water body conditions, and topographic features, also affect the formation and
182
development of cumulus clouds. Additionally, large-scale wind shear and thermodynamic
183
instability all lead to certain boundary layer dynamics affecting cloud patterns.
184
ii. Factors affect the transition to deep convection
185
One of the key challenges continuing to impede understanding of how convective clouds
186
deepen is the inability to determine the characteristics of convective updrafts and downdrafts
187
and their interactions with the environment (especially via entrainment) and microphysical
188
processes. Statistics of the intensity, size, and variation of drafts with the height and width of
189
convective clouds are inadequate to non-existent under many key environmental conditions. In
190
addition, little information exists on the internal turbulent characteristics of the drafts. As a
191
result, the factors determining the behavior of drafts in convective clouds at all stages of
192
development remain far from clear. Aircraft data (e.g., Zipser and LeMone 1980) and TRMM
10
193
radar observations (Zipser et al. 2006; Houze et al. 2015) highlight the relationship between the
194
nature of precipitating convection and environmental conditions that differs between land and
195
ocean and from one climatic regime to another. For example, even though CAPE is often very
196
large over tropical oceans, observations from field programs show that undiluted ascent from
197
the boundary layer is extremely rare over tropical oceans (Zipser 2003). The most powerful,
198
nearly undiluted towers occur mainly over a few land areas (i.e., relatively dry regions near
199
major mountain ranges, Zipser et al. 2006).
200
iii. Upscale growth from deep convective cells to MCSs
201
The tendency toward upscale growth of convective entities to form mesoscale units
202
differs among open oceans, arid lands, rainforests, and monsoons (Houze et al. 2015), yet MCSs
203
are important rain producers over both land and ocean. About 30–70% of warm season rain over
204
the U.S. east of the Rocky Mountains (Fritsch et al. 1986; Nesbitt et al. 2006) and 50–60% of all
205
tropical rainfall (Yuan and Houze 2010) is produced by MCSs. An MCS often begins when
206
convective clouds rooted in the boundary layer aggregate into a unit that is 1–2 orders of
207
magnitude larger in area than an individual convective cloud. However, "elevated MCSs" that
208
develop with no connection to the boundary layer whatsoever are also important over
209
continental regions such as the U.S. (Marsham et al. 2011; Schumacher 2015). Whether MCSs
210
develop from boundary-layer-rooted convection or as elevated systems, net heating by the
211
aggregated convection eventually induces mesoscale circulations in the form of broad sloping
212
layers of up and downdraft circulations (Moncrieff 1992; Pandya and Durran 1996). The
213
induced mesoscale circulation is not generally connected to the boundary layer; rather, a lower
214
tropospheric layer up to several kilometers in depth feeds the sloping updraft, and the sloping
11
215
downdraft begins in the mid troposphere. No parameterization scheme yet addresses the nature
216
of overturning induced by MCSs.
217
A key question is what determines the scale of MCSs? More specifically, how does a
218
non-uniform grouping of convective cells begin growing? One popular theory is that of "self-
219
aggregation" (e.g., Wing and Emanuel 2013) whereby small convective clouds initially
220
randomly dispersed over a broad area concentrate into a mesoscale unit that intensifies through
221
radiative/dynamic feedback. The resulting mesoscale cloud unit can then develop the mesoscale
222
dynamical circulation of an MCS. Houze et al. (2015) have noted that observations of
223
precipitating cloud populations seen by the TRMM radar suggest that conditions are especially
224
suitable for self-aggregation over tropical oceans; it is not yet clear whether self-aggregation
225
occurs in the same way over land, where there is significant diurnal variation in surface forcing
226
and a variety of land-surface and topographic conditions come into play. In realistic
227
environments over both land and ocean, the growth and propagation of MCSs is affected by
228
vertical wind shear that modifies the baroclinic generation of vorticity by the horizontal gradient
229
of convective heating (Moncrieff 1992). Additionally, the heating by aggregated convection in
230
an MCS induces a gravity wave response in the form of the intertwined layers of mesoscale
231
ascent and descent upon which convective-scale elements continue to be superimposed (Pandya
232
and Durran 1996).
233
As introduced in Section 3.a.i, another factor that plays a role in upscale growth of
234
convection to form MCSs is cold pools. As the mode of communication between precipitating
235
convective elements (e.g., Johnson and Houze 1987), cold pool dynamics and evolution need to
236
be understood within the context of the precipitating systems and their environment. Cold pools
237
are affected both by environmental shear and thermodynamic profiles and internal cloud
12
238
microphysics, which in turn feed back on the precipitating system. For example, the horizontal
239
extent of an MCS depends in part on the fallout trajectories of ice particles in the overtopping
240
layer of stratiform cloud. As these fall, cooling by evaporation and melting of hydrometeors
241
determine the intensity, frequency (Hagos et al. 2014), and subsequent extent of cold pools
242
(Feng et al. 2015). More specifically, cold-pool depth partially controls gust-front speed
243
(Wakimoto 1982) and subsequent updraft formation (Feng et al. 2015). As precipitating
244
convective cells deposit cold pools in the boundary layer, they trigger new convection in the
245
vicinity of aging convection, contributing to the development, propagation, and longevity of
246
MCSs depending, in part, on lower-tropospheric vertical shear (Thorpe et al. 1980; Rotunno et
247
al. 1988).
248
Several questions related to cold pools constitute remaining uncertainties. Among those
249
are: what determines whether or not convection is initiated at a cold pool boundary? How long
250
do cold pools last? How deep are they? How strong are the updrafts they induce? What is the
251
inter-relationship among the natures of primary updrafts, downdrafts, cold pools, and secondary
252
updrafts in the process of organization? What are the relative roles of microphysical processes
253
that determine cold pool dynamics (hydrometeor loading, melting, evaporation, riming, ice
254
multiplication, graupel-hail production)? The relative importance of these factors is partially
255
known in regard to microburst downdrafts (Srivastava 1985, 1987; Kessinger et al. 1988) but
256
not for broader cold pools associated with MCSs. What are the roles of dynamical-
257
microphysical feedbacks affecting vertical velocity, especially with regard to ice processes?
258
Concurrent observations of cold pool characteristics with kinematics, microphysics, and
259
environmental conditions are therefore required to address these important questions.
13
260
In summary the key challenges in our understanding of evolving cloud populations are: 1)
261
how boundary-layer processes evolve in a way that leads to cloud populations containing deep
262
and mesoscale convection including cold pool dynamics; 2) Size, intensity, and internal
263
variability of convective drafts; 3), microphysical feedbacks; 4) aggregation of convection; 5)
264
inducement of mesoscale circulation, especially gravity wave response to aggregated
265
convective elements.
