Advanced
data assimilation systems for community use (top)
MM5/WRF 3DVAR Release
In June 2003, a combined MM5/WRF three-dimensional variational
(3DVAR) data assimilation system was released to the research
community after several years of development. Dale
Barker coordinated the release and created a web page
containing links to the software, documentation, etc. The
3DVAR system has been released to both MM5 (http://www.mmm.ucar.edu/3dvar)
and WRF (http://www.wrf-model.org/WG4/wg4_main.html)
communities. An initial tutorial attended by over 50 researchers
from the U.S. and abroad was held at NCAR in conjunction
with the release. A second tutorial was requested by members
of the Italian data assimilation research community and was
presented over two days in L’Aquila, Italy in July
2003 by Barker.
AFWA implementation of MM5/WRF 3DVAR
In late 2002, the MM5/WRF 3DVAR system was implemented operationally
in worldwide MM5-based domains at the US Air Force Weather
Agency (AFWA) in Omaha, Nebraska. This follows an earlier
operational implementation at the Taiwanese Civil Aeronautics
Administration (CAA). These achievements follow several years
of effort, supported by AFWA and CAA, to ensure the accuracy,
efficiency, portability, and flexibility requirements of
the operational and research communities. Areas of research
conducted by Barker include
the following: 1) the impact of Quikscat scatterometer data
on mesoscale NWP; 2) the use of bogus surface pressure observations
as a typhoon bogus vortex implementation system in 3DVAR;
3) the use of observation and variational diagnostics to
improve the usage of observations in 3DVAR; and 4) assessment
and tuning of the performance of 3DVAR and MM5 in mesoscale
North American, European and Asian domains.
|
|
| Figure 12: Verification
against radiosonde observations for height (above) and
relative humidity (below) for AFWA’s Europe 45km
domain. Forecast ranges are 0hr (red), 12h (green) and
24hr (blue). Triangles are MM5 forecast using 3DVAR,
squares use OI. |
Figure 12 is taken from AFWA verification
of the European application of MM5 3DVAR, compared with the
optimum interpolation (OI) scheme that the 3DVAR system was
intended to replace. Verification of both height and relative
humidity (against sonde observations) indicates a significant
reduction of 3DVAR forecast error relative to that of OI.
Improvements (not shown) are also seen in temperature and
precipitation field.
Real-time mesoscale ensemble forecasting sytem
James Bresch developed
a real-time mesoscale ensemble forecasting system. The system
consists of five members, run twice daily, with a 30-km grid
covering the continental United States. The 48-h forecasts
are used to compare various analysis methods, data sensitivities,
and physics packages. Specifically, for the first time, GPS
precipitable water measurements were routinely assimilated
into the real-time MM5 3DVAR system and their impact assessed.
Plots are available on http://rain.mmm.ucar.edu/mm5.
Simulations of Indian monsoon weather systems and the impacts
of various data assimilation methods
Jimy Dudhia, Mitchell Moncrieff, and Barker hosted
Someshwar Das (National Center for Medium Range Weather Forecasting,
India) as part of a collaboration initiated by the Indo-US
Forum on Weather and Climate Modeling held in New Delhi in
2002. The initial studies involved simulations of Indian
monsoon weather systems and the impacts of various data assimilation
methods and physical parameterizations. Das has started applying
NCAR's 3DVAR system in tests with his Indian real-time forecast
system based on MM5. Its daily use will be a valuable test
of this data assimilation method in a new region with local
Indian and satellite data.
Comparison of 4DVAR and EnKF techniques for convective-scale
data assimilation
Alain Caya, Juanzhen Sun, and Chris
Snyder continued to compare the performance of the
4DVAR and EnKF techniques using simulated data of a supercell
storm. 4DVAR and EnKF are the two analysis techniques capable
of accounting for the evolution of the atmospheric flow
represented by a numerical model. The 4DVAR technique was
examined for the convective-scale in the past and showed
promising results. The EnKF is a relatively new technique
with some attractive features. One advantage of EnKF is
that it takes less effort to implement compared with the
4DVAR technique. Using a simple cloud model and radar observations
extracted from the simulation of a supercell storm, identical-twin
experiments were conducted by both the 4-D variational
Doppler radar analysis system (VDRAS) and a recently developed
EnKF analysis system. A number of experiments were conducted
to examine the performance of the two techniques under
various conditions. The results suggested that both techniques
are able to retrieve the unobserved wind and thermodynamic
and microphysical variables with good accuracy when the
systems are carefully tuned. The 4DVAR generally shows
better performance in terms of RMS error reduction in the
first few cycles. Given enough cycles, the EnKF scheme
produces similar or sometimes superior results than 4DVAR.
