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.

Recent shifts in the behaviour of natural watersheds suggest acute challenges for water planning under climate change. Shifts towards less annual streamflow for a given annual precipitation have now been reported on multiple continents, usually in response to a multi-year drought. Future drying under climate change may induce similar unexpected hydrological behaviour, and 15 this commentary discusses the implications for water planning and management. Commonly-used hydrological models poorly represent the shifting behaviour and cannot be relied upon to anticipate future shifts. Thus, their use may result in underestimation of hydroclimatic risk and exposure to “surprise” reductions in water supply, relative to projections. The onus is now on hydrologists to determine the underlying causes of shifting behaviour and incorporate more dynamic realism into 20 operational models. Main points 1. Drought-induced hydrological shifts towards less streamflow for a given precipitation have been reported across multiple continents. 2. Future drying under climate change may induce similar unexpected behaviour. 25 3. Such behaviour creates additional uncertainty in runoff projections, and may lead to ‘surprise’ reductions in future streamflow. Main text In a recent article, Peterson et al. (2021) reported shifts in hydrological behaviour induced by the “Millennium” drought (1997-2010) in Australia and persisting years after the drought ended. 30 Reductions in water resources during and after this drought were far more extreme than expected, even given low rainfall (Saft et al., 2015), because many watersheds shifted into a seemingly different state of streamflow behaviour. Concerningly, some watersheds remain in this state despite a return to near-average climate conditions, so that a year of average rainfall now produces less streamflow than it did before the drought (Peterson et al., 2021). With similar hydrological 35 shifts reported elsewhere in the world, including the USA (Avanzi et al., 2020), China (Tian et al.,