Forecast

Forecasts and basic forecasting methods

General overview

Data for the images displayed on the myWeather forecast page is received from the Global Ensamble Forecast System (GFS) run by the National Centers for Environmental Preciction (NCEP) in the USA. This global model is run four times daily (00z00, 06z00, 12z00 and 18z00) and produces a six hourly forecast at a grid resolution of half a degree (about 50 km). Raw GFS data from NCEP is received and processed at CSAG every six hours and the images generated for the website.

The variables

The table below gives a very brief explaination of some of the variables shown on the web site. All variables except cloud have an associated colour bar with from which to read off intensities.

Rainfall	- Measured in millimeters of accumulated rainfall for the six hour forecast period. - This is the most difficult variable to forecast as it depends on the correct forecasting of many others.
Cloud Cover	- Shows the extent of cloud cover. - Opacity gives an idea of cloud thickness. - Cloud cover does not necessarily mean there will be rain associated with it!
Sea level pressure	- Contours (isobars) represent pressure at sea level; "sea level pressure" over land is an interpolated value. - Shows the structure and intensity of synoptic systems over the forecast region.
700 hPa height	- Height contours of the 700 hPa level. - Temperature determines its altitude but is usually around 3000 meters in our location.
500 hPa height	- Height contours of the 500 Hpa level. - As for the 700 hPa level, temperature determines its altitude and is usually around 5500 meters in our location. - Upper air features (like cut-off lows) are seen at this level.
Wind	- Unlike with the weather stations, arrows point in the direction of the wind. - Multiply by 3.6 to get to km/h.

Forecasting tips

[Please read the SA Weather page to make full use of this section as information presented there is assumed.]

1. Know your synoptic types, learn to recognise the pressure systems and their related weather conditions.

2. It is important to look at the upper air pressure (500hPa) as well as the surface pressure as (a) the surface features are generally driven by the upper air and (b) the intensity of the system can be gauged. For example, if a low pressure extends from the surface into the upper air, it is a strong system and there is a strong possibility of high windspeeds and rainfall; if there is a low pressure at the surface (usually associated with rainfall) and a high in the upper air, it will probably not rain as the high pressure opposes the flow of moisture into the region....two surface low pressures, but different weather because of what is happening in the upper air.

3. Isobars close together mean winds will be stronger than widely spaced isobars. Wind flows along the isobars (as a result of the Coriolis effect) in a clockwise direction for a low pressure system and in an anticlockwise direction for a high pressure system (in the southern hemisphere).

4. The characteristics of a system will be different in different seasons eg. cold-fronts in summer may not have a true "northwester" before-hand and rainfall amounts are lower; cut-off lows can be devastating in summer while in winter only light showers may occur.

5. When trying to determine the likely weather in your location, you need to know the topography around you and what effect this may have on the local weather e.g. the influence of a water mass (lake) on weather; temperature drop with altitude (especially if wind is blowing); topographically modified wind flow and speed as well as rainfall. The large scale synoptic processes are modified by the local topography.

6. If you are in Cape Town, check the station data on the weather stations page to get an idea of pressure tendancies, wind direction changes etc over time.

7. The forecast for tomorrow will be better than the forecast for 3 days time which will be better than the 7 day forecast which will be better than the 10 day forecast. A long-term forecast for a small area (eg your suburb) should be taken with a tablespoon or more of salt.

Why did the forecast get it wrong!?

As mentioned above, the forecast model is run on a half degree (~50x50 km) grid. This means the entire globe (as this is a global model) is broken up into blocks (or grid points) of ~50x50 km and the model is said to have a half degree resolution. The vertical structure of the atmosphere is then also broken up into a number of non-equal "sigma" levels (60 levels for this model). So the model sees the atmosphere as a series of 50x50xheight cubes (see figure below: Image sourced from http://www.atmo.arizona.edu/~barlage/climatology/). Very simplistically then, when the forecast is run, each cube is given an observed value for temperature, wind, pressure, and a host of other variables. Mathematical equations which describe the atmospheric processes in each cube as well as the interaction between cubes are solved and this eventually produces the global weather forecast.....a computationally very expensive exercise which is performed every six hours!

Although this is a very high resolution for a global model, at the regional scale it is often not able to resolve "sub-grid scale" features such as topography, lakes, vegetation and soil type boundaries, energy transfers etc, introducing problems due to model resolution.

The problem of model resolution is perhaps best explained using an example, and since we are Cape Town-based and we have "our mountain", let's have a look at Cape Town. With a forecast model whose grid scale is 50 x50 km, the northern chain of the Cape Peninsula (with a maximum height of just over 1000 m), the Cape Flats (close to sea level), False Bay and Table Bay would all need to be accounted for in one grid cell. Simplifications therefore need to be made in order to account for all the features in the one grid cell and it is these simplifications that account for many of the inaccuracies in the forecast model. Of these the most important would include "flattening" the peninsula, which would cause the loss of all the complex valleys and peaks that highly modify the wind, temperature and rainfall characteristics in Cape Town. Thus the model used here will not capture rainfall that is caused by the peninsula (orographic rainfall) as well as the finer temperature characteristics.

The resolution problem also applies to resolving small atmospheric processes, for example isolated thunderstorms. As these storms and their processes occur in a space much smaller than the grid size, it is difficult for the model to resolve them. Hence it is not possible to accurately forecast where and when it will storm, but only that there is a chance of isolated thundershowers in a region.

Another problem lies in the globally observed data used to start up the model. Weather data is continiously collected worldwide and the model is initiated (started) with this. Sometimes there are errors in the observational data that make it through automatic and human filters which cause subsequent errors in the forecast. The observational network is also very sparse, especially in South Africa and other non-first world countries, which means the forecast is not optimally initiated in our region and may lead to further inaccuracies in the forecast.

Other problems lie with the actual model:

not all atmospheric processes are understood by scientists so so cannot accurately be taken into account by the model
the equations that describe the atmospheric processes are only solved approximately (for computational cost reasons)
many other approximations are made by the model, again for computational reasons.

For the above reasons there is no such thing as a perfect forecast, it does not exist. However, a two-day forecast is accurate about 90% of the time and it is perhaps more beneficial than no forecast at all. We hope this helps in understanding why the forecast is sometimes wrong. Should you have any other questions, please contact us.

PS. Interestingly, there is a forecast method called the "persistence method" which does not use any computers, equations or forecasters. Essentially this method says "Tomorrows weather will be the same as todays!". Statistically, this method is correct more often than a modelled forecast! So there you have it...