A journal paper is being prepared, and will be submitted
to the Monthly Weather Review.
Initialization and forecasting of a hailstorm using radar
observations from STEPS
Forecasting of individual convective systems presents great
challenges to numerical weather prediction. In order to examine
qualitative prediction of precipitation in warm season convective
systems, Sun and Jay
Miller initialized a warm rain cloud-scale model using
single-Doppler observations of a hailstorm from a WSR-88D
radar and performed short-term forecasting experiments. The
model was initialized by assimilating the radar observations
for a period of 30 minutes using a 4DVAR technique. A series
of data assimilation and forecasting experiments were conducted
to examine the sensitivity of the forecast with respect to
environmental, model, and data uncertainties. Results indicate
that the gross structure, location, and propagation speed
of the predicted storm were in good agreement with the observations
up to two hours. Good correlation between the forecast storm
and the observed storm was obtained. It was found that the
general characteristics of the forecast storm resembled those
of the storm obtained from dual-Doppler analysis. Some of
the results were included in a book chapter on thunderstorm
forecasting.
Enhancement of MM5/WRF Three Dimensional Data Assimilation
System (3DVAR)
Syed Rafat Husain Rizvi’s research
enhanced the current capability of the MM5/WRF 3DVAR system,
which was released recently for community use. These enhancements
include the following:
1) Two additional observational types were implemented into
3DVAR: (a) Buoy: useful surface observations collected over
the ocean that provide information over the vast data sparse
region. (b) Wind Profiler: Very high-resolution vertical
profiles of wind observations are mainly available over land.
Unlike most upper air observations, these observations are
available at very high frequency and have the potential benefit
of resolving important mesoscale weather systems.
2) Implementation of Conjugate Gradient (CG) minimization
method. The potential benefit of implementing this method
in 3DVAR lies with the fact that it is much more efficient
than the existing Quasi Newton method as seen in Fig.
13.
|
|
| Figure
13: Implementation of Conjugate Gradient (CG)
minimization method to enhance the MM5 and WRF 3DVAR
system. |
3) Implementation of incremental 3DVAR approach. An outer
loop has been added to the current 3DVAR. This approach produces
good analyses as it makes use of additional observations
during the subsequent outer loop iterations.
4) Efficient calculation of background errors (BE): The
role of background errors is very important in any analysis
scheme as it decides the spread of information contained
in the observations. An efficient way of computing BE has
been implemented. The new code saves about 80-85% of CPU
without losing much accuracy. Interpolated BE does not give
good results, so whenever an analysis for any new region
is run, respective BE have to be computed. With the new code
it is possible to generate BE for any region within a reasonable
time frame. See Fig. 14.
|
|
| Figure
14: Efficient calculation of background errors
(BE) is very important in any analysis scheme as it
decides the spread of information contained in the
observations. An efficient way of computing BE has
been implemented. The new code saves about 80-85% of
CPU without losing much accuracy. |
Optimal
use of existing observations & the potential benefits
of new observing systems (top)
Assimilation of surface observations with EnKF data assimilation
system
Surface-layer (screen-height) observations, such as exist
in a typical mesonet, are under-utilized in current data
assimilation (DA) algorithms because of weak coupling with
the free atmosphere aloft. But simulation and short-range
forecasts of near-surface conditions could benefit from these
data. The Ensemble Kalman Filter data assimilation algorithm,
which uses anisotropic and flow-dependent covariance information
to spread the influence of an observation, is appropriate
for this task. Joshua Hacker (ASP)
and Snyder used a column
PBL model to successfully assimilate simulated observations.
|
|
| Figure 15: Significant
error reduction for a) temperature, b) wind, and c) moisture
is obtained by in inclusion of Surface-layer (screen-height)
observations which are assimilated using the Ensemble
Kalman Filter data assimilation algorithm. |
The results (Fig.15) showed significant
error reduction for a) temperature, b) wind, and c) moisture.
This places an upper bound on the advantage of using surface
observations to specify the state of the PBL. Hacker and Snyder also
performed parameter estimation experiments to mitigate the
negative effects of simulated model error, showing that the
observations can be useful to correct erroneous parameters.
The successful assimilation suggests that the potential exists
for surface observations to improve numerical simulations
and forecasts of air-pollution events, convective outbreaks,
and cyclogenesis where PBL preconditioning is important.
