Predictions of weather and climate are inherently uncertain: initial conditions can never be known perfectly, and our weather and climate simulators are necessarily imperfect representations of the underlying laws of physics. Given this, how can we quantify reliably the uncertainty in weather and climate predictions, from one day ahead to one century ahead, and what is needed to reduce current levels of uncertainty?
These are tough problems. Answering them will be crucial for society worldwide and will involve research across many disciplines, including the physics of weather and climate, nonlinear dynamical systems theory, and the theory of stochastic processes.
A second area of interest is adapting our models to utilise new computer hardware that trades precision for improved performance and reduced energy consumption. This enables us to optimise our use of the available computer resources, ultimately improving our forecasts.
Fig: Ensemble forecasts in the Lorenz attractor