Explainer: computer weather models
29 March 2019
It’s no exaggeration to say computer weather models have revolutionised weather forecasting in the modern era—and they’re getting smarter all the time! So, what are they, how do they work and what can they tell us about our dynamic weather and climate?
What is a computer weather model?
Computer weather models are the main tool we use to forecast weather—as well as a key tool to research the processes that drive weather and climate. The models, also known as numerical weather prediction (or NWP) models, essentially create a virtual planet Earth, simulating the atmosphere, ocean, land surface and sea ice, and use mathematical equations to predict future weather.
Weather models have revolutionised the science of weather prediction. Improvements in models, and a vast increase in the observation data that feeds them, are a big part of why our seven-day forecasts today are about as accurate as a three-day forecast in the early 2000s. The map below shows how steadily improving weather models between 1974 and 2010 would have forecast the path of severe tropical cyclone Tracy, coming closer and closer to the path it actually took.
How do they work?
Weather models carve up the atmosphere into a large number of grid boxes, vertically and horizontally. We take what we know about the weather now—readings from all levels of the atmosphere and under the ocean known as 'observations'—and feed it in. As the models step forward in time, the temperature, humidity and wind in each grid box are updated by calculating heat, moisture and momentum (wind) transfer into and out of the grid box. This takes into account the way the atmosphere interacts with the ocean and land surfaces, such as sunlight heating the ocean and snow cooling the land.
Many processes in the atmosphere (such as thunderstorms and clouds) happen at scales much smaller than the size of the grid boxes used in most models—although this is changing as models improve. Depending on the model, we approximate these elements in each grid box using additional sets of equations.
Video: Short animation illustrating the concept of how weather models divide the Earth into grid boxes.
Taking all the mathematical equations that explain the physics of the atmosphere and calculating them at hundreds of millions of points around the Earth is an enormous job. That's why the models need vast computing power to complete their calculations in a reasonable amount of time, meaning they must use some of the most powerful supercomputers in the world. Our supercomputer can handle more than 1600 trillion calculations per second!
We don't just run the equations once either—under some circumstances we run them multiple times, each with slightly different starting conditions. This is called ensemble forecasting. It tells us whether small changes in the atmosphere or ocean are likely to cause a shift on the forecast. To calculate climate outlooks (a forecast covering the next three months) we run them 165 times. If most of the runs come out with the same result, for example a wetter- or drier-than-average season, then the odds are higher for that outcome. If the model runs show a range of outcomes—some suggest wetter and some drier—then there's no strong signal one way or the other, which means there's a roughly equal chance of wetter or drier conditions.
What can they tell us?
The models can help us predict weather and ocean behaviour from time scales of hours to days to weeks to seasons. They can also help us with prediction and trajectory of severe events, such as tropical cyclones, rainfall and flooding, ocean waves, storm surges and tsunamis.
We also use models in research, to increase our understanding of the physical processes that drive weather and climate and the role played by interactions between the atmosphere and elements such as land, oceans and ice. An example is modelling how pyrocumulonimbus clouds (caused by bushfires) develop and behave. This should lead to better ability to forecast these dangerous conditions in the future—and the insights we gain are also used to improve the models.
Models can also help us understand the changes we are seeing in our climate over time. Estimating the relative contributions of natural variability and climate change to past extreme events such as record-breaking heat is one way of investigating this. It's done by retro-modelling the weather systems that caused the heat, but with less CO2 in the environment than was historically the case and seeing how that changes the outcome.
What does the future hold?
All the time scientists are working to make weather models more accurate. Basically, this boils down to being able to input better data about current conditions such as temperature, humidity and air pressure at the start of the process, and being able to run the models at a ‘finer resolution'. This means that each of the grid boxes that the atmosphere is divided into is smaller, which allows greater detail in the modelling. It's like the difference between looking at a road map at the regional level, showing highways and main roads, and zooming into your town or suburb, with every street showing.
The resolution in forecast models is ever-improving. Our current global weather forecasting model has 25 km grid squares, but our forthcoming model has double the resolution, with 12 km grid squares. Our model for short-range forecasting over more limited areas (e.g. covering eastern NSW) has grid squares 1.5 km wide.
As the models improve, the time-steps between model calculations decrease, so we also get a faster result.
Of course, all that adds up to more and more equations, so being able to run the improved models depends upon enough computing power being available to crunch the numbers.
When the models are calculated at a more detailed level, forecasts overall become more accurate and our ability to forecast things that happen at a very local level should also improve. For instance, fog that may affect airports, or thunderstorms. These are examples of weather phenomena that are influenced by the local details at the surface, such as the shape of the terrain, as well as the larger-scale weather patterns.
Image: Australia, seen from the Japan Meteorological Agency's Himawari satellite. In the southern hemisphere, satellites are a key source of the observations that feed into weather models.
Working with multiple models
Australia has its own weather and climate model, developed by the Bureau and CSIRO. It's called the Australian Community Climate and Earth System Simulator ('ACCESS' for short). But there are several other models around the world that we also look at when making weather and climate forecasts. These include models from meteorological agencies in Japan, the US, the UK, France and Canada—and one that is a collaboration between more than 30 European nations (called ECMWF).
Models around the world have common elements but also differ, particularly in the additional equations they use to approximate the smaller-scale weather processes. It's useful to watch models from other regions as well as our own because if more models are showing a particular outcome (such as a developing low-pressure system that may form into a tropical cyclone) we can have increased confidence in the likelihood that it will happen. And when we blend the output of the various models, the results are more accurate than any single model. This is also relevant for predicting climate drivers such as the El Niño–Southern Oscillation and the Indian Ocean Dipole, where we provide an average of international models as part of our own outlook.
Subscribe to this blog to receive an email alert when new articles are published.