Numerical Models of Weather Forecasting
As part of studing energy issues (for more info see my gas blog
I have been looking at the weather forecasts.
It is quite interesting how much reasonably reliable information is available on the web. Sites such as Net Weather
provide this information as well as referencing other sites such as Wetter Zentrale
which is in German.
Forecasters generally start with numerical models of the weather. The charts generated by those models are then available on various net sites. One of the systems most frequently reviewed is the US Global Forecast System. (GFS)
This system takes in standard data for four runs. The four runs are indicated by the GMT time that they take the data from. They are then available about 5 hours later.
One aspect that varies for GFS is that different sets of data are input for each run.
The variations are:
00z - Weather buoy, satellite data, shipping data, country data, NOAA data
06Z - Weather buoy, satellite data, shipping data
12Z - Shipping data, Satellite data ONLY
18Z - Weather buoy, satellite data, shipping data, country data, NOAA data
The models divide up the atmosphere into subunits and then perform numerical calculations based upon atmospheric physics. This suffers from the "butterfly wing" effect whereby a small change in one place can have a major impact.
The models are generally reported on every 6 hours. Because errors in input to start out with and other factors give unreliability over time the models are frequently run with the data given small adjustments for each run. This gives sets of "ensembles" whereby there is a control run and a number of ensembles and an average for all members of an ensemble.
The interesting thing semantically is that experience indicates that the result which comes from the average for all members of an ensemble is more reliable than the control when forecasting more than 3.5 days in the future.
Different forecasters then use a combination of the numerical models and their own experience to judge how to forecast the weather. It has been interesting that it is now possible to see things such as a cold snap developing via the models reasonably reliably about 7 days in advance.
The more general issues such as climate, however, arise in a different way. A good source for more information about this is the fora on Net Weather (see above).