Enhancing Single-Precision with Quasi Double-Precision: Achieving Double-Precision Accuracy in the Model for Prediction Across Scales-Atmosphere (MPAS-A) version 8.2.1

Abstract. The development of numerical models are constrained by the limitations of high performance computing (HPC). Low precision computations can significantly reduce computational costs, but inevitably introduce rounding errors, which affect computational accuracy. Quasi double-precision algorithm can compensate for rounding errors by keeping corrections, thereby achieving the low numerical precision while maintaining result accuracy. This paper applies the algorithm to the Model for Prediction Across Scales-Atmosphere (MPAS-A) and evaluate its performance across four test cases. The results demonstrate that, after reducing numerical precision to single precision (from 64 bits to 32 bits), the application of quasi double-precision algorithm can achieve results comparable to double-precision computations. The round-off error of surface pressure is reduced by 68 %, 75 %, 97 %, 96 % in cases, the memory has been reduced by almost half, while the computation increases only 2 %, significantly reducing computational cost. The work substantiates both effectiveness and inexpensive computation in numerical models by using quasi double-precision algorithm.