Transcript An Introduction to Forecast Models

An Introduction to Forecast Models
Outline
1.
Important Considerations: Atmospheric Science, Physical Processes.
2.
Weather Forecasting and Creating a Forecast Model.
3.
Model Construction and Resolution.
4.
Initialization and Model Run.
5.
Verification.
6.
Basics to Model Viewing, Time and Types of Data.
7.
Model Types: Operational, Model Output Statistics, Ensembles.
8.
Forecast Ranges: Short-Range, Medium-Range, Long-Range.
9.
Model Access (sources of data).
http://geocities.com/quincyq03/0207PPT.ppt
Why are models important to
weather forecasting?
• Weather is governed by laws of physics that are present in
space, our atmosphere and at the Earth’s surface.
• Equations have been derived and theorized to explain weather.
• These equations are often very complex and the linear aspect
does not sufficiently describe our atmosphere.
• Calculus/differential equations are necessary for calculations.
• These calculations take far too long to solve by hand and there
are many variables that must be considered.
• Forecast models have helped drastically improve
forecasting skill, accuracy and verification.
Atmospheric Science
• The study of Meteorology and forecasting is complex.
• There are many processes taking place, in addition to
many variables that affect weather patterns and events.
• Global circulation causes radiative forcings, that due to
the earth’s shape, creates wind and thermal gradients.
• Daytime and nighttime due to sunrise and sunset also
affect diurnal variances in wind and temperatures.
• The shape of the earth’s orbit around the sun also
creates seasons and variability in weather conditions,
since there are times when parts of the earth’s surface
are closer or farther away from the sun.
Atmospheric Science
• Solar heating and radiation from the earth also take place.
• Uneven surface heating due to many factors, including
terrain and bodies of water, creates a further imbalance.
• Drag and turbulence also play a role in the imbalance.
• Storm systems and air masses are advected (transported)
but are constantly evolving as well.
• Storm systems spin up, develop, mature and weaken,
while air masses are constantly modifying and moving.
• Convergence and divergence of storm systems and air
masses takes place…there is a LOT going on!
Processes and Variables to Consider
• Even the previous slides only begin to graze the surface,
as there is much more that drives weather!
• Introductory Meteorology courses aim to introduce the
student to just some of the basic fundamentals of
Meteorology! Further courses, especially Dynamics and
Thermodynamics get into more detailed and
mathematical analyses of these processes.
• The previous processes and variables all have
calculated and/or theoretical equations to describe them.
• Assumptions are made and without absolute knowledge,
the equations are often approximate.
• With that said, scientists have come to fairly accurate
equations and methods of forecasting.
(From Holton’s
An Introduction
to Dynamic
Meteorology)
This image will
be referred to
again later in
the
presentation.
Weather Forecasting
• Weather forecasting, up into the early 20th century, was
largely very poor and often controversial.
• Early forecasts were done by hand and were often fairly
inaccurate, even within the short-range.
• Some forecasts, such as short-term ones of pressure
and temperature were decent, but scientists were
beginning to realize that weather forecasting would
improve with time, research and technological advances.
• Weather prediction, through numerical calculation would
be that next step. …Numerical Weather Prediction…
Creating a Forecast Model
• Numerical Weather Prediction (NWP) was first
proposed by Lewis Fry Richardson in 1920.
"Richardson's
Forecast Factory"
Richardson knew that
the amount of data that
would need to be
processed would be
enormous, to create
forecasts with accuracy
and practical value.
Today, computers are
used to handle all of the
information needed.
Creating a Forecast Model
• Networks of upper-air observation were introduced in the
1940s, allowing for tracking atmospheric data.
• Equations of atmospheric motion were studied and
simplified, and by 1950, the first NWP experiments began.
• The first forecast computer model was created and it used
the Barotropic Equation of Atmospheric motion to create
500hPa height forecasts. (hPa = millibar(mb))
• Those forecasts out to 24 hours were significantly more
accurate than any previous forecasts, but aside from the
scientists, were not very practical or easy to understand.
Creating a Forecast Model
• Further atmospheric equations were simplified and
entered into forecast models. (numerous complex
equations)
• Computer technology evolved and allowed for more
complex and accurate equations to be used over time.
• As further research was done in the 1960s and 1970s,
physicists and meteorologists were able to create more
realistic, detailed and useful model forecasts.
• With time, computer models have become and are
becoming more able to process enormous amounts of
data, which are necessary to create accurate forecasts.
Model Construction
• The atmosphere is three-dimensional,
the Earth is spherical and the surface
is uneven.
• Significant amounts of data must be
input into forecast models to account
for the variables to be considered.
