1. 452 NWP 2019

2. Major Steps in the Forecast Process
Data Collection
Quality Control
Data Assimilation
Model Integration
Post Processing of Model Forecasts
Human Interpretation (sometimes)
Product and graphics generation

3. Data Collection
Weather is observed throughout the world and the data is distributed in real time.
Many types of data and networks, including:
Surface observations from many sources
Radiosondes and radar profilers
Fixed and drifting buoys
Ship observations
Aircraft observations
Satellite soundings
Cloud and water vapor track winds
Radar and satellite imagery

4. Observation and Data Collection

8. Atmospheric Moisture Vectors

12. Weather Satellites Are Now 99% of the Data Assets Used for NWP
Geostationary Satellites: Imagery, soundings, cloud and water vapor winds
Polar Orbiter Satellites: Imagery, soundings, many wavelengths
RO (GPS) satellites
Scatterometers
Active radars in space (GPM)

13. Quality Control
Automated algorithms and manual intervention to detect, correct, and remove errors in observed data.
Examples:
Range check
Buddy check
Comparison to first guess fields from previous model run
Hydrostatic and vertical consistency checks for soundings.
A very important issue for a forecaster--sometimes good data is rejected and vice versa.

15. 3 March 1999: Forecast a snowstorm … got a windstorm instead
Eta 48 hr SLP Forecast valid 00 UTC 3 March 1999
3 March 1999: Forecast a snowstorm … got a windstorm instead

16. Pacific Analysis
At 4 PM
18 November
2003
Bad Observation

17. Forecaster Involvement
A good forecast is on the lookout for NWP systems rejecting bad data, particularly in data sparse areas.
Quality control systems can allow models to go off to never never land.
Less of a problem today due to satellite data everywhere.

18. Objective Analysis/Data Assimilation
Observations are scattered in three dimensions
Numerical weather models are generally solved on a three-dimensional grid
Need to interpolate observations to grid points and to ensure that the various fields are consistent and physically plausible (e.g., most of the atmosphere in hydrostatic and gradient wind balance).

19. Objective Analysis
Interpolation of observational data to either a grid (most often!) or some basis function (e.g., spectral components)
Typically iterative (done in several passes)

20. Objective Analysis/Data Assimilation
Often starts with a “first guess”, usually the gridded forecast from an earlier run (frequently a run starting 6 hr earlier)
This first guess is then modified by the observations.
Adjustments are made to insure proper balance.
Often iterative

22. An early objective analysis scheme is the Cressman scheme

23. 3DVAR: 3D Variational Data Assimilation
Used by the National Weather Service today for the GFS and NAM (called GSI)
Tries to create an analysis that minimizes a cost function dependent on the difference between the analysis and (1) first guess and (2) observations
Does this at a single time.

24. 3DVAR Covariances: Spreads Error in Space

25. 4DVAR: Four Dimension Variational Data Assimilation
Tries to optimize analyses at MULTIPLE TIMES
Tries to duplicate the observed evolution over time as well as the situation at initialization time.
Uses the model itself as a data assimilation too.

26. 4DVAR Components Full non-linear model
Tangent linear version of the full model (linearized version of the forecast model)
Adjoint of the tangent linear model -which allows one to integrate the model backwards. Tells sensitivity of final state to the initial state.

27. 4DVAR
Typical runs the model back and forth during an initialization period (6-12 hr), roughly ten times.
Substantial computational cost.
Need to have adjoint and TL version of the model.
Currently used by ECMWF, CMC, UKMET, and US Navy. NOT NCEP.

