Improvement of the computational efficiency in SVD-3DEnVar data assimilation scheme and its preliminary application to the TRAMS 3.0 model

Abstract

Although the Singular Value Decomposition-three Dimensional Ensemble Variational (SVD-3DEnVar) data assimilation scheme has achieved successful application in real case simulations with comprehensive numerical weather prediction models, its computational efficiency still falls short of meeting the demands of operational numerical prediction. The main limitations lie in the generation of three-dimensional perturbations and the implementation of parallel calculations. To bridge this gap towards operational readiness, this study introduces key computational optimizations: a new three-dimensional perturbation field generation scheme that supports multi-process parallelism and can directly generate any specified grid, and an efficient parallel implementation scheme tailored for the local patch assimilation in the SVD-3DEnVar scheme. Results from Observing System Simulation Experiment (OSSE) based on the Tropical Regional Atmospheric Model System (TRAMS) indicate that after computational efficiency optimization, the time required to generate a 3D perturbation field was reduced from 22 min to 2.2 s, while the runtime of the assimilation process decreased from costly serial execution, to 1700 min under single-node parallel execution (with 64 cores), and further to less than 15 min (using 150 nodes in parallel). Finally, we conducted an assimilation experiment using actual observational data of sea surface wind fields to preliminarily validate the reasonableness of the assimilation results from the optimized SVD-3DEnVar scheme.