Thursday, October 18, 2012

Rolling Dice with Ensembles

Source:  Of course the Wasatch Weather Weenies do not endorse gambling, except when it comes to the weather.
Our inability to know exactly what the atmosphere looks like at a given time limits how well we can forecast the weather.  Modelers call this "initial condition uncertainty." Because the atmosphere is a chaotic system, errors in the initial conditions can grow with time, and wreak havoc on the weather forecast.  As a result, there are a range of possibilities in any forecast, and typically that range gets larger with increasing forecast lead time.

Ensemble forecast systems (see previous post "Forecast Tools: Ensemble Forecast Systems") produce  many forecasts with slightly different initial conditions to get a handle on the range of future possibilities.  For example, below are all the 168 hour 500-mb geopotential height forecasts produced by yesterday afternoon's run of the NCEP Global Ensemble Forecast System.

Source: Penn State E-wall
These forecasts, which are valid at 0000 UTC Thursday 25 Oct all produce a trough over either the western US (e.g., middle right) or along the west coast (most of the other members).  The consistency across these forecasts gives us some confidence that a large-scale pattern change will occur, but the details of what happens in northern Utah is less clear since our weather depends strongly on the position and intensity of the trough, as well as smaller scale effects related to the positioning of surface fronts and the like.

Unfortunately, we still have a problem.  First, since we don't know the initial conditions precisely, we don't know how to generate all the differing initial conditions for the ensemble precisely either, so there is a very real possibility that the range of possibilities differs from that indicated above.  Further, the model used for those forecasts isn't perfect either, and that is an additional source of error.  As a result, we tend to use ensembles of ensembles for long-range forecasts.  So, in addition to the GEFS, we look at the ensemble produced by the Canadian Meteorological Center (and in practice, others as well).

In the Canadian Meteorological Center ensemble forecast valid for the same time, we see a bit more spread.  A few of the forecasts put a trough over western North America or along the west coast, but there are others that put a ridge over western North America (e.g., upper right).

Source: Penn State E wall
Using these ensembles, meteorologists try to put odds on the weather.  One could think of this like rolling the dice, except how strongly the dice are weighted varies from day to day.  Some patterns are more predictable than others and we can have greater confidence in longer range forecasts.  Other patterns are less predictable, and the future is more uncertain.

1 comment:

  1. Jim, A very good discussion on the difficulties the models have in handling these large scale pattern changes. I linked to it on my blog this morning with regards to the first real winter threat to Star Valley.