FCST

1. Weather Forecasting (AK13)
Methods of Forecasting
Persistence, Climatology, Trend, Analog, and
Numerical Weather Prediction (NWP)
NWP Process:
-Weather Observations
-Data Assimilation
-Forecast Model Integration
-Interpretation
Modern Numerical Weather Prediction Models
Short-Range Forecast Models
Medium-Range Forecast Models
Nowcasting
Forecasts of Forecast Accuracy: Ensemble Forecasting

2. Vilhelm Bjerknes is considered to be the father of modern meteorology. He led the
Bergen School, a talented group of young scientists who made fundamental
advances in understanding the weather.
1904
Vilhem Frimann Koren Bjerknes (1862-1951)
Vilhelm Bjerknes first recognized that numerical weather prediction was possible in
principle in 1904. He proposed that weather prediction could be seen as essentially
an initial value problem in mathematics, since equations govern how
meteorological variables change with time, if we know the initial condition of the
atmosphere, we can solve the equations to obtain new values of those variables at a
later time (i.e., make a forecast).
From Prof Chuck Wash

3. NWP Primitive Equations
The primitive equations describe flow on the sphere under the assumptions
that vertical motion is much smaller than horizontal motion and that the layer
depth is small compared to the radius of the sphere..

5. Lewis Fry Richardson
(1881-1953)
Lewis Richardson, following Bjerknes’ suggestion, devises numerical methods to
solve the system of equations.
Richardson, a brilliant British mathematician, spent almost three years
developing the techniques and procedures to solve the equations.
It took six weeks to produce a six hour forecast, and the results were
very poor. Most of the calculations were performed in a hay loft where
Richardson was assigned as an ambulance driver in France during
The First World War. A dedicated pacifist, Richardson left
meteorology for good when the British government transferred the
meteorology office to the War Department shortly after the war.
Richardson published a book describing
his experiment in 1922. In it, he imagined
that, someday, vast numbers of people
working on parts of the equations in a
huge building could produce weather
forecasts in a timely fashion.
1913-1916
From Prof Chuck Wash

6. 1947-1948 – Jule Charney and Ragnar Fjortoft
developed a simplified, filtered system of equations for weather
forecasting.
1948 – John Von Neumann developed the first stored
program computer (ENIAC).
1950 – Charney, Fjortoft, and von Neumann produce the first
successful computer weather forecast. The Birth of computer
based Numerical Weather Prediction (NWP)
Jule Gregory Charney
1917-1981
Von Neumann with ENIAC…. which occupied a 30’X50’ room…
but can now be
replicated on a
single chip…
“ENIAC-on-a-chip”
1947-
1950
This project was funded by the U. S. Navy
through the Office of Naval Research
From Prof Chuck Wash

7. • Installation of the Navy's First Supercomputer
The CDC 1604 pictures show the installation and use of the Navy's first
supercomputer, which was installed in 1960 by a team including Seymour Cray
himself. "World's first all-solid-state computer -- Model 1, Serial No. 1 of Control
Data Corporation's CDC1604 -- designed, built and personally certified in the
lobby of Spanagel Hall (room 101) by the legendary Seymour Cray."
LTJG Harry Nicholson 1960
Spanagel Hall Room 101
Spanagel Hall room 101

8. Methods of Forecasting
• Climatology (Long term average)
• Persistence
(What happens today will happen tomorrow)
• Trend (short period NOWCAST)
(speed, size, intensity, and direction unchanged)
• Analog
(Assumes that history repeats itself and weather changes over time)
• Conceptual Models(Dynamic description of evolution of weather
phenomena)
• Numerical Weather Prediction (NWP) Models

9. Forecast Error
The current NWP predictability barrier for the atmosphere is ~14
days.
TIME
Error
Persistence
Climatology
Forecast Models NWP
Human Value added
(decreasing)
Better Remote sensing data and smarter climo
Ensemble Models
Smart
Climo
7 days
14 days

