Prioritize...
After you have read this section, you should be able to define ensemble, discuss various visuals for ensembles, and list the strengths and weaknesses of an ensemble.
Read...
Over the past 20-30 years, forecasting techniques have advanced from each model making a single forecast, to several runs of the model yielding an ensemble of forecasts. An ensemble, multiple runs for the same forecast time, provides more information for a forecaster. In particular, by examining several different model runs, a forecaster can build confidence in the predictions if they coincide or question the outcome if the model runs are spread out. Read on to learn more.
What is an Ensemble and how do we visualize?
An ensemble consists of multiple model runs for the same forecast valid time. Each model run is called a “member” of the ensemble. Most often, but not always, all the member model runs start at the same initial time.
A spaghetti diagram plots a mission-critical contour for each member forecast. Take a look at the example below:
The diagram above shows the contours of an ensemble forecast at specific geopotential heights at 500 hPa. Note the two contours selected, one roughly corresponding to the edge of really cold weather and the other the edge of tropical weather.
Spaghetti plots allow us to visualize the amount of uncertainty in the forecast by looking at the spread among ensemble members. There is high confidence in the forecast when the members tend to coincide and low confidence when they are spread apart. This is highlighted in the figure above; the forecast has higher confidence along the Gulf Coast (New York) than in the Pacific Ocean (West Coast) as seen in the right panel (left panel).
Instead of viewing all the member forecasts on one plot, we can plot the output from each member on a ‘postage stamp’ sized image. This is useful for more complex forecast situations where a small number of contours just won’t suffice.
Take a look at the example below:
Each image is the output of an individual member of the ensemble. In this example, we are looking at the mean sea level pressure and temperature at 850 hPa. This type of diagram allows the forecaster to examine all the individual outputs side by side and make comparisons.
Why ensemble?
Why should we use ensembles? Think back to earlier lessons when we discussed probabilistic forecasts. We’d really like the joint PDF of all the weather variables we are interested in. This lets us compute the odds of any combination of outcomes and therefore make decisions based on expected cost. Ensembles can be used to compute those PDFs.
The spread in the PDF can vary with the weather pattern – some patterns are much easier to predict than others. In theory, ensembles can quantify this forecast uncertainty since they know what weather pattern we are in and how it is likely to evolve. One of our big tasks will thus be to determine the spread/skill relationship of our ensemble so we can calibrate it into a good approximation of the true PDF.