Numerical Weather Prediction (NWP) Model Fundamentals: A...
Transcript of Numerical Weather Prediction (NWP) Model Fundamentals: A...
Numerical Weather Prediction (NWP)
Model Fundamentals: A review(Plus 1/2 slide on climate models)
William R. Bua, UCAR/COMET
NCAR ISP Summer colloquium on African Weather and Climate
27 July 2011
Outline
• What is the land-ocean-atmosphere system and
its connection to weather and climate?
• What is in an NWP system?
• What are the shortcomings of NWP models?
• Ensemble Forecast Systems: Mitgating the
shortcomings of NWP models
The Land-Ocean-Atmosphere System
• Conservation of momentum,
heat, moisture
• Conservation of mass
• Hydrostatic approximation
• Dynamical equations are
coupled to
– The earth’s land/ocean surface
(friction/ turbulence, surface
evaporation/ evapotranspiration
and precipitation)
– Sub-grid scale physical/diabatic
processes (radiation, evaporation/
condensation, water phase
changes in precip processes,
cloud/radiation interaction, etc.)
Equations of Motion (Eulerian/Pressure coordinate form)
Simplified Equations
The Land-Ocean-Atmosphere System
• Radiation processes– Incoming solar radiation
– Outgoing terrestrial radiation
• Microphysics– Condensation/evaporation/
sublimation
– Collision/coalescence, mixed phase processes, phase changes
• Convection (shallow *and* deep)
• Turbulent processes
• Land surface processes– Vegetation, soil moisture,
snow, surface energy balance and fluxes
Land and
topography
Precipitation
microphysicsConvection
Vegetation, soil moisture,
surface energy balance/fluxes
Shortwave
scattering
Incoming
shortwave
rad.
Reflection
Parameterized Land/Atmosphere Physical Processes
Longwave Radiation
Longwave Rad.
Climate and Weather Prediction Models
General Circulation
(Climate) models
• Interested in climate details (means,
anomalies, standard deviations) at long
time scales
• Long, lower resolution runs
– Climate drift must be corrected
• Physical processes are simplified
• Slowly varying processes must be
accounted for
– A fully coupled system
– For multi-decadal climate change
• Interactive vegetation adapts to
changing climate
• Carbon cycle/slowly varying
atmospheric chemistry
Numerical Weather
Prediction (NWP) Models
• Interested in short time scales
and weather details
• Short, high resolution runs
– Climate drift not important,
especially for short range
• Physical processes are more
realistic (e.g. microphysics)
• Atmosphere/land coupling; slow
processes held fixed
– Fixed ocean (SSTs)/sea ice
– Fixed vegetation
– Fixed atmospheric composition/
greenhouse gases
NWP MODELS: DYNAMICS
NWP Models: Dynamics
• Horizontal coordinate
system
– Equations computed
either by
– Breaking down the
horizontal direction into
grid points and taking
differences from point to
point …. or
– Breaking down the large
scale flow into a series
of increasingly small
sine and cosine waves
and plugging those into
the equations to do the
calculations
…+
= Shortest wave
NWP Models: Dynamics• Numerical problems
decrease with improved
horizontal resolution
– 2-point wave: poor
depiction, disperses without
advecting
– 7-point wave: better
depiction, disperses and
advects
– 20-point wave: well-depicted
and forecasted
NWP Models: Dynamics
• Vertical coordinate
– Upper left: terrain-
following sigma
– Second: step-
mountain
– Third: hybrid sigma-
isentropic (theta)
– Fourth: hybrid sigma-
pressure (transition to
pressure complete at
about 100-hPa)
NWP Model: Dynamics• Topography
– Only as good as the resolution of
the model
– Can choose representation of
topo in each grid box
• Envelope: valleys and passes
filled, blocking effect enhanced
• Silhouette: averages tallest
features, more valley details
• Mean: averages all features, trims
mtns, diminishes mtn blocking
– Standard deviation of topo in
grid box used for physical
processes
• Land/sea mask depends on
resolution also
NWP Model: Non-hydrostatic Dynamics
• Add an equation for vertical accelerations (below)
• Use in high-res models (< about 5-10 km)
– Will result in mesoscale details of convective systems,
including outflow boundaries and cold pools
– Requires sophisticated physics, esp. for precipitation
– Costs more to run, usually small domain and short-range
forecast only
T-storms, mtn. waves ↑ for warm
moist air
relative to env.
weight of
precip. “pulling
on the air”
1-km Simulated Radar Reflectivity
NSSL-WRFNCEP-WRF
Actual radar
valid at about
same time
NWP MODELS: PHYSICS
NWP Models: Radiation (SW)• Actual SW scatter/
reflection/ abspt.
btw. TOA and sfc.
