第二届未来大气科学论坛

The accuracy of weather forecasts depends critically on the quality of the initial conditions, which are usually obtained by integrating multi-source observational data and numerical forecast products within data assimilation systems. This study involved the development of a four-dimensional ensemble-variational (4DEnVar) data assimilation system within the Model for Prediction Across Scales (MPAS) framework. The system exhibits four distinctive characteristics: First, it uses an advanced unstructured-grid numerical model that can perform both quasi-uniform and variable-resolution simulations, which enhances its ability to capture multi-scale atmospheric processes. Secondly, it incorporates an enhanced nonlinear least squares four-dimensional variational (NLS-4DVar) assimilation algorithm that uses ensemble samples to estimate the background error covariance matrix. This eliminates the need for tangent linear and adjoint models. Third, the enhanced NLS-4DVar algorithm introduces a shrinkage factor to resolve the conflict that arises from using ensemble sample perturbations for both constructing the background error covariance matrix and approximating the tangent linear model. This achieves a precise approximation of both components and consequently improves numerical stability and assimilation accuracy. Fourthly, the system can be readily coupled with the Unified Forward Operator (UFO) module to facilitate the efficient assimilation of multi-source observations. Meticulous, comprehensive one-month observing system simulation experiments (OSSEs) demonstrate that the system has considerable potential and substantial advantages in reducing analysis errors and improving MPAS forecast accuracy. In addition, a one-week real observation experiments comparing the proposed system with JEDI-4DEnVar further demonstrate its performance advantages over existing MPAS-based data assimilation systems.

Data Assimilation, NLS-4DVar, 4DEnVar, NWP, MPAS