FlowReconBench evaluates ten learned and five classical methods for reconstructing full flow fields from sparse sensors. It covers laminar cylinder wakes, two-dimensional Kolmogorov turbulence, and three-dimensional isotropic turbulence, using leakage-controlled splits and shared sensor draws.
Why it matters
It clarifies when learned models beat classical estimators for fluid reconstruction, and whether their uncertainty estimates can be trusted.
Analysis
Physics and fluid modelling may favour hybrid methods that combine learned priors with calibrated classical estimators.
Source: arXiv
AI disclosure: this article was produced with AI assistance and reviewed for accuracy.