– Blue vs. brown
lines
• RRTM model:
– UV (3 bands, 0.2-
0.4 μm)
– Visible (2 bands,
0.44 – 0.76 μm)
– Near IR (9 bands,
0.778 – 12.2 μm)
… 12.2
NWP Models: Radiation (LW)
• Long (IR) wave radiation absorption/reemission in real
atmosphere (actual spectrum shown, with absorption
bands labeled with gaseous absorber)
– Many absorption lines in evidence
• RRTM scheme breaks LW spectrum into 16 bands for
calculations from about 4 μm to 400 μm wavelength
NWP Models: Radiation and Clouds
• Real atmosphere
• Clouds reflect,
scatter, and absorb
SW radiation; some
SW reaches surface
• Clouds absorb and
reemit LW radiation
• Cloud layers, cloud
fraction, water phase
(liquid and/or ice), cloud
overlap all should be
addressed in NWP
models
• Actual atmosphere
– Very small scales (mm - μm)
– Condensation/evaporation/sublimation
– Collision/coalescence (rain)
– Aggregation (snow, riming)
– Bergeron process (ice crystals grow
preferentially in mixed phase clouds)
– Fall rates depend on precip. type
• Models
– Bulk processes based on forecast T,
RH, vertical motion
– Precipitation sometimes assumed to
fall out instantaneously
NWP Models: Precip. Microphysics
• Convection: Real atmosphere
– Conditional instability drives updrafts
(small scale, <1 km)
– Moisture condenses latent
heating, clds./precip.
– Downdrafts from precip. evap.
cooling and precip. drag
– End result: PBL cools/dries, free
atmosphere warms/moistens
• Conv. Param., NWP models
– Can’t resolve thunderstorms;
unresolved updrafts taken into acct.
– Impact on model variables estimated
• Convective trigger
• Vertical exch. of heat/moisture/
momentum at grid scale
– Shallow conv. treated separately
NWP Models: Convection
NWP Models: Surface Processes
• Surface water balance
– Precipitation minus evaporation
as input
• Evaporation controlled by soil
moisture, vegetation, and local
weather conditions (wind, RH,
PAR)
• Surface energy balance
– Incoming minus outgoing
energy fluxes
– Sfc. water and energy balances
coupled via evaporation
0LESHLWGLWSWnet
NWP Models: Turbulent Processes• Observed planetary boundary
layer from surface upward:– Contact and surface layers
– Mixed layer (day) or stable BL with
overlying residual layer (night)
– Capping inversion (night) or
entrainment zone (day)
• NWP version (sub-grid scale):– Contact layer: Fluxes depend on
wind, moisture, temperature forecasts
– Surface layer = constant flux layer
– Mixed and residual layer mixing
depends on wind shear, lapse rate,
diffusion coefficient
– PBL top • Found using forecast stability
• Moisture/momentum/heat exchange w/
free atmosphere modeled, sometimes w/
shallow convection
• Free atmosphere sub-grid
scale mixing/turbulence
– Rate determined by lapse rate
and horizontal/ vertical wind
shear
– Aviation concerns where wind
shears are strong
• Typically near jet stream
• NWP
– Lapse rate and adjacent layer
and grid box wind shears used to
mix air
– Richardson number used as
proxy
NWP Models: Turbulent Processes
• Mountain blocking and
gravity wave drag
– Depends on stability of flow
over topo, angle of wind
relative to topo, topo variability
– More stable: More blocking,
less gravity wave breaking
• NWP:
– Uses resolved topo height and
sub-grid scale topo standard
deviation
– Forecast stability partitions
flow between gravity wave
drag and mountain blocking
NWP Models: Turbulent Processes
Blocked flow around mtn.
