Can reducing precision improve accuracy in weather and climate models?

Abstract

Better weather and climate forecasts are needed to maximise the ability of societies worldwide to prepare for climate change. This paper summarises work regarding the hypothesis that increasing the resolution of global models to achieve this aim could be done with existing computer technology by reducing the precision of the model variables to increase efficiency. A method of testing this idea carried out using computer simulations of the idealised Lorenz '96 atmosphere is explained. Preliminary results suggest that high-resolution, reduced precision forecasts outperform low-resolution ones. If these results also apply to real-world models, the technique could yield substantial benefits.