Numerical Weather Forecast Models

1
Numerical Weather Forecast Models

• Mr. Patrick Dixon
• Senior Meteorologist/Ship Routing Officer
• Naval Maritime Forecast Activity - Norfolk

2
Overview

• An atmospheric prediction model is a system of
  simulating the atmospheric physical processes.
  The atmosphere is simulated with a discrete
  number of grid points in specific levels, that
  begin from the ground, and reach the upper
  atmospheric layers. By this way, a three
  dimensional grid is produced, which is used for
  all the necessary mathematical calculations

3
A three dimensional grid of a numerical weather
prediction model
4

• The time is proceeding with very small time steps
  of the order of few seconds. The continuously
  repetition of estimations in each time step leads
  to the weather prediction for next day or next
  week. The denser the grid the more realistic the
  atmospheric simulation by the models.

5

• The execution of numerical weather models
  requires the use of supercomputers for two main
  reasons
• Weather prediction models are generally large
  code and requires an execution of a large amount
  of data.
• The results of operational prediction model has
  to be available in an operational time.
• The weather prediction models are divided in two
  categories global and regional

6
Global Models

• In these models the grid points extend to the
  whole earth and the equations are integrated in
  the three dimensional atmosphere (north and south
  hemisphere).
• Examples GFS, NOGAPS, ECMWF, UK MET

7
Regional Models

• Also known as Limited Area Models
• The limited area models are used for the
  prediction of the small scale disturbances. These
  models are based in almost the same equations as
  the global models . The main differences is that
  LAM are executed in a limited area, their grid is
  denser, and they are capable of simulating
  accurately the small scale disturbances.
• In global models the equations are integrated in
  the whole earth, with a sparse grid. In limited
  area models the attention is focused in a small,
  specific area (eg Europe or Greece) and the grid
  is much denser.

8
Regional Models (contd)

• Examples COAMPS, MM5, NAM, NGM

9
NOGAPS 4.0

• On 18 September 2002, NOGAPS 4.0 was upgraded
  from T159L24 to T239L30. The corresponding
  Gaussian grid resolution went from 0.75 to
  0.50.
• In September 2003, the NOGAPS MVOI (optimal
  interpolation) data assimilation system was
  upgraded to the NRL Atmospheric Variational Data
  Assimilation System (NAVDAS), a 3-dimension
  variational system. Additionally, in November
  2003, NOGAPS implemented the use of terrain
  fields from USGS Global Land One-kilometer Base
  Elevation (GLOBE) database and changed the
  gravity wave drag scheme to Webster et al.
  (2003).
• The result of these changes has been an average
  increase in verification scores and fewer bad
  forecasts. Further increases are anticipated when
  assimilation of satellite retrievals is replaced
  by assimilation of raw satellite radiances.

10
NOGAPS Model Tendencies

• Surface lows
• 1) NOGAPS is slightly deep with forecasts of
  mid-latitude surface lows of the North Atlantic,
  North Pacific, and Eurasia by about (0.5 mb).2)
  Rapid deepening cyclones defined as those
  deepening at a rate of 7.5 mb per 12 hours or
  greater are underforecast by at least 3.0 mb
  beyond 48 hours, with a substantial 5.5 mb error
  at 96 hours . Rapidly deepening cyclones, account
  for about 6.6 percent of the dataset. This weak
  bias is largest over eastern North America.3)
  Deepening cyclones defined as surface lows
  deepening at a rate of 1 mb per 12 hours or
  greater, accounting for about 45 percent of the
  dataset, show a weak bias at all forecast times,
  with the smallest error (0.35 mb) at 12 hours and
  the largest (2.1 mb) at 96 hrs. Deepening
  cyclones are increasingly underforecast with
  forecast length4) Filling cyclones defined as
  those weakening at a rate of 1 mb per 12 hours or
  greater, account for about 47 percent of
  occurrences. They have a (deep) central mslp
  bias, for all forecast times. Largest deep bias
  values (-2.5 to -3.0 mb), are found in the latest
  forecast times5) Developing oceanic lows tend to
  be slightly underforecast and slow to deepen with
  an ACPE near zero through 72 hours. Mature,
  filling oceanic lows tend to be -2 to -3 mb
  overforecast and slow to fill. 6) NOGAPS
  tendencies for land and ocean basin surface lows
  are a) Atlantic developing low ACPE is slightly
  weak and slow to deepen by 48 hours. Atlantic
  mature lows are mostly deep and slow to fill.
  Filling low ACPE is -3 to -4 mb by 48/72 hours.
  b) Pacific developing low ACPE is slightly weak
  and slow to deepen by 72 hours. Pacific mature
  lows are -3 to -4 mb deep and slow to fill by 72
  hours. c) In view of the general NOGAPS model
  tendency to underforecast oceanic developing SLP
  lows and overforecast oceanic mature, filling
  surface lows associated surface wind speed
  forecasts also exhibit similar biases in the
  areas of higher wind speeds. Surface wind
  forecasts associated with deepening (filling)
  lows are underforecast (overforecast).

