Towards Surrogate Modeling of the Hydrodynamics of Historic Canals

Abstract

Artificial waterways are globally ubiquitous structures, serving not only as important conduits of water and transport, but also as cultural heritage sites and potential ecological refuges. Despite the ubiquity and importance of these structures, little research has studied the conservation value of historic canals in particular (Lin et al. 2020), which requires rigorous characterization of waterway physical design and hydrology, such as seasonal variation in flow (Clifford and Hefferman 2018). To address this need, a procedure for deploying hydrodynamic surrogate models was developed, using the Lake Biwa Canal in Kyoto, Japan as a case study. These models use machine learning to predict and infer core hydrodynamic factors such as eddy-viscosity instead of solving for them numerically with the aim of decreasing computational complexity while maintaining high fidelity. The Lake Biwa Canal was chosen due to its cultural heritage, engineering importance, and extensive history. Photogrammetric surveys of sections of the Lake Biwa Canal provided 3D models upon which ground-truth numerical fluid models were constructed and data-driven turbulence surrogate models were deployed. The resulting surrogate models saw mixed accuracy and performance compared to a ground-truth numerical hydrodynamic model, but provide a basis and workflow for further model improvement.