Gravity wave-inducing flow over mtn.
NWP MODELS: DATA
ASSIMILATION
NWP Models: Data Assimilation (DA)
• Procedure:
– Start with short-range
forecast (1st guess)
and observations
– QC obs., combine
w/short-range forecast
– Weight fcst. and obs.
based on typical error
– Create new analysis
• Analysis minimizes
total error from all
sources
NWP Models: Data Assimilation
• Advantages
– Uses short-range fcst. as 1st guess
• Short-range fcst. is usually good
– Analysis consistent with what model
can fcst. (no unrepresentative obs.!)
– Error characteristics “known” for
each observation type and 1st guess
• Limitations
– 1st guess error not flow-dependent
(or not flow-dependent enough)
– Errors usually assumed symmetric
around error location (unrealistic
where there are gradients)
– 1st guess not always good
– NWP models cannot correctly
forecast all high impact phenomena
NWP MODELS: POST-
PROCESSING FORECAST DATA
NWP Models: Model-Derived
Products
• Post-processing model-resolution data to
another grid resolution
• Statistical guidance
• Model assessment tools– Verification (will be covered in more detail later)
NWP Models: Model-Derived Products
• Horizontal conversion
– Grid-point vs. spectral
• Raw data (either from native
grid-space or spectral space)
intermediate grid
• Derive parameters, then …
• Vertical conversion
– From native vertical coordinate
to standard output levels
• Derive Parameters, then …
• Horiz. interpolation to
dissemination grids
• Station data is taken from
native grid
– Interpolate to station or use
nearest grid point or grid column
• Advantages of post-processed grids
– Can remove unneeded detail through averaging or
other smoothing
– Smaller, easier to send than native grid data
– Availability of derived products (e.g. stability indices,
tropopause data, freezing level)
• Limitations
– For some fields, degradation of data (e.g. static
stability diagrams like Skew-T may not be accurate)
or loss of detail (e.g. precipitation in regions of
rugged terrain)
NWP Models: Model-Derived Products
• Stat. post-processing/MOS
– Relate NWP vars. to obs. wx.
via stepwise linear regression
(pt.-by-pt. or grouped by region)
• Requires sufficient model data to
get stable statistics
– Find variable that best-
minimizes residual fcst. error
– Stepwise, find each variable that
best-minimizes remaining error
– Stop when additional vars. do
not improve fcst.
– Apply to future forecasts
NWP Models: Model-Derived Products
NWP Models: Model-Derived Products
• Statistical post-processing
– Model Output Statistics (MOS) used in S Africa for seasonal
forecasting
• Used in conjunction with regional climate models (RCM) nested
within a long-range forecast from general circulation model (GCM)
• Statistical post-processing (Landman et al., 2009) outperforms RCMs
nested in GCMs
– Not aware (yet) of MOS used in Africa for medium-range
forecast guidance
• Main use for MOS in America is in the short- to medium range
NCEP OPERATIONAL GLOBAL
FORECAST SYSTEM (GFS)
GFS and GEFS Dynamics
• Equations of motion (advection, continuity)
calculated in spectral space (sines and cosines)
– Exact mathematics for however many wavelengths
are calculated
– Truncation error from limiting the minimum
wavelength for calculations
– Operational T574 (~30 km) through 192 hours,
T190 (~ 90 km) from 192-384 hours
– Ensemble at T190 through 384 hours (~ 90 km)
GFS Dynamics
• Vertical coordinate– Sigma-pressure (σ-p)
hybrid
– Levels placed as at right
• Advantage of hybrid (σ-p):– Sigma levels tilted too
much above 500-hPa; adverse for pressure gradient force calc.
