GEOPHYSICAL RESEARCH LETTERS, VOL. 34, L24810, doi:10.

1029/2007GL031723, 2007

Impact of cloud microphysics on hurricane track forecasts

Robert G. Fovell1 and Hui Su2
Received 18 August 2007; revised 29 October 2007; accepted 6 November 2007; published 27 December 2007.
[1] Simulations of Hurricane Rita (2005) at operational microphysics has either not been found in these studies or
resolutions (30 and 12 km) reveal significant track has gone unreported.
sensitivity to cloud microphysical details, rivaling [4] Herein, we demonstrate that microphysical assump-
variation seen in the National Hurricane Center’s multi- tions can dramatically impact forecasted track in Weather
model consensus forecast. Microphysics appears to directly Research and Forecasting (WRF) [Skamarock et al., 2007]
or indirectly modulate vortex characteristics including size model simulations of Hurricane Rita at horizontal resolu-
and winds at large radius and possibly other factors tions of 30 km and 12 km, as typically used for operational
involved in hurricane motion. Idealized simulations made (real-time) forecasts. A more idealized model at finer
at higher (3 km) resolution help isolate the microphysical resolution is used to examine the generality of the results.
influence. Citation: Fovell, R. G., and H. Su (2007), Impact of
cloud microphysics on hurricane track forecasts, Geophys. Res.
2. Operational Ensembles (30 and 12 km
Lett., 34, L24810, doi:10.1029/2007GL031723.
Horizontal Resolution)
1. Introduction [5] Most of these simulations employed WRF version
2.1.2 and a spatially extensive domain centered on the
[2] Official National Hurricane Center (NHC) statistics northern Caribbean. Four microphysical parameterizations
show that Atlantic hurricane position forecasts have (MP) were explored: the Kessler (‘‘warm rain’’), the Lin et
improved markedly in recent decades. Yet, the 2005 season al. (LFO), and the three and five class WRF single moment
demonstrated that much progress remains to be made. On (WSM3 and WSM5) options. All but the Kessler scheme
September 24, 2005, Hurricane Rita made landfall near the incorporate frozen water in some fashion. The Kain-Fritsch
Texas/Louisiana border as a Saffir-Simpson Category 3 (KF), Grell-Devenyi and (from WRF 2.0.3.1) Betts-Miller-
storm with 54 m s 1 maximum winds. This location Janjic (BMJ) convective parameterizations (CPs) were
was correctly identified in the NHC forecast issued 36 h tested. Runs were also made with MP and/or CP schemes
prior to landfall but their 54 h forecast had the highest deactivated. The influence of subgrid-scale turbulent mixing
probability landfall located west of Houston, a >130 km was also explored, and found to affect intensity more than track
shift that prompted a hurried and ultimately unnecessary in this experiment. Other physics schemes were held fixed.
evacuation. This position discrepancy was about average [6] The operational simulations employed 31 vertical
when compared to recent years, but looms very large indeed levels with a 50 mb model top. Usage of additional levels
when weighted by population. or a higher model top were not found to materially affect
[3] Weather forecasts in general have made great use of hurricane motion. Initial and boundary data were provided
ensemble forecasting, in which different models, model by National Centers for Environmental Prediction Global
physics options and/or initializations are applied to the same Forecast System forecasts at one degree resolution, com-
event, yielding an objective measure of forecast uncertainty. mencing either 06 UTC or 18 UTC Sept. 22nd. In the 30 km
Previous work involving both real-data and idealized ensemble, an LFO/KF simulation started at 18 UTC, about
modeling has shown that the choice of cumulus parameter- 39 hours prior to landfall, yielded accurate predictions (not
izations, boundary layer and/or cloud microphysics schemes shown) of landfall location, storm width, timing and inten-
can dramatically influence hurricane simulations, especially sity (936 mb), in good agreement with the contemporaneous
with respect to intensity and intensification rate, rainfall NHC forecast.
production and inner core structure [e.g., Willoughby et al.,
1984; Lord et al., 1984; Braun and Tao, 2000; Wang, 2002; 2.1. The 30 km Ensemble Results
McFarquhar et al., 2006; Zhu and Zhang, 2006]. Regarding [7] The physics-based ensemble experiment was con-
microphysics, Lord et al. [1984] found including ice pro- ducted for the earlier initialization time. Figure 1 presents
cesses resulted in a significantly stronger storm, while Wang a sea-level pressure (SLP) track plot for the LFO/KF
[2002] and Zhu and Zhang [2006] showed that disallowing (control) run. At each point depicted, the lowest SLP
melting and evaporation permitted substantially more rapid recorded during the final 27 hours of the 54 hour simulation,
intensification and lower central pressures. However, sen- based on hourly data, is plotted. Similar to the contempo-
sitivity of hurricane track or propagation speed to cloud raneous NHC forecast, the control run’s hurricane made
landfall near Houston. It deepened to 924 mb before
1
Department of Atmospheric and Oceanic Sciences, University of weakening prior to reaching land; this minimum pressure
California, Los Angeles, California, USA. was slightly lower than actually measured at or subsequent
2
Jet Propulsion Laboratory, California Institute of Technology, to this time (931 mb). The track followed by the model
Pasadena, California, USA. hurricane’s eye is traced by the open circles, representing
three-hourly positions ending at 12 UTC Sept. 24th, about
Copyright 2007 by the American Geophysical Union.
0094-8276/07/2007GL031723 three hours after the actual storm reached the coast.