266
b) Paths toward improving the treatment of convection in high-resolution
267
global models
268
As noted in the introduction, the representation of convection in traditional global
269
models, with grid spacing ~100 km, implicitly or explicitly relies on the assumption that the
270
combined area covered by convective drafts is much smaller than those of a grid column. In that
271
case, scale-separation and statistical quasi-equilibrium are assumed between the resolved
272
circulation, which may destabilize an atmospheric column, and the aggregate of convective
273
drafts, which work to stabilize the column (Arakawa and Schubert 1974). Importantly, such
274
parameterizations implicitly require the grid column to be large compared to the mean distance
275
between updrafts in order for the column to contain a meaningful sample size of updrafts.
276
However, it has been known since the Global Atmospheric Research Program’s Atlantic
277
Tropical Experiment (GATE, Houze and Betts 1981) that long-lasting MCSs are important, and
278
scale-separation is not present in either time or space, even in traditional models. The
279
breakdown of the mean-distance assumption means that stochastic effects start to be relevant
280
(e.g., Plant and Craig 2008). The breakdown of the area assumption means that convection
281
enters a so-called “gray zone,” further discussed below. Additionally convection is often
14
282
assumed to respond to forcing almost instantaneously and deterministically with little memory
283
or internal variability of its own.
284
Advances in computational resources have made possible operational global weather
285
and experimental climate models with spatial resolution ≤ 10 km, which allows the larger
286
aspects of circulations associated with convection to be partially resolved. As a result, we need
287
a fundamental rethinking of the objective of, approach to, and assumptions in convection
288
parameterizations. First of all, simulations at scales ≤ ~10 km could easily have MCSs with
289
areas of updrafts and downdrafts comparable to and even greater than the grid spacing and
290
could last longer than the often assumed adjustment time-scale of cumulus convection. In such
291
cases, convection cannot be treated as an aggregate response to the environment but is partially
292
included in the resolved dynamics. This partial resolution of mesoscale convection does not
293
obviate parameterization; instead it gives it new purpose and definition. Parameterizations must
294
represent the sub-grid states, processes and transitions discussed in the last section, and their
295
interactions with resolved dynamic and thermodynamic processes, which include mesoscale
296
processes. Thus, the resolved dynamics, sub-grid states, and transitions are parts of a continuum,
297
and the scale separation is arbitrarily imposed by the model grid spacing. As scale-separation
298
(i.e., the foundation of the statistical quasi-equilibrium assumption) disappears, an important
299
and stringent constraint emerges: the requirement for resolution awareness. That is, model
300
results should be insensitive to arbitrary changes in grid spacing, especially when grid spacing
301
varies across a model domain. Parameterizations must be aware of the processes that are
302
unresolved, partially resolved, or fully resolved and adjust their operations accordingly.
303
Resolution awareness means that the sums of resolved and parameterized parts of key quantities
304
(e.g., mass flux) should not vary with resolution for the range of resolutions of interest.
15
305
In the absence of scale separation, the need for resolution awareness cannot be treated as
306
an afterthought, which can be addressed by tuning some parameters in order to produce a
307
desired outcome at a given resolution; rather (like statistical quasi-equilibrium before it),
308
resolution awareness must be built-in as a fundamental constraint on the design of robust
309
parameterizations.
310
With the increase of the resolution of models, attempts at resolution awareness have
311
been progressing along several lines, each with its unique strengths and challenges. These
312
methods are briefly summarized below.
313
i. Modifying quasi-equilibrium mass flux schemes
314
As mentioned above, one of the key assumptions in statistical, quasi-equilibrium-based
315
mass flux schemes is that updraft area fraction, σ , satisfies σ << 1 and the compensating
316
subsidence area fraction is
317
MCSs, and with increased resolution it breaks down even for convective-scale drafts. Arakawa
318
et al. (2011) proposed an approach to address this issue by requiring that the parameterized
319
mass flux be rescaled as
320
equilibrium adjustment value M adj and it gradually decreases to zero for situations when the
321
mass flux is fully resolved, i.e., σ=1. Implementation and evaluation of quasi-equilibrium
322
parameterizations with this and similar methods are currently actively being pursued (Grell and
323
Freitas 2013; Wu and Arakawa 2014; Liu et al. 2015; Xiao et al. 2015). Key issues arising for
324
this approach are how one should actually diagnose the value of σ within partially resolved
325
convection, and moreover, how one should diagnose M adj given that the grid-scale atmospheric
326
state traditionally supplied input to a scheme cannot by construction be considered as
. This assumption has never been accurate in the presence of
. Thus, when σ  1 ,
matches the quasi-
16
327
representative of a quasi-equilibrium state. While, by design, such models attempt to account
328
for the changes in the overall mass flux with resolution, they do not address underlying issues
329
related to the resolution dependence on the nature of the interactions between the environment
330
and convection, or on the differences (physical and dynamical) in the resolved and unresolved
331
components of convective clouds. Essentially, this approach attempts to treat the issue of spatial
332
scale separation but not the breakdown of temporal separation, as current implementations of
333
Arakawa’s scaling of the mass flux still assume convection responds fully to the environment
334
within the model time step.
335
ii. Prognostic parameterization of processes
336
The quasi-equilibrium-based approach discussed above is essentially diagnostic. A
337
prognostic approach aims to account for physical processes involved in convection, such as
338
convective up- and downdrafts, and the sub-grid circulations associated with them, such as cold
339
pools. Park (2014) has proposed a methodology of this type, which incorporates the dynamics
340
of multiple convective plumes within a grid column; predicts the initiation, evolution, and
341
advection of plume and environmental properties; allows convection to propagate; and includes
342
aspects of cold pools. This approach is scale adaptive in that it represents only sub-grid scale
343
motion with respect to the resolved motions, thus guaranteeing that the parameterized mass flux
344
vanishes as
345
include convection originating from above the boundary layer or the impacts of wind shear on
346
convection, both important factors in MCS dynamics (Rotunno et al. 1988; Moncrieff 1992;
347
Marsham et al. 2011; Schumacher 2015). Furthermore, the effects of the parameterized sub-grid
348
circulations on surface fluxes are not included.
approaches 1. Currently neither this approach nor quasi-equilibrium approaches
17
349
Other studies that have experimented with prognostic approaches include Pan and
350
Randall (1998) (recently revisited by Yano and Plant 2012), Gerard et al. (2009), and Grandpeix
351
and Lafore (2010). A key issue in this area is to establish which quantities need to be treated as
352
explicitly prognostic in order to capture the relevant dynamical effects, and which quantities
353
may continue to be handled diagnostically. In addition, none of these schemes represent the
354
unique dynamics of an MCS (viz., sloping layered ascent and descent, as described by
355
Moncrieff 1992 and Kingsmill and Houze 1999). Those features would need to be explicitly
356
predicted as part of the resolved dynamics.
