Estimating divergence‐free flows via neural networks

We apply neural networks to the problem of estimating divergence-free velocity flows from given sparse observations. Following modern trend combining data and models in physics-informed networks, we reconstruct flow by training a network such manner that not only matches observations but also approximately satisfies condition. The assumption is balance between two terms allows obtain model has better prediction performance than usual data-driven network. this approach reconstruction truly noiseless synthetic wind fields over Sweden.