By: Annika Reintges
Imagine you are planning a birthday party in 2 weeks. You might check the weather forecast for that date to decide whether you can gather outside for a barbeque, or whether you should reserve a table in a restaurant in case it rains. How much would you trust the rain forecast for that day in 2 weeks? Probably not much. If that birthday was tomorrow instead, you would probably have much more faith in the forecast. We all have experienced that weather prediction for the near future is more precise than prediction for a later point in time.
A forecast period of 2 weeks is often stated to be the limit for weather predictions. But how then, are we then able to make useful climate predictions for the next 100 years?
For that, it is important to keep in mind the difference between the terms ‘weather’ and ‘climate’. Weather changes take place on a much shorter timescale and also on a smaller scale in space. For example, it matters whether it will rain in the morning or the afternoon, and whether a thunderstorm will hit a certain town or pass slightly west of it. Climate however, are weather statistics averaged over a long time, usually over at least 30 years. Talking about the climate in 80 years, for example, we are interested whether UK summer will be drier. We will not be able to say whether July of the year 2102 will be rainy or dry compared to today.
Because of this difference between weather and climate, the models differ in their specifications. Weather models have a finer resolution in time and space than climate models and are run over a much shorter period (e.g., weeks), whereas climate models can be run for hundreds or even thousands of years.
Figure 1: ‘Weather’ refers to short-term changes, and ‘climate’ to weather conditions averaged over at least 30 years (image source: ESA).
But there is more to it than just the differences in temporal and spatial resolution:
The predictability is based on two different sources: Mathematically, (1) weather is an ‘initial value problem’, (2) climate is a ‘boundary problem’. This is related to the question: how do we have to ‘feed’ the model to make a prediction? In other words, which type of input matters for (1) weather and (2) climate prediction models. A weather or climate model is just a set of code full of equations. Before we can run the model to get a prediction, we have to feed it with information.
Here we come back to the two sources of predictability:
(1) Weather prediction is an ‘initial value problem’: It is essential to start the model with initial values of one recent weather state. This means several variables (e.g., temperature and atmospheric pressure) given for 3-dimensional space (latitudes, longitudes and altitudes). This way, the model is informed, for example, about the position and strength of cyclones that might approach us soon and cause rain in a few days.
(2) Climate prediction is a ‘boundary value problem’: For the question whether UK summers will become drier by the end of the 21st century, the most important input is the atmospheric concentration of greenhouse gases. These concentrations are increasing and affecting our climate. Thus, to make a climate prediction, the models needs these concentrations not only from today, but also for the coming years, we have changing boundary conditions. For this, future concentrations are estimated (usually following different socio-economic scenarios).
Figure 2: Whether a prediction is an ‘initial value’ or ‘boundary value’ problem, depends on the time scale we want to predict (image source: MiKlip project).
And the other way around: For the weather prediction (like for the question of ‘will it rain next week?’), boundary conditions are not important: the CO2 concentration and its development throughout the week do not matter. And for the climate prediction (‘will we have drier summers by the end of the century?’), initial values are not important: it does not matter whether there was a cyclone over Iceland at the time we started the model run.
Though, hybrid versions of weather/climate prediction exist: Say we want to predict the climate in the ‘near’ future (‘near’ in climate timescales, for example in 10-20 years). For that, we can make use of both sources of predictability. The term used in this case would be ‘decadal climate prediction’. With this, we will of course not be able to predict the exact days when it will rain, but we could be able to say whether the UK summers in 2035-2045 will on average be drier or wetter than the preceding 10 years. However, when trying to predict climate beyond this decadal time scale, the added value of adding initial values to climate prediction is very limited.