Multi-Channel satellite image analysis using a variational approach

Abstract

Currently, meteorological satellites provide multichannel image sequences including visible, temperature and water vapor channels. Based on a variational approach, we propose mathematical models to address some of the usual challenges in satellite image analysis such as: (i) the estimation and smoothing of the cloud structures by decoupling them into different layers depending on their altitudes, (ii) the estimation of the cloud structure motion by combining information from all the channels, and (iii) the 3D visualization of both the cloud structure and the estimated displacements. We include information of all the channels in a single variational motion estimation model. The associated Euler-Lagrange equations yield to a nonlinear system of partial differential equations that we solve numerically using finite-difference schemes. We illustrate the performance of the proposed models with numerical experiments on two multichannel satellite sequences of the North Atlantic, one of them from the Hurricane Vince. Based on a realistic synthetic ground truth motion, we show that our multichannel approach overcomes the single channel estimation for both the average Euclidean and angular errors.