Wavelet Optical Flow for 2-Component Wind Fields from Aerosol Backscatter Lidar Data

This poster presents the latest progress on the recovery of two-dimensional, two-component wind fields derived from horizontally scanning aerosol backscatter lidar data by a new wavelet-based optical flow algorithm named "Typhoon". Optical flow is a family of computer-vision methods dedicated to motion estimation; its various implementations play critical roles in experimental fluid dynamics (i.e., particle image velocimetry, PIV). For PIV, the apparent displacement of small particles advected by the flow is used to infer the underlying fluid motion. For lidar backscatter images, the wind is derived from the observable motion of macroscopic aerosol features (e.g. organized changes in particle concentration that correspond to plumes or puffs) moving across the scan plane. While cross-correlation methods have been used to recover wind vectors from atmospheric aerosol lidar data in the past, these particular methods suffer from issues related to the required use of an interrogation window. We wish to resolve smaller scale turbulent velocity perturbations than the traditional cross-correlation appears to be able to achieve. Therefore, we have been developing and testing alternative optical flow methods. The modern methods do not involve an interrogation window and they provide a global numerical solution, resulting in a vector at every pixel for a pair of scans. Early attempts using optical-flow class methods on the CHATS dataset proved promising (Mayor et al., 2010). New work conducted in Chico, California, in 2013 features a more recent optical flow algorithm that harnesses the multiscale resolving power of wavelet decomposition. In addition the algorithm has been cast in a framework adapted to real-time wind estimation from PPI scans of aerosol backscatter, such as provided by the Raman-shifted Eye-safe Aerosol Lidar (REAL). Statistical analysis of the 2013 Chico dataset is ongoing at the time of this writing, yet early results obtained in real-time during the measurement campaign already suggest a better ability to capture small motion scales than the cross-correlation method. The figure below shows a streamlined vector flow field derived by Typhoon from one pair of consecutive REAL scans. The copper shades in the background correspond to backscatter intensity. The altitude of this horizontal cross-section is approximately 100 m AGL and the data were collected during a daytime convective boundary layer. The data reveal a microscale vortex circulation. Although the characteristic diameter of this vortex is approximately 300 m, the resulting vector flow field also shows the algorithm has the ability to resolve smaller changes in the flow on the order of 50 m. While this resolution may vary due to the presence or absence of aerosol features to track, this example does demonstrate the efficacy of the technique to resolve 2-component turbulent flow fields at temporal and spatial meteorological microscales.