• Weather balloons are used for upperair observation and information from
many vertical levels are derived.
• These observations are taken across
the world at 12-hour intervals and
cover both land and bodies of water.
Model Construction
• Other data is also obtained from satellites, radar,
ground/soil observations, the oceans and elsewhere.
• Supercomputers are used to take all of this information in
and from equations and theories, compute forecasts.
Model Construction (continued)
• Think of a rectangular
chunk of the world.
• There is are east-west
and north-south directions,
but also an upward
direction of altitude.
• Computer models need
three-dimensional
coverage, so data at
points across the earth
surface (n, s, e, w) are
input, but information is
also added for specific
heights above the ground.
Model Construction (continued)
• Each grid point represents the
• Grid Points are equally spaced
average value for the air
locations that used to forecast the
(volume) surrounding it, which
weather through models.
can be considered an air
parcel.
• The grid points make up cubes
that cover the earth’s surface,
as well as extensions upward
into the atmosphere to be 3dimensional.
• Atmospheric variables are
represented through these grid
points and since there is a
finite number of them, finite
difference approximations
must be used to calculate
variances.
Model Construction (continued)
• A grid cell is the 3-dimensional cube that the
grid cells combine to create.
• The greater the
number of grid points in
a model, the smaller
the grid cells become.
• Smaller grid cells
leads to greater model
resolution and more
detailed forecasts.
G
R
I
C
D
E
L
L
S
Model Resolution
• The resolution of a model, is determined by the distance
that separates the grid points.
• For example, if the distance between two grid points on
a model is 15km, the model resolution is 15km.
• With research and technological capabilities, model
resolution continues to improve.
• The Nested Grid Model (NGM) has not been updated
recently and is one of the lower resolution models today.
• Mesoscale models, such as the MM5 and WRF/NAM
have some of the highest resolutions.
Model Resolution
• Model resolution does not necessarily mean more
accurate forecasts…
• In some ways, the lower resolution models give the best
representation of the general weather patterns.
• For more detailed forecasts, however, the lower
resolution models tend to miss out on details, especially
in the boundary layer (lower levels and surface).
• All models have their uses and it depends on the
forecaster viewing them and the particular weather event
or pattern to determine which are best to use.
Model Initialization
• The models take in observations, and through finite
difference approximations, create a model initialization.
• This initialization is a 00hr “forecast” that represents the
current data that the model run will forecast from.
• Depending on the model resolution and other parameters,
that will determine what the 00hr data looks like.
• Initializations tend to be similar between models at a
given time, with only minor variations.
• However, as will become more evident in subsequent
discussions, sometimes bad data or poor initializations
can lead to lower quality forecasts from that model run.
Example HPC Model Discussion
MODEL DIAGNOSTIC DISCUSSION
NWS HYDROMETEOROLOGICAL PREDICTION CENTER CAMP SPRINGS MD
1230 PM EST MON FEB 04 2008
VALID FEB 04/1200 UTC THRU FEB 08/0000 UTC
MODEL INITIALIZATION...
...SEE NOUS42 KWNO ADMNFD FOR STATUS OF UPPER AIR INGEST...
MODEL INITIALIZATION ERRORS DO NOT APPEAR TO HAVE A SIGNIFICANT IMPACT ON
THE FCST.
...TROF PUSHING INTO THE W D1-2 AND THRU THE CNTRL CONUS D3... THE NAM IS
TOO COLD WITH H85 TEMPS OVER WRN TX AND THE ADJACENT RIO GRANDE VLY BY
UP TO 6 C. THE GFS HAD SIMILAR ISSUES...BUT NOT TO THE DEGREE OF THE
NAM...WITH TEMPS OFF BY APPROX 2-3 C IN THE SAME LCNS. THE NAM AND GFS
MAY BE TOO COLD WITH THE H85 TEMPS AHEAD OF THE TROF OVER THE SRN
PLAINS/LWR MS VLY INTO THE FORECAST PD AS A RESULT. TO THE N...THE
NAM...AND TO A LESSER EXTENT THE GFS...DID NOT DO A GOOD JOB
INITIALIZING THE STRENGTH OF THE H85 TEMP GRADIENT OVER NERN CAN/ERN
AK...WITH THE NAM AND GFS UP TO 3-4 C TOO WARM WITH THE COLD POOL OVER
THE YUKON AND UP TO 4 C TOO COLD OVER NRN B.C./ALBERTA. SOME OF THIS
AIRMASS IS XPCTD TO PHASE WITH THE TROF OVER CONUS LATER IN THE PD. THE
NAM AND GFS MAY NOT BE ACCURATELY DEPICTING THESE AIRMASSES AND THE TEMP
GRADIENT IN THE FORECAST PD.