28. Many of the next generation data assimilation approaches are ensemble based
Example: the Ensemble Kalman Filter (EnKF)

29. Mesoscale Covariances
12 Z January 24, 2004
Camano Island Radar
|V950|-qr covariance

30. Surface Pressure Covariance
Land
Ocean

31. An Attractive Option: EnKF
Temperature observation
3DVAR
EnKF

32. Hybrid Data Assimilation: Now Used in GFS
Uses both 3DVAR and EnkF
Uses EnkF covariances from GFS ensemble in 3DVAR.

34. Next Advance ENVAR
Use temporal covariances to spread impact of observations over TIME.
Now operational.
Has some of the properties of 4DVAR (adjusts model evolution)

35. Grid Point Models
Horizontal (and vertical) variations describd on a 3-D grid
Computer resources needed increase by roughly 8 times for doubling resolution
Can have computational instability, particularly when time step is too long for the grid spacing used.
CFL stability criterion:
Cdt/dx <=1

36. Galerkin Approach (e.g., spectral)
Represents dependent variables (e.g., u, v, T) as a sum of basis functions.
Fourier analysis is an example:

37. GFS and ECMWF Use Spherical Harmonics to Represent Variation in Horizontal
Legendre Polynomials

38. Vertical Coordinate Systems
Originally p and z: but they had a problem…Boundary conditions (BC) when the grid hit terrain!
Then sigma p and sigma z, theta
Increasingly use of hybrids– e.g., sigma-theta, sigma-p

39. Sigma

40. Sigma-Theta

41. New Sigma-P in WRF

42. Nesting

43. Why Nesting?
Could run a model over the whole globe, but that would require large amounts of computational resource, particularly if done at high resolution.
Alternative is to only use high resolution where you need it…nesting is one approach.
In nesting, a small higher resolution domain is embedded with a larger, lower-resolution domain.

46. Nesting Can be one-way or two way.
In the future, there will be adaptive nests that will put more resolution where it is needed.
And instead of rectangular grids, other shapes can be used.

47. Next Generation Global Models
Will use different geometries

48. MPAS: Hexagonal Shapes

49. MPAS

50. FV3-Replacement for GFS

52. Cube Sphere Grid

53. Major U.S. Models Overview
Global Models
Global Forecast System Model (GFS). Uses spectral representation rather than grids in the horizontal. Global, resolution equivalent to 13 km grid model. Run out to 384 hr, four times per day.
Navy Global Environmental Model (NAVGEM) hour, four times a day. 50 levels, 37 km grid
UKMET Unified Model just taken on by U.S. Air Force. They call it the Global Air-Land Weather Exploitation Model (GALWEM)

54. Major International NWP Centers
ECMWF: European Center for Medium-Range Weather Forecasting. The Gold standard. Their global model is considered the best. 9 km resolution
UK Met Office Unified Model: An excellent global model slightly superior to GFS
Canadian Meteorological Center
Other lesser centers

55. Global Forecast System (GFS) Model
Previous called the Aviation (AVN) and Medium Range Forecast (MRF) models.
Spectral global model and 64 levels
Relatively primitive microphysics.
Sophisticated surface physics and radiation
Run four times a day to 384 hr (16 days!).
Major increase in skill during past decades derived from using direct satellite radiance in the 3DVAR analysis scheme and other satellite assets.
13 km grid spacing equivalent over the first 10 days of the model forecast and 35 km from 10 to 16 days (384 hours). Thus, it now essentially a global mesoscale model

56. GFS Hybrid Data Assimilation

57. GFS
Vertical coordinates are hybrid sigma/pressure… sigma at low levels to pressure aloft.

59. GFS Data Assimilation (GDAS)
Global analysis every 6 hr
Has a later data cut-off time than the mesoscale models…and thus can get a higher percentage of data.
Uses much more satellite assets..thus improve global analysis and forecasts.
Major gains in southern hemisphere
Hybrid Data assimilation based on 3DVAR (called GSI) and GFE ensemble (next slide)

60. In addition to the high resolution GFS, there is a lower resolution GFS ensemble called GFE
34 km grid space equivalent
20 members
Stochastic physics to help produce diversity
16 days

