Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Persistent Contrails and Contrail Cirrus. Part I: Large-Eddy Simulations fromInception to Demise
D. C. LEWELLEN, O. MEZA,* AND W. W. HUEBSCH
West Virginia University, Morgantown, West Virginia
(Manuscript received 2 October 2013, in final form 10 March 2014)
ABSTRACT
Large-eddy simulations with size-resolved microphysics are used to model persistent aircraft contrails and
contrail-induced cirrus from a fewwing spans behind the aircraft until their demise aftermany hours. Schemes
for dynamic local ice binning and updating coupled radiation dynamically as needed in individual columns
were developed for numerical efficiency, along with a scheme for maintaining realistic ambient turbulence
over long times. These capabilities are used to study some of the critical dynamics involved in contrail evo-
lution and to explore the simulation features required for adequate treatment of different components. A
‘‘quasi 3D’’ approach is identified as a useful approximation of the full dynamics, reducing the computation to
allow a larger parameter space to be studied. Ice crystal number loss involving competition between different
crystal sizes is found to be significant for both young contrails and aging contrail cirrus. As a consequence, the
sensitivity to the initial number of ice crystals in the contrail above a threshold is found to decrease signifi-
cantly over time, and uncertainties in the ice deposition coefficient andKelvin effect for ice crystals assume an
increased importance. Atmospheric turbulence is found to strongly influence contrail properties and lifetime
in some regimes. Water from fuel consumption is found to significantly reduce aircraft-wake-induced ice
crystal loss in colder contrails. Ice crystal shape effects, coupled radiation, and precipitation dynamics are also
considered. An extensive set of simulations exploring a large parameter space with this model are analyzed in
a companion paper.
1. Introduction
It is not uncommon for portions of the upper tropo-
sphere to be apparently cloud free (or containing only
thin cirrus) yet highly supersaturated with respect to ice
(e.g., Gierens et al. 2012). For such conditions ice clouds
seeded by the passage of aircraft (contrails) can persist
and grow into significant cloud cover that at late times
could easily be mistaken for natural cirrus (Minnis et al.
1998). Given the projected increases in air traffic in the
coming decades and potential impact of increased cloud
cover on climate, understanding the formation, prop-
erties, and effects of persistent contrails is of clear im-
portance (Penner et al. 1999). Despite recent progress in
evaluating global impact of contrails (Ponater et al.
2002; Rap et al. 2010; Frömming et al. 2011; Chen and
Gettelman 2013), contrails remain the most uncertain
component of aviation impact on climate (Lee et al.
2010).
Contrail evolution is a complex process dependent on
many properties of the aircraft and of the ambient at-
mosphere. Exploring these sensitivities involves dy-
namics on length scales ranging from nanometers for the
ice microphysics to kilometers for the late-time atmo-
spheric dispersion. This makes numerical simulation of
contrails challenging. A large number of simulations are
required to fully represent the phenomena given the size
of the parameter space, but with an a priori unknown
level of complexity required to make these representa-
tions suitably faithful.
Persistent contrail evolution involves the growth of ice
within an expanding wake-plume volume in an ice-
supersaturated environment. In the young contrail the
volume expansion is driven by aircraft-induced dynam-
ics: engine jets, trailing vortex interactions, and residual
Brunt–Väisälä oscillations set up by the wake-vortex
*Current affiliation: Inter American University of Puerto Rico,
Bayamon, Puerto Rico.
Corresponding author address: David C. Lewellen, Department
of Mechanical and Aerospace Engineering, P.O. Box 6106, West
Virginia University, Morgantown, WV 26506-6106.
E-mail: [email protected]
DECEMBER 2014 LEWELLEN ET AL . 4399
DOI: 10.1175/JAS-D-13-0316.1
� 2014 American Meteorological Society
descent. Effects with longer time scales take over lateron: atmospheric dispersion, ice sedimentation, andbuoyant forcing due to radiative or latent heating. Pre-vious simulation studies (e.g., Gierens 1996; Chlond
1998; Gierens and Jensen 1998; Jensen et al. 1998;
Khvorostyanov and Sassen 1998; Sussmann and Gierens
1999; Lewellen and Lewellen 2001a; Chen and Lin 2001;
Paoli et al. 2004; Paugam et al. 2010; Wong and Miake-
Lye 2010; Naiman et al. 2011; Unterstrasser and Sölch2010) have generally been limited to specific age regimes
of contrail evolution, typically with greatly simplified
ice microphysics and/or fluid dynamics and sampling
only small regions of the relevant parameter space.
In a more extensive study (Unterstrasser et al. 2008;
Unterstrasser and Gierens 2010a,b) contrails have been
simulated from ;20 s to several hours over a wider
sampling of parameter space, though with bulk ice mi-
crophysics and 2D, prescribed (rather than simulated)
wake dynamics, and ambient atmosphere. In another
approach, Schumann (2012) describes a contrail-cirrus
prediction model (CoCIP), designed to model the full
contrail life cycle but at a dramatically simplified level to
allow the modeling of large populations of contrails.
Here we describe a more comprehensive approach,
simulating contrails starting at 1-s age from realistic
aircraft conditions and following the evolution with 3D
fluid mechanics, size-resolved (‘‘binned’’) microphysics,
and coupled radiation through the wake and ambient-
turbulence-dominated regimes out to the demise of the
resulting contrail cirrus many hours later. Novel strate-
gies are used to improve numerical efficiency. These
capabilities are used to explore some of the critical dy-
namics involved in contrail evolution and to assess what
is required for adequate simulation of different facets—
for example, whether detailed initial conditions, 3D fluid
mechanics, binned microphysics, and/or realistic ambi-
ent atmospheric motions are required to reproduce
contrail-averaged quantities. A ‘‘quasi 3D’’ simulation
mode is established as a useful approximation, which
makes practical the coverage of large regions of pa-
rameter space with the simulations.
This paper describes the model and highlights some
general features of the simulation results that have not
appeared elsewhere.A companion paper (Lewellen 2014,
hereafter Part II) analyzes a large set of full-lifetime
simulations for a single aircraft including significant
shear and coupled radiation, emphasizing late-time
and lifetime-integrated behavior. An extensive para-
metric study focused on young contrail behavior will be
presented elsewhere, including different aircraft and
physically based parameterization of the results. A sig-
nificant and surprising result from the present work is
the importance in both young and old contrails of ice
crystal loss involving competition between different
crystal sizes in the ice spectrum. An analytic treatment
of this effect along with a more general analysis of dif-
ferent regimes of ice-size spectra development has been
presented in Lewellen (2012).
Themodel and simulation procedures are described in
section 2, with further details given in the appendix.
Sample simulation results are given in section 3. Some
key physical results are discussed at more length in
section 4, followed by concluding remarks in section 5.
2. Contrail model
a. Fluid mechanics and earlier model
The starting point for the present work was the large-
eddy-simulation (LES) model used for studies of wake
dynamics, wake-plume chemistry, and contrails de-
scribed in Lewellen and Lewellen (1996, 2001a,b); only
a brief summary is given here. A 3D finite-difference
implementation of the incompressible Navier–Stokes
equations is employed, second-order accurate in space
and time, with a direct solver for the pressure and
a variable time step. The piecewise parabolic method
(PPM) is used for advection of the liquid-ice potential
temperature Qli, total water mixing ratio qt, ice number
densities, and passive species concentrations. The sub-
grid model uses a quasi-equilibrium second-order tur-
bulence closure scheme with a prognostic equation for
the subgrid turbulence kinetic energy (TKE) and di-
agnostic equation for the subgrid turbulence length scale
L. The physical damping of turbulence by stable tem-
perature or rotation gradients is included by appropriate
reductions in L. The latter helps prevent unphysically
large subgrid diffusion in vortex cores if they are only
modestly resolved and was developed and used exten-
sively for tornado simulations (Lewellen et al. 2000;
Lewellen and Lewellen 2007).
The aircraft wake–contrail is initialized as an in-
stantaneous line source in the downstream (y) direction.
There are competing needs of relatively large domains
in the cross-stream (x) and vertical (z) directions to
handle sheared and precipitating contrails and fine grid
resolution to treat both the wake-vortex cores and spe-
cies dispersal from the exhaust jets. To meet these effi-
ciently we employ 1) stretched grids in x and z to
concentrate fine grid spacing where it is needed, 2) ver-
tical and cross-stream grid tracking to keep the falling
vortices or drifting contrail within the fine grid region,
and 3) different grids and domains for different stages
of the evolution as the needs change. The x and y
boundary conditions are chosen periodic. TheBoussinesq
approximation and periodic boundary conditions
4400 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
in z are used for early simulation segments on smaller
domains (with imposed shifts to account for mean
temperature or velocity gradients); the anelastic ap-
proximation and a closed domain in z are used for later
segments on deep domains.
Themain limitations of the contrail studies of Lewellen
and Lewellen (2001a) were the use of simple bulk ice
microphysics and omission of coupled radiation or any
means of maintaining ambient atmospheric turbulence.
We describe the model improvements to surmount
these limitations below, with further details given in the
appendix.
b. Ice microphysics with dynamic local binning
Binned microphysics schemes are more numerically
costly than their bulk counterparts. For a given appli-
cation, however, there is often no definitive way to test
the adequacy of a bulk scheme other than to compare its
results with those of a size-resolved scheme. This moti-
vated the development of the numerically more efficient
bin-resolved scheme described here. The main LES
code handles the advection, diffusion, and sedimenta-
tion of ice species. For ice crystal diffusional growth and
sublimation we adapted subroutines from the National
Aeronautics and Space Administration (NASA) Ames
Community Aerosol and Radiation Model for Atmo-
spheres (CARMA; e.g., Jensen et al. 1994; Ackerman
et al. 1995). For later reference, the basic equation em-
ployed for the growth of a crystal of massm is (see, e.g.,
Pruppacher and Klett 1997)
dm
dt5
4pCfy(11 dSi 2 eak/d)"L2s
KRyT2 1
RyT
ei(T)D*
# , (1)
where dSi is the supersaturation, ei is the vapor pressure
with respect to ice,Ls is the latent heat of sublimation,Ry
and D* are the gas constant and diffusivity for water va-
por (the latter modified by kinetic theory effects for small
crystals),K is the thermal diffusivity, and fy is a ventilation
coefficient (51 for small spheres). The C and d are the
crystal capacitance and characteristic length, both taken
equal to the radius r for spherical crystals, and ak 5 2si–y/
(riRyT) is taken in analogy with the Kelvin equation for
droplets but using the surface tension for an ice–water
vapor interface si–y, and ice density ri. The growth rate
varies for different sized crystals, via the C factor and
Kelvin term, and the size spectrum naturally develops
over time, in contrast with a bulk model where the de-
velopment of only one or more moments of the spectrum
are followed. For example, in the simple bulk model used
in Lewellen and Lewellen (2001a), the expression
analogous to (1) is for the time evolution of the average
mass in the local spectrum, m, ak is set to 1, and Cfy is
replaced by a parameterization of the form amb chosen to
try to incorporate some of the spectral and shape effects.