– σ-p reduces this problem considerably
GFS Dynamics• The physics grid
– Sub-grid scale physical process calculations done at grid points and transformed into “spectral space”
– Grid is 0.31 -0.38 resolution over southern Africa domain for operational, about 0.9 -1.1resolution for ensemble GFS
• Topography– T574 topo at right
• Highest point resolved is in Lesotho (2725-m)
– T190 topo next
• Highest points in Kenya and Lesotho (2096-m)
– Land-sea mask
• T190 loses islands, some lakes, shoreline resolution
GFS Precipitation and Clouds
• Precipitation and clouds– “Grid-scale precipitation”
• Simple microphysical processes are modeled (“simple cloud”)
• Precipitation hydrometeors NOT tracked; fall out instantaneously
• Cloud water (in both liquid and solid phases) is tracked and used to determine radiative qualities
– Convective scheme• Simplified Arakawa-Schubert
(SAS)
• Physically realistic, includes observed convective processes
T382
GFS Vegetation Type and Fraction
• Vegetation type and greenness fraction– Required to tap sub-
surface soil moisture• 13 types
• Climatological seasonal cycle for green vegetation fraction
• Vegetation canopy can retain up to 2-mm of water and drip-through is modeled
– Greenness fraction from climatology
• If excessive drought or wetness, may result in surface energy balance problems
Veg fraction Jan-Apr
Veg fraction, May-Aug.
Veg fraction, Sep.-Dec.
GFS Soil Model• Soil moisture model
– Surface layer (0-10cm)
– Root zone layer 1 (10-40cm)
– Root zone layer 2 (40-100 cm)
– Deep soil layer (100-200 cm)
– Diffusion and gravitation act sub-sfc
water, movement depends on soil
type (9 soil types)
• Soil thermal model
– Additional layer (200-800 cm) with
deep soil temp (~avg annual
temperature) constant (bottom
boundary condition)
– Diffusion of heat through layers with
top boundary condition provided by
surface (skin) temperature
GFS Radiation
• Short wave (Chou, 1990,
1992)
– Predicted ozone (O3),
water vapor (H2O)
– Prescribed CO2
– Prescribed O2
– Aerosols
• RRTM long wave
– CO2, H2O, O3, CH4, N2O,
CCl4, chloro-
fluorocarbons
GFS Radiation and Clouds
• Cloud radiative properties
depend on water phase (liquid
or solid), cloud water mixing
ratio
• Cloud fraction dependence
– For grid-scale clouds, cloud
water mixing ratio and RH
– For convective cloud, convective
precipitation amount
• Clouds are overlapped
randomly
GFS surface layer
• Transport of heat and moisture in surface layer (treated as 1st model layer) depends on vertical gradients and winds
• Surface roughness affects the wind speed and depends on vegetation type
• Gradient of pot temp, q, wind determines sensible, latent heat fluxes, momentum flux
GFS Planetary Boundary Layer and
Free Atmosphere Turbulence
• A “non-local scheme”
• PBL top set to where Bulk
Richardson number Ri is
first > 0.5
• Vertical diffusion coeff. fit
to flux at PBL top and
surface, which
determines the diffusion
rate through the PBL
• In free atmosphere, local
wind shear and stability
determine turbulent
vertical transports
Ri >0.5
GFS: Data Assimilation System
• Gridpoint statistical interpolation system (GSI)
– 6-hour cycle
– 6-hour forecast is background (1st guess) for new analysis
– Observations weighted by relative accuracy then GSI
minimizes error taking all obs into acct.
