Producing a weather forecast

How computers predict weather (continued)

While there are limits to model resolution, forecasters would like to include many of the processes that occur at scales smaller than the grid size. These 'sub-grid processes' include convection and the turbulent transfer of heat, moisture and momentum near the surface of the earth. For example, individual cumulus clouds may be around one kilometre across, and cumulonimbus clouds are less than about 10 km across.

The only way to include sub-grid processes is to approximate or 'parameterise' their statistical effects.

Numerical weather prediction faces much greater difficulties in the tropics than in the middle and high latitudes. In particular, tropical models require an extremely good initial analysis of water vapour distribution for useful prediction to be useful. This requires a dense network of observations, something not yet available.

Limitations

The accuracy of predicted maps is inherently limited by the scale at which we can make detailed atmospheric measurements, the finite power of the computers, and our incomplete understanding of some small-scale processes in the atmosphere.

We must remember that the atmosphere is a 'chaotic' system whose chief characteristic is extreme sensitivity to initial conditions. Small-scale motions in the atmosphere are random and complex; small errors are magnified as the length of the forecast period increases. There may well be a definite time limit for which specific forecasts can be made - probably about two weeks.

Notwithstanding the increasing sophistication of numerical weather prediction, meteorologists are still needed, as they can apply other tools and skills generally unavailable to the computer, including current weather charts, satellite images and local knowledge.