<p>Dike-break floods, characterized by high flood peaks, large volumes, and sudden onsets, seriously threaten the flood control and safety of river basins. In addressing the computationally intensive and time-consuming problem of numerical modeling of large-scale outburst floods, this study proposed a novel hybrid alternative modeling approach. The proposed methodology integrates a low-fidelity (LF) hydrodynamic model with a sparse Gaussian processes (SGP) model, incorporates rotated empirical orthogonal functions (REOF) to address high-dimensional data challenges, streamlines the model structure, and enhances computational efficiency. The SGP model uses training data from the high-fidelity (HF) model to rectify LF model results, enhancing computational efficiency while ensuring precise reproduction of HF model results. Validation in the Yongding River floodplain demonstrates that the hybrid model significantly improves flood extent and depth predictions compared to the LF model, with substantially lower computational costs than the HF model. The results indicate that the REOF-SGP model achieved probability of detection (POD) values higher than 0.8 and rate of false alarm (RFA) values lower than 0.2 within the 120-hour simulation period. The prediction error for inundation depth in the floodplain generally fell within the range of (−0.1 m, 0.1 m). The computational efficiency was 11 times higher than that of the HF hydrodynamic model. This method enhances large-scale flood inundation calculation efficiency while ensuring refined simulation of dynamic flood area changes, aiding rapid prediction of nonlinear flood evolution and water depth distribution.</p>