High computational cost is often the most limiting factor when running high-resolution hydrodynamic models to simulate spatial-temporal flood inundation behaviour. To address this issue, a recent study introduced the hybrid Low-fidelity, Spatial analysis, and Gaussian Process learning (LSG) model. The LSG model simulates the dynamic behaviour of flood inundation extent by upskilling simulations from a low-resolution hydrodynamic model through Empirical Orthogonal Function (EOF) analysis and Sparse Gaussian Process (Sparse GP) learning. However, information on flood extent alone is often not sufficient to provide accurate flood risk assessments. In addition, the LSG model has only been tested on hydrodynamic models with structured grids, while modern hydrodynamic models tend to use unstructured grids. This study therefore further develops the LSG model to simulate water depth as well as flood extent and demonstrates its efficacy as a surrogate for a high-resolution hydrodynamic model with an unstructured grid. The further developed LSG model is evaluated on the flat and complex Chowilla floodplain of the Murray River in Australia and accurately predicts both depth and extent of the flood inundation, while being 12 times more computationally efficient than a high-resolution hydrodynamic model. In addition, it has been found that weighting before the EOF analysis can compensate for the varying grid cell sizes in an unstructured grid and the inundation extent should be predicted from an extent-based LSG model rather than deriving it from water depth predictions.

Accurate flood inundation modelling using a complex high-resolution hydrodynamic (high-fidelity) model can be very computationally demanding. To address this issue, efficient approximation methods (surrogate models) have been developed. Despite recent developments, there remain significant challenges in using surrogate methods for modelling the dynamical behaviour of flood inundation in an efficient manner. Most methods focus on estimating the maximum flood extent due to the high spatial-temporal dimensionality of the data. This study presents a hybrid surrogate model, consisting of a low-resolution hydrodynamic (low-fidelity) and a Sparse Gaussian Process (Sparse GP) model, to capture the dynamic evolution of the flood extent. The low-fidelity model is computationally efficient but has reduced accuracy compared to a high-fidelity model. To account for the reduced accuracy, a Sparse GP model is used to correct the low-fidelity modelling results. To address the challenges posed by the high dimensionality of the data from the low- and high-fidelity models, Empirical Orthogonal Functions (EOF) analysis is applied to reduce the spatial-temporal data into a few key features. This enables training of the Sparse GP model to predict high-fidelity flood data from low-fidelity flood data, so that the hybrid surrogate model can accurately simulate the dynamic flood extent without using a high-fidelity model. The hybrid surrogate model is validated on the flat and complex Chowilla floodplain in Australia. The hybrid model was found to improve the results significantly compared to just using the low-fidelity model and incurred only 39% of the computational cost of a high-fidelity model.