Abstract: Under active ionospheric conditions, the correction performance of existing GNSS broadcast ionospheric models generally suffers from degradation to varying degrees. To address this common challenge, this study focuses on the system-level parameter estimation and mathematical representation of the BeiDou Global Ionospheric Delay Correction Model (BDGIM), the broadcast ionospheric model adopted by BeiDou-3, and proposes a multi-dimensional refinement strategy to improve its accuracy and robustness. The results indicate that, by updating the ionospheric background field on a semiannual basis, the mean root mean square error (RMSE) is reduced from 11.78 TECU to 7.84 TECU, representing an improvement of approximately 33%. After incorporating globally distributed overseas stations, the RMSE is further reduced from 9.92 TECU to 8.69 TECU, with the most pronounced improvement observed in low-latitude regions. By further introducing Tianmu low Earth orbit (LEO) radio occultation observations, the global mean error decreases from 15.16 TECU to 10.45 TECU. Moreover, a 9-parameter regional modified spherical harmonic model is developed, achieving an accuracy improvement of approximately 26.8%. In addition, for the negative values occasionally occurring in some high-latitude regions, the application of a minimum-value constraint is recommended. These results demonstrate that the proposed methods can significantly improve the accuracy, stability, and regional applicability of BDGIM.