• Background for analysis is assumed to be good quality, typically has
the heaviest weighting
• All obs moved to the analysis time for assimilation
• All obs are quality controlled before assimilation
– Balance constraint makes analysis internally consistent
between mass and wind
CANADIAN GLOBAL
ENVIRONMENTAL MULTISCALE
MODEL (GEM)
Canadian Global Environmental
Model (GEM)• Equations of motion
(advection, continuity)
calculated on a grid
– Truncation error from grid
length limitations
– 800x600 points
• 33x33 km at 49°N
• 33x50 km at equator
– Run at 00 and 12 UTC (to 240
and 144 hours, respectively)
GEM Vertical Coordinate
• Hybrid vertical coordinate
– Flatter surface less
PGF error
• 80 vertical levels
– Model top at 0.1 hPa
• Best resolution in PBL,
tropopause/jet-stream
level and in stratosphere
– Improves assimilation of
satellite radiances
GEM Topography
• Uses “mean orography” (average over grid
box)
– Data from U.S. Geological Survey 30” data
set
• Parameterizations related to topography
– Gravity wave effects on flow
– Mountain blocking
GEM Physics
T382
• Precipitation and clouds– “Grid-scale precipitation”
• Simple microphysical processes are modeled (“simple cloud”)
• Precipitation hydrometeors NOT tracked; fall out instantaneously
• Cloud water (liquid and solid phases) tracked and used for radiation parameterization
– Convective scheme• Deep
– Kain – Fritsch conv. scheme
• Shallow
– Kuo-Transient
• Physically realistic, estimates observed convective processes
GEM Vegetation Type and Fraction
• Interactive Soil-Biosphere-Atmosphere (ISBA)
– Vegetation derived from USGS vegetation type data
set
• 24 vegetation types
• Canopy water immediately available for evaporation
• Each type has unique evapotranspiration parameters
– Can have mixed land-water-sea ice-glacial ice grid
boxes; each has its own unique surface energy
balance
• Energy fluxes are area-weighted average
GEM Vegetation Type and
Fraction
• All vegetation types in each grid box
accounted for
– Parameters are averaged for all types that
appear in grid box
– Land surface heat and moisture fluxes are
predicted from these *averaged* parameters
GEM Soil Model
• Soil is divided up into clay and
sand fractions
– Clay strongly holds onto water
– Sand is more porous
• For moisture, two layers
– Surface layer 10-cm thick directly
evaporates
– Deep layer is accessed by
vegetation roots
• For temperature, two levels
– Surface skin level
– Deep soil level
Surface layer
Eva
po
tran
sp
iratio
n
GEM Radiation Schemes
• New implementation in 2009
– Long- and shortwave radiation schemes
• K-distribution technique (Li and Barker 2005)
based on line-by-line calculations (accurate and
fast!)
– Cloud-radiation interaction
• Cloud water content in each model layer predicted,
phase diagnosed
• Optical depth of layer determined by clear air
radiatively active gases + cloud liquid/ice content
GEM Planetary Boundary Layer
and Free Atmosphere Turbulence
• Vertical diffusion of heat, energy, and moisture by turbulence in PBL– Diffusion based on amt of turbulent kinetic energy in each layer and
…
– The distance a representative parcel from the layer can travel up and down before buoyancy stops its vertical motion (including distance from the ground)
– Includes buoyancy due to lapse rate, vertical wind shear (mechanical turbulence) and moist processes
• Non-topographic gravity waves accounted for in areas of convection, instabilities, and where geostrophic adjustment is occurring
GEM Data Assimilation System
• Atmosphere
– 4-D VAR (x,y,z *and* time)
• No longer a simple snapshot of the atmospheric
conditions
• Now a time evolution of atmospheric conditions
during the assimilation done in “batches”
– Land surface
• Optimal interpolation of skin temperature and soil
moisture based on analyzed 1.5-m RH and air
temperature
– Not actual soil moisture data, but makes soil moisture
and skin temp consistent with screen temp and RH at
time of day when PBL is well-mixed
GEM Data Assimilation SystemObservations Used
MODEL SHORTCOMINGS:
ERROR IN NWP MODELS
The rationale for Ensemble Forecast Systems (EFS)
Initial Conditions
• Initial condition (IC)
uncertainty
– Atmosphere is a chaotic system
with multiple flow regimes
– Lorenz (1963): Sensitive
dependence to ICs
• Varies based on atmospheric flow
• NWP models and IC
uncertainty
– Example: 500-hPa height
• Initial differences about 10-
20 meters
• Sensitive dependence to ICs
leads to large errors (150+
meters) by 96-h
Model-specific Sources of Error
• Model uncertainty
– Dynamics truncation
error (because calculated
on grid, or up to “N” waves
in spectral models)
– Flows that cannot be handled well by the GFS
• Tight gradients
• Sharply curved flow
• Blocking and cut-off flows
Grid point truncation error
Model-specific Sources of Error
• Physics– Convective
parameterization
– Topography (Orographic precipitation? Errors of representativeness for locales in areas of rough terrain?)