357
iii. PDF-based turbulent schemes
358
The PDF-based approach (Golaz et al. 2002; Larson and Golaz 2005) involves a three-
359
step process beginning with the second-order turbulence equations that solve for the second-
360
order moments (correlations) of vertical velocity, moisture, and potential temperature, as well as
361
their covariances. Predicted moments are used to construct PDFs of model variables. The PDFs
362
are then used to calculate the third-order moments (triple correlations), which are necessary to
363
bring the method to closure. The current version of this parameterization uses a family of
364
double Gaussian PDFs. This approach has been successfully implemented in GCMs: CAM and
365
ACME with the Zhang and McFarlane (1995) deep convection scheme (Bogenschutz et al. 2013)
366
and GFDL AM3 with the Donner (1993) deep convection scheme. The approach performs well
367
for shallow cumulus and stratocumulus clouds (Guo et al. 2014). Its development into a unified
368
scheme that does not specifically refer to convective cloud categories (i.e., shallow vs. deep) is
369
in progress. However, it is computationally expensive, and the strategies for implementing
370
organization mechanisms (such as shear and cold pool dynamics) in such a scheme are only
371
beginning to be developed (e.g., Storer et al. 2015; Griffin and Larson 2016). One feature of this
18
372
approach is that it predicts largely prescribed probability distributions of vertical velocity,
373
temperature, water vapor, and hydrometeor mixing ratios. The hydrometeor concentrations in
374
deep convection are rarely directly obtained in observations, except for limited aircraft in-situ
375
measurements in field programs. However, useful statistics of closely related quantities can be
376
derived from polarimetric radars and possibly other remote sensors. An observation of these
377
variables concurrently with updraft/downdraft measurements, to the extent possible, is
378
necessary so that covariances can be derived. This requirement implies that some quantities
379
need to be measured at higher frequency than routine observations (esp., temperature, wind, and
380
water vapor content), and others such as vertical velocity may require as-yet unavailable
381
platforms or instruments. LES models can provide further information on the nature of PDFs
382
down to very small scales. Having empirical and high resolution simulation-based knowledge of
383
the PDFs will provide the natural forms of the PDFs used in these schemes. Gaining such
384
information empirically and from high resolution models is crucial for the PDF methods to
385
work.
386
iv. Explicit approaches: Superparameterization and global cloud permitting models
387
In the superparameterization approach, 2-D cloud resolving models are embedded in
388
model grid elements. The GCM provides the large-scale forcing and the CRM runs at ~1–4 km
389
resolution with periodic lateral boundary conditions within each grid element (Randall et al.
390
2013) to provide the cloud and radiative tendencies to the GCM. Evaluation and improvement
391
of this approach, including extending it into three dimensions (i.e., two perpendicular CRM
392
columns) is an active area of research. It has been shown to improve the progression of MCSs
393
over the central U.S. in comparison to the same model using traditional parameterizations
394
(Prichard et al. 2011; Kooperman et al. 2013; Elliott et al. 2016). Furthermore, a
19
395
superparameterization version of NCAR’s CAM model (SPCAM) has been shown to have
396
better skill in representing the MJO than several other models (Kim et al. 2009) including the
397
conventional version of CAM (Benedict and Randall 2009). However, as is often the case with
398
changes in cumulus parameterizations (Kim et al. 2012), the improvement comes with biases in
399
the mean state and in the boundary-layer interactions. Lack of communication among the
400
embedded CRMs is a challenge for the superparameterization of convection and convective
401
organization. For instance, when MCSs are generated in the CRM grid elements within a GCM
402
column, they are confined to that GCM column due to the CRM’s periodic lateral boundary
403
conditions. Although MCSs can be generated across contiguous global model domains on the
404
parent global model grid as a result of the joint effects of latent heating and vertical shear,
405
circulation structure of the MCSs is compromised (Pritchard et al. 2011). The 2-D assumption is
406
also limiting because most real MCSs are not 2-D. Although 3-D CRMs can be utilized, it has
407
only been attempted in very small domains (e.g., Khairoutdinov et al. 2005) owing to the added
408
computational expense.
409
The most physically realistic and mathematically consistent approach to including
410
convection in a global model is to employ a global cloud-permitting model (GCPM). The
411
simulated mesoscale cloud systems are three-dimensional and not confined by periodic lateral
412
boundary conditions. The pioneering global CPM simulations conducted on Japan’s Earth
413
Simulator had 3.5-km, 7-km, or 14-km computational grids according to the length of the
414
simulation (e.g., Miura et al. 2005; Satoh et al. 2008). When compared to TRMM observations,
415
the 7-km grid spacing those Nonhydrostatic icosahedral atmospheric model (NICAM)
416
simulations successfully captured not only the MJO but also the clusters of MCSs within it
417
(Miyakawa et al. 2012). Additionally, while they are computationally extremely expensive,
20
418
such models are now being run for short periods at cloud resolving sub-kilometer grid spacing
419
(Miyamoto et al. 2013).
420
v. Dynamically based parameterization for mesoscale convection
421
The main message from the foregoing sections is that we are at the crux of a new era of
422
cloud-permitting global weather and global models where we can no longer neglect mesoscale
423
convection. This situation points to the need to integrate mesoscale dynamics into
424
parameterizations where it is presently conspicuous only by its absence. Moncrieff (1992)
425
pointed out how a large class of MCSs can be characterized by a simultaneous adjustment to the
426
thermodynamic and wind shear profiles. Two parameterization developments are underway that
427
build on this idea: the multi-cloud model (Khouider and Majda 2006) and the slantwise-layer-
428
overturning model (Moncrieff 2010; Moncrieff and Waliser 2015). The multi-cloud model
429
represents the diabatic heating and the associated circulations as three cloud types observed to
430
dominate the diabatic heating in tropical convection: congestus, deep precipitating convection,
431
and precipitating stratiform cloud associated with MCSs (Johnson et al. 1999; Mapes 2006).
432
Slantwise layer overturning is a computationally efficient paradigm for the parameterization of
433
mesoscale convection based on multiscale coherent structures in a turbulent environment.
434
c) Observational data needs
435
The scientific and parameterization challenges discussed in the previous section
436
highlight the need for understanding how the size, intensity, and internal turbulent structure of
437
updrafts/downdrafts relate to one another, to boundary-layer processes, to microphysical and
438
cold pool processes, and to large-scale and mesoscale context. The corresponding observational
439
requirement includes continued advancement in methods of observation, further collection, new
21
440
analysis methods, and delivery of observational data in ways that will inform process studies
441
and parameterization. Datasets and analyses need to indicate how aspects of drafts relate to one
442
another to determine the interactive processes of convection. It is important, therefore, for
443
information to be examined, collected, and retrieved in the form of concurrent and collocated
444
observations/retrievals rather than isolated time series of single quantities. Some especially
445
pressing needs are discussed below, and some future short- and long-term observation strategies
446
and investments are suggested.
447
i. Merged products from existing data and infrastructure
448
Relationships between environmental water vapor and precipitation, precipitation type
449
and latent heating, cloud structure and radiative processes, and between microphysical processes
450
and cold pool formation have all been previously examined mostly individually—but they must
451
be obtained concurrently with up- and downdraft statistics in order to be most useful in
452
parameterization development. Field projects are the best venue for providing such information.