61. GFS is not the best global model

68. New Global Model: FV3 developed by NOAA’s GFDL part of the Next Generation Global Prediction System project (NGGPS)

69. Grid based

70. Cubed Sphere

71. How Good is FV-3?

74. Higher Resolution Operational Models

75. Major U.S. High-Resolution Mesoscale Models (all non-hydrostatic)
WRF-ARW (developed at NCAR)
NMM-B (developed at NCEP Environmental Modeling Center)
COAMPS (U.S. Navy)
MM5 (NCAR, old, replaced by WRF)
RAMS (Regional Atmospheric Modeling System, Colorado State)
ARPS (Advanced Regional Prediction System): Oklahoma

76. Operational Mesoscale Model History in US
Early: LFM, NGM (history)
Eta (mainly history)
MM5: Still used by some, but mainly phased out
NMM- Main NWS mesoscale model, updated Eta model. Sometimes called WRF-NMM and NAM.
WRF-ARW: Heavily used by research and some operational communities.
NMM replaced by NMM-B

77. NMM-B also called NAM (North American Mesoscale) model
Hybrid sigma-pressure vertical coordinate
60 levels
Betts-Miller-Janjic convective parameterization scheme
Mellor-Yamada-Janji boundary layer scheme

78. NMM-B Nests
One-way nested forecasts computed concurrently with the 12-km NMM-B parent run for
CONUS (4 km to 60 hours)
Alaska (6 km to 60 hours)
Hawaii (3 km to 60 hours)
Puerto Rico (3 km to 60 hours)
For fire weather, moveable 1.33-km CONUS and 1.5-km Alaska nests are also run concurrently (to 36 hours).
Call them HRW-High Resolution Windows

79. NMMB 4-km Conus

80. NCEP NAM Generally less skillful than GFS, even over U.S.
Generally inferior to WRF-ARW at same resolution (more diffusion and smoothing, worse numerics)

83. Navy COAMPS (Coupled Ocean/Atmosphere Mesoscale Prediction System)
Sigma-Z
Atmosphere
and Ocean

85. Rapid Refresh NWP A Powerful Tool For Nowcasting Two Resolution: RUC: 13 km HRRR: 3 km

87. Rapid Refresh (RR). A.K.A RUC
A major issue is how to assimilate and use the rapidly increasing array of off-time or continuous observations (not a 00 and 12 UTC world anymore!
Want very good analyses and very good short-term forecasts (1-3-6 hr)
The RUC/RR ingests and assimilates data hourly, and then makes short-term forecasts
Uses the WRF model…which uses a hybrid sigma/isentropic vertical coordinate
Resolution: Rapid Refresh: 13 km and 50 levels, High Resolution Rapid Refresh (3 km)

89. NOAA hourly updated models 13km Rapid Refresh (RAP) (mesoscale)
Rapid Refresh and HRRR
NOAA hourly updated models
13km Rapid Refresh (RAP) (mesoscale)
Version 2 – scheduled
NCEP implementation
Q2 (currently 28 Jan)
RAP
3km HRRR (storm-scale)
HRRR
High-Resolution
Rapid Refresh
Scheduled NCEP
Implementation Q3 2014
NCEP Production Suite Review
Rapid Refresh / HRRR
3-4 December 2013

90. Lightning (proxy reflectivity)
Observations Used
Hourly Observations
RAP 2013 N. Amer
Rawinsonde (T,V,RH)
120
Profiler – NOAA Network (V) 21
Profiler – 915 MHz (V, Tv) 25
Radar – VAD (V)
125
Radar reflectivity - CONUS
1km
Lightning (proxy reflectivity)
NLDN, GLD360
Aircraft (V,T)
2-15K
Aircraft - WVSS (RH)
0-800
Surface/METAR
(T,Td,V,ps,cloud, vis, wx)
Buoys/ships (V, ps)
GOES AMVs (V)
AMSU/HIRS/MHS radiances
Used
GOES cloud-top press/temp
13km
GPS – Precipitable water
260
WindSat scatterometer
2-10K