Design of optimal global observing systems
Because resources for meteorological observations are limited,
choices must be made among proposed enhancements to the weather
observing system. Rebecca Morss analyzed
choices among observing systems from a public policy research
perspective. She explained the role and importance of problem
definition in policy research to a meteorological audience,
using five alternate definitions of the observing system
design problem to demonstrate how different problem definitions
can lead to different results. The problem definitions presented
build towards an appropriate problem definition for observing
system design, an important prerequisite for designing a
more cost-effective, integrated global observing system.
Designing optimal observing systems requires balancing the
costs and benefits of different observations. Although the
meteorological community has discussed the question of optimal
investment in observations for more than three decades, it
still lacks a practical, systematic framework to analyze
the issue. Morss collaborated
with two economists, Kathleen Miller (ESIG) and graduate
student Maxine Vasil (University of Colorado) to develop
such a framework, using an economics approach. To demonstrate
the framework, they are analyzing the appropriate level of
investment in observations for an idealized example, based
on previous research results.
Effect of incorporating total ozone from the Total Ozone
Mapping Spectrometer (TOMS) into the initial condition of
MM5 forecasts
Christopher Davis and Simon
Low-Nam continued their investigation of the effect
of incorporating total ozone from the Total Ozone Mapping
Spectrometer (TOMS) into the initial condition of MM5 forecasts.
They have identified cases of large forecast error at time
ranges of 48-72 h over the eastern Pacific Ocean and western
North America from the period January-February 2000. Ozone
data are incorporated using the well-known correlation
between potential vorticity (PV) and ozone and statistics
of the vertical structure of PV. The effect of ozone data
was found to be minimal in the case with the largest error
(31 January 2000). Examination of sensitivity using the
MM5 adjoint model suggests that the most sensitive region
is the middle troposphere, not the tropopause in this particular
case. Thus ozone, whose fluctuations are dominated by the
tropopause and lower stratosphere on synoptic time scales,
cannot improve this particular forecast. Other cases are
being examined to assess the generality of this result.
Optimal
strategies for obtaining targeted observations in data
sparse regions (top)
Research with ensemble Kalman filters
Mesoscale data assimilation is hampered by two factors.
First, observations that are plentiful (e.g., Doppler radar
measurements of wind and reflectivity) involve only a subset
of atmospheric variables, while observing platforms such
as radiosondes that measure all relevant variables are sparse
and resolve mesoscale motions poorly. Second, the balances
between variables, such as geostrophy, that pertain at large
scales in the atmosphere are questionable at the mesoscale;
these balances are an important component of more traditional
assimilation schemes such as 3DVAR. To overcome these difficulties,
there has been substantial effort within MMM and the Data
Assimilation Initiative to explore the potential of the ensemble
Kalman filter (EnKF) for mesoscale assimilation.
EnKF used in the analysis and prediction of convective
scale motions in a simple cloud model
Chris Snyder and Fuqing
Zhang (Texas A&M University) applied the EnKF to the
analysis and prediction of convective scale motions using
simulated observations in a simple cloud model. They have
shown that a 50-member EnKF is able to estimate tangential
and vertical velocity and temperature, given four to six
scans (or about 20 to 30 minutes) of simulated observations
of radial velocity extracted from a reference simulation
of a supercell thunderstorm (Snyder and Zhang 2003). They
have also explored how the assimilation is influenced by
changes in the available observations or in the quality of
the initial estimate of the storm (Zhang et al. 2003). These
results are the first for the EnKF outside of global atmospheric
models and hold substantial promise for the application of
the EnKF to meso- and convective scales.
|
|
| Figure
16: The EnKF was applied to the analysis and
prediction of convective scale motions using simulated
observations in a simple cloud model, showing that
a 50-member EnKF is able to estimate tangential and
vertical velocity and temperature, given four to six
scans (or about 20 to 30 minutes) of simulated observations
of radial velocity extracted from a reference simulation
of a supercell thunderstorm. These results are the
first for the EnKF outside of global atmospheric models
and hold substantial promise for the application of
the EnKF to meso- and convective scales. |
A numerical technique to correct the time-lag and calibration
errors in Vaisala radiosonde humidity measurements
Relative humidity (RH) measurements from the Vaisala RS80-H
radiosondes used at most National Weather Service (NWS) sites
are known to be inaccurate in the upper troposphere (UT)
due to measurement errors that result from slow sensor response
and an inaccurate calibration at low temperatures. These
measurement errors have implications for parameterizing water
vapor and cloud initiation in the UT in numerical models,
assessing the strong dependence of longwave radiative transfer
in the UT on the water vapor concentration, and validation
of remote sensor water vapor retrievals from the next generation
of satellite sensors. Larry Miloshevich developed
a numerical technique to correct the time-lag and calibration
errors in Vaisala radiosonde humidity measurements. Comparison
of corrected radiosonde measurements with simultaneous measurements
from the reference-quality NOAA cryogenic hygrometer demonstrates
that the corrections remove the temperature-dependent dry
bias and substantially reduce the variability.
