Ensemble Prediction Flashcards
What is the traditional method of making a weather forecast?
Take the best model available and run it until it loses its skill due to the growth of small errors in the initial conditions
Skill is typically lost after ____ days or so, depending on the season
6
An alternate method that produces forecasts with skill up to ____ days after the initial forecast uses what is called “____ _____”
15 days
“ensemble forecasting”
Instead of using just one model run, many runs with slightly different ____ conditions are made
initial
An average, or “____ _____” of the different forecasts is created
ensemble mean
Why will the ensemble mean likely have more skill?
- It averages over the many possible initial states
- Smoothes the chaotic nature of climate
Is now possible to forecast probabilities of different conditions because of the ____ ensemble of forecasts available
large
Every day, at __Z, ___Z, __Z, and __Z global weather observations are collected, transmitted to major weather centers, and are used to produce a snapshot of the global atmosphere
00Z, 06Z, 12Z, and 18Z
What does the “snapshot” of the global atmos (i.e. the model analysis or 0-hour forecast) include?
Winds, pressure, moisture, temperature, etc. at multiple vertical levels in the atmos and at grid intersections of about 100 km
Why does an individual model run merely “ sample” one of many possible current and future states of the atmospheric circulation?
- Because of the uncertainty in the initial conditions
What are some examples of uncertainty in initial conditions?
- instrument error
- spatial/temporal sampling error
- lack of data
- erroneous influences of the first guess background field
How does NCEP (and other weather centers) make many forecast runs out of a model out to 15 days?
- Make initial conditions of the individual runs slightly perturbed
What is the most skillful numerical forecast that can be made given the chaotic nature of the atmospheric circulation?
The average of all these model integrations
Somewhere between ___ and ___ numerical model runs are executed daily using slightly altered initial conditions
12 and 30
These slight alterations are known as “____”.
perturbations
These 12 to 30 model runs using perturbed initial conditions are called the ____ ____, and theoretically represent the current _____ ____ of the distribution of atmos states expected in the atmos out to forecast times of 15 days
ensemble members
best estimate
One can also use the members to estimate probabilities of certain events, such as ____ ____ or ____ ____ normal temperatures
much below
much above
Such probabilities are dependent on the ____ in the forecasts of individual members
spread
The ensemble prediction approach attempts to define the ____ ____ ____ of atmosphere variables
Probability Density Function (PDF)
This forecast scatter between ensemble members ____ with forecast lead time and should eventually approach the ____ observed in everyday weather
increases
scatter
The scatter between separate ____ ____ should approach the ____ _____ scatter of the observed weather
model forecasts
normal climatological
What are the 5 steps of the NWP Process?
1) Gather Observations
2) Data Assimilation
3) Numerical Weather Predictions
4) Forecast Postprocessing
5) Issue forecasts and evaluate
What are the 2 categories of problems with ensemble predictions?
1) Chaos
2) Model Error
As forecast lead time ____ the ensemble mean maps get _____
increases
smoother
This increasing ____ reflects the fact that there is more _____ in the predictions at ___ lead times or further into the future
smoothness
uncertainty
long lead times
The “____ or _____” distribution of states at the initial time is becoming more ____ out as individual members sample different but plausible future solutions of the atmospheric circulation
tight or peaked
more spread out
Even when there are only slight differences, it is these differences that will result in a very different forecast of the future atmospheric state (due to ____)
chaos
As one goes further in to the future, the lines start showing more and more spread or scatter so that they look like ______
spaghetti
Why is the ensemble mean a smoother damped wave pattern?
Troughs and ridges which are superimposed on one another from different ensemble members cancel each other out in the mean
With ensemble means, information on individual storms is what?
Essentially lost so that one member may be predicting high pressure in a region while another is predicting low pressure
_____ smoothes out these differences. However, there may still be useful information at these “long-lead” forecast times, especially for the ____ scales of motion
Averaging
larger scales of motion
What do the two heavy green lines represent on spaghetti plots? And how often do they change?
The two climatological lines of 500 mb height that correspond to the red and blue forecast lines.
They change slightly every day and are based on about 30 years of global data
By comparing the high latitude ___ line with the ___ lines and the lower latitude ___ line with the ___ line, one can determine how far removed from climatology a particular region is observed or predicted to be
green with blue
green with red
This suggests that the ensemble mean is approaching ____
climatology
Sometimes, even for predictions beyond ___ days, most of the predicted lines may remain ___ or ____ their respective climatological lines
10 days (240 hours) remain above or below
(due to ocean SST departures due to El Nino)
At the initial time the spread is small so that the distribution must be strongly ____
i.e.: there is relatively ____ confidence about the current state of the atmos
peaked
high confidence
Any spread is due to the small _____ introduced in the initial conditions of individual ensemble members. Are generally within the range of observational errors
perturbations
Early on in the forecast integration or cycle, many locations over the hemisphere are ____ removed from their climatological value
far
As the forecast lead time ____, the departures from climatology get ____ and the spread gets ______
increases
smaller
larger
This means that the probability distribution is becoming less ____ and at very ____ lead times it should approach the climatological distribution
peaked
long lead times
Once this climatological distribution is reached, any predictive skill present in the initial conditions of the atmosphere has been ____ and the “forecast” would be __ ____ than using climatology
lost
no better
Model errors and the manner whereby the initial conditions are perturbed produces a ____ distribution that may deviate from the ____ distribution
probability distribution
climatological distribution
Individual ___ ____ can systematically contaminate all the ensemble forecasts run on that model, regardless of how the initial conditions are perturbed
model biases
One solution is to generate ensembles from a ____ of different models.
Name one example
cluster
i.e.: the FSU Superensemble
What is a big contributor to model error?
Parameterizations
Much of the weather occurs at scales ____ than those resolved by the weather forecast model.
smaller
Models must treat, or _____ the effects of the sub-gridscale on the resolved scale
parameterize
What are five examples of parameterization?
1) land surface characteristics and processes
2) cloud microphysics
3) turbulent diffusion and interactions with surface
4) Orographic drag
5) radiative transfer
What is a historical weather event which is very similar or closely parallel to the current event?
weather analog
Even if NWP centers produce calibrated ensemble forecasts at ____ scale, much of the important weather happens at the _____
grid scale
sub-gridscale
Can solve this problem by finding analogs from the ___, and then use ____ weather observations from these analogs
past
actual
______ _______ help to compensate for some of the effects of chaos
ensemble forecasts
ensemble forecast technology is ____; better probabilistic forecasts with each passing year
maturing
Still can’t expect ____ ____ data to provide reliable weather input to drive extended local forecasts without some adjustment
raw ensemble
