High Resolution WRF Model Output
High resolution Numerical Weather Prediction (NWP) models have been widely used
to help with short range forecasting and warning operations. Recent advances in NWP and
in computational efficiency have resulted in an improvement in and the availability of
high resolution model forecasts on the convective scale.
The High Resolution Rapid Refresh (HRRR) model is a 3-km hourly updated nest inside of the Weather
and Research Forecast (WRF) Rapid Refresh (RR) model. WRF RR is the next-generation
version of the 1-h cycle system and will replace in 2010 the Rapid Update Cycle (RUC) which
currently supports NWS operations. The RR uses a version of the WRF model and the
Gridpoint Statistical Interpolation (GSI) analysis largely developed at NWS National Centers for
Environmental Prediction (NCEP). The HRRR is initialized with latest 3-d radar reflectivity
via 13km backup RUC at ESRL/GSD, which includes radar reflectivity assimilation via its
radar-DFI (digital filter initialization) technique. The HRRR
is believed to be the only hourly updated, radar-initialized, storm-resolving
model running at this time over the US. Real-time HRRR data can be viewed at the ESRL HRRR page
The NCEP 4 km WRF-NMM run for SPC uses the WRF-NMM model code very similar to that
used in the operational HiresW. It is initialized from interpolated NAM model output. The basic
physics selection the same as the NAM (Ferrier microphysics, GFDL radiation, MYJ PBL, and the NOAH land surface model)
except for no parameterized convection. Also, the microphysics are tweaked to allow for larger raindrops and thus more
intense simulated radar signals.
Real-time NCEP 4 km WRF-NMM data can be viewed from the following web pages:
00 UTC run or the
12 UTC run.
During the past few years, NWS Raleigh has been developing a local high resolution
modeling capability. We expect to use the model in both an operational and research
capacity to help refine phenomena related to smaller scale weather events that are
common to our area. Currently the NWS Raleigh is running a 4km WRF-NMM four times
a day using 12km NAM initial/boundary conditions and no convective parameterization.
Real-time WFO RAH 4KM WRF-NMM data can be viewed at the NWS Raleigh page
Real time high resolution convective modeling simulations have become more and
more common in recent years. The growing exposure to and participation in
modeling has resulted in an improved understanding of the role this data should
provide. The Experimental Forecast Program at the
Hazardous Weather Testbed (HWT) focuses on the application of
cutting edge numerical weather prediction models to improve severe weather
forecasts. Experience at local NWS Weather Forecast Offices, journal publications, and results from the
HWT has provided many insights and lessons learned which are shown below:
- The output from high resolution models remains strongly correlated to
that of the parent model used to initialize.
- High resolution models regularly struggle when the synoptic scale forcing
is weak. In addition, if the synoptic scale conditions are not reasonably well
forecast, the modeling of mesoscale details will struggle.
- WRF simulations that explicitly handle convection typically take 2-3
hours to "spin up" convection, which significantly degrades the accuracy and
utility of the first few hours of model data. Forecasters need to recognize
this limitation and balance the notion that the most recent model run is the
best with what model run is likely the most appropriate.
- Given the spin up issues, model initialization sensitivity, and
other issues, the timing of convection can struggle, but often times the
pattern and big picture depiction is handled well.
- Recent studies have shown that reducing the horizontal grid spacing
to 1 or around 2km provides limited improvement over 4km simulations with
increased computational cost.
- A participant in the HWT notes that while there have been many
changes and improvements to modeling in recent years (microphysics schemes,
PBL schemes, radar assimilation, etc), several longstanding problems still
exist (sensitivity to initial conditions, sensitivity to model physics,
parameterization of features, upscale growth, etc).
- Despite the limitations, recent experience shows that high resolution
models still provided valuable guidance to forecasters including a general
depiction of convective coverage, geographical location, storm initiation,
convective mode, and precipitation pattern to forecasters. Although the location/timing
of features may not be exactly correct, seeing the overall “character of the
convection” can still be of great utility to forecasters especially considering
they are not available in the current suite of operational models (i.e. NAM/GFS).
- Participants in the HWT note that one of the big challenges in the
future will be how to best incorporate high resolution model guidance
into the forecast process. Many forecasters already feel that they are at
or even past the point of data overload, they need proof that these models
are useful and good use of their time. In addition, an efficient way to use
and view the data is needed.
- Another lesson learned from the HWT is that training is a critical
issue with using high resolution models to ensure that the data is used
effectively and efficiently. Given the wide range in methods to view the
data (single deterministic run, multiple deterministic runs, probabilistic
guidance) and the numerous new (simulated reflectivity, updraft helicity,
updraft/downdraft strength), significant training is going to be required
regarding both what to view, and how to view it.
- The future of high resolution modeling likely won’t be driven by
single, deterministic model forecasts but rather an ensemble of convective
resolving models and probabilistic guidance from storm scale ensembles to
address the uncertainty that accompanies all weather forecasts.
It is interesting to note the success the 5 hour forecast from
the 21 UTC High Resolution Rapid Refresh (HRRR) valid at 02 UTC on 4/26
had in the placement and intensity of the isolated convective cell east of Raleigh.
The images to the right show the HRRR forecast of reflectivity and
updraft helicity were near the location of the supercell thunderstorm
that produced the Zebulon tornado.
The top portion of the image to the right shows
the HRRR 1km AGL reflectivity product valid at 02 UTC which placed a small area of convection to
the east of Raleigh. This forecast is very similar to the observed
regional reflectivity product valid at 2358 UTC on 4/25 shown in the
bottom portion of the image to the right. While the timing is nearly
2 hours later then observed, the placement is impressive.
The timing issue could result from numerous factors including the time it takes for the
HRRR to develop or "spin up" convection after the initialization time.
The HRRR one hour maximum updraft helicity product shown as
the second segment of the image to the right is very close to
the observed tornado track shown in the bottom segment of the
image to the right. The location, length and orientation of the track
are very similar.
The chart below allows the viewer to compare low level, simulated
radar reflectivity forecasts from various sets of high resolution model guidance
described above and the observed radar reflectivity. Multiple runs of the
HRRR are shown in the first 5 rows, the next two rows show the
NCEP 4 km WRF-NMM run for SPC, with the NWS Raleigh 4km WRF-NMM shown next.
The horizontal axis begins at
22 UTC and continues through 02 UTC and is centered at around the
hour in which the tornado is observed (00 UTC). Move your cursor over the thumbnails in the matrix below to
view popup windows with enlarge imagery of forecast and observed reflectivity or
click on the individual pane to open a larger version of that image.