How do we work out the weather forecast?

One of the favourite parts of my job (besides the hair/makeup/wardrobe for TV ;) ) is analysing the next run of the weather models.

The main 'runs' are available twice a day, and I actually don't like talking to people in the hour or so before they are ready, because I feel the information I have is 'old'. If anyone asks me a question about the weather at that point, I will usually back away and mutter something about "I'll tell you in an hour" - even if its the news director!

This analysis leads to the forecast that you read or hear, how we come up with the words to describe what is most likely to happen.

Weather models have three steps to them: input, crunch, and the answer.


The current state of the atmosphere.

All the observations from satellites, radars, weather stations, buoys, aeroplanes, ships, weather balloons, soil probes - all of it - are mapped into a current state of the atmosphere.

Image courtesy of ECMWF, one of the international weather models

It would be wonderful if we could grab every bit of information about every inch of the earth and its atmosphere, but we can't. 

For example, weather stations measure what is happening at that exact location, but they may be a long distance from the next weather station. Melbourne Olympic Park is 16 km from Essendon Airport, the next closest station, while a station called Carnegie in outback Western Australia is more than 300 km from the nearest station. 

This means that we have a very good set of data for certain locations, but are missing information in between. 

This is supplemented by satellites, but its not every minute. In fact, satellites now record once every ten minutes, and radar's once every six minutes. Good but not absolutely every piece of information


The model takes all this information and produces an approximate current state of the atmosphere.

This is then used to feed a set of mathematical equations that describe how the atmosphere works - its dynamics and physical processes. 

The earth is huge, and that's a lot of data (even with the missing parts), so the numbers are crunched by supercomputers that can take anywhere from two to eight hours to come up with the solution.


When its all processed, 'the answer' is produced. 

What went in at the start was an approximation, so what comes out at the end won't be the exact answer of what is going to happen. 

And this gets worse with time.

Say they missed a part of a cold front, or thought it was 200 km away from where it was actually located. That front's location could then become out of sync with the rest of what has been modelled, and this gap will grow with each day that we look ahead. This is why I limit my forecasts to 8 days, anything after that can be quite a poor representation of day to day conditions.

But, the technology gets better every year, and the next model run may have a better idea of exactly where that front is, because it affected a passing ship, or went through a denser part of the observation network. 

The answer arrives looking like this:

ECMWF model output from Metservice NZ

This is a sample of the output from a European model (the ECMWF). It shows:

  • MSL - the dark lines - each line joins observations of constant pressure at sea level. You can see a line of 1028hPa pressure south and east of Tasmania. This is an area of high pressure. Well southwest of Perth are two deep lows, shown by lines with numbers less than 970hPa. hPa is a unit of measurement for pressure, the hectopascal.
  • 1000-500 thickness - the grey dashed lines, and the highlighted levels of 5760 in red, and 5400 in blue (for more see Thickness). This is how tall or short a column of air is, between the ground and the 500mb pressure level up in the atmosphere. Higher numbers are taller = warmer air, lower numbers are shorter = colder air. 
  • Rainfall - the coloured areas, using the legend in millimetres on the right hand side. The time step shows that this is rainfall over the last six hours.

We analyse lots of different maps like these, from all the different models (which could be saying rather different things, based on a different crunch equation, and a different approximation of current conditions) and come up with the forecast.

The example above shows a time step of six hours, but there are some in the short term (next few days) that have a resolution of one hour in a 12 by 12 km square. 

But have a think about how a shower moves through (for more see Showers vs Rain). They can last for less than 30 minutes, and be less than one km wide. So, we can't see every shower in these weather models. This is why we describe it as a chance of precipitation - and can't guarantee that you will see wet weather.

The main models have 'the answer' by about 6am and 6pm. On my radio crosses to 3aw each morning, and in the 6pm news on Channel 7, I will have analysed the latest run of these models. The BoM need to have their forecast issued ahead of these times, so they are ready for all the rest of the media to use. This is why my forecasts can sometimes differ. 

Coming soon...
Stylised, easy to use maps, of current and future weather conditions, based on the best of the weather models.
More details to follow.