Incorporating Texture Features into Optical Flow for Atmospheric Wind Velocity Estimation

Wind velocity estimation is a critical component of weather prediction, but accurately estimating atmospheric winds with in-place instruments is expensive, inefficient, and provides only sparse coverage. Consequently, methods which can extract wind patterns from remote sensing data would offer significant benefits—satellites and satellite-derived data in particular, provide comprehensive coverage of global weather phenomena. In this paper, we develop an optical flow method which successfully extracts dense wind velocity estimates from water vapor data, a commonly estimated quantity by satellites equipped with infrared or microwave instruments. The test-data comes from direct numerical simulations of a mesoscale convective system over the eastern Pacific and provides accompanying ground truth wind velocities, allowing us to quantitatively measure the performance of our methods.