Impact of Microphysics on Hurricane Track and Intensity Forecasts
Connecting state-of-the-art satellite observations and hurricane forecast models
Contacts: Prof. Robert Fovell (UCLA); Dr. Hui Su (JPL)
National Hurricane Center (NHC) statistics show that hurricane track errors have decreased steadily in the last few decades.
Yet, much progress remains to be made. Hurricane track forecasts have benefited from ensemble forecasting, in which different models,
model physics options and/or initializations are applied to the same event, yielding an objective measure of forecast uncertainty. In
this study, we demonstrate that cloud microphysical assumptions can exert a dramatic impact on forecasted track. This was revealed in
model simulations of Hurricane Rita using the Weather Research & Forecasting Model (WRF) at 30 km horizontal resolution. Several
bulk microphysical schemes were tested in conjunction with a number of convective parameterizations and found to provide an ensemble
spread for landfall location comparable to that provided by the NHC multi-model ensemble which incorporates over a dozen models of different
types and levels of complexity. With the JPL's supercomputing power, we aim to examine the sensitivity of hurricane forecasts to microphysics
at both low and high resolution simulations, and for different hurricane cases. The model simulations are then compared with available
satellite observations. We will assess the impact of the microphysical schemes, uncover avenues for their improvement, and hopefully
lead to a more accurate hurricane track and intensity forecasts.
Sensitivity of hurricane track to microphysics
A cloud is composed of billions of condensed water particles, which can include free-floating cloud droplets and ice crystals as well as snowflakes,
moderate density graupel particles and higher density ice like hail. Numerical models cannot track every particle, so the physics of condensate creation,
growth, destruction and motion has to be parameterized. "Cloud microphysics" refers to how these processes are handled in a model. There are many
microphysical schemes, employing different assumptions regarding condensate number, habit, history and behavior. One of the simplest and most computationally
efficient schemes, the Kessler parameterization, admits only two forms of condensate: cloud droplets and rain. The LFO scheme, in contrast, adds three forms
of frozen water: ice crystals, snowflakes and graupel.
At coarser horizontal resolutions, models cannot even resolve individual clouds, much less their constituent condensate. "Cumulus parameterizations"
attempt to compensate for the thermodynamic effects of these subgrid-scale clouds. The variation among convection schemes are even greater than for microphysics.
Three of the most commonly employed parameterizations include the Kain-Fritsch, Grell-Devenyi and Betts-Miller-Jancic schemes. We have found that the
combination of microphysics and cumulus scheme can have a dramatic impact on the track, and also the intensity, of hurricanes simulated using the WRF model.
An ensemble of simulations, consisting of various combinations of parameterizations and a variety of horizontal resolutions, was constructed for 2005's Hurricane Rita.
Rita is an interesting and important case because it was originally expected to make landfall in the Houston/Galveston area, but veered off to a far less densely
populated area. The figure below left (click here to open a larger version in a new window) illustrates the track sensitivity with respect to moist processes in
runs employing 30 km resolution. The combination of LFO microphysics and the Kain-Fritsch cumulus scheme yielded a storm that struck Houston, in agreement with the
contemporaneous official NHC forecast. Yet, by varying these schemes, and important assumptions inherent in them (such as hydrometeor fallspeeds, for example), we
were able to obtain an landfall spread that mimicked that generated by NHC's multi-model ensemble shown below right. That ensemble includes over a dozen different
models of varying complexities, from sophisticated dynamical models to statistical models.
Left (Top): WRF model simulation of Hurricane Rita tracks. The model resolution is 30km. The colored field shows the lowest sea-level pressure (SLP) recorded during the
last 27 hours of a 54 hour control simulation of Rita using LFO (5 class) microphysics and the Kain-Fritsch (KF) convective scheme. The superposed black line traces
the model hurricane, which strikes Houston. Also shown are tracks of minimum SLP for runs using the Kessler (warm rain) scheme, the WSM3 simple ice scheme (with the
Betts-Miller-Jancic convective scheme), the Kessler with reduced rain fallspeed, and WSM3 with enhanced ice fallspeed.
(Bottom): The spread of NHC multi-model ensemble forecast at 06 UTC, 22 September. Note a similar ensemble spread was obtained from a single
model simply by varying the model microphysics and convective schemes. Image from Jonathan Vigh, Colorado State University.
WRF modeling of hurricanes
Comparison with satellite observations
The ensemble shown above has been expanded to include higher resolution simulations, but additional cases must be examined to assess the robustness of the
moist process sensitivity. Another important part of this project, just underway at this time, is to use satellite-derived data to validate the WRF model
simulations, to suggest improvements to the model microphysics and for assimilation into the model. The images shown below are from 2006's Hurricane Ernesto,
which was observed by NASA A-Train satellites. Among other data, these satellites will provide information regarding cloud tops and condensate concentrations,
important for tuning microphysical schemes.
Left (Top): Hurricane Ernesto imaged by Aqua MODIS on 29 August, 2006. CloudSat and Aura orbits passed through the hurricane cloud system.
(Middle): CloudSat radar reflectivity profile along the orbit track on 29 August, 2006.
(Bottom): Aura Microwave Limb Sounder (MLS) ice water content measurements in the upper troposphere.
These satellite observations will be use to evaluate WRF model simulated cloud structure and identify the optimal microphysics schemes for hurricane forecasts.
Microphysics exerted a profound influence on track and landfall location in the Hurricane Rita simulations. This was found not only in the 30 km runs,
but also in experiments using a finer, 12 km grid spacing (not shown). The latter resolution is high enough to support strong tropical cyclones without
recourse to convective parameterizations, permitting a considerable expansion of our ensemble parameter space. Yet, even though we can simplify simulations
somewhat by excluding convective schemes, they are still profoundly complex. Another potentially important part of this work will involve the use of substantially
simplified numerical models, which will study hurricane formation and movement in realistic but more highly controlled circumstances.
An example of this is the "Waterworld" experiments, typified by the animation below. One of the strengths of the WRF model is that it can also be used in an
idealized framework. "Waterworld" is a WRF real/ideal hybrid, a strategy first proposed by Prof. Gary Lackmann of North Carolina State University. In this
hybrid, the WRF domain is cast on a spherical, rotating Earth, but devoid of land and initialized with but a single sounding like idealized models. A perturbation
is introduced in the domain; within the first day of the simulation, the resulting convection organizes into a tropical cyclone. The animation below shows near
surface winds and SLP for one of the Waterworld runs, over a seven day period. The initial winds are calm in this simulation, so the motion of the hurricane is
completely self-driven. These simulations are being analyzed for clues as to why microphysics can play such a large role in simulated hurricane track and intensity.