Deep learning model for hydraulic engineering

My research focuses on advancing hydrologic and hydraulic applications by predicting riverbed topography, which is crucial but often inaccessible data. I devised a conditional generative adversarial network (CGAN) to predict river cross-sections. The model significantly reduces error compared to traditional conceptual methods so that it can enhance simulations in sediment transport, flooding mapping, and watershed management. Through this research, I explore the potential of neuron networks to improve our understanding and management of water resources.

Figure 1 The LiDAR technique cannot detect the topography below the water surface, so a hydro-flatten area is shown in the channel.

Figure 2 Leveraging the deep learning model, we can generate reseaonable riverbed surface (the mesh in the figure above) for further analyses or applications.