453
While many field experiments have been carried out, these projects have primarily documented
454
the synoptic and mesoscale environments of convective drafts with insufficient ability to
455
observe the drafts themselves. Further field efforts for obtaining draft statistics in highly
456
documented environmental settings remain a paramount need and objective.
457
Even with improved airborne capability, field programs have a major shortcoming,
458
which is their short duration—typically a few months or less. Statistically robust datasets over
459
longer time periods are needed. One possible solution is to use the DOE SGP site in a field-
460
experiment mode. For example, instead of passively obtaining measurements at the site with
461
standard scanning strategies of the radars, lidars, sounding launches, and other measurements, a
462
new approach would be to adjust the scan strategies and other measurement procedures (such as
22
463
sounding frequency) to the forecasted weather situation and real-time conditions. The default
464
pre-planned modes would be altered to ones best suited to sample properties of shallow clouds
465
when deep clouds are absent, to a deepening cloud population, and to expected MCSs
466
occurrence, depending on the forecast. This adaptive operation procedure would collect the
467
most relevant information for the type of weather that is occurring. Decisions could be made by
468
scientists monitoring the weather forecasts and the scientists could implement different
469
strategies quickly through online communication with engineers operating the instruments. This
470
mode would adapt a field campaign approach to a permanent observational facility to optimize
471
its ability to provide concurrent observations of the details of convective systems that are
472
necessary for parameterization development.
473
ii. Long-term improvement of observational infrastructure
474
Some of the most-needed measurements for parameterization development are not only
475
unavailable but also may be difficult or impossible to obtain with existing resources and
476
observational platforms. They require sustained, coordinated investments from interested
477
national and international agencies. There are some especially important directions for future
478
observational work that will support development of convective parameterization:
479
•
Airborne platforms for in-situ measurements and radar technology for remote detection
480
of draft properties have been limited to date, resulting in a critically insufficient amount
481
of information on updraft/downdraft intensities, dimensions, and internal turbulent
482
characteristics, which need to be determined concurrently and statistically. Multi-
483
Doppler radar techniques are sometimes offered as a substitute for airborne
484
measurements; however, limitations of sampling, resolution, and uncertainty in
485
converting Doppler data to air motion velocities make Doppler radar inadequate by
23
486
themselves. Aircraft are not nimble because of flight planning restrictions, and other
487
logistics such as safety concerns. Nevertheless, there appears to be no substitute for the
488
need for in-situ targeting by suitable aircraft, with strong airframe, high-altitude
489
capability, and instrumentation to obtain information on drafts of all strengths,
490
turbulence, and cloud microphysics at multiple altitudes. A state-of-the-art convection-
491
penetrating research aircraft is needed in the atmospheric sciences community, and
492
multi-agency cooperation could allow it to be used in connection with the
493
aforementioned SGP observational program as well as in shorter-term field programs
494
using advanced radars, lidars, profilers, soundings, and other observations to provide the
495
environmental context.
496
•
Flexible S-band dual-polarization scanning radars as dedicated research facilities are
497
critical to support aircraft measurements. Specifically, these long-wavelength radars are
498
the most important instrumentation to provide microphysical context. NEXRAD radars
499
operate in a pre-defined, full-volume scanning mode that is optimized for nowcasting
500
but does not provide sufficient vertical resolution to obtain precise distributions of
501
microphysical characteristics indicated by dual-polarization radar technology. Highly
502
sensitive S-band scanning radars such as NSF's S-Pol and NASA's NPOL are essential
503
to provide the necessary microphysical context because these radars are not constrained
504
to operational scanning strategies. They can provide increased vertical resolution
505
through frequent, adaptable Range Height Indicator (RHI) scan sectors, which is critical
506
because microphysical processes and updraft characteristics have a very fine-scale
507
variability with height (or temperature) that cannot be captured by routine operational
508
tilt-sequence scanning of NEXRAD radars. S-band information is also critical because
24
509
W, Ka-, X- and C-band scanning radars can be severely, and at times completely,
510
attenuated by heavy precipitation associated with MCSs, where the up- and downdrafts
511
are most intense, thus limiting the full spectrum of observations needed to understand
512
upscale growth or precipitating systems. For these reasons, it is very important to
513
continue carrying out field programs that employ S-band research radars with flexible
514
scanning strategies in concert with aircraft direct measurement of up- and downdraft
515
properties. The primary obstacle is that these radars are expensive to maintain and
516
deploy, so interagency cooperation might be needed to maintain or even expand the
517
number of S-band scanning radar facilities.
518
•
Most of the vertical re-distribution of heat by convection occurs at low latitudes,
519
especially over the tropical warm oceans, the Maritime Continent, and monsoon regions
520
of Asia and Africa. Although GATE, TOGA-COARE, MONEX, and AMIE/DYNAMO
521
have provided critical information over the world's largest oceans, these projects did not
522
fully address the scientific questions discussed in foregoing sections because of limited
523
observational technology—especially aircraft unable to document up- and downdraft
524
statistics. Satellite-based radar reflectivity data indicate that the nature of convection
525
varies from one regime to another throughout low latitudes (Houze et al. 2015).
526
However, the satellite measurements are not capable of documenting dynamical
527
differences from one region to another (Maritime Continent, monsoons, western vs.
528
eastern Atlantic and Pacific ITCZs, and the South Pacific Convergence Zone). Field
529
campaign data will be needed ultimately to address the key science questions in order
530
for parameterizations to accurately distinguish among the various forms of tropical and
25
531
subtropical convection. Interagency, international programs will be needed to
532
accomplish this large challenge.
533
d) Integration
534
In the last three sections, key scientific and parameterization challenges, as well as
535
observational needs, have been discussed individually. However, even if the technical aspects of
536
process modeling, parameterization development, and observations are addressed, progress is
537
not guaranteed unless the challenges of effectively using observations to inform
538
parameterization development and validate modeling are met. For that, integration of
539
observations, improved process understanding, and model development will be required.
540
Among the many challenges for integration are:
541
1)
Observed and modeled quantities not being the same,
542
2)
Spatiotemporal scales represented by the measurements being different than what the
543
544
545
model represents, and
3)
Uncertainties associated with observations not well quantified, thus introducing
additional uncertainties when evaluating model processes.
546
Two specific approaches for meeting these challenges are presented below.
547
i. Instrument simulators
548
Model variables are usually in the form of temperatures, mixing ratios, and wind
549
components, averaged over a grid-cell volume. Many instruments, especially remote sensors,
550
measure other types of atmospheric variables, such as radar reflectivity, light scattering, or
551
radiative flux. To connect the observations to model output requires simulators, which are
552
software designed to calculate the observable quantities from model output. Remotely sensed
26
553
quantities are generally electromagnetic or optical fields that respond to complex moments of
554
the particle distributions (cloud, precipitation, air molecules), but which are not computed
555
directly within models due to the differences in measurement volume to grid-cell volume and
556
assumptions due to attenuation, cloud and aerosol size distributions, and other instrument
557
specific technical details that generally are unknown when running a model. Nonetheless, model
558
output can be used to estimate the observable quantities via the simulators. The simulators
559
involve a range of physical assumptions, are challenging to design, and require further research.