In contrail evolution the range of crystal sizes that is
significantly populated varies greatly with location and
time (e.g., a spectrum of small crystals in the cores of
young contrails versus a spectrum of larger crystals in
precipitation streamers). The number of ice bins re-
quired locally is typically a modest fraction of the global
requirements for the full contrail evolution. Accordingly
a dynamic local ice binning, as shown in Fig. 1, was
employed to reduce the computational costs. For each
grid cell only the local ice spectrum and thermodynamic
variables are passed to the CARMA routines at each
time step to update the ice spectrum. In each grid cell the
local bin space is dynamically shifted up or down within
the global space during the simulation as needed for best
following the evolving crystal spectra. Details of the
implementation are given in the appendix.
By default we assume spherical ice crystals, but some
effects of crystal shapes have been explored as will be
discussed later. Pressure perturbations due to fluid mo-
tions are accounted for in computing local saturation
conditions. These have a small effect within the trailing
vortex cores but are negligible elsewhere. Buoyancy due
to latent heat release is included; fragmentation and
aggregation of ice crystals are not (as the latter is
expected to be very small at contrail temperatures; see,
e.g., Connolly et al. 2012).We have developed a new and
numerically efficient scheme for including spectrally
resolved homogeneous freezing nucleation but defer its
description elsewhere; this process is not generally en-
countered during contrail evolution unless a persistent
updraft is imposed, and so is not included in the simu-
lations considered here. The crystals in this work are
treated as pure ice; that is, the aerosol core options in
CARMA are not utilized here.
FIG. 1. Dynamic local binning schematic. A global space of ng ice
mass bins defined with fixed mass ratio between neighboring bins
(mj11/mj5Rm) such that the full range of ice crystal mass expected
during a contrail’s evolution is encompassed. In the simulation,
only a subset of this bin space with nl, ng bins is employedwith the
choice, specified by the starting index within the global space, al-
lowed to vary with space and time [is 5 is(x, t)] so as to best rep-
resent the local evolving ice-size spectrum. The overall CPU and
memory needs are effectively reduced by the ratio nl/ngwith no loss
in simulation accuracy.
DECEMBER 2014 LEWELLEN ET AL . 4401
c. Coupled radiation
Radiation processes can play an important role in ice-
cloud evolution (e.g., Dobbie and Jonas 2001). The ef-
fects of radiative heating or cooling of the contrail have
been included in the LES by adapting the two-stream
radiation routine from CARMA, which is spectrally
resolved, using 26 solar and 18 infrared wavelength bins.
The work of Gounou and Hogan (2007) comparing
a fully 3D radiative transfer model or the independent
column approximation (ICA) applied to idealized con-
trails suggests that the ICA is not an unreasonable
choice for contrail modeling, where the concern is the
effect of radiation on the contrail dynamics itself
(though for computing the radiative forcing of the con-
trail on Earth, the uncertainty is likely greater; Forster
et al. 2012). Taking advantage of the ICA, we choose
the frequency for updating the radiative heating rates
in each vertical column independently, dynamically
adjusting each as is needed during the evolution. In
implementing this scheme, the modification of crystal-
vapor exchange rates due to direct radiative heating of
crystals are not included (since themicrophysics updates
occur more frequently); however, these contributions
are minor corrections for crystal sizes of a few tens of
microns or less (Gierens 1994) and so should not sig-
nificantly impact the contrail dynamics. Further details
are discussed in the appendix. Given that the contrail is
generally optically thin, localized spatially, and that the
radiative heating rates typically change more slowly
than other dynamics in the simulation, this scheme
dramatically reduces the CPU requirements without
compromising the accuracy of the computation. In-
clusion of both coupled longwave and shortwave radi-
ation now generally increases the simulation time over
the same case with no radiation by 20% at most, com-
pared with an overall order of magnitude increase if the
radiation update is computed everywhere at each time
step.
d. Late-time turbulence maintenance
Our treatment of ambient turbulence for young con-
trails was discussed in Lewellen and Lewellen (1996) and
Lewellen et al. (1998). Beyond a few Brunt–Väisälä pe-riods aircraft wake-induced dynamics die away leavingambient turbulence, wind shear, and large-scale uplift/subsidence as primary drivers of contrail evolution. Themost relevant conditions are stably stratified with a mix-ture of low-amplitude gravity waves and turbulence sinceaircraft avoid regions of strong turbulence for safety. Theturbulence levels are expected to be lower than generallyfound in natural cirrus since the latter usually requires anactive source of uplift for its nucleation whereas contrails
do not. The generation, maintenance, and properties ofturbulence in the upper troposphere and lower strato-sphere are complex and not well understood (see, e.g.,Quante and Starr 2002). Accurately simulating the in-
teraction of gravity waves producing intermittent patches
of turbulence with LES would require both very large
domains (for the spectrum of traveling waves) and rela-
tively fine grids (for resolving turbulence generation in
breaking waves); however, for useful sensitivity studies
we assume that it is sufficient to simulate quasi-steady
ambient turbulence with realistic, statistically re-
producible behavior across the range of conditions likely
to be encountered at cruise flight level, regardless of how
the motions are initially generated or maintained.
Unstratified homogeneous isotropic turbulence is
completely categorized statistically given the turbulent
dissipation rate. This is not true for the stratified case
where the mix of waves and turbulence, along with the
energy transfer rate, can differ significantly on different
scales and in time. Moreover, in this case there can be
significant energy loss through wave transport out of the
domain or temperature mixing within. The parameter
space is multidimensional, including at least stratifica-
tion, mean shear, gravity wave spectrum, turbulence
strength on different length scales, and level of in-
termittency. To provide representative solutions we
perform a separate LES of decaying turbulence to pro-
duce velocity perturbation fields that are then added
periodically to the contrail LES. Varying the amplitude
of the initial temperature perturbations Tamp, the age of
the added field fage, and time between additions taddalong with the mean shear and stratification allow dif-
ferent quasi-steady turbulence states to be achieved.
The fields produced have a realistic patchy character as
expected since in between regenerative additions they
represent LES Navier–Stokes solutions for the desired
stratification and shear. This is in contrast to the scheme
employed in Paugam et al. (2010), where velocities from
a simulation with an unstable Richardson number are
imposed on the desired stable conditions. Further details
and sample results are given in the appendix. Including
turbulence regeneration is less important if either cou-
pled radiation or mean wind shear drive more significant
levels of turbulent diffusion in and around the contrail.
e. Quasi-3D simulation
For exploring some sensitivities, a full 3D LES should
not be required. A caricature of themost important fluid
motions—the strong downward motion of the wake
vortices and the expansion of the plume through mixing
with the ambient environment—should suffice. For this
purpose we conduct some simulations with only a mini-
mal number of downstream grid points (typically six
4402 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
independent points plus two additional for imposing
periodic boundary conditions). On larger scales the
simulation is effectively 2D, but 3D eddies are crudely
resolved in a narrow spatial range of scales determined
by the downstream resolution (and hence the choice
of downstream domain size). This 3D eddy scale was
roughly chosen in the range expected to be most
important—on the order of a fraction of the vortex spacing
for the wake-dominated contrail and of the contrail
depth (before precipitation) for older contrails—with
the results proving to be rather insensitive to the precise
choice. The dramatic reduction in memory and CPU
time requirements of this quasi-3D approximation
(Q3D) allows a much larger parameter space to be ex-
plored as well as the ability to test bin or grid resolutions
that might be prohibitive in 3D. Because of the small
downstream extent, there is a larger spread in results
between different turbulent realizations than arises with
full 3D simulations (cf. Fig. 9), so averaging over a small
ensemble of Q3D simulations is sometimes desirable;
this is not an issue for studying sensitivities that do not
strongly affect the fluid dynamics (e.g., changing ice
microphysics).
As will be shown below, the Q3D approximation
proves to be a much better physical approximation to
the full dynamics than one would expect a priori. It is
sometimes true that only a limited subset of degrees of
freedom need be included to represent the physics that
determines some averaged quantities. A discussion of
why that might be the case here is given in section 4a.
The success of such a reduced treatment is not entirely
unexpected—we have seen similar in reproducing some
mean aspects of atmospheric boundary layer behavior
using LES with greatly restricted degrees of freedom
(e.g., Lewellen and Lewellen 1998)—but ultimately the
utility in any given case can be confirmed only by com-
parison with fully 3D simulations.
f. Initialization and modeling choices
The initialization, modeling, and grid choices are
partly aircraft specific. Those used here for a B-767 with
wingspan of 47.25m, flight speed 237m s21 at altitude of
10.7 km, total circulation in the wake of 391m2 s21, fuel
flow 5.78 kg km21, specific combustion heat 43MJkg21,
and assumed propulsion efficiency of 30% are given for
illustration. Table 1 lists the domains and grid resolu-
tions used for different time periods. The simulations
are initialized from 2D Boeing wake rollup calculations
(Czech et al. 2005) in two steps: an initial high-resolution
short-domain run without ice for the early turbulent
spread of the exhaust jet followed by the introduction of
ice crystals distributed spatially like an exhaust tracer
species. Only cases satisfying theAppelman criterion for
contrail formation (see, e.g., Schumann 1996) are con-
sidered. It is expected that essentially all of the contrail
ice crystals are produced during the initial rapid mixing
and cooling of the exhaust jets. Accordingly, the total
crystal number is set consistent with the fuel flow rate
and an assumed ice crystal number emission index
EIiceno. This quantity is effectively an aircraft variable in
our treatment.We consider variations in the simulations
around a nominal value of 1015 particles per kilogram
of fuel, consistent with in situ aerosol measurements
(Schröder et al. 2000; Anderson et al. 1998) and jet-
plume ice nucleation modeling (Kärcher and Yu 2009).Delayed nucleation tied to the contrail is expected to be
of secondary importance at most. In the wake-dominated
regime vertical displacements are almost entirely down-
ward (relative to the ambient atmosphere at rest). At
late times radiative heating of the contrail can produce
sustained updrafts, but supersaturations are depleted by
the existing contrail itself, limiting the possibilities for
new ice nucleation [a result supported by the simulation
tests of Unterstrasser and Gierens (2010b)]. Further
details on simulation initialization and some sensitivity
tests are given in the appendix.
3. Contrail evolution from seconds to hours
Contrails exhibit a wide range of behavior, as even
casual observation of the sky under different conditions
will attest. The primary sets of 3D simulations considered
here are summarized in Table 2. Temperature T and rel-
ative humidity with respect to ice (RHi) are the two most
important ambient variables governing contrail evolu-
tion. Persistent contrails are found within sizable ranges
of both (Iwabuchi et al. 2012). To best illustrate different
dynamical elements that arise in different regimes we
highlight in particular ‘‘warm’’ cases with T 5 225K
(toward the upper end of the observed range) and ‘‘cold’’
cases withT5 205K (toward the low end of the observed
range) along with a more typical temperature for cruise
altitude of 218K. Note that for fixed RHi warm and cold
also correspond respectively to ‘‘relatively moist’’ versus
TABLE 1. Grid and ice binning parameters for 3D simulations.
