Forecasting the weather over the coming week is a so-called ‘initial value problem’. The goal is to take information about the current state of the atmosphere and propagate it forward in time using our forecast models. In other words, the initial state gives us all the information we need to know about the weather over the coming week.
Forecasting the climate over the coming century is a so-called ‘boundary value problem’. The biggest issues are to firstly make informative estimations about future forcing on the system (e.g. greenhouse gas concentrations), and to secondly, build an Earth-system model that will respond correctly to that imposed forcing. The precise starting conditions of the climate model are unimportant – only the transient forcing gives information.
On intermediate timescales, both initial conditions and boundary conditions can play a role. But because the Earth-system has components with different time scales, a forecast time scale which is a boundary value problem for one component could be an initial value problem for a second component.
For example, on timescales of a few months to years, the atmospheric problem is one of correctly supplying the boundary conditions. However, the oceans vary much more slowly than the atmosphere. Over a few months, the ocean still has information from the initial conditions. Since the ocean state provides a boundary condition for the atmosphere, we find we can predict aspects of the atmospheric state many months in advance.
This project seeks to unpick sources of atmospheric predictability on timescales from days to decades. What boundary conditions provide information to the atmosphere, and for how long in advance are these boundary conditions predictable? And finally, how well can forecast models reproduce these sources of predictability?