Global atmospheric models have been continuously improved in the development process,and have gradually adopted unstructured computing grids,such as the atmospheric component of Model for Prediction Across Scales (MPAS-A) characterized by the spherical centroid Voronoi mesh.In order to improve the initial value of MPAS-A model,relevant data assimilation research is actively carried out simultaneously.In order to meet the needs of rapid assimilation of multi-source observation data by using variational method,this paper takes the Gridpoint Statistical Interpolation (GSI) system used in NCEP business in the United States as the assimilation module,handles transformations between unstructured and structured spherical grids based on conservation remapping method,constructs the GSI-MPAS assimilation and prediction framework,and carries out grid transformation experiments and assimilation prediction experiments.The grid transformation tests show that the transformation errors of model physical quantities are closely related to their distribution patterns,and the second-order precision conservative remapping transformation results are better than the first-order precision transformation results.The continuous rolling cycle assimilation and prediction experiment for one week shows that the GSI-MPAS assimilation and prediction framework based on the conservation remapping method can effectively assimilate multi-source observation data,improve the quality of initial value field,make each variable obtained by MPAS-A prediction more accurate,and have a positive effect on precipitation prediction.Further analysis shows that the improvements in the Northern Hemisphere are obviously better than those in the Southern Hemisphere and the equatorial region due to the assimilation of more observation data in the Northern Hemisphere.