11
NOGAPS Model Tendencies (contd)

• Surface Lows (contd)
• 7) Former West Pacific tropical cyclones are
  typically slow to move during and after
  transition to extra-tropical. 8) Secondary
  cyclogenesis continues to be underforecast. Lee
  cyclogenesis off Southeast Kamchatka and
  Greenland is underforecast. 9) NOGAPS continues
  to merge complex lows into one, usually deeper
  low pressure system, especially at the extended
  forecast period. 10) Surface lows forming south
  of the polar jet under weak synoptic-scale
  forcing are slow to deepen. 11) Surface lows
  north of the polar jet at high latitude (gt50N
  latitude) tend to be too deep. These are usually
  mature lows which have bottomed-out and tend to
  be slow to fill. 12) Sea-level pressure analyses
  and forecasts over the very high terrain of
  Greenland, Himalayas, and Antarctica are suspect
  and should be used with caution. 13) In the warm
  seasons, late Spring to early Fall, a spuriously
  deep surface low is observed in the analysis and
  forecasts over the very high terrain of the
  Himalayas (vicinity 30N-090E). This "lock-in"
  feature is caused by model reduction of station
  pressure to sea-level, and the warm season
  surface air temperatures.

12
NOGAPS Model Tendencies (contd)

• Tropical cyclones 1) NOGAPS exhibits several
  model tendencies associated with a tropical
  cyclone's stages of evolution a) Tropical
  cyclone genesis NOGAPS tends to generate
  spurious tropical cyclones in the extended
  forecast periods (tau 72 and beyond). This is
  most pronounced in the Indian and West Pacific
  Oceans. b) Tropical cyclone phase Due to the
  resolution of the NOGAPS global model, sizes of
  tropical cyclones in the NOGAPS analyses and
  forecasts are almost always too large in areal
  extent and this may cause false interaction with
  nearby tropical cyclones. c) Transition and
  extra-tropical phase For tropical cyclones
  undergoing transition to extra-tropical, the
  forecast surface low is usually overforecast
  (deep), slow to fill, and slow to move. In the
  re-deepening extra-tropical phase, former
  tropical cyclones are underforecast (weak) and
  slow to move. Directional bias is usually behind
  and to the left of the analysis track (slow to
  move and toward the U/L cold air), especially is
  zonal flow. 2) Track errors On average, NOGAPS
  TC forecasts tend to be east and south of the
  verifying position in a cartesian coordinate
  framework. In a storm-relative sense ("following
  the storm"), NOGAPS TC forecast are behind and to
  the left of the verifying position.

13
NOGAPS Model Tendencies (contd)

• Upper-level 1) Forecast upper level troughs and
  associated surface lows moving in strong zonal
  flow tend to be fast to move, especially at
  extended forecast periods. 2) Upper level highs
  south of the polar jet are slightly strong. 3)
  NOGAPS wind speed forecast variability is
  greatest in the 300 to 250 mb jetstream region of
  the upper troposphere. Intense jet level winds
  may be underforecast due to the limited vertical
  resolution of the model.

