Characterising and predicting the movement of clouds using fractional-order optical flow

Estimating cloud motion with complex background through the sequence of satellite images plays an important role in weather forecasting. This motion can be used for characterization of clouds and predicting storms. Optical flow is used here for motion estimation which gives horizontal and vertical velocity components. Velocity field vector alone is not sufficient to analyze the cloud behavior for predicting extreme weather conditions and there is a need to develop some visual features for enhancing the weather prediction strategies. In this paper, we utilize the optical flow to localize the high alert regions. To have a better localization and motion signatures, we develop a fractional order technique to compute optical flow. Also, the localization is characterized by brightness of image, magnitude, directions, vorticity and irrotational components of the optical flow. We did analysis on sequence of images for Mumbai, India heavy rain that happened during August 28-29, 2017, cyclonic data sets for May 16, 2018, September 19-20, 2018 and October 14, 2018. Visual features show different patterns for extreme and normal weather situations. A study on interpolation and extrapolation of the image sequence is also presented. Optical flow based interpolation and Advection-anisotropic-diffusion based extrapolation model give promising results.