Grid
Time
period
Domain
(x 3 y 3 z km3)
Fine grid spacing
(Dx 3 Dy 3 Dz m3)
1 0–2 s 0.84 3 0.052 3 1 0.4 3 0.4 3 0.4
2 2–90 s 0.84 3 0.26 3 1 0.6 3 2. 30.6
3 90–300 s 0.84 3 0.26 3 1 1.25 3 1.6 3 1.25
4 5–15min 1.52 3 1.04 3 1.6 5 3 4.8 3 5.
5 15–90min 6 3 2.08 3 4 16 3 16 3 16
6 1.5–12 h 12 3 4.16 3 4 25 3 28 3 25
Ice bins 48 global, 20 local
50-nm–2.6-mm radius; mass ratio 2
DECEMBER 2014 LEWELLEN ET AL . 4403
‘‘relatively dry.’’ Both higher (130%) and lower (110%)
values of RHiwithin the observed rangewere considered.
In the naming convention used for the simulations the
first number indicates RHi, the second number indicates
T,T and the trailing letter indicates the ambient-turbulence
level. In some cases multiple turbulent realizations of
a given simulationwere performed by choosing different
random number seeds governing the perturbations for
wake and ambient-turbulence initializations; this can
give some indication of the statistical uncertainty in the
results. The qualitative behavior from these sample ca-
ses is representative of that we have seen in much larger
sets of simulations (e.g., in Part II).
For each simulation the results constitute a five-
dimensional space: time-evolving ice-size spectra in
three spatial dimensions. To assess the effects of dif-
ferent physical parameters or modeling choices it is
convenient to distill this to a few simple metrics. For this
purpose we choose the time-evolving total (i.e., in-
tegrated over the x and z directions, averaged over y) ice
crystal numberN(t), massM(t), and surface area S(t) per
meter of flight path (Figs. 2 and 3) and the contrail-
averaged ice crystal size spectrum at selected times. In
the absence of crystal growth/sublimation, pure fluid
dynamic motions would leave these measures invariant.
It is natural to characterize the radiative effects of cloud
layers in terms of an optical depth, but this is an am-
biguous measure for contrails given their limited hori-
zontal extent and nonuniform cross section. The
quantity S(t) provides an approximate measure of op-
tical depth integrated for the contrail as a whole,
unchanged by these ambiguities. It is approximately
twice the total extinction E defined in Unterstrasser and
Gierens (2010a,b). For a given orientation and width
definition one can recover a mean optical depth from S
in the limit of approximating the extinction efficiency as
independent of crystal size, Qext ’ 2; for example,
t51
WDy
ðDy0
dy
ðdx t
51
WDy
ðDy0
dy
ðdx
ðdz�
jpr 2j NjQ
jext’ S/2W (2)
for optical depth t on a vertical path, contrail width W,
and crystal number densities Nj for crystal radius rj. We
consider only properties of the contrail itself; to de-
termine the radiative impact of the contrail on Earth’s
surface, one would need to fold in information about the
incoming radiation and environments above and below
the contrail as well.
Figures 4–6 illustrate some of the contrail spatial
evolution. The main events of the wake-dominated re-
gime (cf. Fig. 5)—exhaust entrainment–detrainment
into the trailing vortex system, vortex fall, linking, ring
dynamics and decay, and Brunt–Väisälä oscillation ofthe wake–—affect the contrail appearance in the same
ways as was described when simulating with bulk mi-
crophysics (Lewellen and Lewellen 2001a) or, for many
facets, just a passive exhaust tracer (Lewellen and
Lewellen 1996). We emphasize here structure evolution
at later times and changes that depend on spectrally
resolved microphysics.
For much of the evolution most of the contrail volume
is driven close to ice saturation because crystal number
densities are typically large in young contrails and time
scales long in older contrails. As a consequence, for
conditions with constant ambient RHi, saturation mix-
ing ratio qs, and air density r,M(t) scales essentially with
the growth of the plume cross-sectional area, A(t):
M(t)’ (RHi2 1)qsrA(t) . (3)
At very low temperatures the water from the aircraft
exhausts becomes competitive in the young contrail and
should be added to (3). Until precipitation becomes
important, the atmospheric pressure p enters the dy-
namics significantly only through qs and r but drops out
in the combination in (3); the insensitivity to p that this
predicts for young contrail dynamics was confirmed in
simulation tests.
The value of M(t) increases with plume growth dom-
inantly driven by exhaust jet mixing (until ;10 s de-
pending on aircraft), vortex fall (;300 s depending on
TABLE 2. Ambient conditions at flight level for principal 3D
simulations. Three further variations of the simulations with as-
terisks were performed with coupled longwave radiation, varying
the upwelling blackbody radiation temperature for each caseTsfc5258, 278, and 298K. The low and moderate ambient-turbulence
levels are quantified in Fig. A1. Unless noted otherwise, we assume
an air pressure of 238.5 hPa, EIiceno 5 1015 kg21, potential tem-
perature gradient of 2.5Kkm21, and nomean uplift, subsidence, or
wind shear (though the ambient-turbulence fields do introduce
such motions locally). To illustrate precipitation dynamics, we
consider deep supersaturated layers with initial profiles with con-
stant RHi above flight level down to 1 km below, linearly dropping
to RHi 5 50% at 1.5 km below flight level and below.
Simulation RHi (%) T (K) Turbulence
S30T225m 130 225 Moderate
S30T225l 130 225 Low
S10T225m* 110 225 Moderate
S30T218m 130 218 Moderate
S10T218m* 110 218 Moderate
S30T205m 130 205 Moderate
S10T205m 110 205 Moderate
S10T205l 110 205 Low
4404 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
aircraft and stratification), Brunt–Väisälä oscillation(;20min depending on stratification), ambient turbu-
lence, and mean wind shear (which is not included in the
simulations here but considered at length in Part II).
Eventually crystals around the plume edges become
large enough that sedimentation velocities exceed tur-
bulent mixing velocities and precipitation becomes the
dominant means of bringing ice crystals into contact
with supersaturated air. The resultant growth in ice
crystals and increase in fall velocities provides a positive
feedback—essentially a ‘‘precipitation instability.’’ A
rapid increase in vertical extent of the contrail ensues
[(e.g., from;(40 to 90) min in Figs. 4 and 6] until the full
depth of the supersaturated layer is reached. Eventually
the total ice mass falls as ice is lost to subsaturated
regions below. Often a rough equilibrium is established
between new ice growth on the contrail edges (driven
by turbulent mixing) and precipitative ice loss within the
older parts in the center. The precipitation growth of
the contrail occurs much sooner in the warm cases than
the cold. Consideration of a large set of simulations (Part
II) shows that the time scale for this initial growth of the
contrail precipitation plume downward is essentially that
required for a single crystal growing in the given ambient
conditions to fall through the supersaturated-layer depth,
depending strongly on T and RHi but much less on
EIiceno, turbulence level, shear, or coupled radiation. At
late-enough times (not reached in all of the simulations),
the entire remaining plume precipitates downward and
N(t) and S(t) drop off rapidly (cf. Figs. 3 and 6).
Ambient-turbulence levels or coupled radiation can
affect the contrail evolution significantly over time
scales longer than ;30min (Figs. 2 and 3). Increased
turbulence increases plume dispersion and with it M(t)
and S(t). In the absence of precipitation the growth seen
in the simulations is roughly a power law,M(t); t0.8 or t1
for the ‘‘low’’ and ‘‘moderate’’ turbulence, respectively.
This is consistent with the rough ;t0.8 dilution rate for
exhaust species out to a few hours found from in situ
measurements by Schumann et al. (1998). Including
coupled radiation can either increase or decrease the
FIG. 2. Time histories of total (a) ice crystal number, (b) icemass,
and (c) ice surface area per meter of flight path from sample 3D
simulations. Line colors indicate different combinations of RHi
and T, and dashed lines indicate different ambient-turbulence
levels as follows: S30T225m (black solid lines, three different tur-
bulent realizations included), S30T225l (black dashed), S10T225m
(green), S30T205m (blue), S10T205m (red solid), and S30T205l
(red dashed).
FIG. 3. Time histories of total (a) ice crystal number and (b) ice
surface area per meter of flight path from cases S10T225m with
varying coupled radiation: none (dashed line), coupled longwave
with Tsfc 5 258, 278, 298K (thin to thick solid lines, respectively).
DECEMBER 2014 LEWELLEN ET AL . 4405
ice crystal number, ice surface area, and contrail lifetime
(e.g., Fig. 3), depending on the regime. Different regimes
of coupled radiation behavior are explored in Part II.
Early in the evolution the wavelength of the most
prominent downstream periodicity in the ice distribu-
tion (e.g., Fig. 5) is set by the Crow instability (Lewellen
and Lewellen 1996). In the simulations there is a natural
clumping that occurs so that the dominant wavelength
tends to increase in time. If conditions are such as to
foster more rapid precipitation, these structures will be
directly reflected in the precipitation patterns as well.
Since the lowest precipitation streamers originate from
FIG. 4. Mean vertical profiles (z5 0, indicating flight level) of (a) engine tracer concentration, (b) ice crystal number, (c) ice mass, and
(d) ice surface area, per meter of flight path from simulation S10T225m. Times shown are 10, 90, 300, 900, and 1800 s; and 1, 1.5, 3, 6, 9, and
12 h (colors progressing through the spectrum from red to blue).
FIG. 5. Drift plot of cross-stream-integrated ice surface area showing young contrail development for cases (top) S10T205m and
(bottom) S10T218m. The horizontal axis sweeps over both time and downstream distance (drift velocity 5 12m s21) to illustrate both
temporal and spatial evolution. The plot format as described in Lewellen et al. (1998) is motivated by scanning lidar measurements.
4406 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
the younger contrail at flight level, and subsequent
streamers precipitate from two sides of the parent pop-
ulation, there is a natural tendency for the precipitation
streamers to exhibit a shorter periodicity than the ice
mass distribution at flight level for the mature contrail.
This has been reported in some contrail observations
with large ice mass (Atlas et al. 2006). Over time am-
bient turbulence, waves, and shear can destroy all rem-
nants of the periodic structure inherited from the young
contrail.
In the absence of ice nucleation, N(t) monotonically
decreases. Themost dramatic losses are tied to adiabatic
heating in the falling trailing vortex system producing
subsaturated conditions (Lewellen and Lewellen 2001a;
Sussmann and Gierens 1999), but there are also signifi-
cant number losses during this period and throughout
the later evolution (discussed more below) that do not
depend on the local qt falling below the ice saturation
mixing ratio qs as well as losses at later times through
precipitation. Initially the spatial evolution of ice crystal
number parallels that of a passive exhaust tracer (e.g.,
Fig. 4)—driven by advection and diffusion. The differ-
ences in these distributions grow in time because of
sedimentation and evaporative loss of ice. The spatial
distribution of crystal number is also seen to be far less
uniform during the contrail evolution than that of the ice
mass, with the uniformity of the ice surface area distri-
bution lying in between. In particular the number den-
sities in the precipitation streamers are typically orders
of magnitude less than those near flight level for most
of the evolution. In the contrail-averaged size spectra
(Fig. 7) the precipitation plume contribution adds only
a small shoulder to the number distributions; nonethe-
less this sometimes dominates the ice mass (cf. Fig. 7c).