Model Run
• Once an initialization takes place, models then use the
initialization data and equations to extrapolate a forecast.
• Models need a method for time-stepping. (extrapolation)
• Explicit Time Differencing involves the a predicted value
to be determined at a given point for time step S+1, from
the previous time step, S.
• Implicit Time Differencing is more complicated and
includes the steps of S+1 and S-1.
• The latter system is more complicated and takes up
much more computer power, but has many advantages.
Model Run
• As a model continues a forecast from the starting time,
00(hr) and continue out from there. (+6hr, +12hr, etc)
• Models are run out under their limitations and setups,
until the process is adjusted or ends.
• For example, with the Global Forecasting System (GFS),
the model resolution is downscaled after 180 hours.
• Model accuracy tends to decrease over time.
• Since approximations are used at initialization and
throughout the model run, such a decrease in accuracy
with time can be expected. Also, the model is forecast
potential scenarios, so that must be considered as well.
Model Verification
Model Run Delay
• Models have so much data to ingest and work
with, that there is typically a long delay for their
forecasts.
• Depending on the model, this delay may be
anywhere from around 1 hour to even 6 or more
hours!
• For those of you who do view some model
output, this explains why you have to wait a few
hours after the initialization time to view the data!
Basics to Model Viewing
• The time scale that the vast majority of the
models use is Universal Time (UTC)
• During Standard Time (fall/winter), UTC is
5 hours ahead of Eastern Standard Time.
• During Daylight Savings Time, UTC is only
4 hours ahead of Eastern Savings Time.
• UTC is also known as Zulu (z) time.
Main Types of Models
• Operational models
– The ones that are most common & numerous.
• Model Output Statistics (MOS)
– Adjusted operational model runs, based on
statistics.
• Ensembles
– Several members with slight variations.
The Operational Model
• This model is the standard run for most
models.
• The NWS Forecast Discussions will often
refer to these runs as the “OP run”.
• The GFS, NAM, ECMWF, GGEM, etc all
have an operational model – the one used
“the most”.
Model Output Statistics (MOS)
• This model is based off of its operational
counterpart, but has many adjustments to it.
• MOS forecasts are used to fine-tune and adjust
the operational output.
• MOS will output more detailed information, than
what can generally be derived from the OP run.
• Statistical analyses are used to further adjust the
forecasts for individual stations, seasons, etc.
More on MOS
• The most common example of a MOS model
would be the GFS MOS (MAV and MEX).
MOS Wrap-up
Ensemble Forecasts
•
Ensembles are created from the operational
models in two different ways:
1. Different physical parameterizations
2. Various initial conditions.
•
Physical Parameterizations have to do with
how models develop clouds, transfer energy,
handle the boundary layer, etc.
•
Initial Conditions are the conditions that the
models initialize from.
“The Two Ways”
• Why different physical parameterizations?
– Under some situations, the operational model may be
more biased towards a certain, eventual outcome.
– By viewing & comparing runs from different physical
parameterizations, they can be analyzed better.
• Why alter the initial conditions?
– Consider a model run…the initialization is almost
always off very slightly with all of the variables.
– The model approximates initial conditions, so there is
always some error across the board.
Physical Parameterizations
• “The approximation of unresolved processes in terms of
resolved variables is referred to as parameterization”.
Holton 474
• Physical parameterizations in forecast models are very
difficult, complex and aim to increase forecast accuracy.
• The atmospheric (including boundary layer and surface)
processes must be approximated and simulated.
• Aside from Ensembles, other models also have various
physical parameterization schemes.
• This explains, in part, why some models seem to handle
certain weather patterns or events differently.
Models
have
various
schemes
of
evaluating
and
forecasting
these
processes.
Physical Parameterizations
• The operational model run has set parameterizations, but
each ensemble member may have slight variations.
• The method of explaining/forecasting the processes is
already approximate, so minor adjustments are added.
• Each ensemble member will have slight variations, so as
the model runs out, further forecast variations will arise.
• Depending on the situation and the individual member,
different scenarios will unfold.
• Forecasters can compare and assess the ensemble
members, to evaluate the operational run and also see
what kind of spread and mean forecast is forecasted.
Altering Initial Conditions
• The “Butterfly Effect” can be considered.
• Instead of a butterfly slightly flapping its wings to throw
off a forecast, slight deviations from the model
initializations will cause a similar chain reaction.
• This chain reaction explains why models are not perfect.
• As a forecast model generates longer and longer
forecasts, each proceeding interval has less and less
accurate information to extrapolate from.
• Although the initial conditions are important, there are
often many minor errors in a given initialization.
