Overview of Deterministic Computer Models

Overview of Deterministic Computer Models
Spring 2016
Kyle Imhoff
What do we mean by deterministic?
 The computer model solves the equations of motion
(discussed on the first day of class) for each grid point.
 At each grid point, model output provides information on
exactly what the temperature will be, how much rain/snow
falls, what wind speeds will be, etc.
 NO probabilities involved, no statistical techniques
 Strictly physically-driven
Basic Idea (review from last week)
 Weather is governed by laws of physics that we understand
and can approximate through the use of equations
– The key word there is approximate
 System of equations are very complex and not easily solved
by humans – use of calculus/different equations
 Computer power used to solve the equations and to provide
useful guidance tools to a human forecaster
 We will get into physics and atmospheric processes later in
the class – this lecture will focus on how a model works
Step 1: Observational Data
 The atmosphere works in 3-dimensions (i.e. in the
horizontal and vertical) – we must measure atmospheric
processes both at the surface and aloft
 In order for the model to even have a chance at forecasting
accurately, significant amounts of data (and GOOD data)
must be provided to the model
 The use of all of the surface observational data and weather
balloon/satellite/aircraft data for the upper atmosphere
discussed last week is provided as initial input to the model
Step 2: Data Assimilation
 This essentially means the model brings in the data and sifts
out the bad from the good.
 Once the model has completed this process, the model can
use that data to start solving the equations
 There are many different ways to “sift” through the data –
we will discuss this in some more detail later in the class
Step 3: Solving the equations
 The model solves the equations in time steps
 Recall from one of our first lectures a sample equation from the governing set of equations:
 w, in this case, represents vertical motion (wind speed)
 The model will take initial observations and solve the equation in time steps
– So, the model solves the equation in increments of time (e.g. every 3 hours, every 6 hours,
etc.)
– This technique is called “finite differencing”
 Thus, by solving the equation in steps, in this example you would get values of vertical wind
speed for future hours
Step 4: Bring numbers to life
 The computer model must
now interpret numerical
output and place it on a
grid system
 The sketch to the right
shows the basics of how a
model estimates the
atmosphere through a series
of grids
Step 4: Bring numbers to life (cont’d)
 The model solves the equations at
each grid point (as shown to the
right)
 Each grid point represents the
average value for the surrounding air
(we call this a “parcel” of air in
meteorology
 So, the model approximates the
atmosphere as a series of points and
cubes of air
Step 4: Bring numbers to life (cont’d)
 An example of a grid cell
(or “parcel” of air) is
shown to the right
 The greater the number of
grid points, the smaller the
grid cell
 Smaller grid cells mean the
model has a “higher”
resolution
Step 5: Model Output Displays
Graphical Output
Tabular Output
Computer Model Resolution
 The resolution of a model refers to how far apart the grid
points are from one another
 The higher the resolution, the smaller the difference
between grid points
 When grid points are closer together, finer details can be
seen in the model
 Model resolution plays a key role in how accurately the
model can forecast certain types of phenomena
Computer Model Resolution Differences
Lower resolution (12 km between grid points)
Higher resolution (4km between grid points)
Computer Model Types – Global vs. Regional
 Global-scale models:
– United States: Global Forecast
System (GFS)
– Europe: European Centre for
Medium-Range Weather
Forecasts (ECMWF)
– Canada: Canadian
Meteorological Centre (CMC)
– United Kingdom: United
Kingdom Meteorological Office
(UKMET)
 Regional-scale models:
– Many different models
– United States: North
American Model (NAM)
– United States: High
Resolution Rapid Update
(HRRR)
– United States: Rapid Update
Cycle (RUC)
Computer Models – Operational Structure
 Computer models are typically run every 6 or 12 hours




– i.e. 00z, 06z, 12z, 18z
Weather balloons are launched at 00z and 12z everyday worldwide – this makes 00z and 12z runs of
models the most accurate because they contain better upper-level data
Models produce forecast output for as little as 12-16 hours in their entirety or as much as 384 hours out
from the model initialization time
Resolution affects how long a model can run
– The higher the resolution, the more grid points must have the equations solved, thus more time it takes
the model to run and complete each time step
– So, higher resolution models tend to be short-range forecast models (within 72-84 hours)
– Lower resolution models forecast in the long-range (beyond 84 hours)
The same holds true for special extent of the model
– global-scale models have more grid points, thus take longer for each time step – they are lower
resolution compared to regional-scale models
Why create so many model “runs” each day?
 Running consistent model runs allows the more recent run
of the model to include the most recent observations
– The idea is that including newer observational data will create
more accurate forecasts
 So, the 18-hour forecast in the 00z run of the model should
not be as accurate as the 12-hour forecast of the 06z run of
that same model
One Main Problem
 The use of multiple models and multiple runs of each model creates an
information overload to the forecaster
– Which model should you focus on?
– Should you average all of the models together?
– If all of the models are radically different, where should they even begin?
– If the GFS did better for the last snowstorm, will it do well for the one that is
forecast for next week?
 This problem has driven the industry to the point where weather forecasters
must now be thought of as professional interpreters of computer model data
Summary
 Computer models have vastly improved our ability to
predict the weather – and more accurately farther out in
time
 They simulate what the real atmosphere looks like and what
it will look like in the future – it is not the real thing by any
stretch!
 They provide guidance tools for forecasters
 With so many tools available, weather forecasters can have
a hard time trying to find the signal in all of the noise