The ice crystal size spectrum shifts to larger sizes in
time as crystals are lost. There is often a long period in
which the spectral shape (displayed logarithmically) is
nearly constant, just changing in magnitude and mean
crystal size (cf. Fig. 7a); the dynamics responsible are
discussed more below. The local spectra differ signifi-
cantly in different spatial regions: around the original
flight level small crystals and high number densities
dominate, at the plume edges larger crystals arise, and in
the precipitation plume the crystal size progressively
increases, and number densities decrease, with distance
from flight level. These differences between core and
FIG. 6. Drift plot of (top) vertically integrated and (bottom) cross-stream-integrated ice surface area showing long-time contrail de-
velopment for case S10T225m with coupled longwave radiation with Tsfc 5 298K. The drift format is as in Fig. 5, but over much longer
times with a slower drift velocity (2m s21).
DECEMBER 2014 LEWELLEN ET AL . 4407
precipitation regions are well known already from in situ
measurements (e.g., Heymsfield et al. 1998; Lawson
et al. 1998; Atlas et al. 2006) and 2D simulations (Jensen
et al. 1998; Unterstrasser and Gierens 2010b); the latter
have demonstrated as well that the vapor pressure is
driven near ice saturation in the contrail core as is found
here. The evolution of the spectral widths in Fig. 7 are in
at least rough agreement with the measurements within
contrails at different times of Schröder et al. (2000),though differences in (and incomplete knowledge of)
the ambient conditions in the measurements precludes
a more quantitative comparison.
Idealized simulations with some physics turned off
help to isolate the different physical mechanisms influ-
encing each contrail metric. Figure 8 provides an ex-
ample. Among the four simulations only two distinct
behaviors ofN(t) arise, depending onwhether theKelvin
term in the microphysics is included or not, and only two
behaviors of M(t), depending on whether or not
precipitation is allowed. The crystal size spectra are
paired depending on the former effect for smaller size
crystals but on the latter for larger. Of the two effects
precipitation dominates the influence on the large
crystal spectra and M(t) evolution, while the Kelvin
effect strongly influences the small crystal spectrum and
N(t) evolution. The surprising importance of the Kelvin
effect even at late times is discussed below.
4. Further discussion
a. Quasi-3D simulation
Contrails exhibit significant 3D structures at different
ages that are readily apparent in the simulations (e.g.,
Figs. 5 and 6). Turbulence in the exhaust jets leads to
FIG. 7. Evolving spectra of ice crystal numbers vs crystal radius
from (a) S10T205l and (b) S30T225m. (c) Spectra of total icemass for
S30T225m. The times and line colors are as in Fig. 4.
FIG. 8. Sample results from Q3D simulations for RHi 5 130%,
T 5 218K, and low ambient turbulence with precipitation present
(solid lines) or turned off (dashed lines), and Kelvin effect in the
microphysics present (thin red lines) or turned off (thick black
lines). (a) Ice crystal number, (b) ice mass, and (c) ice-size distri-
bution at contrail age 5 h.
4408 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
small-scale downstream clumping, the Crow instability
produces characteristic periodic downstream puffs with
spacing of a few wingspans, these in turn influence
subsequent patterns of precipitation streamers, and
ambient wind shear and wave fields modulate different
patterns in older contrails and contrail cirrus. Which if
any of these 3D dynamics significantly affect contrail-
averagedmetrics? Figure 9 compares the cold and warm
B-767 simulations from an ensemble of turbulent re-
alizations of Q3D simulations for the same conditions.
The agreement is quite respectable and is representative
of results we have found when comparing Q3D simu-
lation with 3D across the range of conditions we have
considered. In contrast, a 2D simulation run within the
LES (not shown) compares poorly. The Q3D simula-
tions differ from the 2D in effectively allowing one scale
of 3D motions to be crudely resolved, chosen on the
order of a fraction of the vortex spacing for the wake-
dominated contrail and of the contrail depth (before
precipitation) for older contrails. The Q3D simulation
excludes 3D eddies on larger scales. The wake dynamics
and exhaust dispersion for the first tens of seconds do
not critically involve suchmodes, nor does aging contrail
cirrus (where shear typically dominates the large-scale
dispersion). The 3D component omitted from Q3D
simulations that plausibly is of greatest quantitative
importance is the wake-vortex dynamics at intermediate
times. The depth reached by the wake-vortex system
affects the fraction of ice crystals surviving the wake
regime and the later contrail spread (because sub-
sequent vertical growth is inhibited by stratification). A
priori the Crow instability dynamics would seem criti-
cally involved in setting the average depth, but the
success of the Q3D approximation over a large param-
eter space suggests otherwise. The results are consistent
with the average vertical extent being determined by the
balance between the downward wake momentum and
the positive buoyancy acquired in falling through the
ambient stratification, together with smaller-scale en-
trainment (detrainment) into (out of) the vortex system.
Both Q3D and 3D simulations represent these elements
in approximately the same way. Fully 3D simulations
permit reconnection and reorientation of vortex seg-
ments, which affect the contrail appearance greatly (e.g.,
Fig. 5) but its average properties to a much lesser extent.
Mixing scales smaller than the vortex spacing are im-
portant, however. Different 2D treatments of the de-
scent of a wake-vortex in a stably stratified atmosphere
can give widely varying predictions for the evolution
[see, e.g., Table 1 of Widnall (1975)] depending largely
on the entrainment–detrainment assumptions made. In
our 2D simulations (not presented here) the mixing into
and out of the falling vortex system is underestimated
(the subgrid model being designed for 3D rather than
2D applications). This leads to a slowly decreasing
spacing of the vortex pair, a downward acceleration, and
a greatly overestimated total vertical extent. The Q3D
simulations develop some resolved smaller-scale mixing
leading to an essentially similar trajectory of descent of
the vortex system as that found in the 3D simulations.
b. Ice crystal number loss
Figure 8 shows that there are crystal losses dependent
on the Kelvin effect that persist even to ages of several
hours. This component, which we dub ‘‘in situ loss,’’ can
occur even without conditions becoming subsaturated
and can be the dominant loss mechanism at late times
until precipitation into subsaturated layers wins out.
Figure 10 shows that in situ loss is critical in the wake-
dominated regime as well. Crystal-loss fractions are
significant even for large RHi and greatly exceed the
losses found with the Kelvin effect turned off or with the
bulk microphysics of Lewellen and Lewellen (2001a).
This irreversible loss leads to significant differences in
contrail properties that persist in time.
The in situ loss rates and the shape of the ice crystal
spectrum that results are derived analytically from the
governing equations for diffusional growth of spheri-
cal crystals in Lewellen (2012); here we limit ourselves
FIG. 9. Comparison for cases S10T205l and S30T225m of 3D
(dashed lines, threedifferent turbulent realizations for S30T225mcase)
and quasi-3D (solid lines, four different turbulent realizations for each
case) simulation results. Time histories of total (a) ice crystal number
and (b) ice mass.
DECEMBER 2014 LEWELLEN ET AL . 4409
to summarizing the basic physical mechanism. For
small spherical crystals, surface energy (Kelvin) effects
shift the critical vapor supersaturation level that sepa-
rates sublimation from growth away from zero to
a value depending on crystal radius, 12 eak/r [cf. (1)].
Given that ak; 23 1023mm is much less than themean
crystal sizes encountered in the contrail, the magnitude
of the Kelvin-dependent effect seen in the figures is at
first surprising. Its origin is in the local competition
between different sized crystals together with dynamics
forcing the system close to saturation. In the young
descending contrail adiabatic heating forces the water
vapor mixing ratio qy below qs but the high number
densities ensure that conditions never get far from
equilibrium (the simulations rarely show vapor satu-
ration levels in these regions below 97% and generally
above 99%). Even the small differences due to the
Kelvin effect for modest crystal sizes are relevant in
comparison to the small levels of subsaturation main-
tained in most regions experiencing crystal loss. This
broadens the size spectrum leading to more rapid
crystal loss.
At late times conditions are naturally driven toward
equilibrium (qy / qs) so they appear slightly super-
saturated to the larger crystals but, because of the
Kelvin effect, slightly subsaturated to the smallest.
Large crystals scavenge moisture from small ones
leading to a slow but persistent crystal loss. This rep-
resents an example of the more general phenomenon
known as Ostwald ripening [as pointed out to us by
J. Ross (2009), personal communication] or ‘‘spectral
ripening,’’ which has also been considered for liquid
water clouds (Çelik and Marwitz 1999; Wood et al.
2002). This effect is amplified in contrail evolution by
long integration times allowing a small effect to accu-
mulate and low turbulence and vertical velocities pre-
venting other dynamics from taking over (in contrast to
most natural-cirrus evolution). It is the combination of
the growth in mean crystal size and the spectral
broadening due to the moisture scavenging that permits
the nearly constant spectral shape (displayed logarith-
mically) noted earlier in Fig. 7a to arise.
The mean in situ loss rate is unmodified by modest
vertical oscillations but can be eliminated by suffi-
ciently large plume mixing or ascent or low-enough
number concentrations (Lewellen 2012). While the in
situ loss rates can be quantitatively examined in the
simulations, the rates for actual contrails are more
uncertain. The Kelvin effect is quantitatively more
poorly understood for ice crystals than for water drops
and, as noted below, uncertainties in the ice deposition
coefficient factor in as well. While a shift in the critical
vapor saturation level is certainly present for small
crystals, and the basic effect of large crystals growing at
the expense of smaller ones is likely robust, there may
be significant quantitative changes owing to crystal
growth and shape effects. There are also smaller re-
sidual crystal losses that can remain even in the absence
of the Kelvin effect (barely evident in Fig. 8) depending
on the combined effects of gravity wave magnitude and
parcel mixing.
c. Sensitivity to EIiceno
The changes in crystal-loss mechanisms seen with
spectrally resolved microphysics alter the sensitiv-
ities to EIiceno. Figure 11 shows sample Q3D results
varying EIiceno. Because of the in situ crystal losses the
fraction of the total crystals lost in the wake-
dominated regime rises with EIiceno faster than the
expectations from bulk treatments (e.g., Lewellen and
Lewellen 2001a; Unterstrasser et al. 2008). Moreover,
this sensitivity remains for older contrails, contrary to
the expectations based on bulk treatments that the
fraction of crystals lost would be approximately in-
dependent of EIiceno (e.g., Unterstrasser and Gierens
2010a). One consequence is that the sensitivity to
EIiceno . ;1014 kg21 diminishes significantly over
time; in Fig. 11 an initial factor of 1000 difference in
crystal number is reduced to ;3 within a few hours.