L24810 1 of 5
L24810 FOVELL AND SU: MICROPHYSICS IMPACT ON HURRICANE TRACK L24810

Figure 1. Sea level pressure track plot for the 30 km control (LFO/KF) run commencing 06 UTC Sept. 22nd, covering the
simulation’s final 27 h; contour interval 8 mb. For selected ensemble members, markers denote eye positions every 3 h
ending 12 UTC Sept. 24th. Dashed tracks represent sensitivity runs described in section 2.2. Only part of the domain is
shown. Landfall area encompassed by NHC ensemble members reaching the coast by 12 UTC Sept. 24th is highlighted.

[8] Superposed are the best and worst results from this simulated hurricane landfall to the west of Houston, shown
ensemble, as determined by position error. The WSM3/BMJ on Figure 1 as the short-dashed line, a considerable increase
combination (solid squares) correctly simulated both land- in position error.
fall location and timing. In contrast, the Kessler/KF member [11] The long-dashed line on Figure 1 shows what tran-
(solid circles) produced a weaker (minimum SLP 944 mb), spired when the rainwater terminal velocity was set to zero in
more westward moving storm. Tracks from remaining the Kessler scheme, effectively removing precipitation. This
members (not shown) generally fell between these two run’s position error was no worse than that of the control run.
extremes. Taken together, this physics-based ensemble Without precipitation, there is little to no evaporation cooling
possessed a similar spread with respect to landfall as the in the boundary layer. However, another modified Kessler
NHC’s multi-model ensemble did at this same time (also run lacking only evaporation of rainwater possessed the same
indicated on Figure 1). The NHC ensemble consists of over track as the original Kessler/KF storm. These results suggest
a dozen models of various types and levels of complexity. that, at least for this particular situation, considerable sensi-
tivity can be excited via manipulating hydrometeor fall
2.2. The 30 km Ensemble Sensitivity Tests speeds.
[9] Many microphysical parameterizations treat hydrome-
teors in bulk, based on presumed particle size distributions, 2.3. The 12 km Ensemble Results and Sensitivity Tests
types and densities. Even the simplest schemes contain [12] To ascertain whether the microphysical influences
numerous assumptions and ‘‘knobs’’ that might lack observa- found in the 30 km runs persist when the resolution is
tional or theoretical justification, and thus can be a source of altered, a full physics-based ensemble was conducted using
uncertainty. To be useful uncertainty, each scheme has to have 12 km grid spacing. Model hurricanes with realistic inten-
a reasonable chance of producing the most skillful result in any sity without CP schemes were obtained at this resolution, so
given situation, something a more extensive experiment might only those members are considered herein. Figure 2 shows
reveal. Since model physics can interact in complex and SLP track plots for the 12 km Kessler, LFO and WSM3
potentially unpredictable ways, the performance of various runs. As in the 30 km ensemble, the Kessler scheme
MP schemes and the impact of their inherent assumptions are produced the weakest and most westward propagating
likely case- and even resolution-dependent. hurricane, still making landfall well west of Houston. This
[10] We have attempted to identify MP scheme ‘‘knobs’’ was also clearly the widest vortex of the three. The WSM3
that excite the sensitivity seen above. The most significant simulation again yielded the most accurate landfall while
difference between the Kessler scheme and any MP that LFO microphysics maintained the deepest storm (929 mb).
considers ice is that the average particle fall speed is likely All of the 12 km runs made without CP schemes tended to
different when frozen condensate is included. Fall speed make landfall a few hours late.
assumptions directly and indirectly influence particle [13] While the basic MP dependencies are similar at this
growth rates, the horizontal spread of condensate and higher operational resolution, some of the specific sensitiv-
vertical heating profiles, potentially interacting strongly ities differ from their 30 km counterparts. For example, at
with how and where CP-based adjustments are triggered. 12 km and without active CP, altering the rainwater terminal
To explore the role of hydrometeor fall speed on the velocity in the Kessler scheme had less impact on the
ensemble spread, we took the most accurate member landfall location (not shown). At 30 km, that alteration
(WSM3/BMJ) and forced ice to share the terminal velocity was perhaps exaggerated owing to interaction with the
of raindrops having equivalent mass. This resulted in a cumulus parameterization, which is more critical to vortex