|
|
| Figure
17: Mean (dots) and standard deviation (bars)
of the percentage change in measured water vapor that
results from correcting the sensor time-lag and calibration
errors, shown as a function of altitude for all soundings
from the Miami NWS site during the July 2002 NASA/CRYSTAL-FACE
experiment. Vertical bar is the standard deviation
of the tropopause altitude. |
Application of the humidity corrections to NWS radiosonde
data (Figure 17) shows that the
relative humidity in the UT is increased by 20% on average,
and more than 40% for some soundings. The corrections also
decrease the water vapor in the lower stratosphere by 20%
on average, thereby steepening the humidity profile in the
troposphere-stratosphere transition region, with implications
for studies of stratospheric dehydration.
Extended EnKF tests with cloud models including WRF
William Skamarock and Snyder extended
tests of the EnKF to other cloud models (including a prototype
implementation for WRF). They successfully used simulated
observations from an idealized supercell, thus confirming
the results of Snyder and Zhang (2003), as well as from an
idealized squall-line simulation. These latter experiments
extend the results to a situation in which individual convective
cells are not quasi-steady state and the assimilation period
of two hours covers several cell lifetimes. Skamarock and Snyder also
completed experiments, again in the context of a simulated
squall line, in which the initial ensemble mean used in the
EnKF includes error in the environmental sounding. They found
that assimilation severely degraded for cases in which errors
are comparable to likely uncertainty in soundings (several
m s-1), and they began exploring the possibility
of estimating the environmental soundings based on the radar
observations.
Developing an EnKF system for regional mesoscale data assimilation
based on WRF
The past year also saw progress in developing an EnKF system
for regional mesoscale data assimilation based on WRF. Lateral
boundary conditions are an important source of uncertainty
in regional forecasts and this uncertainty must be accounted
for in the EnKF. To this end, Ryan Torn, Gregory Hakim (both
of the University of Washington), Alain
Caya, and Snyder began
exploring various treatments of the lateral boundary conditions
for short-range limited-area ensemble forecasts.
Incorporation of WRF into the Data Assimilation Research
Testbed (DART)
In addition, Caya and Skamarock led
an effort, together with Jeffrey
Anderson, Dale Barker, and Snyder,
to incorporate WRF into the Data Assimilation Research Testbed
(DART, developed by the NCAR Data Assimilation Initiative)
in order to facilitate further development and testing of
the EnKF for WRF. Caya and Snyder began
assimilation experiments with WRF using the EnKF and simulated
observations on a continental US domain. Preliminary results
suggest that the EnKF is an effective assimilation scheme
for this problem. While regional mesoscale assimilation and
ensemble forecasting is important in its own right, these
results also hold the potential for a unified approach from
meso- to convective scales, in which the ensemble produced
by the EnKF on regional domains will supply boundary conditions
for convective-resolving nested domains.
Doppler radar data assimilation using the MM5/WRF 3DVAR
system
Qingnong Xiao worked in
the area of Doppler radar data assimilation using the MM5/WRF
3DVAR system. He modified the 3DVAR system to include capabilities
of producing vertical velocity (w) increments, as well as
cloud water (qc) and rainwater (qr) increments. Doppler radial
velocity and reflectivity data can be assimilated into 3DVAR
analyses with his effort. He demonstrated that the 3DVAR
system for Doppler velocity assimilation is stable and robust
in a cycling mode making use of high-frequency radar data.
Assimilation of Doppler radial velocities improves the prediction
of the rain-band movement and intensity change. As a result,
an improved skill for the short-range heavy rainfall forecast
is obtained. Xiao found
that continuous assimilation with 3-h update cycles is important
in producing an improved heavy rainfall forecast. This work
is a collaboration among MMM, ATD, the Korean Meteorological
Administration and Seoul National University, Korea.
|
