Weather is a favourite topic of conversation. Its constantly changing patterns influence us all. With the vast Atlantic Ocean to the west, forecasting for Ireland presents a particularly challenging problem. Agriculture, fishing, tourism and transport are all strongly affected by adverse weather. Accurate forecasts are essential for efficient planning and operations, and are of considerable financial value. Indeed, the guidance provided by the Meteorological Service results in economic benefits far above the cost of producing it.

During recent decades there has been significant progress in our ability to forecast the weather. This is largely thanks to the development of advanced numerical models running on powerful computers. The Meteorological Service has an ongoing programme of research and development with the aim of producing forecasts of higher quality, valid for longer periods and tailored for specialised applications. The Met Service has recently acquired an SGI Challenge-L, a computer capable of running a new, sophisticated forecasting system. This will provide more detailed, accurate and timely guidance, and increase the range and value of the forecast products issued to industry and the public.

How can forecast accuracy be improved? The atmosphere is a sheet of fluid bound by gravity to the rapidly rotating earth, its motion driven by solar energy. The tropics receive more heat than the polar regions. The resulting temperature gradients are hydrodynamically unstable, so that small disturbances amplify into large weather systems in a few days. This process is difficult to predict and imposes a limit on the predictability of the weather: the system is essentially chaotic, and detailed forecasts beyond about two weeks may never be possible. However, shorter range predictions may be made with increasing accuracy by modern methods. Scientific forecasting is based on application of the fundamental laws of physics to predict the evolution of the atmosphere.

The principles which govern the atmosphere, conservation of mass and energy and Newton's laws of motion, have been known for a long time, but their use in forecasting requires extensive calculation and is feasible only using a powerful computer. The physical laws are expressed as mathematical equations for each variable: pressure, temperature, humidity, winds and so on. These quantities vary continuously in space and time; to calculate how they change, one approximates them by their values at a finite set of points, dividing the atmosphere into a series of cells and considering the central value to be representative throughout the cell. This process is called discretisation. It has the drawback that one cannot represent phenomena smaller than the cell size. But its great advantage is a simplification of the mathematical system of partial differential equations to a large but soluble algebraic system.

The computer program to solve this system is called a model. A new forecasting model, HIRLAM (High resolution limited area model), is currently being implemented at the Met Service. The research and development work is being done in collaboration with the meteorological Institutes of Denmark, Finland, the Netherlands, Norway and Sweden. The model calculates how variables such as pressure evolve in time, using the basic laws of motion and also allowing for the effects of radiation, friction, stratiform and cumuliform clouds, air-ocean interactions and various other physical processes.

The most important weather phenomena are associated with water, and these are also the hardest to simulate. Even the largest thunder clouds are smaller than the computational cell, so that they cannot be modelled directly; they are represented in an empirical way using a so-called parameterisation scheme. The accuracy of precipitation forecasts improves as the cell size is reduced, but the computation time also rises sharply, so a compromise choice is necessary.

The HIRLAM model uses a cell size of 50 km, covering Europe and the North Atlantic with a grid of about 10 000 cells. With some 20 vertical levels and five dependent variables (temperature, two wind components, humidity and cloud water) there are one million degrees of freedom: thousands of calculations are required to compute the changes of each variable, and many short time-steps must be made to complete a one-day forecast. The computational task is truly enormous. The time required for this task on the new computer (with two processors) is about 40 minutes. Early availability is critical if the forecaster is to give maximum advance notice of impending severe weather.

Improvement in computer modelling of the atmosphere during recent years has been spectacular. Older models were capable only of simulating the flow in the free atmosphere. The latest models, such as HIRLAM, give a good indication of future conditions at the earth's surface. Guidance provided by the model includes predictions of surface temperature and winds. However, it is still essential to interpret the computer predictions in terms of local weather conditions, and this is where the expertise and experience of the forecaster comes in. The best forecasts are produced by this 'man-machine-mix' and human intervention in the production of forecasts will be required for the foreseeable future.