Altering Initial Conditions
• Ensemble models that use this method, take the “initial
conditions” and adjust them slightly in several ways.
• There are many ensemble members that now have
slightly different initializations to work with.
• The ensembles then extrapolate their forecasts and
commonalities can be noticed – “Ensemble Mean”.
• By viewing all of the different ensembles, we can evaluate
the operational run and determine if the run looks
accurate, or if it is more likely to be an outlier.
• However, a difficulty with the ensembles is that
sometimes the mean is far off from the actual verification
and outliers may verify. Occasionally, the ensemble
spread may not even cover the actual scenario!
Forecast Ranges
• We will consider the short-range
forecasts to include generally forecasts to
go out to 72 hours or less.
• The medium-range forecasts will be from
3 to 7 days.
• Long-range forecasts include those
beyond 7 days.
• Different models have various forecast ranges.
Short-Range Forecasts
• Short-range forecasts tend to focus on the exact
details, such as temperature gradients,
precipitation, and mesoscale phenomena.
• These forecasts tend to originate from models
with higher resolution.
• When forecasting severe weather or precipitation
types, these are useful for specifics.
• However, a forecaster must also consider that
models are not flawless and real-time data should
also be considered.
Examples of SR Models
• WRF/NAM: Forecasts out to 84 hours.
• SREF: Short-Range Ensemble Forecasts!
• RUC: Rapid updates for out to 12 hours.
• NGM: Lower resolution, but out to 48 hours.
• MAV MOS: 72 hour forecasts.
• RGEM: Regional GEM out to 48 hours.
SREF for 2m temps in the NE
Examples of MR/LR Models
• GFS: Forecasts out to 180 hours and 384 hours.
• ECMWF: Euro Centre for MR WX Forecasts!
• UKMET: UK model that forecasts for the MR.
• GGEM: Known as the “Canadian” (MR).
• NOGAPS: Developed by the US Navy (MR).
• Ensembles from various models to mainly MR.
• CFS: Climate Forecast System (LR).
GFS Long-Range Forecast
Model Access
• There is a substantial amount of model data that
can be accessed for free online.
• Some models are restricted, but most aren’t.
NCEP: http://www.nco.ncep.noaa.gov/pmb/nwprod/analysis/
This site is commonly used and has several American models.
E-Wall: http://www.meteo.psu.edu/~gadomski/ewall.html
This site has a lot of various models, from SR to MR, American to
Foreign and is a personal favorite of mine!
• Attend next weeks’ session on Online Weather
Resources for many more links and information!
INCOMPLETE Works Cited
•
•
•
http://www.tpub.com/content/aerographer/14269/css/14269_75.htm
http://en.mimi.hu/meteorology/
http://www.booty.org.uk/booty.weather/metinfo/models/NWP_basics.htm
•
http://www.ncep.noaa.gov/nwp50/Presentations/Tue_06_15_04/Session_1/Lynch_NWP50.pdf
• Information has been paraphrased or has been included through
previous experience and knowledge by Quincy Vagell, unless noted
through footnote.
• Any and all slides and information within them can be shared and
used, though proper referencing is necessary.
• For more information, details or to ask any questions you may have,
contact Quincy Vagell.
Additional Considerations
• The next three slides were not presented,
but have been left in the PowerPoint.
• The first two discuss some of the variables
and processes considered by the models.
• The third and final slide shows a view of
how one particular model processes data
and creates forecasts.
Analytical Meteorology
• MTR 175, Introduction to Analytical Meteorology is a
course that focuses on physical concepts and
elementary problems in Meteorology. Processes and
variables that must be considered with modeling include:
Radiation – The transfer of energy. cooling, heating, etc
Advection – Transport from one location to another.
Conduction – Heat transfer through physical objects.
Turbulence – Irregular fluctuation in (atmospheric) flow.
Drag – Frictional force between the air and the ground.
Latent heat – Heat transfer through change of state.
Saturation – Deals with moisture content.
Stability – Deals with resistance to vertical movement.
Buoyancy – The ability of an object to rise or sink.
Analytical Meteorology
Precipitation – Formation, size, intensity, velocity; type.
Gradient – The rate of change of a quantity. (P, T, etc)
Development and dissipation. (fronto- and cyclo-genesis)
Divergence – Outflow/separation of a physical quantity.
Convergence – Inflow/coming together of a quantity.
Tendency - Trend. (pressure, temperature, etc)
Orographics – How land and water affect weather.
Vorticity – A measure of rotation. (ie. Air parcel)
Condensation – The change from vapor to a liquid.
Evaporation – The change from liquid to a vapor.
Random probability – Must be considered, since it is
impossible to create perfectly precise forecasts.
Data Assimilation