Figure 6
From: A scalable convolutional neural network approach to fluid flow prediction in complex environments
Error propagation in large domains. (A) Domain decomposition of a single \(1280 \times 384\) domain into set of \(128 \times 128\) subdomains. Error bars are scaled by a factor of \(10^3\) for visual clarity. (B) Pixel level mean-square error (MSE) of each subdomain, evaluated on CFD data in a \(1280 \times 384\) domain resolution. As expected, subdomain level errors compound as distance increases from the inlet of constant flow (here on the left hand vertical boundary). (C) Mean pixel-level MSE of the central row for each model, comparing Contiguous model predictions (blue) against Continuous model predictions (purple). Inlet direction is from left to right.
