Atmosphere Ocean Science Friday Seminar

---

Estimating Surface-Layer Momentum, Sensible Heat, and Latent Heat Fluxes by Improving Bulk Flux Estimates + Utilizing Machine Learning

Speaker: Nathan Mitchell, NYU

Location: Warren Weaver Hall 1314

Date: Friday, September 11, 2026, 4 p.m.

Synopsis:

 

Monin-Obukhov Similarity Theory (MOST), developed by Monin and Obukhov (1954), uses assumptions about the surface layer to derive formulas to replace eddy covariance fluxes. These so-called “bulk” estimates are widely used in climate models today to predict the exchange of momentum, heat, and moisture between the surface and the atmosphere. These fluxes drive the climate system, so it’s imperative we estimate them accurately. This project has two components. First, we combine the work of numerous past authors and introduce our own correction factor to improve the bulk estimates as much as possible. Second, we set the bulk formulas aside and utilize machine learning, namely feedforward neural networks, to predict the fluxes directly. What we find is that the machine learning algorithms outperform the bulk estimates to varying degrees.