An application for incorporating pattern recognition into the lake effect snow forecasting process a

1
An application for incorporating pattern
recognition into the lake effect snow forecasting
process at WFO Binghamton, NY.

• Mike Evans

2
Outline

• Lake effect snow band types in central NY.
• Modeling lake effect snow.
• Combining model output with pattern recognition.
• A pattern recognition improvement application.
• November 30th and December 1st, 2007.

3
WFO BGM County Warning Area
4
Single Bands
5
Multi-bands
6
Multi-bands
7
Modeling

• Large single bands can be modeled realistically
  using models with resolutions of 5 to 20 km
  (Ballentine 1998, Niziol 2003).
• Smaller multi-bands require higher resolution
  (Watson et. al. 1998).

8
20 km NAM February 5, 2007
9
Smaller multi-bands from a 2.5 km MM5 (Watson
et. al. 1998)
10
12 km NAM vs. 6 km WRF
11
Modeling Summary

• Medium resolution models can model lake effect
  snow bands.
• Small multi-bands best modeled at resolutions of
  5 km or less.
• Placement / intensity errors still occur.

12
Given that high resolution models are not
perfect, the following forecast methodology has
been developed

• Examine low-resolution models to forecast the
  meso-scale environment (flow direction, moisture
  depth, inversion height, ect).
• Use pattern recognition / experience to develop a
  set of expectations on outcomes.
• Use high resolution models to test expectations.
• Final forecast utilizes a combination of pattern
  recognition and explicit output from high
  resolution models.

13
Experience / pattern recognition

• Forecasters have a wide range of experience.
• Forecasters have a range of abilities regarding
  recall of previous similar events (pattern
  recognition).
• The best forecasters use utilize a vast wealth of
  experience to develop expectations and modify
  explicit model output.

14
An application to make anyone a pattern
recognition expert

• Application grabs current 12 km NAM forecast
  soundings from user selectable points.
• Compares forecast data with data from a large
  historical sounding data base.
• Runs an algorithm to determine the most similar
  historical soundings at each hour.
• Provides users with information on these
  historical events.

15
More on the application

• At each hour radar data, model soundings, and
  snowfall can be displayed.
• Data from the top 5 similar events can be
  displayed a range of possibilities.

16
Example November 30, 2007
17
Most similar events - hourly
18
Example December 1, 2007
19
Most similar events - hourly
20
Summary the BGM LES pattern recognition
application

• Analyzes a current meso-scale forecast compares
  to a large historical data base.
• Returns data from the most similar historical
  cases.
• Anyone can be a pattern recognition expert.
• A range of possible outcomes is displayed.
• Output from high resolution models can be
  examined critically.

21
Now available online

• http//www.lightecho.net/les/
• http//www.lightecho.net/lesBGM/ (under
  development)

22
References

• Ballentine R. J.,  A. J. Stamm,  E. E. Chermack, 
  G. P. Byrd, and D. Schleede 1998 Mesoscale Model
  Simulation of the 45 January 1995 Lake-Effect
  Snowstorm Wea. Forecasting,13, 893920.
• Niziol, T., 2003 An analysis of
  satellite-derived Great Lakes surface
  temperatures in regards to model simulations of
  lake effect snow. Preprints, 10th Conference on
  Mesoscale Processes, Portland Or, June 2003,
  Amer. Meteor., Soc., CD-ROM, P1.9.
• Watson, J.S., Jurewicz, M.L, Ballentine, R.J.,
  Colucci, S.J. and Waldstreicher, J.S.,1998 High
  resolution numerical simulations of finger lakes
  snow bands.. Preprints, 16th Conference on
  Weather Analysis and Forecasting, Phoenix, Az.,
  January 1998, Amer. Meteor., Soc. 308-310..