Reading the spaghetti plot

This is “the famous spag plot“.  I posted this primer over at the graphs collection, but some may have missed it there.

Current spaghetti animation


The spaghetti plot is a quite old (1996?) “experimental” product of the Earth System Research Laboratory of the US National Oceanic and Atmospheric Administration, using output from the US National Weather Service’s Environmental Modeling Center and National Centers for Environment Prediction.  It looks like a mess, but it’s really pretty simple, and useful.

Easy read:

The plot shows an animated US weather model prediction of the state of the southern polar circulation for the next 15 days. Simply put, we need at least the main spaghetti of red lines — preferably the blue lines — to sweep our alps in nice big looping pattern for us to get snow. (Just because they do doesn’t guarantee snow; but if they don’t, snow is unlikely, because the lines are most directly an indication of average atmospheric temperature.)

Detailed read:

The plot shows contours of geopotential height from a special implementation¹ of the US National Weather Service’s GFS model. The contours are of the height above sea level at which 500 mb (500 hPa) atmospheric pressure occurs. That’s about halfway through the atmosphere by mass (average sea level pressure is ~1000 hPa and the pressure at the edge of space is zero). Geopotential height reflects two main effects:

  • The prevailing pressure of the circulation pattern. If a low pressure zone is passing (surface or upper), the geopotential height will be lower.
  • The average temperature of the atmosphere below. Our atmosphere is effectively a free column of gas bound by gravity. It expands and contracts with temperature according to the universal gas law, so even without circulation, geopotential height would rise and fall with the average temperature of the gas column below.

The plot shows contours of two elevations: in winter 5820 m in red and 5700 m in blue. Both are heights to 500 hPa; the blue ones are closer to the pole than the red because it’s colder there. The green lines show the average positions of the two contour levels for the time of year. (The contour elevations chosen to make this plot are adjusted through the year as the atmosphere warms and cools with the annual cycle².)

There are lots of lines at each contour elevation because this is a plot of an ensemble of model runs. The model has been run 42 separate times: 21 starting at 00z (midnight universal time) and another 21 starting at 12z (midday). The 21 runs comprise one using a best estimate starting condition (the “control runs”, plotted in brown for 00z and grey for 12z), plus another 20 starting from very slightly differing initial conditions³. This approach gives an indication of the chaotic nature of weather systems. It’s particularly useful in long range forecasting.

(Once in a while this chart will show many more contour levels in multiple colours — the “weird spag” form. That appears to be by design, even though that version is much more difficult to interpret.)

ECMWF version

ECMWF produces a similar (perhaps higher skill, for us) ensemble forecast, available here. Instead of using multiple lines, ECMWF contour the average of the ensemble of results⁴, and show coloured zones to indicate the spread. In the pair of plots, the ensemble average is on the left and the “deterministic” result (the best estimate, in this case from their high resolution model) is on the right (sample only, click for link):


BOM version

A similar (but deterministic, not ensemble) animated view can be obtained from the Australian Bureau of Meteorology’s ACCESS model. The coloured zones there indicate wind speed at the 500 hPa level, i.e. high-altitude “jet streams” (sample only, click for link):



The broad state of the polar 500 hPa circulation used to be⁵ referred to as the “LWT” (long wave trough), a concept which maybe shouldn’t quite die. Basically if you filter the isobars or geopotential contours at the 500 hPa level to show just the very broad scale pattern, it’s often possible to discern a multi-lobed arrangement that rotates very slowly clockwise around the pole, taking about 4 weeks to get right around (if it does at all). The behavior of the LWT may provide some additional guidance for long range (multi-week) forecasting. Here is Weatherzone’s version, based on GFS (sample only, click for link):


1. A lower resolution version of the model optimised for high execution speed, to permit multiple runs.

2. I don’t fully understand the basis of the choices they make. They appear to use slightly lower plotted contour levels in summer, even though the average geopotential heights then are substantially higher:

Summer average geopotential heights

Winter average geopotential heights

Presumably their intention is to choose contours representative of the likelihood of cold, snowy weather.

3. The method of selecting the variations in initial conditions is a huge topic in itself. Suffice to say they’re far from just arbitrary small changes. It’s necessary to choose variations that are internally consistent and fit withing the real uncertainty in the data assimilation used to derive the best-estimate starting analysis.

4. An Environmental Modelling Center discussion of the benefits of spaghetti ensemble plotting over mean-and-variation plotting is here.

5. The Australian Bureau of Meteorology stopped producing LWT charts in 2010.