This insensitivity is one sided, however: for small-
enough EIiceno the crystal number densities are re-
duced sufficiently so that in situ losses are absent over
FIG. 10. Total ice crystal number evolution in young contrails
simulated with bulk microphysics (long dashed lines), binned mi-
crophysics (solid lines), or binned without the Kelvin effect (short
dashed lines). Results from Q3D simulations with ambient T 5218K and RHi 5 110% (thin lines) or 130% (heavy lines). The
number losses around 100 s are forced by wake-vortex descent and
those around 700 s are forced by descent in a Brunt–Väisälä oscil-lation of the wake plume.
4410 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
the time scales available. The dramatic changes in
contrail evolution suggested in Fig. 11 from reductions
in EIiceno below this threshold are considered more in
Part II. It is interesting that the EIiceno suggested by
both jet-plume ice nucleation modeling for current
aircraft [the soot-dominated regime of Kärcher andYu (2009)] and in situ aerosol measurements (Schröderet al. 2000; Anderson et al. 1998) is of the same order of
magnitude (;1015 kg21) as the lower range over which
in situ losses are found to be large in the wake-
dominated regime here. Even if the mechanisms lim-
iting the crystal numbers in the initialization stage
considered by Kärcher and Yu (2009) were absent, the
ice crystal numbers in contrails would tend to be forced
to the order of magnitude consistent with in situ
measurements anyway by the crystal losses during the
wake-dominated regime. The quantity M(t), governed
dominantly by plume volume and ambient conditions, is
relatively insensitive to EIiceno unless it drops below
about 1014 kg21 (Fig. 11b); the time scale for reaching the
equilibrium state in (3) exceeds plume expansion time
scales if ice number densities are small enough.
d. Sensitivity to water from fuel consumption
The ice crystal number losses in the young contrail are
greatly reduced at colder temperatures (e.g., Figs. 2 and
5). Reduced crystal loss for dryer conditions and thus
smaller crystals seems counter intuitive, but it proves to
be an effect of water from fuel combustion. The water
from the engines raises qt/qs locally in the contrail
(particularly in those regions with highest crystal num-
ber densities) to a much greater extent the smaller qs,
and hence the smaller T. For example in S10T205 the
engine water raises qt/qs in the young (;30 s) contrail
core from the ambient 110% to over 200% while in
S30T225 the corresponding increase is only from 130%
to ;138%—explaining the enhanced crystal survival in
the former case despite the decreased ambient RHi. As
discussed more in Part II, this is one factor in raising the
potential significance of cold contrails despite their rel-
atively low ice water content. That very low temperature
conditions lead to greatly reduced ice number losses was
noted in Unterstrasser et al. (2008) but attributed to
changes in growth–sublimation rates with temperature.
e. Further microphysics uncertainties
1) ICE CRYSTAL SHAPE EFFECTS
For the present simulations we have assumed spherical
ice crystals for simplicity. In situ sampling of contrails
suggest that the shapes of small (;20-mm diameter or
less) crystals that are found in young contrails or contrail
cores are quasi spherical (e.g., Gayet et al. 1998; Schröderet al. 2000; Febvre et al. 2009) while the shapes of the
larger ones found in the contrail periphery or in pre-
cipitation streamers are more complex (e.g., bullet ro-
settes, extended columns, or irregular) (e.g., Lawson et al.
1998; Heymsfield et al. 1998; Gayet et al. 1998). Iwabuchi
et al. (2012) found results consistent with this picture
from a large survey of the optical properties of persistent
contrails. Crystal shapes can affect growth rates, sedi-
mentation velocities, and optical properties. To assess
their potential importance for contrail evolution Meza
(2010) added to the current LES model the option of
using nonspherical shapes: hexagonal columns, plates,
bullet rosettes, or a mass-dependent combination of
shapes (near spherical for small sizes, hexagonal columns
FIG. 11. Effects of varying EIiceno on total (a) ice crystal number,
(b) ice mass, and (c) ice surface area, for Q3D simulations with
RHi 5 110%, T 5 218K, and low ambient turbulence [dashed
patterns from top to bottom in (a) indicate 1017, 1016, 1015, 1014,
1013, 1012, and 1011 per kilogram of fuel]. The twoEIiceno5 1011 per
kilogram of fuel cases shown differ by the choice of initial radius of
the ice crystals (0.2mm for double crosses and 8.0mm for double
asterisks).
DECEMBER 2014 LEWELLEN ET AL . 4411
of varying aspect ratio for midrange, and bullet rosettes
for the largest). The shape effects were incorporated
through modifications of C, fy, and d in the ice growth
[(1)] and of sedimentation velocity as a function of crystal
mass; the effects on coupled radiation were only partially
included. The precise implementation and comparison of
results for different shapes are given in Meza (2010), but
we summarize the main findings here.
For realistic shape assumptions and in the absence
of coupled radiation none of the differences in N(t) and
M(t) for different conditions due to shape effects were
found to be large in comparison to the Q3D realization
spread. Somemore significant differences were found in
S(t) but were consistent with being primarily geometrical
rather than dynamical, representing the change in surface
area for fixed number density and mass given by shape
change alone. This suggests that while itmay be important
to consider shape differences when assessing the radiative
impact of a contrail, it might be sufficient to perform
simulations with spherical crystals to determine N(t) and
M(t) and then assess shape effects on the radiative impact
afterward. The exception would be cases where the radi-
ative heating–cooling of the contrail itself strongly couples
into its dynamics, together with a heating–cooling rate
that is strongly shape dependent. Preliminary sets of
simulations with coupled radiation inMeza (2010) did not
identify any dramatic crystal shape-dependent effects, but
more complete treatment of the radiation for nonspherical
shapes is required before drawing final conclusions.
The relative insensitivity of the contrail evolution to
crystal shape that was found is perhaps not unexpected:
small ice crystals in the contrail do not differ much from
spheres, while large crystals have significant fall veloci-
ties shortening lifetimes and so limiting their contribu-
tion to contrail-integrated metrics.1 For any compact
shape, such as a hexagonal column with aspect ratio
around one, the shape-dependent factors that might
directly affect contrail evolution (e.g., the capacitance
factor or sedimentation velocity) do not vary much from
the spherical results (even though the optical properties
might be quite distinguishable). Further, within the
contrail core, the growth of ice mass is generally gov-
erned by mixing rates of ambient air into the contrail or
vertical displacement of it, rather than growth rates for
individual crystals that can be affected by shape. Meza
(2010) found large differences in contrail evolution due
to shape only when the smallest crystals were assumed
(unphysically) to be bullet rosettes: the reduced d for
a given crystal mass that results amplifies the Kelvin-
dependent in situ crystal loss, thereby reducing N(t)
significantly relative to the result for spheres.
2) ICE DEPOSITION COEFFICIENT
The ice deposition (or accommodation) coefficient bi
is poorly determined experimentally, with results in the
literature ranging over more than two orders of magni-
tude (e.g., Pruppacher and Klett 1997, their Table 5.5).
The default choices made by different cirrus modelers
covers nearly as large a range and have a significant
impact on the crystal number densities produced via
homogeneous nucleation in natural-cirrus simulations
(Lin et al. 2002). The quantity bi appears in the ice
growth [(1)] effectively through modifying the water va-
por diffusion coefficient D (Pruppacher and Klett 1997):
D*5D/(11 lb/r) with lb ’D
bi
�2p
RyT
�1/2
. (4)
There is negligible effect as bi approaches 1 or as crystals
grow large compared to the mean free path of water
vapor (a fraction of a micron for contrail conditions). As
already noted, contrail dynamics are such that ‘‘small
crystal’’ effects can have an extended range of influence.
Figure 12 shows results from simulations varyingbi away
from the CARMA assumed value of 0.93. There is in-
creasing sensitivity seen in crystal losses as bi drops
below 0.1 and profound differences in M(t) as well as
N(t) when the values become small enough. The sensi-
tivity is dominantly through the effect of bi on in situ
crystal loss (Lewellen 2012), although significant sensi-
tivity remains even with Kelvin effects excluded since
bi � 1 reduces overall ice growth rates. Recent labo-
ratory measurements designed specifically for cirrus
conditions (Skrotzki et al. 2013) recommend the value of
bi 5 0.7 and may, if confirmed, significantly reduce this
uncertainty, though it remains unexplained why some
measurements give much lower values and remains
a possibility that bi may vary with crystal size, habit,
supersaturation, and/or temperature.
5. Concluding remarks
In this work, contrails have been simulated from a few
wing spans behind the aircraft through the wake-vortex
evolution, transition to contrail cirrus, and eventual
demise through precipitation of the constituent ice
crystals into subsaturated air. The use of dynamic local
ice binning, column by column dynamic updating of
radiative heating, a scheme for providing resolved am-
bient atmospheric turbulence, and the surprising success
of a quasi-3D simulation mode make practical the
1 Further discussion on the potential effects of nonspherical
crystal shapes integrated over the contrail’s lifetime are given in
Part II.
4412 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
exploration of a large parameter space of detailed con-
trail simulations out to many hours of simulated time.
While the simulations are more numerically costly than
the approach of Unterstrasser and Gierens (2010a,b),
they have the advantage of permitting a direct assess-
ment of the effects on contrail dynamics of explicit
binned ice microphysics, 3D fluid dynamics, realistic
resolved aircraft wake dynamics, and resolved ambient
atmospheric turbulence/waves.
Some of these effects prove more critical than others.
While 3D dynamics can dramatically affect contrail
appearance, comparing 3D and Q3D results suggests
that including a very limited range of scales of 3D tur-
bulence may be sufficient for mean contrail properties.
Inclusion of resolved ambient turbulence/waves dem-
onstrates their importance to older contrail extent and
lifetime; it is not appropriate to consider only a single
turbulence level as if it were universal, particularly in the
absence of other turbulence production (e.g., frommean
shear or coupled radiation). More importantly, the in-
clusion of binned microphysics significantly increases
rates of ice crystal number loss in both young and old
contrails because of local competition between large and
small crystals for available moisture and how this is
modified by the Kelvin effect. This adds to previous
contrail work suggesting the need for improved micro-
physics (Huebsch and Lewellen 2006; Unterstrasser and
Sölch 2010). It might be possible to approximately in-
corporate these effects through suitable modifications of
2D fluid dynamics and bulk microphysics models a pos-
teriori, but they will not naturally be present.
For lower ambient temperatures the water from the
aircraft engines becomes important in boosting qt/qs in
the young contrail and thereby significantly reducing
crystal number losses in the wake-dominated regime.
The fractional rate of ice crystal number loss is also
strongly dependent on the crystal number densities
themselves (in contrast to simple expectations from bulk
models). As a consequence, the sensitivity to changes in
effective ice crystal number emission index are dra-
matically reduced over time for EIiceno . ;1014 kg21.
Given the sensitivity of the simulations to the Kelvin
effect and the ice deposition coefficient bi through their
impact on crystal-loss rates, improving the observational
constraints on these would be useful.
