How climate scientists develop climate models
Submitted by astone on Thu, 06/07/2012 - 14:52
When commentators dismiss climate models as “merely models” it means they have failed to grasp how important models of all kinds have become to many parts of our daily life.
We can see their importance in the way the word “model” appears in the media every day. Models are used to predict how the Australian economy will change, how traffic lights affect traffic flow, how the heart works and how an earthquake might trigger a tsunami. Models are everywhere, helping design new cars, wings for aeroplanes, improving manufacturing processes and exploring how climate will change in the future. There are many types of model.
Conceptual models are things you and I use every day. You use a conceptual model to connect things that matter to you. They are basically a logical connection of ideas that when packaged together help you understand something you care about.
Statistical models use statistical methods to try to understand data in a way that helps you understand something you care about. You may have heard about regression models, or understand that just because two data sets are highly correlated does not mean they are necessarily connected.
Conceptual and statistical models are very different from the models used to understand our climate. Climate models are physical models. They are built on mathematical and physical principles. For example, if I give you $10 and take away $5 you expect to be left with $5. You would be terribly surprised if I gave you $10, took $5 away and you had $500 in your pocket. This example is like the “conservation” laws of physics used in climate models. These laws state energy, mass and momentum are “conserved”. These are fundamental laws; mass, energy and momentum cannot be miraculously created or lost – you have to be able to exactly tell where it all goes. Economic models, for example, are not built on these kinds of fundamental laws – so if you hear that climate models cannot work because they are models and “just look at how bad economic models are” its best to conclude that the person saying this does not understand what they are talking about.
Climate models are built in two key steps. First, scientists find the mathematical expressions for the physical laws that explain climate. Then, scientists implement these mathematical expressions on a computer. This can be extremely complex and a large community in the mathematics, physics and computer science disciplines have developed what are called numerical methods to help with this.
To this end, climate models use a three-dimensional grid. Imagine for a moment a chessboard wrapped around the Earth and another chessboard wrapped around the Earth a little higher in the atmosphere, and then another and another. Hopefully, you can imagine the atmosphere around the entire planet represented by rectangular boxes (we do the same for the oceans of course).
The climate system is among our most complex natural systems, so modelling what happens in and across these boxes requires seriously big computers.
Now, because computers are not infinitely fast, these rectangular boxes have to be of a particular size to be able to complete the vast number of calculations. Currently for climate science the most common size for these boxes is about 300 x 300km each. The size of these rectangular boxes imposes some important constraints on how a climate model is built.
First, climate scientists want to make the rectangular boxes smaller. With a smaller box size, scientists can simulate some important things more realistically. Eddies in the ocean, patterns of vegetation and clouds are examples of important things that exist at scales smaller than rectangular boxes. Unfortunately, as you decrease the size of the boxes, from say 300 x 300km to 100 x 100km, it dramatically increases the amount of computer time needed to such an extent that few computers exist that are big enough.
For this reason scientists have to define the processes that exist at scales below the size of the rectangular box in different ways – this is called parameterization.
Parameterization is very complex and challenging. It requires an intimate understanding of the processes you are trying to parameterize. Climate scientists, physicists, mathematicians, chemists, biologists, geologists, soil scientists, crop scientists, oceanographers and many other areas of science work together to parameterize these processes.
Let's look at an example – the land surface. There are some fundamental laws land surface scientists have to adhere to, such as the conservation of mass and energy.
Rainfall reaching the Earth’s surface cannot disappear – it must be accounted for. Some is re-evaporated from leaves that catch the rainfall and some hits the soil surface where it flows into streams. Some accumulates in the soil and is sucked up and transpired by plants.
Researchers have to parameterize how leaves intercept rainfall, how water moves through soil and how plants suck up water and transpire it. These parameterizations are based on observations made by soil scientists, plant physiologists etc. The mathematical expressions derived from these observations are translated into computer code.
Once scientists have done this, there is a continual process of comparing the results of the models to real world observations. Researchers continually expose their models to new observations as they become available. There is a range of well-organized international programs of observations that are continually reviewed and used to see if any weaknesses appear in the models. Other groups modelling clouds, aerosols, ocean biology or one of the many other disciplines using models all do the same.
The final step in the model development process is to combine all the components into a full climate model and then evaluate it as a whole. The whole model is compared to the current climate, the climate of the 19th and 20th century, how it responds to volcanic eruptions, how it responds to changes in the sun and how it captures ancient climates. We also compare our models with results from other models built in the US, France, Germany, UK, Japan, China, and several other countries, continuously looking for better ways to simulate the climate. This process of building, evaluating and refining a comprehensive climate model is very complex and time-consuming and typically requires many years and dozens of scientists and software engineers.
No model is perfect but many models are useful. Climate models are now so good that a climate scientist could not distinguish between the simulated temperature pattern over the northern hemisphere and the observations of those temperatures taken from the field. The models respond to known impacts such as volcanic eruptions and trends through the 19th and 20th century in ways that are consistent with the observations.
While the models are not perfect, they are outstanding tools for predicting climate at the scales they were designed for. So, if you hear someone on the radio dismissing climate models as “merely models” it is worth reflecting that these “merely models” helped design your car, the aeroplane you flew in on and the understanding of how your heart or brain work.
A key to understanding the reliability of climate models is to know that they are based on laws of physics. They are not like the economic models or models of human behaviour. These are, of course, useful and legitimate models, but because they are not dependent on the fundamental laws of physics they suffer from a disadvantage that climate models avoid. Ultimately, dismissing climate models is like dismissing gravity when deciding whether or not to use a rope for bungee jumping. It’s a pretty silly attitude really and it reflects a limited understanding of how most 21st Century science works.
Article by Director of COECSS Prof Andy Pitman and Deputy Director Prof Christian Jakob.
Want to know more?