Atmospheric Motion Vector Retrieval Using the Total Variation-Based Optical Flow Method
Igor Yanovsky, Derek J. Posselt, Longtao Wu, Svetla M. Hristova‐Veleva, Hai Nguyen, Bjorn Lambrigtsen, Xubin Zeng · 2023
Atmospheric motion vector (AMV) retrieval from water vapor measurements is important in climate research and weather forecasting. However, conventional feature tracking methods for AMV retrievals generate velocity fields with gaps and large errors. In this work, we test the optical flow algorithm by generating a nature run of a convective weather phenomenon, which provides water vapor variables and wind vector fields at various pressure levels. We show that our optical flow algorithm generates superior performance when compared with traditional feature tracking algorithms used in operational centers, generating dense AMVs with no gaps and significantly improving AMV accuracy. The optical flow algorithm performs well down to very low wind speeds and does not require a low-wind cutoff threshold. In our studies, we considered various measurement configurations, including water vapor retrievals at different temporal resolutions and found that the optical flow algorithm is not sensitive to the time interval between images.