Numerical models of the atmosphere and ocean have proved to be immensely valuable forecasting tools for short time-scale weather and longer time-scale seasonal and climate prediction. As the decades pass, these models have been improving due to increased computing power, improved modelling of the dynamics, improved parametrisation of sub-grid scale processes and improved use of observations. These modelling improvements may be slowing and further large increases in computing power will almost certainly emerge from heterogenous computing architectures configued in even more massively...