Quantitative differences aside, most of our conclu-
sions on contrails of more than a few minutes’ age are
consistent with other studies (e.g., Jensen et al. 1998;
Lewellen and Lewellen 2001a; Unterstrasser andGierens
2010a,b), including the critical importance of ambient
RHi, the primary role of temperature in determining
crystal size, the very small fraction of crystal number but
significant fraction of ice mass found in precipitation
streamers and their efficiency in removing moisture
from large volumes, and the critical importance of
coupled radiation in some regimes but not others. Sed-
imentation plays a critical role in the vertical spread of
the contrail and its ultimate lifetime. The time scale for
the initial onset of significant precipitation growth is
dominantly governed by RHi and T. The contrail/
contrail-cirrus lifetime is generally shortened by fac-
tors leading to larger crystal sizes, including higher RHi
orT, lower ice numbers, increased turbulent mixing with
supersaturated air, and increased wake-induced or in
situ crystal loss. An assessment of the effects of different
physical variations on overall contrail significance
therefore requires covering a sizable parameter space
with simulations over the full lifetime. The model de-
scribed here is efficient enough for such studies, taken
up in a companion paper (Part II).
FIG. 12. Effects of varying ice deposition coefficient: bi5 1 (solid
line), 0.5 (short dashed), 0.3 (long dashed), 0.1 (long–short dashed),
0.01 (starred–dashed). Results from Q3D simulations with RHi 5110%, T 5 218K, and low ambient turbulence.
DECEMBER 2014 LEWELLEN ET AL . 4413
The simulations here show a clear potential for long-
lived, large-volume contrails. Even if the ice cloud inmost
of this volume is not optically apparent it could still have
significant radiative impact. At the same time the aging
contrail can sometimes transform a very large volume
that was highly supersaturated into one that is near zero
supersaturation, extending from just above flight level
down through the depth of the supersaturated layer in the
vertical and over several kilometers in the horizontal
even in the absence of strong mean wind shear. This
could possibly provide subsequent aircraft with a corridor
in which to fly without producing significant contrails
themselves. While this would be a drifting corridor,
conditions for which it might be toughest to track (e.g.,
large cross-stream shear) would tend to give wider targets
of subsaturated horizontal extent. By removing the su-
persaturation in this region the corridor could also lead to
a reduction in natural cirrus. Because of the extensive
interaction of a single contrail over its lifetime with both
the horizontal (particularly with wind shear) and the
vertical extent of the ice-supersaturated region, these
effects may need to be accounted for explicitly in regional
and global simulations of contrail effects. Because of the
potential depletion of ice supersaturation, this may be
particularly important in regions of heavy air traffic or for
future projections of increases in air traffic.
Acknowledgments. This research has been supported
by the Boeing Company. We thank Steve Baughcum
and Mikhail Danilin for important contributions
throughout the course of this work, Eric Jensen and
Andy Ackerman for providing access to CARMA,
Michael Czech and Jeffrey Crouch for providing 2D
wake results for initialization, and Tim Rahmes and
John Ross for useful input and discussions. We thank
three anonymous reviewers for comments leading to
improvements in the presentation of this work.
APPENDIX
Additional Model and Simulation Details
a. Ice microphysics
At each time step in each grid cell where the ice
numbers are not numerically insignificant, the local ice
spectrum (cf. Fig. 1) is passed to the CARMA routine
for updating the diffusional growth, performed in a bin
space extended from the local space by one bin on either
end. Advection and diffusion updates in each spatial
direction are conducted on the subspace of the global
bin space that is populated along the given spatial line.
At the end of each time step the ice spectrum at each
spatial point x is checked to see whether the spectrum
would be better represented in local bin space if is(x, t)
were shifted up or down by 1. The spectrum is then
‘‘folded back’’ from the expanded bin space into the
appropriately shifted local space. The crystals outside
the new local space are simply added to the nearest size
bin that is included. Both total ice number and qt are
conserved in the folding process; ice mass is not. Ideally
the number of local bins should be chosen large enough
so that the number of crystals folded back is always
negligible. The latter is monitored by the code so that
a simulation may be terminated if an increase in number
of local bins is required. The simple shifting criterion of
minimizing the number of crystals at the local bin ends is
found to work well. When crystals evaporate out of the
smallest global bin they are removed completely from
the simulation. Physically crystals evaporating out of the
smallest local (but not global) bin should sometimes be
removed as well, in regimes where large crystals are
growing at the expense of small ones so that down-
shifting of the local space is not appropriate. This is di-
agnosed by crystals piling up in the lowest local bin after
the spectrum is folded back; accordingly, crystals evap-
orating out of the lowest local bin are removed if its bin
population exceeds that of the next larger bin.
Based on several tests we choose for the simulations
here Rm 5 2, ng 5 48, nl 5 20, and smallest global bin
corresponding to crystal radius of 50 nm. The local
binning scheme was critically tested by running it
concurrently with a conventional global binning within
the same simulation for different contrail conditions.
Contrail simulations were also performed varying the
ice bin resolution. In addition, in Lewellen (2012) dif-
ferent ice binning choices in parcel model tests were
compared against exact analytic solutions of ice spec-
trum development for numerically challenging regimes
with in situ crystal loss. The tests suggest that for con-
trail simulation a bin mass ratio of 3.0 is probably
marginally adequate, while a ratio of 2.0 generally re-
produces results of a much finer binning well (though
somewhat overestimating in situ crystal loss; Lewellen
2012). Parcelmodel tests showed that, for a binmass ratio
of 2.0, even as few as 10 local bins faithfully represents the
evolution in different growth regimes. Within contrail
simulations the concurrent local–global tests showed al-
most indistinguishable results (much less than typical
Q3D realization differences) using 15 local bins. An ad-
vantage of the dynamic local binning is that the largest
and smallest crystal sizes represented in the global space
may be selected conservatively without significant in-
creases in numerical cost. Sensitivity tests showed no
apparent change in results when the choice of 50-nm
minimum radius was shifted up or down by a few bins.
4414 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
Early in the contrail dynamics the restriction on the
time step set by the wake fluid mechanics is generally
more stringent than that set by the ice physics. At late
times the microphysics is substepped on a shorter in-
terval than the fluid mechanics if necessary (an existing
feature of CARMA). When populations of crystals with
large fall velocities become significant, the LES time
step is reduced if necessary so that the fall velocity sat-
isfies the Courant condition. If the population is physi-
cally insignificant, then its fall velocity is simply reduced
to maintain stability with the current time step.
b. Coupled radiation with dynamically adjustedindependent column updates
For each grid column in the domain, vertical profiles
ofT, qy, and ice particle concentrations are passed to the
CARMA radiative transfer routine to compute a radia-
tive heating rate for each point in the column. The
contributions from all columns are stored and used to
update local conditions at each time step. For a single
column denote the vertical heating rate profile HR(z, t)
and time between updates dtR. Then dTE [ dtR 3maxz[HR(z, t1 dtR)2 HR(z, t)] provides an estimate of
the maximum error in temperature due to the delay in
updating the heating rates. Since the coupled radiation
mainly affects the contrail dynamics through buoyancy-
induced changes in parcel heights, a natural temperature
difference scale to consider in the simulation is dT0 [gDz, where g is the potential temperature gradient and
Dz is the finest vertical grid spacing. After each update if
dTE. fRdT0 then dtR is halved; if dTE, fRdT0/3 then it is
doubled. The tolerance fR is an input for the scheme. The
dtR’s are allowed to vary for each column independently,
with the exception of requiring that the values for
neighboring columns differ by no more than a factor of 2.
This avoids potential problems associated with horizontal
advection of sharp-edged clouds. In addition to these
updates, the heating rates in all columns are computed for
the first time step and for eachLESoutput cycle (typically
every 300 s at late times). A value of fR 5 0.01 has been
chosen for the simulations here. A set of contrail test
cases showed that changing to fR5 0.001 or 0.1 produced
little or no differences in mean contrail development.
Other components of the radiation scheme are taken
essentially as is fromCARMA. For our sensitivity studies
here and in Part II we take the upwelling blackbody ra-
diation temperature Tsfc as our primary variable. Condi-
tions above the LES domain are specified via a blackbody
temperature of downwelling infrared radiation at domain
top and an assumed integrated water column above the
domain (taken as 130K and 0, respectively, for the sim-
ulations here and in Part II). Between the LES domain
and the surface, 10 evenly spaced layers (in pressure
coordinates) are added, with the pressure and tempera-
ture interpolated between the surface values and those at
the bottom of the LES domain, and a constant relative
humidity with respect to water assumed (taken as 30%
for the simulations here). A constant CO2 concentration
of 350ppmv is assumed throughout the column and the
ozone mixing ratio taken from the U.S. standard atmo-
sphere midlatitude profile; no other trace gasses are in-
cluded in the radiation computations. When incoming
solar radiation is included (as for some runs in Part II),
latitude, date, surface albedo, and either time-varying or
constant solar zenith angle can be specified.
c. Late-time turbulence maintenance
For each of the domains in Table 1 except grid 1
(where ambient atmospheric turbulence plays no sig-
nificant role) a preliminary LES to produce ambient-
turbulence fields was performed with periodic boundary
conditions, uniform grid spacing in each direction and
the desired ambient stratification, mean shear, and do-
main size. These were initialized with temperature
perturbations �1�2Tamp with �1 and �2 randomly chosen in
each grid cell and distributed uniformly between21 and
1. Fields from these decaying turbulence simulations
were taken at a chosen age fage and these added to the
contrail LES every tadd seconds to maintain some am-
bient turbulence in the main simulations over long
times. Before adding the aged turbulence fields to the
contrail LES the horizontal mean averages were re-
moved and the domain shifted by a random amount in
x, y, and z so that the sequential additions were effec-
tively independent of each other even though for each
domain only a single field from the corresponding de-
caying turbulence LES was utilized. The accompanying
subgrid TKE field was also added but the temperature
perturbations were not, since this would interact un-
physically with the ice dynamics. The initially imposed
mean vertical shear in the contrail LES persists on its
own over the course of a simulation (without any
nudging or imposed pressure gradients required) since
only perturbations away from the horizontal average are
added, periodic horizontal boundary conditions are
used, and the shear and stratification levels are chosen
such that the mean shear does not directly drive a shear
instability.
After repeated field additions the turbulence statistics
of the simulation, averaged over periods long compared
to tadd, become quasi steady. The state is determined by
the choice of stratification, shear, Tamp, fage, and tadd as
well as the domain size and grid resolution. The turbu-
lence and gravity wave amplitudes can be increased by
increasing Tamp and/or decreasing tadd; the relative im-
portance of gravity waves compared to turbulence is
DECEMBER 2014 LEWELLEN ET AL . 4415
increased by increasing fage and/or tadd; and the level of
intermittency increases with increased tadd. The nu-
merical overhead of this scheme is minimal.