91. RAPv2 Hybrid Data Assimilation
13 km
RAP
Cycle
13z
14z
15z
ESRL/GSD RAP 2013
Uses GFS 80-member ensemble
Available four times per day
valid at 03z, 09z, 15z, 21z
80-member GFS EnKF Ensemble forecast valid at 15Z (9-hr fcst from 6Z)
Obs
Obs
Obs
GSI
Hybrid
GSI
Hybrid
GSI
Hybrid
1 hr fcst HM
Obs
1 hr fcst HM
Obs HM
Obs
GSI HM Anx
GSI HM Anx
GSI HM Anx
Refl
Obs
Refl
Obs
Refl
Obs
Digital
Filter
Digital
Filter
Digital
Filter
18 hr fcst
18 hr fcst
18 hr fcst

92. Rapid Refresh: 13 km and larger domain

93. High-Resolution Rapid Refresh: 3 km, 1 hr, smaller domain

100. Accessing NWP Models
The department web site (go to weather loops or weather discussion) provides easy access to many model forecasts.
The NCEP web site is good place to start for NWS models.
The Department Regional Prediction Page gets to the department regional modeling output.

102. A Palette of Models
Forecasters thus have a palette of model forecasts.
They vary by:
Region simulated
Resolution
Model Physics
Data used in the assimilation/initialization process
The diversity of models can be a very useful tool to a forecaster.

103. RTMA (Real Time Mesoscale Analysis System)
NWS Mesoscale Analysis System for verifying model output and human forecasts.

106. Real-Time Mesoscale Analysis RTMA
Downscales a short-term forecast to fine-resolution terrain and coastlines and then uses observations to produce a fine-resolution analysis.
Either a 5-km or 2.5 km analysis.

107. RTMA
The RTMA depends on a short-term model forecast for a first guess, thus the RTMA is affected by the quality of the model's analysis/forecast system
CONUS first guess is downscaled from a 1-hour RR forecast.
Because the RTMA uses mesonet data, which is of highly variable quality due to variations in sensor siting and sensor maintenance, observation quality control strongly affects the analysis.

108. Why does NWS want this?
Gridded verification of their gridded forecasts (NDFD)
Serve as a mesoscale Analysis of Record (AOR)
For mesoscale forecasting and studies.

109. TX 2 m Temperature Analysis

112. The End

113. WRF and NMM

114. History of WRF model
An attempt to create a national mesoscale prediction system to be used by both operational and research communities during 1990s.
A new, state-of-the-art model that would have good conservation characteristics (e.g., conservation of mass) and good numerics (so not too much numerical diffusion)
A model that could parallelize well on many processors and easy to modify.
Plug-compatible physics to foster improvements in model physics.
Designed for grid spacings of 1-10 km

115. NWS goes its own way ARW (Advanced Research WRF) developed at NCAR
Non-hydrostatic Mesoscale Model (NMM) Core developed at NCEP
NMM
ARW

116. The NCAR ARW Core Model: (See: www.wrf-model.org)
Terrain following vertical coordinate
two-way nesting, any ratio
Conserves mass, entropy and scalars using up to 6th order spatial differencing equ for fluxes. Very good numerics, less implicit smoothing in numerics.
NCAR physics package (converted from MM5 and Eta), NOAH unified land-surface model, NCEP physics adapted too

117. NWS NMM-B1—Used in the NAM RUN
Run every six hours over N. American and adjacent ocean
Run to 84 hours at 12-km grid spacing.
Uses the Grid-Point Statistical Interpolation (GSI) data assimilation system (3DVAR)
Start with GDAS (GFS analysis) as initial first guess at t-12 hour (the start of the analysis cycle)
Runs an intermittent data assimilation cycle every three hours until the initialization time.
1-Non-hydrostatic mesoscale model, NAM: North American Mesoscale run