14
GFS

• GFS stands for the Global Forecast System. The
  GFS incorporates all codes that support the
  production of the GFS suite of products,
  including a medium range forecast model (MRF) and
  a global data assimilation system (GDAS). We will
  generally use GFS to refer to both.
• The predecessor to the GFS was developed
  experimentally during the late 1970s (Sela 1980)
  and implemented as the global forecast model at
  the National Meteorological Center (NMC, now the
  National Centers for Environmental Prediction or
  NCEP) on 18 March 1981 at the following
  resolutions
• Triangular truncation at 30 waves with 12 levels
  (T30L12)
• T24L12 from 48 to 84 hours
• T24L6 from 84 to 144 hours (TPBs 282A and 282B)

15
GFS (contd)

• The 1981 global spectral model was developed as a
  result of increased computing power, which
  enabled spectral models to become competitive
  with global operational grid point models. In
  fact, the model replaced the seven-level, 191-km
  grid point primitive equation model used in
  various configurations since the late 1960s.
• Major changes were made to the global spectral
  model in 1985 (TPB 351), at which point it was
  renamed the Medium Range Forecast (MRF) Model.
  These changes included new physics packages, an
  increase in the number of waves resolved to
  rhomboidal truncation at 40 waves (R40), and an
  increase in the number of equally spaced layers
  from 12 to 18. A history of further changes that
  have taken place from 1991 to the present can be
  found at the NCEP GFS Changes Web site.
• Currently, the GFS is run four times per day (00
  UTC, 06 UTC, 12 UTC, and 18 UTC) out to 384
  hours. The initial forecast resolution was
  changed on May 31, 2005 to T382 (equivalent to
  about about 40-km grid-point resolution) with 64
  levels out to 7.5 days (180 hours). At later
  forecast times, the GFS has a resolution of T190
  (equivalent to about 80-km resolution) and 64
  levels beyond to day 16 (384 hours). All GFS runs
  get their initial conditions from the Spectral
  Statistical Interpolation (SSI) global data
  assimilation system (GDAS), which is updated
  continuously throughout the day.

16
COAMPS 3.0

• On the synoptic scale, COAMPS consistently
  performs as well as other models (NOGAPS/GFS) in
  forecasting synoptic scale events. On the
  mesoscale, COAMPS frequently outperforms other
  models in predicting mesoscale meteorological
  events, particularly close to land in the
  littoral zone. The strongest feature of the
  COAMPS 27-km nest is its ability to capture
  localized winds and small-scale effects. The
  Mistral, for example, is well depicted in shape,
  size and duration. One of the noted tendencies of
  all COAMPS areas is the inherent cold-bias in the
  low level maximum temperature forecast. This is,
  on average, about 1C cold.

17
COAMPS Model Tendencies

• COAMPS EUROPE (Nest2 27 km resolution)
• Nest2 demonstrates poor skill in most narrow
  straits, such as the Straits of Bonifacio and
  Gibraltar, due to the current model spatial
  resolution. Specifically, Nest2 tends to under
  forecast the funneling effect by 5 -15 kts.
• Precipitation guidance is generally good as far
  as shape of the precipitation area but is low on
  the amount. Nest2 is very good at depicting
  orographic induced precipitation, however remains
  light for lighter precipitation amounts.
• A cold bias exists over land, in general, and
  specifically notable in Europe with the least
  being at tau 12 at the surface and the greatest
  at tau 48 at 250 mb.
• COAMPS performs well through 66 hours on
  predicting the Mistral.  Mistral winds were
  almost always observed when the model predicted
  them, and were almost always correctly forecast
  when they were observed.  Winds at the eastern
  and western edges of the Mistral were not as well
  forecast, indicating that the model does not
  necessarily predict the exact shape of the
  Mistral.

18
COAMPS Model Tendencies (contd)

• COAMPS EASTERN PACIFIC (Nest2 27 km resolution)
• Strong prefrontal wind events associated with
  Pacific cyclones tend to be overforecast
  particularly for forecasts beyond 24 hours. 
  However, when a prefrontal wind event is actually
  observed the model does a good job with both the
  position and magnitude of the event. Post-frontal
  northerlies show similar but less severe
  tendencies. There is a modest overprediction of
  northerly wind events, but again, when a wind
  event actually exists, the model handles it well
  through most of the forecast. The northerlies
  exhibited very little phase error.
• Since most EPAC cyclones are in the filling
  stage, the wind overpredictions suggest the same
  "slow to fill" trend exhibited by NOGAPS. If the
  storm is weakening more rapidly than is
  predicted, the associated wind maxima will likely
  be stronger than what is observed. Also, since
  northerlies are more anchored to the coast by the
  synoptic gradient, phase errors are less likely
  to occur for these winds