Demonstrations of New Dense Optical Flow Applications for Geostationary Satellite Imagery

The science of computing brightness motion in imagery pairs and sequences at every image pixel, or so-called "Dense Optical Flow" (DOF), has advanced considerably in the last four decades to support applications like objective robotic vision, autonomous driving, augmented reality, and motion picture special effects. DOF derivation is enabled in satellite imagery by the spatial and temporal resolution of new instruments like the Advanced Baseline Imager on the Geostationary Operational Environmental Satellite (GOES)-R series platform. Many of these new techniques have not yet been used for geostationary satellite imagery applications, such as feature tracking, wind (and flow-field) estimation (e.g. Atmospheric Motion Vectors; AMVs), imagery navigation correction, image interpolation, nowcasting, and stereoscopy. These DOF applications improve our ability to objectively understand and interpret imagery sequences and pairs and can be used to improve future decision-making tools and machine-learning artificial intelligence methods. This presentation will overview multiple new DOF derivation techniques now being explored at the Cooperative Institute for Research in the Atmosphere (CIRA) and how they differ from current and operational OF systems (e.g. patch-matching and AMVs). This presentation will also overview applications, validations, and lessons learned for adapting DOF algorithms and their assumptions to handle the unusual fluid-flows present in satellite imagery animations. Early results demonstrate that wind estimations from DOF algorithms present comparable accuracy to operational AMVs when compared to airborne wind-profiling lidar measurements, though accuracy depends heavily on the DOF technique used, and the wind height assignment techniques involved. Improvements and ongoing developments to DOF algorithms to better serve current applications will be discussed.