560
Various investigators are in the process of developing simulators for GCMs using the Cloud
561
Feedback Model Inter-comparison Project (CFMIP) Observation Simulator Package (COSP)
562
framework (Bodas-Salcedo et al. 2011). However, much effort remains to produce the needed
563
wide range of simulators. Nonetheless, currently available simulators have begun to be used in
564
studies using CRMs, LES models, and GCMs (e.g., Varble et al. 2011; Hagos et al. 2014).
565
ii. Cross-scale and hierarchical approaches to modeling
566
High-resolution global atmospheric modeling is often thought of as the ultimate limit to
567
traditional global modeling. Alternatively, one can view a high-resolution global model as a
568
large-domain limit to highly resolved regional models. This latter perspective enables
569
evaluation of model physics in regional and variable resolution global modeling frameworks at
570
a fraction of the computational cost of global high-resolution models. The treatment of
571
mesoscale convection in the gray zone can advance by utilizing advances in knowledge of
572
physical and dynamical processes gained from improved observations and from LES and CRM
573
modeling, as discussed in preceding sections of this article. Thus, high-resolution global
574
modeling activities should be viewed as a hierarchy and designed whenever possible as integral
575
parts of the modeling continuum that includes LES, CRM, variable resolution models, as well
27
576
as operational global cloud permitting models. Consideration is required of what can be learned
577
through evaluation of one approach using a specific choice of observational data that can benefit
578
other approaches up and down the hierarchy where direct evaluation using that specific
579
observational data is not feasible.
580
4.
Conclusion
581
The next generation of global models, with grid spacings as fine as 1–10 km, must be
582
able to represent the entire spectrum of convective clouds regardless of model resolution. Future
583
models will resolve certain features of clouds, while other aspects will remain parameterized,
584
even in the highest-resolution models, and the features parameterized will depend on the nature
585
of the cloud populations in relation to the model resolution. Thus, parameterizations will need to
586
operate seamlessly across all the involved scales and phenomena; they cannot be scale specific.
587
The overview presented here was motivated by discussions at a DOE-supported workshop
588
aimed at devising strategies for addressing these issues. The workshop concluded that accurate
589
representation of convection in global models requires advances in our basic understanding of
590
convection, specifically, the sequence of transitions in convective cloud populations from stable
591
boundary layer up to cloud population states that include mesoscale dynamics and
592
dynamical/microphysical interaction on a range of scales, from turbulent elements within
593
individual drafts to MCSs. Furthermore, high-resolution global modeling can benefit from a
594
hierarchical approach that takes full advantage of progress in other modeling frameworks
595
including LES, limited-area CRMs, variable-resolution, and operational high-resolution forecast
596
models.
28
597
In order to effectively test hypotheses and evaluate models, observations need to be
598
considered in terms of merged products that document concurrent and collocated
599
observations/retrievals of cloud variables as well as environmental context. Furthermore,
600
observational strategies can be proactively adapted in near-real time to effectively sample
601
prevailing cloud populations in near-real time. Based on forecast conditions, scanning
602
procedures and sounding launches can be scheduled to optimize instrument operations. Flexible
603
S-band radars designed for research should continue to be used to conduct specialized dual-
604
polarization scans that will support aircraft sampling. Aircraft platforms must be improved to
605
include robust convection-penetrating aircraft with sufficient altitude capability to study deep
606
convection. Field campaigns remain essential with advanced aircraft and radar instrumentation
607
that can explore all deep convective regimes, including Tropical Ocean, coastal zones, and
608
various types of land surface and topography.
609
29
610
Acknowledgement: Funding for the workshop was provided by the U.S. Department of
611
Energy’s Atmospheric Systems Research Program. We thank the ASR Program Managers,
612
Shaima Nasiri and Ashley Williamson, as well as Jerome Fast, ASR Science Focus Area (SFA)
613
Principal Investigator at PNNL, for the support and encouragement throughout the planning and
614
execution of the workshop. We also would like to thank Emily Davis and Alyssa Cummings
615
who provided logistical support to the workshop. Finally we would like to thank all the
616
workshop participants: Mitch Moncrieff (NCAR), Ed Zipser (University of Utah), Greg
617
Thompson (NCAR), Sungsu Park (Korean National University), Chidong Zhang (University of
618
Miami), Courtney Schumacher (Texas A and M), Russ Schumacher (Colorado State University),
619
Robert Plant (University of Reading), Daehyun Kim (University of Washington), Chris
620
Williams (NOAA), Sue van den Heever (Colorado State University), Yunyan Zhang (Lawrence
621
Livermore National Laboratory), Shaocheng Xie (Lawrence Livermore National Laboratory),
622
Scott Collis (Argonne National Laboratory), Jeff Trapp (University of Illinois Champaign-
623
Urbana ), Chris Golaz (Lawrence Livermore National Laboratory), Steven Rutledge (Colorado
624
State University), Angela Rowe (University of Washington), Jim Mather (Pacific Northwest
625
National Laboratory), Phil Rasch (Pacific Northwest National Laboratory), Jiwen Fan (Pacific
626
Northwest National Laboratory), Jerome Fast (Pacific Northwest National Laboratory), William
627
Gustafson (Pacific Northwest National Laboratory), Steve Klein (Lawrence Livermore National
628
Laboratory), Vince Larson (University of Wisconsin), and Tony Del-Genio (NASA GISS).
629
Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of
630
Energy under Contract DE-AC05-76RLO1830.
631
30
632
References
633
Arakawa A, and W. H. Schubert, 1974: Interaction of a cumulus cloud ensemble with the large-
634
scale environment Part I. J. Atmos. Sci.,. 31, 674–701.
635
Arakawa, A., J.-H. Jung, and C.-M. Wu, 2011: Toward unification of the multiscale modeling
636
of the atmosphere. Atmos. Chem. Phys., 11, 3731–3742.
637
Barnes, H. C., and R. A. Houze, Jr., 2013: The precipitating cloud population of the Madden-
638
Julian Oscillation over the Indian and West Pacific Oceans. J. Geophys. Res., 118, 6996-7023,
639
doi:10.1002/jgrd.50375.
640
Bodas-Salcedo, A., and Coauthors, 2011: COSP: Satellite simulation software for model
641
assessment. Bull. Amer. Meteor. Soc., 92, 1023–1043
642
Bogenschutz, P. A., A. Gettelman, H. Morrison, V. E. Larson, C. Craig, and D. P. Schanen,
643
2013: Higher-order turbulence closure and its impact on climate simulations in the Community
644
Atmosphere Model. J. Climate, 26, 9655-9676.
645
Chen, S. S., and R. A. Houze, Jr., 1997: Diurnal variation and life cycle of deep convective
646
systems over the tropical Pacific warm pool. Quart. J. Roy. Meteor. Soc., 123, 357-388.
