Abstract:

Numerical simulations of chaotic and multiscale systems, such as engineering turbulence and globalclimate, present significant challenges. Even at the highest level of simulation fidelity, due to theassumptions introduced in initial condition, boundary condition, or model parameters, numericalpredictions often deviate from experimental measurements or field tests. Data assimilation provides asystematic framework to address these limitations by optimally infusing limited observations intonumerical simulations, enabling the full access to all the scales. In this talk, l will first focus on canonicalwall-bounded turbulence, where the incompressible Navier-Stokes equations are generally considered aperfect model for describing flow physics. Given limited velocity data, our objective is to construct aNavier-Stokes solution that reproduces these observations and predict the unknown flows. A critical dataresolution is identified that ensures an accurate reconstruction of turbulence, Below this critical thresholdthe origins of measurements become increasingly ambiguous, and only the flows within the domain ofdependence of observations can be accurately decoded.l wil then address the more challengingscenario of biased model equations, using climate simulations as an example. A prediction-correctionstrategy will be presented, where data assimilation is utilized to nudge the simulations towardsobservations. These aligned simulation-observation pairs are then used to train a machine-learning modethat identifies and removes systematic biases. When applied to free-running climate simulations, thisframework demonstrates significant improvements in capturing the statistics of rare events.

Introduction of speaker:

Mengze Wang is a Presidential Assistant Professor in the Department of Mechanical Engineering at CityUniversity of Hong Kong. Before ioining CityUHK in 2026.PostdoctoralResearcher athe WaSMassachusetts Institute of Technology since 2023. He received his Ph.D. in Mechanical Engineering fromJohns Hopkins University (2022) and B.Sc. in Theoretical and Applied Mechanics from Peking Universit(2016). His research focuses on inverse problems in fluid dynamics. He developed eficient and accuratescientific computing methods for data assimilation, uncertainty uantification, and generative deeplearning.These approachs have been applied in aerospace engineering, geophysical flows, andbiomedical fluids. He has published over 10 articles on top journals in fluid mechanics and computationalphysics, and one book chapter. His honors include the Corrsin-Kovasznay Outstanding Paper Award andAndrea Prosperetti Travel Award from Johns Hopkins University.

审核：刘有晟、游小清