Distilling deep optical flow stereo methods to retrieve dense three-dimensional wind fields
Abstract
Geostationary atmospheric motion vectors (AMVs) provide the dense horizontal wind vectors (u,v) and heights ingested into data assimilation systems. Traditional AMVs track features using window-based cross-correlation and estimate heights via infrared brightness temperatures paired with numerical weather prediction (NWP) background states, creating a circular dependency that yields inaccurate heights, high computational cost, and sparse retrievals. Stereo winds from GEO-GEO and GEO-LEO geometrically resolve heights from parallax shifts across different poses, eliminating NWP dependence and improving accuracy, but they remain computationally heavy with limited coverage. In this work, we replace window-based tracking in stereo matching with deep optical flow for efficient, improved retrieval. Fine-tuning balances a self-supervised geometric residual loss with supervised radiosonde reconstruction. To eliminate multi-satellite overlap requirements, we distill the stereo teacher into a single-satellite student model. Chi-square and height uncertainties from the teacher are emulated by the student for quality assurance. The student generates winds across full-disk GEO imagery globally. Validation compares stereo and student models against radiosondes, operational AMVs, ERA5 reanalysis, and EarthCARE cloud profiles. Results through triple collocation show that stereo winds improve performance beyond operational AMVs for water vapor bands (6.2, 6.9, and 7.3 μm), wit degradation in the long-wave infrared (11.2 μm) band.
Related
- Advancing Methods to Monitor Greenland Ice-Marginal Lake Area Fluctuations and Outburst Floods Using Machine Learning and Multispectral Optical Imagery
- ‘envar’: An R package to retrieve and process environmental variables for macroecology and biogeography
- Brief communication: Mass balance of glaciers in the Eastern Hindu Kush from 1979 to 2002 using KH-9 panoramic and SPOT-5 stereo imagery
- Optical mediation of chlorophyll-a trends and radiative heating in coastal waters
- Forward modeling of near-field optical processes in pure rotational Raman temperature lidar
BibTeX
@misc{vandal2026distilling,
title = {{Distilling deep optical flow stereo methods to retrieve dense three-dimensional wind fields}},
author = {Thomas J. Vandal and Dong L. Wu and James L. Carr and Derek J. Posselt and Elise Penn and Tristan Ballard and August Posch and Kate Duffy},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.03100},
eprint = {2609.03100},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.03100},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Distilling deep optical flow stereo methods to retrieve dense three-dimensional wind fields
AU - Thomas J. Vandal
AU - Dong L. Wu
AU - James L. Carr
AU - Derek J. Posselt
AU - Elise Penn
AU - Tristan Ballard
AU - August Posch
AU - Kate Duffy
PY - 2026
DA - 2026/09/02
PB - arXiv
DO - 10.48550/arxiv.2609.03100
UR - https://arxiv.org/abs/2609.03100
AB - Geostationary atmospheric motion vectors (AMVs) provide the dense horizontal wind vectors (u,v) and heights ingested into data assimilation systems. Traditional AMVs track features using window-based cross-correlation and estimate heights via infrared brightness temperatures paired with numerical weather prediction (NWP) background states, creating a circular dependency that yields inaccurate heights, high computational cost, and sparse retrievals. Stereo winds from GEO-GEO and GEO-LEO geometrically resolve heights from parallax shifts across different poses, eliminating NWP dependence and improving accuracy, but they remain computationally heavy with limited coverage. In this work, we replace window-based tracking in stereo matching with deep optical flow for efficient, improved retrieval. Fine-tuning balances a self-supervised geometric residual loss with supervised radiosonde reconstruction. To eliminate multi-satellite overlap requirements, we distill the stereo teacher into a single-satellite student model. Chi-square and height uncertainties from the teacher are emulated by the student for quality assurance. The student generates winds across full-disk GEO imagery globally. Validation compares stereo and student models against radiosondes, operational AMVs, ERA5 reanalysis, and EarthCARE cloud profiles. Results through triple collocation show that stereo winds improve performance beyond operational AMVs for water vapor bands (6.2, 6.9, and 7.3 μm), wit degradation in the long-wave infrared (11.2 μm) band.
ER -