– Surface energy balance considerations
• Soil moisture
• Climatological vegetation fraction (does not vary based on climate anomalies)
Model-specific Sources of Error
• Data assimilation systems
– Bad 1st guess (the 6-
hour forecast)
– Extreme excursions from
balance constraint (data
might be right, but will be
rejected)
– Lack of good data
– Time interpolation of data
– Coarseness of some data
(e.g. satellite radiances in
the vertical)
Question:
• What kind of an NWP system could we
design to show us the impacts of:
– NWP model uncertainty/imperfections
– Initial condition uncertainty/imperfections
– The predictability of the current atmospheric
flow regime (given that the atmosphere is
chaotic)?
ENSEMBLE FORECAST SYSTEMS:
MITIGATING EFFECTS OF FORECAST
UNCERTAINTY
Terminology for Ensembles
• Ensemble Forecast Systems (EFS)
• Familiar EFSs
– National Centers for Environmental Prediction (NCEP, U.S.) :
• Global Ensemble forecast system (GEFS)
– Canadian Meteorological Center (CMC)• Canadian ensemble forecast system (CEFS)
– North American Ensemble Forecast System (NAEFS) GEFS + CEFS
– European Center for Medium-Range Forecasts (ECMWF)
Terminology for Ensembles
• Ensemble member– One from among a full set of ensemble forecasts
• Ensemble control– The ensemble member run from the control initial conditions
• Ensemble perturbation– Initial condition and forecast differing from the control initial
condition and forecast
• Post-processing– Development of meaningful EPS products from the raw
ensemble output using statistical methods (we’ll cover some of those more in depth in this lecture)
EFS: Architecture
• Goal: have as many plausible forecast outcomes as
possible
– IC uncertainty: choose ICs to
• Maximize forecast spread
• Minimize ensemble mean error (center perturbations on IC control, use
GOOD NWP models!)
– Model diversity to account for model imperfections/uncertainty
• Dynamical formulation differences
• Vary parameters in a physical parameterization, use different physical
parameterizations in one model, or use multiple models with different
parameterizations
• EFS usually 2-3 times coarser than high-res. deterministic
model in horizontal and vertical
– Computational constraints
– Higher resolution competes with wanting many forecast
possibilities
EFS and Initial Conditions (ICs)
• Methods
– Bred vectors (NCEP)
• Find fastest-growing errors by
perturbing ICs and using
differences to “breed”
perturbations
– Singular vectors (ECMWF)
• Statistical method to find fastest-
growing errors
– Use EFS to determine 1st guess
flow-dependent uncertainty
(Ensemble Kalman Filter or
EnKF) and makes EFS
perturbations (multitasking)
• Directly links DA system and EFS
• Can be part of a hybrid 3D- or
4D-VAR DA system
ANL
P1 forecast
P4 forecastP3 forecast
P2 forecast
t=t0 t=t2t=t1
Rescaling
EFS and Dynamical Core
• Where EFS has diversity in dynamics
– Use different formulation for dynamical
equations (e.g. spectral versus grid point,
change grid point configuration, etc.)
– Use different numerical methods for
calculations (e.g. parcel-following semi-
Lagrangian versus fixed point Eulerian)
– Use different parameters for calculations (e.g.
vertical diffusion)
EFS and Physical Parameterizations
• Use different parameterizations (e.g. convection as at right)
• Tweak parameters within a parameterization (e.g. change vegetation type or vegetation resistance in a single soil model)
• Add stochastic (random) noise to time tendencies of temperature, moisture, winds from physical parameterizations
EFS: The final product
• EFS samples the probability distribution of forecast outcomes
• Statistical analysis is necessary to post-process the large volumes of data produced by EFS and describe the probability distributions
Initial condition
probability distribution
7-day forecast
probability distribution
GEFS, CEFS, AND NAEFS
ARCHITECTURES
Model GFS (current)
Initial uncertainty ETBV1
Model uncertainty Stochastic physics2
Tropical storm Relocation of model
vortex to analysis
Daily frequency 00,06,12 and 18UTC
Hi-res control
(GFS)
T574L64
Low-res control
(ensemble control)
T190L28
00, 06, 12 and 18UTC
Perturbed members 20 for each cycle
Forecast length 384
Implemented 2010
GEFS Configuration
1 Ensemble Transform Bred
Vectors (with rescaling)
2 Random perturbation of
tendencies from physical
parameterizations every 6 hours
NCEP plans to increase GEFS
resolution to T254 (~55 km) for
the first 192 hours in NH spring
2012.
