Reconstruction of three-dimensional turbulent flow structures using surface measurements for free-surface flows based on convolutional neural network

We present a machine learning model to reconstruct three-dimensional turbulent flow structures using surface measurements for free-surface flows. The proposed model, which is based on the convolutional neural network (CNN) and is trained using the data obtained from direct numerical simulations of turbulent open-channel flows, can predict the velocity fluctuation field near the free surface and large-scale flow structures away from the surface with high accuracy. The CNN model outperforms a traditional linear stochastic estimation (LSE) model considerably. Further analyses of the CNN model and LSE model provide insights into how the two models are related to the flow physics of the subsurface turbulence.