Architecture

A novel lightweight neural network where atmospheric physics is embedded directly into the network structure, not merely the loss function. Four stability-regime expert sub-networks are dynamically gated by a learned Monin-Obukhov classifier, while an analytical backbone provides zero-parameter TKE estimation that the neural residual refines.

Output turbulence corrections are derived via Kolmogorov spectral constraint defining physically consistent energy cascade scaling across all atmospheric regimes from convective boundary layer through stratospheric cruise.

4-Regime Gated MoE
FiLM Density Conditioning
Kolmogorov Spectral Output
NASA POWER Weather

Trajectory

6-DoF missile flight simulation with real-time ML turbulence correction

Turbulence Globe

Global atmospheric turbulence analysis across 8 flight levels

PSTNet v1.0|NASA POWER Weather|Zero Framework Dependencies
Validated: Supersonic M2.8 · High Supersonic M4.5 · Hypersonic M8.0