Given the use of different domains and grids for dif-
ferent simulation segments, a practical consideration is
the extent to which the character of the fields can be
matched across different domains or grids. This was
found empirically to be achieved approximately if the
same Tamp, fage, and tadd are chosen and in addition the
same resolution is used for the preliminary dying tur-
bulence simulation in each case; Fig. A1 provides an
example from the current simulations. For the simula-
tions of Table 2 we chose fage and tadd both equal to
10min (roughly the Brunt–Väisälä period for the givenstratification) and choseTamp5 0.2K and 2K for the low
and moderate turbulence cases, respectively. The hori-
zontal and vertical resolutions for the decaying turbu-
lence simulations were chosen to be 50 and 33m,
respectively. The mean dissipation levels produced (Fig.
A1) fall within the broad range observed in the upper
troposphere or lower stratosphere [e.g., as given in
Table 17.3 of Quante and Starr (2002)].
d. Simulation initialization
The 2D Boeing results are utilized five spans behind
the aircraft (;1-s age) and include velocities, down-
stream vorticity, a passive exhaust tracer, and amodeled
turbulent eddy viscosity. From the latter a turbulent
kinetic energy is estimated, a fraction of which is used to
initialize the subgrid TKE field for the LES and a small
fraction is used for resolved velocity perturbations. The
velocity field from the 2D calculation is combined with
these random perturbations (and a mean ambient shear
profile if desired) to form the LES velocity field. The
spreading exhaust jets rapidly develop their own re-
solved turbulence and the LES subgrid TKE adjusts on
a fast time scale as well so that the simulation results are
not very sensitive to the magnitudes of perturbations or
subgrid TKE chosen. The velocity field is extended from
the smaller domain of the 2D simulation by integrating
the vorticity field using Biot–Savart’s law. The passive
tracer from the 2D model is used directly to initialize
a passive tracer in the LES and is scaled (given a fuel
flow rate, propulsion efficiency, and specific combustion
heat) and added to a desired ambient gradient to form
the LES potential temperature field.
The fine resolution in grid 1 of Table 1 is required to
resolve the small vortex cores provided by the 2D
Boeing simulation as well as being beneficial in rapidly
developing realistic jet turbulence. Comparison be-
tween the 0.4-m-resolution LES and a 0.25-m-resolution
test run out to 15 s supported the adequacy of the coarser
resolution. It is numerically costly to run the full binned
microphysics simulation with sufficient spatial resolu-
tion to maintain the initial tight vortex cores throughout
the vortex-dominated stages of the LES. Our previous
studies initialized with more diffuse cores (Lewellen and
Lewellen 1996, 2001a) suggest that this is not required in
order to simulate the dispersion of the exhaust species
and basic vortex dynamics well. Accordingly, we run the
2-s simulation on grid 1 with the subgrid rotational
damping turned off to modestly diffuse the vortex cores
so that they can be well resolved on grid 2.
Results from the initial LES are used to initialize the
main contrail simulation on grid 2 with ice introduced.
Periodic copies were used to extend from the smaller to
larger downstream domain and an ambient-turbulence
field of desired character added as discussed earlier. The
initialization times represent a compromise. The LES
must be initialized late enough that the approximations of
taking the wake to be uniform downstream, and the hot
exhaust jets to be incompressible, are reasonable. It must
be early enough to generate the large concentration
fluctuations in the exhaust jets that are a product of 3D
FIG. A1. (a) Mean dissipation rate (m2 s23) and (b) vertical
velocity variance (m2 s22) vs time for sample simulated ambient-
turbulence fields such as those employed in the contrail simula-
tions. Thick solid lines are from low-turbulence simulation on 12343 4 km3 domain; other lines frommedium-turbulence simulation
on 123 43 4 km3 domain (solid line), 0.843 0.263 1 km3 domain
(long dashed), and 123 0.323 4 km3 Q3D domain (short dashed).
The small-scale dissipation rate and vertical velocity variance are
chosen to roughly quantify the ambient-turbulence state, as the
most important impact on contrails is through mixing rates and
vertical motions.
4416 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
turbulence in the early stages. This prompted the ;1-s
initialization age and high-resolution LES. Ice initializa-
tion is delayed in part to ensure that the exhaust jets have
cooled so as to become supersaturated with respect to ice.
Amuch longer delay is undesirable because it would slow
the initial development of the Crow instability, which
necessitates the longer downstream domain.
Ice is introduced in a single size bin (generally 0.2-mm
radius). Simulation tests showed results to be insensitive
to the initial ice spectral distribution: in the young wake
plume the number densities are sufficiently high that the
crystals rapidly (in ;1 s) adjust their mean size to
achieve local equilibrium conditions regardless of their
starting point, and the mixing in the exhaust jets is rapid
enough that it quickly governs the shape of the local size
spectrum, washing out the memory of the initial bin
distributions. Figure 11 provides an example showing
that even if the initial crystal numbers are much less and
the initial bin size significantly raised, the effects wash
out after the wake-dominated regime.
For simplicity the ice has been introduced spatially
following a passive exhaust tracer. Given the complex-
ities of ice nucleation in the turbulent cooling jets this
need not be the case physically. Simulations with the ice
number density initially distributed spatially as some
power of the local tracer concentration C with the total
numbers held constant, were compared to test the sen-
sitivity. Varying over the range from C1/3 to C3, the re-
sulting changes in contrail properties still proved small.
This is consistent with changes in results due to engine
placement, which, while more significant, are generally
not dramatic. Note that in these sensitivity tests only
persistent contrails were considered; significant changes
to the evolution of short-lived contrails in subsaturated
conditions have not been ruled out.
Contrails exhibit an approximate periodicity down-
stream governed by the Crow instability. Ideally we
would perform simulations in a long enough domain to
include several Crow periods in order to produce aver-
age statistics that do not vary between different turbu-
lent realizations of the simulation. In previous work
comparing simulations on different length domains
(both without ice and with bulk ice microphysics) we
have found, however, that a single Crow period develops
in essentially similar fashion whether it interacts with
independent realizations downstream or only periodic
copies of itself. This makes simulation of a single Crow
period a physically reasonable and numerically efficient
choice. Running different realizations then gives some
measure of the quality of the averages (cf. Fig. 2).
When a new run segment is initialized on a different
grid during the course of a simulation (cf. Table 1), the
3D field variables are simply linearly interpolated onto
the new grid. The fields are sufficiently well resolved that
any nonconservation of species that this interpolation
could in principle produce are in practice confirmed to
be negligible. The new downstream domain is always
chosen to be an integer multiple of the old one, with the
fields periodically continued into the expanded domain.
A fresh ambient-turbulence field generated for the new
expanded domain is added at the start to break the
symmetry between the periodic copies of the contrail
downstream and to provide local variations in the new
regions added in the cross-stream directions. Given the
presence of the contrail, the domain expansion in the
cross-stream directions is not done as a periodic copy
and so is not required to be an integer multiple of the
old domain. Because the cross-stream boundaries
are always located far from the expanding contrail and
the fresh turbulence addition dominates the decayed
turbulence remaining on the old grid, any discon-
tinuities in the ambient velocity fields produced by this
procedure have negligible effect on the subsequent
contrail evolution.
Given the vertical grid tracking utilized in the code,
it is necessary throughout the simulation segments and
regridding steps to keep track of where the original
flight altitude lies within the grid at each time. Hy-
drostatic balance within the domain is then assumed to
obtain the mean pressure at each level in the grid from
the pressure assumed at flight level. This same capa-
bility is employed if a uniform uplift or subsidence is to
be included in the simulation (as is done for some cases
in Part II).
REFERENCES
Ackerman, A. S., O. B. Toon, and P. V. Hobbs, 1995: A model for
particle microphysics, turbulent mixing, and radiative transfer
in the stratocumulus-topped marine boundary layer and com-
parisons with measurements. J. Atmos. Sci., 52, 1204–1236,
doi:10.1175/1520-0469(1995)052,1204:AMFPMT.2.0.CO;2.
Anderson, B., W. R. Cofer, D. R. Bagwell, J. W. Barrick, C. H.
Hudgins, and K. E. Brunke, 1998: Airborne observations of
aircraft aerosol emissions I: Total nonvolatile particle emis-
sion indices. Geophys. Res. Lett., 25, 1689–1692, doi:10.1029/98GL00063.
Atlas, D., Z. Wang, and D. P. Duda, 2006: Contrails to cirrus—
Morphology, microphysics, and radiative properties. J. Appl.
Meteor. Climatol., 45, 5–19, doi:10.1175/JAM2325.1.
Çelik, F., and J. D. Marwitz, 1999: Droplet spectra broadening by
ripening process. Part I: Roles of curvature and salinity of
cloud droplets. J. Atmos. Sci., 56, 3091–3105, doi:10.1175/
1520-0469(1999)056,3091:DSBBRP.2.0.CO;2.
Chen, C.-C., and A. Gettelman, 2013: Simulated radiative forcing
from contrails and contrail cirrus. Atmos. Chem. Phys. Dis-
cuss., 13, 10 939–10 959, doi:10.5194/acpd-13-10939-2013.Chen, J., and R. Lin, 2001: Numerical simulation of contrail mi-
crophysical and radiative properties. Terr. Atmos. Oceanic
Sci., 12, 137–154.
DECEMBER 2014 LEWELLEN ET AL . 4417
Chlond, A., 1998: Large-eddy simulations of contrails. J. Atmos.
Sci., 55, 796–819, doi:10.1175/1520-0469(1998)055,0796:
LESOC.2.0.CO;2.
Connolly, P., C. Emersic, and P. R. Field, 2012: A laboratory
investigation into the aggregation efficiency of small ice
crystals. Atmos. Chem. Phys., 12, 2055–2076, doi:10.5194/
acp-12-2055-2012.
Czech, M., G. Miller, J. Crouch, and M. Strelets, 2005: Predicting
the near-field evolution of airplane trailing vortices. C. R.
Phys., 6, 451–466, doi:10.1016/j.crhy.2005.05.005.
Dobbie, S., and P. Jonas, 2001: Radiative influences on the struc-
ture and lifetime of cirrus clouds. Quart. J. Roy. Meteor. Soc.,
127, 2663–2682, doi:10.1002/qj.49712757808.
Febvre, G., and Coauthors, 2009: On optical and microphysical
characteristics of contrails and cirrus. J. Geophys. Res., 114,
D02204, doi:10.1029/2008JD010184.
Forster, L., C. Emde, B. Mayer, and S. Unterstrasser, 2012: Effects
of three-dimensional photon transport on the radiative forcing
of realistic contrails. J. Atmos. Sci., 69, 2243–2255, doi:10.1175/
JAS-D-11-0206.1.
Frömming, C., M. Ponater, U. Burkhardt, A. Stenke, S. Pechtl, and
R. Sausen, 2011: Sensitivity of contrail coverage and contrail
radiative forcing to selected key parameters.Atmos. Environ.,
45, 1483–1490, doi:10.1016/j.atmosenv.2010.11.033.
Gayet, J.-F., and Coauthors, 1998: In situ measurements of
the scattering phase function of stratocumulus, contrails
and cirrus. Geophys. Res. Lett., 25, 971–974, doi:10.1029/
98GL00541.
Gierens, K. M., 1994: The influence of radiation on the diffusional
growth of ice crystals. Beitr. Phys. Atmos., 67, 181–193.