2 of 5
L24810 FOVELL AND SU: MICROPHYSICS IMPACT ON HURRICANE TRACK L24810

Figure 2. As in Figure 1, but for selected members of the 12 km ensemble. Lf represents latent heat of fusion. Only part of
the domain is shown.

development and maintenance at that coarser grid spacing. [15] Persistent convective asymmetries can also influence
The substantial influence regarding ice fall speed previously vortex motion by inducing flow across the vortex towards
encountered (in the WSM3/BMJ experiment) was also the enhanced diabatic heating [Willoughby, 1992; Wang and
diminished at higher resolution in the absence of a CP. Holland, 1996]. Superposed on Figure 3 is the asymmetric
Yet, other sensitivities were discovered. Figure 2d shows component of tropospheric average ascent, a good proxy for
what transpired when the WSM3 scheme was modified to convective heating. The negative values (dashed contours)
neglect the latent heat of fusion (Lf), the extra 10% heating in this field represent relatively weaker rising motion. For
that occurs when liquid water freezes. The resulting hurri- both storms, a dipole pattern is revealed, but the WSM field
cane was wider, weaker and made landfall at Houston. The is rotated clockwise relative to the Kessler pattern, possibly
same alteration in the 30 km experiment had little effect on assisting the former’s relatively more poleward motion. In
track, at least when a CP scheme was active. any event, among these simulations anything that is done to
narrow the vortex, whether it becomes more intense as a
2.4. Synthesis result or not, tends to permit the hurricane to propagate
[14] Considering the 30 and 12 km results jointly, we see more northward. In the case of Rita, at least, that resulted in
that microphysical assumptions can exert a significant a more accurate landfall.
influence on hurricane track over relatively short (54 h)
time scales. Microphysics may modulate storm motion by
directly or indirectly influencing characteristics such as
3. Idealized Experiments
depth, radial structure and azimuthal asymmetry known to [16] Hurricanes often move through complex and dynam-
control vortex motion. Among these simulations, vortex ic environments, complicating analysis of the microphysical
depth appeared to vary little but width variations were impacts on simulated track and intensity. To isolate these
particularly pronounced, as demonstrated again in Figure 3’s influences, a modified real-data version of WRF version 2.2
vortex-following composites of 850 mb absolute vorticity for was created which retains Earth’s rotation and (optionally)
the 12 km ensemble’s Kessler and WSM3 members. Among curvature, but has no land, a uniform SST of 29°C and a
these storms, wider vortices tracked relatively more west- calm, horizontally homogeneous base state built on
ward, consistent with Xiao et al.’s [2000] experience. This Jordan’s [1958] hurricane season composite. Three tele-
may partly reflect ‘‘beta drift’’ [e.g., Holland, 1983; Chan scoping domains were used, the outer being a 3240 km
and Williams, 1987] that is sensitive to winds well beyond the square with 27 km resolution and the innermost being
eyewall [Fiorino and Elsberry, 1989a, 1989b]. 669 km on a side with 3 km grid spacing. The outer domain

