APPLICATION OF NUMERICAL MODELS IN THE FORECAST PROCESS FROM NATIONAL CENTERS TO THE LOCAL WFO David

1
APPLICATION OF NUMERICAL MODELS IN THE FORECAST
PROCESS -FROM NATIONAL CENTERS TO THE LOCAL
WFO David W. Reynolds National Weather
Service WFO San Francisco Bay AreaMonterey, CA
NOAAs National Weather Service

• 50th Anniversary of Operational Numerical Weather
  Prediction
  
• June 17, 2004

NOAAs National Weather Service
2
Models and Forecasters

• Focus talk on T12 hrs and beyond
• For these valid times the forecaster is heavily
  dependent on numerical model guidance.
  
• Major paradigm shift is taking place in how
  forecasts are being prepared at local WFOs.
  
• Have transitioned from forecaster spending
  several hours reviewing model output then
  manually preparing text forecasts to nudging an
  existing digital forecast database toward a new
  collaborated model solution, drawing the
  forecast.
• National Centers made significant paradigm shift
  several years ago when transitioning from acetate
  and grease pencils to digital product creation.
  
• NCEP service centers very often initiate guidance
  forecast from new model guidance the forecast
  challenge is picking the right model and value
  adding by knowing how to adjust for model biases.
  

3
Models and Forecasters

• Where do forecasters add value?
• A difficult question and hard to find good
  objective quantifiable data, especially related
  to sensible weather parameters.
  
• Long standing QPF data shows forecasters can add
  value. Equivalent to 10-15 years of model
  development.

4
Models and Forecasters

• Current QPF process in NWS shows value added in
  each step of the forecast process.
  
• However uncertainty still large. Migrate from
  deterministic to probabilistic.
  
• USWRP supported

5

Models and Forecasters

• For other manual forecast parameters such as
  medium range surface pressure it becomes more
  problematic to determine forecaster
  contributions.
• Relates to forecast uncertainty.
• Begs the question of the role of the forecaster
  when uncertainty is high.
  
• Deterministic versus probabilistic forecasts.

6
Models and Forecasters

• UKMET office does show some skill in days 3 and
  4 utilizing special software package called
  On-Screen Field Modification that allows a four
  dimensional adjustment of model output through
  manipulation of surface pressure features.

7
Models and Forecasters The Future

• Interactive Forecast Preparation System the
  new paradigm. Paper by Ruth to follow.
  
• 2.5 or 5 km gridded forecasts at 1, 3, 6, or 12
  hour temporal frequency of sensible weather, wave
  heights out through day 7.
  
• Formatters used to generate text forecasts from
  the grids.
  

8
Models and Forecasters The Future

• More dependence now on statistically post
  processed model output then before (MOS).
  
• Model biases in 2m temperatures, dewpoints, and
  inability to obtain raw model output consistent
  with grid resolution (2.5-5 km).
  
• Shear workload of populating the large number of
  grids required questions how relevant the
  forecaster will be in the gridded forecast
  process.
• Forecasters coping with workload but no reliable
  measure of forecast quality. Lack of analysis
  of record at 2.5 or 5 km.
  
• Grids deterministic implying accuracy. No
  uncertainty provided to the customer.

9
Models and Forecasters The Future

• No gridded verification to date.
• Utilization of point verification would indicate
  little value being added by forecasters over MOS
  for most if not all parameters.
  
• Must have a measure of uncertainty.

10
Models and Forecasters The Future

• How should we proceed?
• Rapid expansion into short and medium range
  ensemble forecasting provides new opportunities.
  
• Multi-model ensembles available from the WRF
  architecture and multi-model medium range
  ensembles (GFS and CMC) will provide the catalyst
  to move in a new direction.
• Statistically post processed ensemble means and
  calibration of ensemble probabilities could
  provide the basis for populating the grids.
  
• Ensemble means and downscaling methods could
  provide the expected value forecasts. (Jun Du)
  
• Calibrated probabilities provide a reliable
  measure of uncertainty (probability density
  function).

11
Models and Forecasters The Future

• So how does the forecaster fit into this new
  picture both nationally and locally?
  
• EMC and MDL prototype use of calibrated ensemble
  based forecasts and downscaling techniques to
  populate sensible weather grids (Test Bed).
  Parallel with the deterministic approach.
  Provides objective measure of uncertainty we
  cant get from forecasters.
• Hand off this methodology to WFOs who will QC
  grids locally.
  
• WFOs primary function will be the interface with
  local partners and customers to educate on
  interpretation and application of output.
  
• EMC and HPC begin testing methods to improve
  model forecasts of significant events via manual
  intervention in the assimilation cycle.
  
• THORPEX provides an opportunity to test this
  concept.
  

12
Models and Forecasters The Future

• Intervention
• Forecaster intervenes with data assimilation
  cycle during significant expected weather events
  in order to maintain critical observations in the
  forecast cycle.
• Methodology has proven useful in MET office for
  major extratropical cyclones.
  
• Applications for both west and east coast rapid
  cyclogenises.
  
• Forecaster now adds value at the front end, not
  the back end, of the process.