——, 1996: Numerical simulations of persistent contrails. J. Atmos.
Sci., 53, 3333–3348, doi:10.1175/1520-0469(1996)053,3333:
NSOPC.2.0.CO;2.
——, and E. Jensen, 1998: A numerical study of the contrail-to-
cirrus transition. Geophys. Res. Lett., 25, 4341–4344,
doi:10.1029/1998GL900151.
——, P. Spichtinger, and U. Schumann, 2012: Ice supersatura-
tion. Atmospheric Physics, U. Schumann, Ed., Research
Topics in Aerospace, Springer Verlag, 135–150, doi:10.1007/
978-3-642-30183-4_9.
Gounou, A., and R. J. Hogan, 2007: A sensitivity study of the
effect of horizontal photon transport on the radiative forcing
of contrails. J. Atmos. Sci., 64, 1706–1716, doi:10.1175/
JAS3915.1.
Heymsfield, A. J., R. P. Lawson, andG.W. Sachse, 1998:Growth of
ice crystals in a precipitating contrail. Geophys. Res. Lett., 25,
1335–1338, doi:10.1029/98GL00189.
Huebsch, W. W., and D. C. Lewellen, 2006: Sensitivity study on
contrail evolution. Proc. 36th Fluid Dynamics Conf. and Ex-
hibit, San Francisco, CA, American Institute of Aeronautics
and Astronautics, 2006–3749, doi:10.2514/6.2006-3749.
Iwabuchi, H., P. Yang, K. N. Liou, and P. Minnis, 2012: Physical
and optical properties of persistent contrails: Climatology and
interpretation. J. Geophys. Res., 117, D06215, doi:10.1029/
2011JD017020.
Jensen, E. J., O. B. Toon, D. L. Westphal, S. Kinne, and A. J.
Heymsfield, 1994: Microphysical modeling of cirrus: 1. Com-
parison with 1986 FIRE IFOmeasurements. J. Geophys. Res.,
99, 10 421–10 442, doi:10.1029/93JD02334.
——, A. S. Ackerman, D. E. Stevens, O. B. Toon, and P. Minnis,
1998: Spreading and growth of contrails in a sheared envi-
ronment. J. Geophys. Res., 103, 31 557–31 567, doi:10.1029/
98JD02594.
Kärcher, B., and F. Yu, 2009: Role of aircraft soot emissions in
contrail formation. Geophys. Res. Lett., 36, L01804,
doi:10.1029/2008GL036649.
Khvorostyanov, V., and K. Sassen, 1998: Cloud model simulation
of a contrail case study: Surface cooling against upper tropo-
spheric warming. Geophys. Res. Lett., 25, 2145–2148,
doi:10.1029/98GL01522.
Lawson,R. P., A. J. Heymsfield, S.M.Aulenbach, andT. L. Jensen,
1998: Shapes, sizes and light scattering properties of ice crys-
tals in cirrus and a persistent contrail during SUCCESS.
Geophys. Res. Lett., 25, 1331–1334, doi:10.1029/98GL00241.
Lee, D., and Coauthors, 2010: Transport impacts on atmosphere
and climate: Aviation. Atmos. Environ., 44, 4678–4734,
doi:10.1016/j.atmosenv.2009.06.005.
Lewellen, D. C., 2012: Analytic solutions for evolving size distri-
butions of spherical crystals or droplets undergoing diffusional
growth in different regimes. J. Atmos. Sci., 69, 417–434,
doi:10.1175/JAS-D-11-029.1.
——, 2014: Persistent contrails and contrail cirrus. Part II: Full
lifetime behavior. J. Atmos. Sci., 71, 4420–4438, doi:10.1175/
JAS-D-13-0317.1.
——, and W. S. Lewellen, 1996: Large-eddy simulations of the
vortex-pair breakup in aircraft wakes.AIAA J., 34, 2337–2345,
doi:10.2514/3.13399.
——, and ——, 1998: Large-eddy boundary layer entrainment. J. At-
mos. Sci., 55, 2645–2665, doi:10.1175/1520-0469(1998)055,2645:
LEBLE.2.0.CO;2.
——, and ——, 2001a: The effects of aircraft wake dynamics on
contrail development. J. Atmos. Sci., 58, 390–406, doi:10.1175/
1520-0469(2001)058,0390:TEOAWD.2.0.CO;2.
——, and ——, 2001b: Effects of aircraft wake dynamics on mea-
sured and simulated NOx and HOx wake chemistry. J. Geo-
phys. Res., 106, 27 661–27 672, doi:10.1029/2001JD000531.
——, and ——, 2007: Near-surface intensification of tornado
vortices. J. Atmos. Sci., 64, 2176–2194, doi:10.1175/JAS3965.1.
——, ——, L. R. Poole, R. J. DeCoursey, G. M. Hansen, C. A.
Hostetler, and G. S. Kent, 1998: Large-eddy simulations and
lidar measurements of vortex-pair breakup in aircraft wakes.
AIAA J., 36, 1439–1445, doi:10.2514/2.535.
——, ——, and J. Xia, 2000: The influence of a local swirl ratio on
tornado intensification near the surface. J. Atmos. Sci., 57, 527–
544, doi:10.1175/1520-0469(2000)057,0527:TIOALS.2.0.CO;2.
Lin,R.-F.,D.O. Starr, P. J.DeMott,R.Cotton,K. Sassen,E. Jensen,
B. Kärcher, and X. Liu, 2002: Cirrus Parcel Model Comparison
Project. Phase I: The critical components to simulate cirrus
initiation explicitly. J. Atmos. Sci., 59, 2305–2329, doi:10.1175/
1520-0469(2002)059,2305:CPMCPP.2.0.CO;2.
Meza, O., 2010: The effect of ice crystal shape on aircraft contrails.
Ph.D. dissertation,WestVirginiaUniversity, 124 pp. [Available
online at http://hdl.handle.net/10450/10872.]
Minnis, P., D. F. Young, D. P. Garber, L. Nguyen,W. L. Smith, and
R. Palikonda, 1998: Transformation of contrails into cirrus
during SUCCESS. Geophys. Res. Lett., 25, 1157–1160,
doi:10.1029/97GL03314.
Naiman, A. D., S. K. Lele, and M. Z. Jacobson, 2011: Large eddy
simulations of contrail development: Sensitivity to initial and
ambient conditions over first twentyminutes. J. Geophys. Res.,
116, D21208, doi:10.1029/2011JD015806.
Paoli, R., J. Hélie, and T. Poinsot, 2004: Contrails formation in
aircraft wakes. J. Fluid Mech., 502, 361–373, doi:10.1017/
S0022112003007808.
Paugam, R., R. Paoli, and D. Cariolle, 2010: Influence of vortex
dynamics and atmospheric turbulence on the early evolution
4418 JOURNAL OF THE ATMOSPHER IC SC IENCES VOLUME 71
of a contrail. Atmos. Chem. Phys., 10, 3933–3952, doi:10.5194/
acp-10-3933-2010.
Penner, J. E., D. H. Lister, D. J. Griggs, D. J. Dokken, and
M. McFarland, Eds., 1999: Aviation and the Global Atmo-
sphere. Cambridge University Press, 373 pp.
Ponater, M., S. Marquart, and R. Sausen, 2002: Contrails in
a comprehensive global climate model: Parameterization and
radiative forcing results. J. Geophys. Res., 107, 4164,
doi:10.1029/2001JD000429.
Pruppacher, H. R., and J. D. Klett, 1997: Microphysics of Clouds
and Precipitation. Kluwer Academic Publishers, 954 pp.
Quante, M., and D. O. Starr, 2002: Dynamic processes in cirrus
clouds: A review of observational results. Cirrus,D. K. Lynch
et al., Eds., Oxford University Press, 346–374.
Rap, A., P. Forster, J. Haywood, A. Jones, and O. Boucher, 2010:
Estimating the climate impact of linear contrails using the UK
Met Office climate model. Geophys. Res. Lett., 37, L20703,
doi:10.1029/2010GL045161.
Schröder, F., and Coauthors, 2000: On the transition of contrails
into cirrus clouds. J. Atmos. Sci., 57, 464–480, doi:10.1175/
1520-0469(2000)057,0464:OTTOCI.2.0.CO;2.
Schumann, U., 1996: On conditions for contrail formation from
aircraft exhausts. Meteor. Z., 5, 4–23.——, 2012: A contrail cirrus predictionmodel.Geosci. Model Dev.,
5, 543–580, doi:10.5194/gmd-5-543-2012.
——, H. Schlager, F. Arnold, R. Baumann, P. Haschberger, and
O. Klemm, 1998: Dilution of aircraft exhaust plumes at cruise
altitudes. Atmos. Environ., 32, 3097–3103, doi:10.1016/
S1352-2310(97)00455-X.
Skrotzki, J., and Coauthors, 2013: The accommodation coefficient
of water molecules on ice—Cirrus cloud studies at the AIDA
simulation chamber. Atmos. Chem. Phys., 13, 4451–4466,
doi:10.5194/acp-13-4451-2013.
Sussmann, R., and K. M. Gierens, 1999: Lidar and numerical
studies on the different evolution of vortex pair and secondary
wake in young contrails. J. Geophys. Res., 104, 2131–2142,
doi:10.1029/1998JD200034.
Unterstrasser, S., and K. Gierens, 2010a: Numerical simulations of
contrail-to-cirrus transition—Part 1: An extensive parametric
study. Atmos. Chem. Phys., 10, 2017–2036, doi:10.5194/
acp-10-2017-2010.
——, and ——, 2010b: Numerical simulations of contrail-to-cirrus
transition—Part 2: Impact of initial ice crystal number, radiation,
stratification, secondary nucleation and layer depth. Atmos.
Chem. Phys., 10, 2037–2051, doi:10.5194/acp-10-2037-2010.
——, and I. Sölch, 2010: Study of contrail microphysics in the vortex
phase with a Lagrangian particle trackingmodel.Atmos. Chem.
Phys., 10, 10003–10015, doi:10.5194/acp-10-10003-2010.
——, K. Gierens, and P. Spichtinger, 2008: The evolution of con-
trail microphysics in the vortex phase.Meteor. Z., 17, 145–156,doi:10.1127/0941-2948/2008/0273.
Widnall, S. E., 1975: The structure and dynamics of vortex fila-
ments. Annu. Rev. Fluid Mech., 7, 141–165, doi:10.1146/
annurev.fl.07.010175.001041.
Wong, H.-W., and R. C. Miake-Lye, 2010: Parametric studies of
contrail ice particle formation in jet regime using microphys-
ical parcel modeling. Atmos. Chem. Phys., 10, 3261–3272,
doi:10.5194/acp-10-3261-2010.
Wood, R., S. Irons, and P. R. Jonas, 2002: How important is the
spectral ripening effect in stratiform boundary layer clouds?
Studies using simple trajectory analysis. J. Atmos. Sci., 59, 2681–2693, doi:10.1175/1520-0469(2002)059,2681:HIITSR.2.0.CO;2.
DECEMBER 2014 LEWELLEN ET AL . 4419