Fig. 22From: Deep convolutional architectures for extrapolative forecasts in time-dependent flow problems2D River problem: Interpolative auto-regressive predictions using the CNN model for 20th, 30th, and 40th timestep after the first initial 30 time-steps used in the lookback window (i.e. timestep = 50, 60, and 70 respectively); the end of training is at time-step = 80Back to article page