A digital ecohydraulic twin for riverine habitat prediction and optimization

Rivers sustain aquatic ecosystems by providing suitable physical habitats. Ecohydraulic models are commonly used to assess ecological suitability; however, they are computationally intensive. Data-based models are efficient but often lack spatial resolution, which restricts the spatially explicit management of environmental issues. To bridge the gap, this study develops a reduced-order digital ecohydraulic twin (RETwin), a novel framework that delivers rapid, accurate, and spatially resolved habitat prediction and optimization. The model couples physically informed dimensionality reduction with machine learning (ML)-driven compressive sensing to preserve dominant hydraulic patterns and improve computational efficiency. It represents a practical decision-support tool for sustainable river management and ecosystem conservation. The RETwin is tested using a real-world river reach, targeting European grayling (Thymallus thymallus) and brown trout (Salmo trutta). It generates accurate and fast habitat predictions at both local and reach scales. For the habitat suitability index (HSI), the model achieves a mean coefficient of determination of 0.81, with a root mean square error of 0.13 and a mean absolute error of 0.07. The model produces reliable estimates with mean errors of less than 1 % for the normalized weighted useable area. Habitat optimization experiments demonstrate that the RETwin can identify near-optimal operating conditions within minutes of computation. Training time is reduced from hours for benchmark ML models to minutes, and predictive times per scenario are on the order of seconds (∼300 times faster than physics-based simulations). This combination of accuracy and efficiency makes the model valuable for ecological modeling and operational decision-making.