Supplementary data for: Comparison of optical flow derivation techniques for retrieving tropospheric winds from satellite image sequences

This study introduces a validation technique for quantitative comparison of algorithms which retrieve winds from passive detection of cloud- and water vapor-drift motions, also known as Atmospheric Motion Vectors (AMVs).  The technique leverages airborne wind-profiling lidar data collected in tandem with 1-min refresh rate geostationary satellite imagery.  AMVs derived with different approaches are used with accompanying numerical weather prediction model data to estimate the full profiles of lidar-sampled winds which enables ranking of feature tracking, quality control, and height-assignment accuracy and encourages meso-scale, multi-layer, multi-band wind retrieval solutions.  The technique is used to compare the performance of two brightness motion, or “optical flow,” retrieval algorithms used within AMVs, 1) Patch Matching (PM; used within operational AMVs) and 2) an advanced Variational Optical Flow (VOF) method enabled for most atmospheric motions by new-generation imagers.  The VOF AMVs produce more accurate wind retrievals than the PM method within the benchmark in all imager bands explored.  It is further shown that image regions with low texture and multi-layer-cloud scenes in visible and infrared bands are tracked significantly better with the VOF approach, implying VOF produces representative AMVs where PM typically breaks down.  It is also demonstrated that VOF AMVs have reduced accuracy where the brightness texture does not advect with the mean wind (e.g. gravity waves), where the image temporal noise exceeds the natural variability, and when the height-assignment is poor.  Finally, it is found that VOF AMVs have improved performance when using fine-temporal refresh rate imagery, such as 1-min versus 10-min data.

Methods
Details for data creation and processing can be found in Apke, J. M., Y. Noh, and K. Bedka, 2022: Comparison of Optical Flow Derivation Techniques for Retrieving Tropospheric Winds from Satellite Image Sequences, J. Atmos. Ocea. Tech., DOI: 10.1175/JTECH-D-22-0057.1

Table 3 contains root mean square vector differences used to benchmark Atmospheric Motion Vector retrieval algorithms. Netcdfs contain atmospheric wind profiles collected by the Doppler Aerosol Wind-Profiling (DAWN) lidar data over the Pacific Ocean during the Aeolus Calibration and Validation field experiment, along with wind profiles estimated by optical flow-based cloud-drift tracking algorithms derived from Geostationary Operational Environmental Satellite (GOES)-R series Advanced Baseline Imager (ABI) 0.64, 3.9, and 6.9 um data blended with Global Forecast System (GFS) numerical weather prediction derived winds.  In this case, the DAWN winds are inferred to be the true state of the atmospheric winds, and the optical flow algorithms are retrievals that can be done anywhere within the ABI field of view. Hence, lower root mean squared vector difference values imply an approach which better retrieves the truth.  Within the manuscript, three optical flow approaches are tested: a 5x5 least squares target tracking method called "Patch Matching," a dense (every image pixel) method based on penalty function minimization called "Variational Optical Flow," and a quality controlled Variational Optical Flow algorithm called "VOF-QC."  For more details on how each algorithm retrieves winds from cloud and water-vapor-drifts in ABI imagery, see the manuscript above.