Model GEM (current)
Initial uncertainty EnKF1
Model uncertainty Multiple physical
parameterizations2
Tropical storm Relocation of model
vortex to analysis
Daily frequency 00 and 12UTC
Hi-res control
(GEM)
33-km, 80 levels
Low-res control
(ensemble control)
~100-km, 28 levels
00 and 12 UTC
Perturbed members 20 for each cycle
Forecast length 384
Implemented 2009
CEFS Configuration
1 Ensemble Kalman Filter (from
data assimilation system)
2 Random perturbation of
tendencies from physical
parameterizations every 6 hours
CEFS Physics Diversity (all use GEM
dynamical core)
Summary (1)
• To forecast weather and climate
– Model the land-ocean-atmosphere-(and
cryosphere (ice)) system
– NWP models are used for the short-to-
medium range
– Climate models (a.k.a. general circulation
models or GCMs) use the same basic
formulation …
• … but deal with longer time scales, so ocean and
sea (and for century-long global change runs, even
land) ice should be considered variable, and
coupled to the atmosphere and
Summary (2)
• Deterministic NWP models include
– Dynamics
• Fcst. resolvable motions with equations (e.g. advection)
– Physics
• “Parameterize” unresolved physical processes through
estimating their impact on forecast (e.g. convection)
– Analysis/data assimilation systems determine the
initial conditions from which to start the forecast
– Post-processing
• Write out forecast data to be assessed
• Relate model data to verification based on statistics
• Compute diagnostics to assess possible high-impact
events
Summary (3)
• The U.S. NCEP Global Forecast System is a global
spectral model
– ~ 30-km equivalent grid point resolution and 64 levels
– 3D-VAR snapshot, obs data moved to analysis time
– Runs to 15 days, 4x per day
– Full model physics over land, but (for now) …
– ~ Fixed SST anomalies, sea ice can change in thickness
• Met. Service of Canada Global Environmental
Multiscale (GEM) model
– 33-km gridpoint model with 80 levels
– 4D-VAR, obs data assimilated at obs time by forecast model
– Runs to 10 days at 00 UTC, 6 days at 12 UTC
– Full model physics over land, but fixed SST and sea ice
Summary (4)
• Sources of forecast error
– Chaotic nature of the atmosphere (“sensitive
dependence on initial conditions”, Lorenz 1963)
– Data assimilation errors (i.e. initial condition
uncertainty) lead to growing forecast errors and
ultimately very different forecasts
– Model imperfections
• Dynamics: Numerical approximations, truncation error
• Physics: Estimate of impact of unresolved processes
• No way to get a perfect single forecast in the
foreseeable future, which leaves us with ….
Summary (5)
• EFSs to leverage IC uncertainty, NWP
imperfections
– “Perturbed” ICs based on forecast sensitivity,
increases range of forecast solutions
• Good to link NWP analysis system to the EFS
– NWP model imperfections addressed by
• Using different models
• Using different physical parameterizations within the
same model
• Modifying parameters in physical parameterizations
• Adding random noise to calculated impact from physical
parameterizations on the forecast variables
For more information …
• MetEd NWP training websitehttps://www.meted.ucar.edu/training_detail.php
Click on topics, choose Numerical Modeling (NWP)
• Course 1 (NWP basics)
– Info on how NWP and EFS work
– Info on how specific models work, including specific EFS
– Introduction to specific new forecast tools
• Course 2
– Using NWP in the Forecast Process (applications to
operations)
The NWP Training Team
• An “Army” of One at present
– Liaison between U.S. Environmental Modeling
Center’s NWP model development staff and
operational meteorologists
– Developing lessons and other training on
NWP models in operational context
– E-mail: [email protected]