3 of 5
L24810 FOVELL AND SU: MICROPHYSICS IMPACT ON HURRICANE TRACK L24810

Figure 3. Vortex following composites for the 12 km Kessler and WSM3 members, constructed between forecast hours
48 and 54 h, inclusive. Colored field is 850 mb absolute vorticity (units 10 5 s 1); contoured is the asymmetric component
of tropospheric average ascent (0.1 m s 1 contours, negative values dashed). Black dot marks eye location.

is intended to capture the entire environmental response to Kessler (LFO) vortex was weakest (strongest), consistent
the hurricane; its boundary conditions are fixed, and thus with the real-data runs. The LFO storm eventually spent
effectively closed. over 2 days at or very near Category 5 strength, while the
[17] The real-data simulations commenced with a pre- warm rain case fluctuated between Categories 2 and 3. As in
existing vortex and no condensation. Idealized simulations the real-data experiment, vortex width rather than depth was
often start off with an artificially imposed circulation. We
elected to ‘‘breed’’ a vortex by placing a synoptic-scale
warm and moist anomaly centered at 20°N and integrating
for a spin-up period (t s) of 24 h with the Kain-Fritsch CP
scheme active and microphysics switched off. During this
period, a coherent and well-resolved cyclone formed,
achieving a central SLP of 969 mb by 24 h. At that time,
the CP scheme was switched off and one of three MP
schemes (Kessler, LFO or WSM3) was activated in all
domains.
[18] Figure 4 shows results from an experiment retaining
Earth curvature (having variable Coriolis parameter f ).
Despite sharing a common startup, the storms quickly
diverged with respect to track, propagation speed and
intensity. Since there was no imposed large-scale flow,
vortex motions represent self-propagation clearly modulated
by microphysics. As in the real-data runs, the Kessler vortex
tracked farthest west and the WSM3 storm moved most
northward. A substantial propagation speed difference is
also evident: at t s + 54 h, the Kessler vortex’ forward
motion was 9 km h 1 and increasing while the LFO and
WSM3 storms were moving 43% and 52% slower, respec-
tively. Figure 4a shows the storms at that time were roughly
following the 850– 200 mb layer average flow they were
responsible for creating. When combined with track varia-
tions, position differences among the simulated storms
eventually became extremely large. The rapid movement
of the Kessler vortex relative to the ice MP storms is the Figure 4. Twelve-hourly positions for the idealized
most substantial difference with respect to the real-data Rita experiments’ Kessler (K), LFO (L), and WSM3 (W) storms,
runs. Quantitatively similar direction and speed disparities starting at t s + 18 h. (a) The 850 – 200 mb layer mean winds
were noted in a lower (12 km) resolution version of this averaged in 500 km square storm-centered domains at t s +
experiment (not shown). 54 h. Storm circulation is effectively removed. (b) The
[19] Figure 4b shows radial profiles of the 10 m wind radial profiles of 10 m wind speed vs. distance from eye at
speed taken at t s + 54 h. With regard to intensity, the that time.

