Abstract: Objectives: Traditional Global Navigation Satellite System (GNSS) vertical time series prediction methods usually model the data solely based on temporal variables, which limits their ability to capture the dynamic characteristics of the sequences. To comprehensively account for the influence of multisource factors—such as terrestrial water storage, temperature, pressure, precipitation, and polar motion—on GNSS station vertical time series, this study proposes a hybrid prediction approach integrating Variational Mode Decomposition (VMD) and Light Gradient Boosting Machine (LightGBM). Methods: The proposed method is evaluated using GNSS vertical time series data from eight stations across Europe and the Americas. Among them, four representative stations are selected for detailed analysis of prediction performance. The predictive capability of the multi-source data-driven LightGBM model is compared with three classical autoregressive neural network models: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Recurrent Neural Network (RNN). Results: The experimental results indicate that the LightGBM model incorporating multi-source data significantly outperforms the three autoregressive models in terms of prediction accuracy. Specifically, the Root Mean Square Error (RMSE) is reduced by 8.5%~40.2%, and the Mean Absolute Error (MAE) is reduced by 8.0%~42.7%. In further prediction experiments, the proposed VMD-LightGBM combined model achieves additional improvements compared to the standalone LightGBM model, with RMSE reduced by 32.5%~40.8% and MAE reduced by 30.2%~37.7%. Conclusions: The proposed VMD– LightGBM combined model exhibits outstanding performance in prediction accuracy, stability, and generalization capability.