Mixed-precision arithmetic in the ENDGame dynamical core of the Unified Model, a numerical weather prediction and climate model code

Abstract

The Met Office’s weather and climate simulation code the Unified Model is used for both operational Numerical Weather Prediction and Climate modelling. The computational performance of the model running on parallel supercomputers is a key consideration. A Krylov sub-space solver is employed to solve the equations of the dynamical core of the model, known as ENDGame. These describe the evolution of the Earth’s atmosphere. Typically, 64-bit precision is used throughout weather and climate applications. This work presents a mixed-precision implementation of the solver, the beneficial effect on run-time and the impact on solver convergence. The complex interplay of errors arising from accumulated round-off in floating-point arithmetic and other numerical effects is discussed. A careful analysis is required. The mixed-precision solver is now employed in the operational forecast to satisfy run-time constraints without compromising the accuracy of the solution.