647
Donner, L. J., 1993: A cumulus parameterization including mass fluxes, vertical momentum
648
dynamics, and mesoscale effects. J. Atmos. Sci., 50, 889–906.
649
Feng, Z., S. Hagos, A. K. Rowe, C. D. Burleyson, M. N. Martini, and S. P. de Szoeke, 2015:
650
Mechanisms of convective cloud organization by cold cools over tropical warm ocean during
651
the AMIE/DYNAMO field Campaign. J. Adv. Model. Earth System, 7, 357–381,
652
doi:10.1002/2014MS000384.
31
653
Fritsch, J. M., R. J. Kane, and C. R. Chelius, 1986: The contributionc of mesoscale convective
654
weather systemsmcws to the warm-season precipitationws0 in the United States. J. Climate
655
Appl. Meteor.,. 25, 1333-1345.
656
Gerard, L., 2015: Bulk mass-flux perturbation formulation for a unified approach of deep
657
convection at high resolution. Mon. Wea. Rev., 143, 4038–4063.
658
Gerard, L., J.-M. Piriou, R.J.-M., Brozkova, J.-F.R., Geleyn, J.-F. and D. Banciu, D., 2009:.
659
Cloud and precipitation parameterization in a meso-gamma-scale operational weather prediction
660
model., Mon. Wea. Rev., 137, 3960-3977.
661
Gerard, L., 2015: Bulk Mass-Flux Perturbation Formulation for a Unified Approach of Deep
662
Convection at High Resolution. Mon. Wea. Rev., 143, 4038–4063.
663
Golaz, J.-C., V. E. Larson, and W. R. Cotton, 2002: (2002a), A PDF-based model for boundary
664
layer clouds. Part I: Method and model description., J. Atmos. Sci., 59,(24), 3540–3551.
665
Grandpeix, J.-V., and J.-P. Lafore, J.-P., 2010: A density current parameterization coupled with
666
Emanuel’s convection scheme. Convection Scheme. Part I: The modelsModels. J. Atmos. Sci.,
667
67, 881–897.
668
Grell, G. A., and S. Freitas, 2013: A scale and aerosol aware stochastic convective
669
parameterization for weather and air quality modeling. Atmos. Chem. Phys. Discuss., 13,
670
23845-23893, doi:10.5194/acpd-13-23845-2013.
671
Guo, Z., M. Wang, Y. Qian, V. E. Larson, S. Ghan, M. Ovchinnikov, P. A. Bogenschutz, C.
672
Zhao, G. Lin, and T. Zhou, (2014:), A sensitivity analysis of cloud properties to CLUBB
32
673
parameters in the single-column Community Atmosphere Model (SCAM5).), J. Adv. Model.
674
Earth Syst., 6, 829–858, doi:10.1002/2014MS000315.
675
Hagos, S. M., Z. Feng, C. D. Burleyson, K. S. Lim, C. N. Long, D. Wu, and G. Thompson,
676
2014: Evaluation of convection-permitting model simulations of cloud populations associated
677
with the Madden-Julian Oscillation using data collected during the AMIE/DYNAMO field
678
campaign. J. Geophys. Res. Atmos., 119, 12052-12068, doi:10.1002/2014JD022143.
679
Hagos, S, C Zhang, Z Feng, C Burleyson, C De Mott, J Benedict, and M Martini. 2016. “The
680
Impact of Diurnal Cycle on the Propagation of MJO Across the Maritime Continent.” In Press.
681
Journal of Advances in Modeling Earth Systems.
682
Hirota, N., Y. N. Takayabu, M. Watanabe, and M. Kimoto, 2011: Precipitation reproducibility
683
over tropical oceans and its relationship to the double ITCZ problem in CMIP3 and MIROC5
684
climate models. J. Climate, 24, 4859-4873.
685
Houze, R. A., Jr., and A. K. Betts, 1981: Convection in GATE. Rev. Geophys. Space Phys., 19,
686
541-576
687
Houze, R. A., Jr., S. G. Geotis, F. D. Marks, Jr., and A. K. West, 1981: Winter monsoon
688
convection in the vicinity of north Borneo. Part I: Structure and time variation of the clouds and
689
precipitation. Mon. Wea. Rev., 109, 1595-1614.
690
Houze, R. A., Jr., K. L. Rasmussen, M. D. Zuluaga, and S. R. Brodzik, 2015: The variable
691
nature of convection in the tropics and subtropics: A legacy of 16 years of the Tropical Rainfall
692
Measuring Mission (TRMM) satellite. Rev. Geophys., 53, doi:10.1002/2015RG000488.
33
693
Hung, M.-P., J. Lin, W. Wang, D. Kim, T. Shinoda, and S. J. Weaver, 2013: MJO and
694
convectively coupled equatorial waves simulated by CMIP5 climate models. J. Climate, 26,
695
6185-6214.
696
Johnson, R. H., and D. L. Priegnitz, 1981: Winter monsoon convection in the vicinity of North
697
Borneo. Part II: Effects on large-scale fields. Mon. Wea. Rev., 109, 1615-1628.
698
Johnson, R. H., and R. A. Houze, Jr., 1987: Precipitating cloud systems of the Asian monsoon.
699
In Monsoon Meteorology (C.-P. Chang and T. N. Krishnamurti, Eds.), 298-353.
700
Johnson, R. H., T. M. Rickenbach, S. A. Rutledge, P. E. Ciesielski, and W. H. Schubert, 1999:
701
Trimodal characteristics of tropical convection. J.Clim., 12, 2397–2418.
702
Kessinger, C.J., Parsons, D.B., Wilson, J.W., 1988. Observations of a storm containing
703
misocyclones, downbursts, and horizontal vortex circulations. Mon. Weather Rev. 116, 1959–
704
1982.
705
Kikuchi, K., and B. Wang, 2008: Diurnal precipitation regimes in the global tropics, J. Clim.,
706
21, 2680–2696, doi:10.1175/2007JCLI2051.1.
707
Kooperman, G. J., M. S. Pritchard, and R. C. J. Somerville, (2013:), Robustness and
708
sensitivities of central U.S. summer convection in the super-parameterized CAM: Multi-model
709
intercomparison with a new regional EOF index., Geophys. Res. Lett., 40, 3287–3291.
710
Larson, V. E., and J.-C. Golaz, 2005: Using probability density functions to derive consistent
711
closure relationships among higher-order moments., Mon. Wea.Weather Rev., 133,(4), 1023–
712
1042.
34
713
Li, G. and S. P. Xie, S.P., 2014:. Tropical biases in CMIP5 Multimodel Ensemble: The
714
excessive Equatorial Pacific cold tongue and double ITCZ problems. J. Problems. Journal of
715
Climate, 27, (4), pp.1765-1780.
716
Liu, Y.-C., J. Fan, G. J. Zhang, K.-M. Xu, and S. J. Ghan, 2015:, Improving representation of
717
convective transport for scale-aware parameterization: 2. Analysis of cloud-resolving model
718
simulations., J. Geophys. Res. Atmos., 120, 3510–3532, doi:10.1002/ 2014JD022145.