10. Factors that determine predictability
Model Initial Condition Uncertainty
- Lack of full observational coverage at every point in model
- Observations are not measured to a infinite degree of precision
Model Uncertainty
- Model formulation: Uncertainties in parameterizations
- Model Processing Resolution precision ≠ accuracy
- Equations used do not fully capture processes in atmosphere
Nonlinear Dynamics in the atmosphere and oceans
- moisture
- tropical influences
Chaos Theory: small differences in initial state can lead to large
differences later. NWP ~ 14 day limit with current methods

11. Characteristics of a Chaotic System
Sensitive to Initial Conditions
Aperiodic: Solutions never repeat
exactly, but may appear similar
Courtesy Maj Eckel

12. Some Reasons for Model Error
1. Model Resolution is not sufficient to capture all
features in the atmosphere.
2. Initial Observations are not available at every point in
the atmosphere.
3. The observational data can not be measured to an
infinite degree of Precision
4. Equations used by a model do not fully capture
processes in the atmosphere.
Model Definition: A description of observed behavior, simplified by
ignoring certain details, which allows complex systems to be understood and
their behavior predicted within the scope of the model.

13. Old Method of Data Assimilation

14. Bring all the data (spread in time and space) into the model
Create the Best Initial Conditions for the Model
Data void
areas use
previous
forecasts
Indirect
observations
Data Assimilation
From Capt. Gunderson

15. Chaff ?

16. Data Assimilation
Filtering out the Bad data
Bad Good
LF Richardson 1922
human computer http://www.earthsimulator.org.uk/launch.php
8 hr fcst took 6 weeks
to calculate & failed

17. 1999 French Wind Storm
Don’t Filter out the Good Data
• Bad Forecast Better Forecast
Observations were 'thrown away' as the computer assumed
that they were 'wrong' ("quality control")!

18. Combination of satellite with ground
based data is providing global data
coverage nearly continuously.
Data counts reached are in the tens of
millions daily.

20. Data Assimilation
-Quality Control & Weights
-Objective analysis
vertical and horizontal resolution selected
Next Start The Model
Initialization of NWP Model
Balance, spatial & temporal

21. Forecast Error
The current NWP predictability barrier for the atmosphere is ~14
days.
TIME
Error
Persistence
Climatology
Forecast Models NWP
Ensemble Models
Smart
Climo
7 days
14 days
Model
Initialization

22. Model Topography
X X
X X
X
X
X X X
Global Model Grid points X

23. 27km and 3km resolution
81 times more grid points 3km grid

24. Here is an comparison of
model terrain (colors) with
actual terrain.
Naturally, one of the limits of
models is to properly depict
the physical processes that
are associated with air flow
over or along complex or
steep terrain.

25. Improved Marine Forecast of SST with higher res.

26. Takes at least 5 grid
points to define a
feature in a grid point
model

28. Vertical Levels:
-More Near Surface for Moisture & Heat transfer
-Less resolution needed ~600-to ~300mb
-Increases resolution needed near tropopause for Jet stream
AFWA MM5
15km grid
resolution

29. Sigma Coordinate
Pressure
---------------------
Surface Pressure

30. Model Parameterization
When models cannot resolve feature and/or processes that
occur within a single grid box they are parameterized.
Model the effect of process rather than process itself

32. IBM Blue Gene L
2008 Lawrence Livermore Labs
106,498 nodes
Linux Cluster
NCEP/NOAA 2015
IBM/CRAY-XC40

33. Types of Models
• Global Models
• Mesoscale-Nested Regional Models
• Statistical Models (MOS)
• Nowcast Models
• Trajectory and Dispersion Models Hysplit
• Ensemble Models
• Wave Models
• Adjoint Models (look back in time)

34. Meteorological Models
• NAVGEM (Navy),
ECMWF (Europe),
GFS (NOAA NCEP),
UKMET, Canada and
many others
• Resolution ~10 to 100km
• 0 to 16 (32) Day Forecast
• 1 - 6 hour output Interval
• Run every 6 or 12 hours
• COAMPS (Navy), MM5
(Portable), RUC/RAP
(NOAA NCEP) , WRF
(NOAA NCEP)
• Resolution 1 to 30 km
• 0 to 4 Day Forecast
• 1 - 3 hour output interval
• Run every 1 to 6 hours
Global Regional/Local

