Scientific research is usually based on theoretical understanding or observations of reality – often the most reliable results come from combining the two.
Climate change research starts with theory – familiar science such as Newton’s laws of motion, energy conservation and so on.
[set next 4 paras as a pop-up box with a link e.g. Here’s an illustration. ?]
Let’s start with the theory. A coin has two sides, so the probability of either side coming up is 50-50. There are only two possible outcomes, and no bias to one side of the coin or the other.
But if you decided to base your prediction on observations, you might expect different results. Let’s say you toss a coin 100 times and note the results. Perhaps you got 60 heads and 39 tails, and once the coin landed on its edge.
So what you might expect to happen when you toss a coin might change depending on whether you are using observations or theory to guide you. Combining the two tells you to roughly expect a 50-50 chance – but that it might take a lot of throws to get that result, and that, very occasionally, something really odd can happen.
Weather forecasters use some (not all) of the same theoretical understanding to build computer models of the oceans and atmosphere to help make their predictions. The models are based on fundamental scientific principles but they are also tested against observations to assess them and to refine their design. There are two really important aspects of the use of observations which makes this process reliable. First, we expect the basic principles that control weather for the next few days to be the same as that which has done so in the past. That means that using past observations to check if the model represents enough of the science for our purposes is okay. Second, everyday we make new weather forecasts and get new weather observations so it is (fairly) easy to check if the forecasting system is reliable. (In statistics these “new forecasts” and “new observations” are called ‘out-of-sample’ observations. They are critical for having confidence in scientific results.)
[Add a page or a box pop-up about insample and outofsample. Probably give a medical eexample.]
But what about predicting climate change?
But we can go further. We can make climate models, based on the theory and our latest understanding we have. These climate models are the main tools we have for working out the consequences in detail. And because we know the future will be different, we also know that we don’t have relevant observations to assess our models against. Observations of the past are immensely valuable but we know enough to know they can’t tell the whole story – like expecting observations of a river like the Thames or the Rhine to help us understand water flowing over a waterfall. Furthermore when trying to predict fifty or hundred years ahead there are no pairs of forecasts and observations to check the system works.
All this menas that our reliance on theoretical understanding is greater than in many scientific disciplines. How we interpret climate model output is one of the biggest challenges in science today. The climate models have huge value for research but how do we make sure that we don’t get mislead by all the apparent detail they produce.
[Last sentence: link to Confidence in ambiguity or Uncertainty and confidence in climate predictions – B: I think the latter is a better link, unless you feel this page is more of a footnote to Flavours/ambiguity, which is where it links from]