719
Marsham, J. H., S. B. Trier, T. M. Weckwerth, and J. W. Wilson, 2011: Observations of
720
elevated convection initiation leading to a surface-based squall line during 13 June IHOP 2002.
721
Mon. Wea. Rev., 108, 322-336.
722
Miura, H., H. Tomita, T. Nasuno, S. Iga, M. Satoh, and T. Matsuno, 2005: A climate sensitivity
723
test using a global cloud resolving model under an aqua planet condition. Geophys. Res. Lett.,
724
32, doi:10.1029/2005GL023672.
725
Miura, H., M. Satoh, H. Tomita, A. T. Noda, T. Nasuno, and S. Iga, 2007: A short-duration
726
global cloud-resolving simulation with a realistic land and sea distribution. Geophys. Res. Lett.,
727
34, doi:10.1029/2006GL027448.
728
Miyakawa, T., Y. N. Takayabu, T. Nasuno, H. Miura, M. Satoh, and M. W. Moncrieff, 2012:
729
Convective momentum transport by rainbands within a Madden-Julian Oscillation in a global
730
nonhydrostatic model with explicit deep convective processes. Part 1: Methodology and general
731
results69, 1317-1338.
732
Moncrieff, M. W., 1992: Organized convective systems: Archetypical dynamical models, mass
733
and momentum flux theory, and parametrization. Quart. J. Roy. Meteor. Soc., 118, 819–850.
35
734
Moncrieff, M. W., 2010: The multiscale organization of moist convection and the intersection
735
of weather and climate, in Climate Dynamics: Why Does Climate Vary? Geophys. Monogr.
736
Ser., Vol. 189, Eds. D-Z. Sun and F. Bryan, pp. 3–26, doi: 10.1029/2008GM000838.
737
Moncrieff, M. W., and D. E. Waliser, 2015: Chapter15. Organized Convection and the YOTC
738
Project, Seamless Prediction of the Earth-System: From Minutes to Months, (G. Brunet, S
739
Jones, P.M. Ruti Eds.), WMO-No. 1156, ISBN 978-92-63-11156-2, Geneva, Switzerland,
740
http://library.wmo.int/pmb_ged/wmo_1156_en.pdf.
741
Moncrieff, M. W., D. E. Waliser, M. J. Miller, M. E. Shapiro, G. Asrar, and J. Caughey, 2012:
742
Multiscale convective organization and the YOTC Virtual Global Field Campaign. Bull. Amer.
743
MeteorMeteorol. Soc., 93,1171-1187, doi:10.1175/BAMS-D-11-00233.1.
744
Neena, J.M., J-Yi Lee, D. Waliser, B. Wang, and X. Jiang (2014), Predictability of the Madden
745
Julian Oscillation in the Intraseasonal Variability Hindcast Experiment (ISVHE), J. ofClimate,
746
27, 4531-4543.
747
Nesbitt, S. W., R. Cifelli, and S. Rutledge, 2006: A. Storm morphology and rainfall
748
characteristics of TRMM precipitation features. Mon. Wea. Rev. 134, 2702-2721.
749
Pan, D.-M. and D. A. Randall, D. A., 1998:. A cumulus parameterization with prognostic
750
closure. Quart, Q. J. Roy. MeteorMeteorol. Soc., 124, 949–981.
751
Pandya, R., Durran, D., 1996. The influence of convectively generated thermal forcing on the
752
mesoscale circulation around squall lines. J. Atmos. Sci. 53, 2924–2951.
753
Park, S., 2014: A Unified Convection Scheme (UNICON). Part I: Formulation. J. Atmos. Sci.,
754
71, 3902–3930.
36
755
Plant, R. S., and G. C. Craig, G. C., 2008:. A stochastic parameterization for deep convection
756
based on equilibrium statistics., J. Atmos. Sci., 65, 87-105.
757
Pritchard, M., M. W. Moncrieff, and R. C. J. Somerville, 2011: Orogenic propagating
758
precipitation systems over the US in a global climate model with embedded explicit convection.
759
J. Atmos. Sci., 68, 1821-1840.
760
Randall, D., M. Branson, M. Wang, S. Ghan, C. Craig, A. Gettelman, and J. Edwards, (2013:),
761
A community atmosphere model with superparameterized clouds, Eos Trans. AGU, 94,(25),
762
221–222..
763
Rasmussen, K. L., A. J. Hill, V. E. Toma, M. D. Zuluaga, P. J. Webster, and R. A. Houze, Jr.,
764
2015: Multiscale analysis of three consecutive years of anomalous flooding in Pakistan. Quart. J.
765
Roy. Meteor. Soc., 141, 1259–1276, doi:10.1002/qj.2433.
766
Romatschke, U., and R. A. Houze, Jr., 2010: Extreme summer convection in South America. J.
767
Climate, 23, 3761-3791.
768
Rowe, A. K., and R. A. Houze, Jr., 2015: Cloud organization and growth during the transition
769
from suppressed to active MJO conditions. J. Geophys. Res. Atmos., 120, 10,324–10,350,
770
doi:10.1002/2014JD022948.
771
Satoh, M., T. Matsuno, H. Tomita, H. Miura, T. Nasuno, and S. Iga, 2008: Nonhydrostatic
772
Icosahedral Atmospheric Model (NICAM) for global cloud resolving simulations. J. Comput.
773
Phys., 227, 3486-3514.
37
774
Schumacher, R.S., 2015: Sensitivity of precipitation accumulation in elevated convective
775
systems to small changes in low-level moisture. Journal of the Atmospheric Sciences, 72, 2507-
776
2524.
777
Srivastava, R.C., 1971. Size distribution of raindrops generated by their breakup and
778
coalescence. J. Atmos. Sci. 28, 410–415.
779
Srivastava, R.C., 1985. A simple model of evaporatively driven downdraft application to
780
microburst downdraft. J. Atmos. Sci. 42, 1004–1023.
781
Storer, R. L., B. M. Griffin, J.B. M., Höft, J. K.., Weber, E.J. K., Raut, V. E.., Larson, M. Wang,
782
and P. J. V. E., ... & Rasch, P. J. (2015:). Parameterizing deep convection using the assumed
783
probability density function method. Geoscientific Model Development, 8,(1), 1-19.
784
Thorpe, A. J., M. J. Miller, and M. W. Moncrieff, 1980: Dynamical models of two-dimensional
785
downdraughts. Quart. J. Roy. Metor. Soc., 106, 463-484.
786
Varble, A., A. M. Fridland, E. J. Zipser, A. S. Ackerman, J.-P. Chaboureau, J. Fan, A. Hill, S. A.
787
McFarlane, J.-P. Pinty, and B. Shipway, 2011: Evaluation of cloud-resolving model
788
Intercomparison simulations using TWP-ICE observations: Precipitation and cloud structure. J.
789
Geophys. Res., 116, doi: 10.1029/2010JD015180.