35. Global or Regional Model

36. Model Output Statistics (MOS)
Creates a subgrid scale forecast
Captures resolution by doing statistics

37. Seattle MOS www.meted.ucar.edu Fcsting in the West (B. Colman)
12hr 30hr

38. Cathy Kessinger et al., NCAR

39. Hysplit & Cal Puff
Trajectory Model and Dispersion Model

40. John Merrill,University Rhode Is.

41. Iceland Volcano halts
Europe flights
16,000 Eur flights canceled on Apr 16th 2010
NASA-Terra 11:39z 15 April 2010

42. T
The true state of the
atmosphere exists as a single
point that we never know
exactly.
Nonlinearities drive apart the
forecast trajectory and true trajectory
(i.e., Chaos Theory)
Encompassing Forecast Uncertainty
12h
forecast 36h
forecast
24h
forecast
48h
forecast
T
48h
verification
T
T
T
12h
verification
36h
verification
24h
verification
An analysis produced to run a model
is somewhere in a cloud of likely states.

43. T
Ensemble Forecasting:
-- Encompasses truth
-- Reveals uncertainty
-- Yields probabilistic information
Courtesy Maj Eckel
T
An ensemble of likely analyses leads to an ensemble of likely forecasts
Encompassing Forecast Uncertainty

44. Hurricane Frances
Ensemble Forecast
Problems with
tropical influences
in mid latitude
forecasts and vice
versa.

45. Frances Mean Ensemble and
Analysis

46. An Example of an Ensemble Forecast

48. NCEP ENSEMBLE FORECAST
Initial
Conditions
6 day
Day 15
NCEP web training material link

49. AVN AVN+GDFL
Floyd and Gert Sept. 1999
Wave Forecasts

50. Katrina WW3 05/08/28/18z
Significant Wave Height = avg height of 1/3rd largest

51. Adjoint Models
www.nrlmry.navy.mil/~langland/adap_about.html
www.nrlmry.navy.mil/projects/adap.html
Small inaccuracies in sensitive regions can amplify rapidly
48 Hour Forecast Sensitivity 72 Hour Forecast Sensitivity
Adjoint Model: Maps a sensitivity gradient vector from a
forecast time, to an earlier time, which can be the initial
time of a forecast trajectory.

52. Checking the models

53. Model Verification
UW
Analysis
Model 3
Model 2
Model 1

56. NWS Traditional Forecasting Process
– Schedule Driven
– Product Oriented
– Labor Intensive
National Centers
Generate Graphical Products
National Centers
Model Guidance
Field Offices
Type Text Products
TODAY...RAIN LIKELY.
SNOW LIKELY ABOVE 2500
FEET. SNOW
ACCUMULATION BY LATE
AFTERNOON 1 TO 2
INCHES ABOVE 2500 FEET.
COLDER WITH HIGHS 35
TO 40. SOUTHEAST WIND
5 TO 10 MPH SHIFTING TO
THE SOUTHWESTEARLY
THIS AFTERNOON.
CHANCE OF
PRECIPITATION 70%.
MARYLAND EASTERN SHORE
EASTON
PTCLDY CLOUDY PTCLDY PTCLDY SUNNY PTCLDY
60/52 63/54 65/47 55/40 55/37 50/33
POP 20 POP 20 POP 20 POP 20 POP 10 POP 10
U.S. Drought
Monitor
Excessive Heat
Products
Threats
Assessments








57. User-Generated Products
New Forecasting Process
– Interactive
– Collaborative
– Information
Oriented
NWS Automated Products
Text
Graphic
Digital
Voice
National Digital
Forecast
Database
Local Digital
Forecast
Database
Field
Offices
National
Centers
Collaborate
Data and Science Focus
National Centers
Model Guidance
Grids

58. The pencil tool allows for quick “stretching”
of values. The GFE then recalculates the grid.
Editing: manually

59. National Digital Forecast Database
(NDFD)
• Contains a seamless
mosaic of NWS digital
forecasts
• Is available to all users
and partners – public and
private
• Allows users and
partners to create a wide
range of text,
graphic, and image
products