4 of 5
L24810 FOVELL AND SU: MICROPHYSICS IMPACT ON HURRICANE TRACK L24810

the obvious discriminating characteristic; Figure 4b reveals References

the Kessler storm actually developed stronger flow at larger Braun, S. A., and W.-K. Tao (2000), Sensitivity of high-resolution simula-
radius, which may explain why it tracked both more quickly tions of Hurricane Bob (1991) to planetary boundary layer parameteriza-
tions, Mon. Weather Rev., 128, 3941 – 3961.
and more westward [Fiorino and Elsberry, 1989a, 1989b]. Chan, J. C.-L., and R. T. Williams (1987), Analytical and numerical studies
In an f-plane version of this experiment (not shown), the of the beta-effect in tropical cyclone motion, J. Atmos. Sci., 44, 1257 –
Kessler vortex was again widest and weakest but, as 1265.
expected, none of those model storms translated significant Fiorino, M. J., and R. L. Elsberry (1989a), Some aspects of vortex structure
related to tropical cyclone motion, J. Atmos. Sci., 46, 975 – 990.
distances owing to the absence of beta gyres (and large- Fiorino, M. J., and R. L. Elsberry (1989b), Contributions to tropical cyclone
scale flow). The important point is that microphysical motion by small, medium and large scales in the initial vortex, Mon.
assumptions can clearly and substantially influence factors Weather Rev., 117, 721 – 727.
Holland, G. J. (1983), Tropical cyclone motion: Environmental interaction
responsible for storm motion, especially in situations lack- plus a beta effect, J. Atmos. Sci., 40, 328 – 342.
ing strong large-scale forcing. Jordan, C. L. (1958), Mean soundings for the West Indies area, J. Meteorol.,
15, 91 – 97.
Lord, S. J., H. E. Willoughby, and J. M. Piotrowicz (1984), Role of a
4. Summary and Conclusions parameterized ice-phase microphysics in an axisymmetric, nonhydro-
static tropical cyclone model, J. Atmos. Sci., 41, 2836 – 2848.
[20] Hurricane track and landfall forecasting is a complex McFarquhar, G. M., H. Zhang, G. Heymsfield, R. Hood, J. Dudhia, J. B.
scientific problem with significant societal import. Herein, it Halverson, and F. Marks (2006), Factors affecting the evolution of
was demonstrated that variation of cloud microphysical Hurricane Erin (2001) and the distributions of hydrometeors: Role of
microphysical processes, J. Atmos. Sci., 63, 127 – 150.
processes, performed in the context of ensemble forecasting Skamarock, W. C., et al. (2007), A description of the Advanced Research
at operational resolutions, can yield an ensemble spread WRF Version 2, Tech. Note NCAR/TN-468+STR, 88 pp., Natl. Cent. for
comparable to multi-model experiments, likely by directly Atmos. Res., Boulder, Colo.
Wang, B., and G. J. Holland (1996), The beta drift of baroclinic vortices.
and indirectly modulating vortex structure. Indeed, it is Part II: Diabatic vortices, J. Atmos. Sci., 53, 3737 – 3756.
possible that the differences among various dynamical Wang, Y. (2002), An explicit simulation of tropical cyclones with a triply
models could chiefly reside in their respective handling nested movable mesh primitive equation model: TCM3. Part II: Model
refinements and sensitivity to cloud microphysics parameterization, Mon.
of microphysics, along with other processes related to Weather Rev., 130, 3022 – 3036.
convection. The uncovered sensitivities were found to vary Willoughby, H. E. (1992), Linear motion of a shallow-water barotropic
somewhat with resolution, possibly owing to a subtle inter- vortex as an initial-value problem, J. Atmos. Sci., 49, 2015 – 2031.
play among model physics, and are probably case-dependent. Willoughby, H. E., H. Jin, S. J. Lord, and J. M. Piotrowitz (1984), Hurri-
cane structure and evolution as simulated by an axisymmetric, nonhydro-
Still, microphysics appears to be one avenue to exciting the static numerical model, J. Atmos. Sci., 41, 1169 – 1186.
inherent propagation sensitivity of hurricane-like vortices Xiao, Q., X. Zou, and B. Wang (2000), Initialization and simulation of a
and should be considered as a valuable part of physics-based landfalling hurricane using a variational bogus data assimilation scheme,
Mon. Weather Rev., 128, 2252 – 2269.
ensemble forecasting. Zhu, T., and D.-L. Zhang (2006), Numerical simulation of hurricane Bonnie
(1998). Part II: Sensitivity to varying cloud microphysical processes,
[21] Acknowledgments. This work was supported by NSF grant J. Atmos. Sci., 63, 109 – 126.
ATM-0554765. We acknowledge the support by the Jet Propulsion
Laboratory, California Institute of Technology, sponsored by NASA, and
The Aerospace Corporation. The single-sounding WRF initialization used R. G. Fovell, Department of Atmospheric and Oceanic Sciences,
in the idealized simulations was designed by Gary Lackmann and Kevin University of California, Los Angeles, Los Angeles, CA 90095 – 1565,
Hill (North Carolina State Univ.). Jonathan Vigh (Colorado State Univ.) USA. ([email protected])
provided the NHC ensemble track data. The authors benefitted from H. Su, Jet Propulsion Laboratory, California Institute of Technology,
constructive comments from the reviewers and Editor. Pasadena, CA 91109, USA. ([email protected])

5 of 5