790
Varble, A.,J. Zipser, A. M. Fridlind, P. Zhu, A. S. Ackerman, J.-P. Chaboureau, S. Collis, J. Fan,
791
A. Hill, and B. Shipway, 2014: Evaluation of cloud-resolving and limited area model
792
intercomparison simulations using TWP-ICE observations: 1. Deep convective updraft
793
properties. J. Geophys. Res., 119, 13,891-13,918, doi:10.1002/2013JD021371.
38
794
Williams, M., and R. A. Houze, Jr., 1987: Satellite-observed characteristics of winter monsoon
795
cloud clusters. Mon. Wea. Rev., 115, 505-519.
796
Wing, A. A., and K. A. Emanuel, 2013: Physical mechanisms controlling self-aggregation of
797
convection in idealized numerical modeling simulations. J. Adv. Model. Earth. Syst., 5,
798
doi:10.1002/2013MS000269.
799
Wu, C.-M., and A. Arakawa, 2014: A Unified Representation of Deep Moist Convection in
800
Numerical Modeling of the Atmosphere. Part II. J. Atmos. Sci., 71, 2089–2103.
801
Xiao, H., W. I. Gustafson Jr., S. M. Hagos, C.-M. Wu, and H. Wan, (2015:), Resolution-
802
dependent behavior of subgrid-scale vertical transport in the Zhang-McFarlane convection
803
Parameterization., J. Adv. Model. Earth Syst., 7, 537–550, doi:10.1002/ 2014MS000356.
804
Yano, J.-I. and R. S. Plant, R. S., 2012:. Finite departure from convective quasi-equilibrium:
805
periodic cycle and discharge-recharge mechanism. Quart, Q. J. Roy. Meteor. Soc., 138, 626-637.
806
Yuan, J., and R. A. Houze, Jr., 2010: Global variability of mesoscale convective system anvil
807
structure from A-train satellite data. J. Climate, 23, 5864-5888.
808
Yuan, J., and R. A. Houze, Jr., 2013: Deep convective systems observed by A-Train in the tropical
809
Indo-Pacific region affected by the MJO. J. Atmos Sci., 70, 465–486.
810
Zhang, C., 2005: The Madden–Julian oscillation. Rev. Geophys., 43, RG2003,
811
doi:10.1029/2004RG000158.
812
Zhang, C., 2013: Madden-Julian Oscillation: Bridging weather and climate. Bull. Amer. Meteor.
813
Soc., 94, 1849-1870.
39
814
Zhang, G. J., and N. A. McFarlane, 1995: Sensitivity of climate simulations to the
815
parameterization of cumulus convection in the Canadian Climate Centre general circulation
816
model. Atmos-Ocean, 33, 407-446.
817
Zhang, X., W. Lin, and M. Zhang, (2007), Toward understanding the double Intertropical
818
Convergence Zone pathology in coupled ocean-atmosphere general circulation models., J.
819
Geophys. Res., 112, D12102, doi:10.1029/2006JD007878.
820
Zipser, E. J., and M. A. LeMone, 1980: Cumulonimbus vertical velocity events in GATE. Part
821
II: Synthesis and model core structure. J. Atmos. Sci., 37, 2458-2469.
822
Zipser, E. J., D. J. Cecil, C. Liu, S. W. Nesbitt, and D. P. Yorty, 2006: Where are the most
823
intense thunderstorms on earth? Bull. Amer. Meteor. Soc., 87, 1057–1071.
824
825
826
827
828
829
830
831
832
40
833
834
835
836
837
Appendix: Acronyms and Abbreviations
838
ACME
Accelerated Climate Model for Energy
839
AMIE
ARM MJO Investigation Experiment, Indian Ocean, 2011-12
840
ARM
Atmospheric Radiation Measurement
841
ASR
Atmospheric System Research program
842
CAPE
Convective Available Potential Energy
843
CAM
Community Atmospheric Model
844
COSP
Cloud Feedback Model Intercomparison Project (CFMIP) Observation
845
Simulator Package (COSP)
846
CMIP5
Coupled Model Inter-comparison Project Phase 5
847
CPM
Cloud Permitting Model
848
DOE
U.S. Department of Energy
849
DYNAMO
Dynamics of Madden-Julian Oscillation field campaign, Indian Ocean,
850
2011-2012
41
851
ENSO
El Niño Southern Oscillation
852
GATE
Global Atmospheric Research Program’s Atlantic Tropical Experiment, 1974
853
GCM
Global Climate Model
854
GFDL AM3
Geophysical Fluid Dynamics Laboratory Atmospheric Model 3
855
GoAmazon
Green Ocean Amazon Field Campaign, 2014-2015
856
IOP
Intensive Observing Period
857
ITCZ
Inter-tropical Convergence Zone
858
LES
Large Eddy Simulation
859
MCS
Mesoscale Convective System
860
MONEX
Monsoon Experiment, India and Malaysia, 1978-1979
861
PECAN
Plains Elevated Convection At Night, Central U. S., 2015
862
PNNL
Pacific Northwest National Laboratory
863
ROCORO
Routine Atmospheric Radiation Measurement (ARM) Aerial Facility (AAF)
864
Clouds with Low Optical Water Depths (CLOWD) Optical Radiative Observations (RACORO)
865
SGP
866
TRMM
867
868
Southern Great Plains, ARM Observational site in Oklahoma
Tropical Rainfall Measurement Mission, U.S./Japan satellite with radar and
radiometers for precipitation measurement, in orbit 1997-2014
RHI
Range Height Indicator, a radar display at constant elevation angle
42
869
SPA
870
TOGA-COARE Tropical Ocean—Global Atmosphere Coupled Ocean Atmosphere Response
871
872
Storm Penetrating Aircraft
Experiment, western tropical Pacific, 1992-1993
WRF
Weather Research and Forecasting Model
873
874
875
43
876
Figures
877
878
Figure 1. A photograph of convective clouds over Africa from the International Space Station
879
(photo credit NASA).
880
881
882
44
883
884
885
Figure 2. The core themes on which progress is required for accurate treatment of convection in
886
the next-generation global models.
887
45
888
889
890
Figure 3. (a) Diurnal cycle of June-July-August precipitation from observations and
891
CMIP5 models (Courtesy of Chengzhu Zhang from Lawrence Livermore National
892
Laboratory) and (b) the annual cycle of surface temperature at the location of ARM’s
893
Southern Great Plains site (Adapted from Zhang et al. 2016). The gray lines represent
894
individual CMIP5 models.
895
46
896
897
Figure 4. Variance of the MJO mode along the equator averaged between (a) 15°N and
898
15°S and (b) 5°N and 5°S (Adapted from Hung et al. 2013). The different line styles
899
represent different CMIP models.
47
900
901
902
48
903
Figure 5. (a) Annual cycle of all-India rainfall derived from satellite observations (black) and
904
from 20 CMIP5 models (blue) and (b) same but normalized by the annual mean precipitation.
905
The dashed red curve represents the multi-model mean.
906
49