Rainfall Advection using Velocimetry by Multiresolution Viscous Alignment

TL;DR

This paper introduces a Bayesian-based multiresolution viscous alignment algorithm for estimating dense, consistent motion fields from satellite imagery, improving over traditional correlation and optical flow methods especially for large time gaps.

Contribution

The paper presents a novel Bayesian formulation for satellite image motion estimation that handles large displacements and small-scale deformations more effectively than existing methods.

Findings

Produces dense displacement fields with spatial consistency

Handles large time-step separations and rapid deformations

Outperforms correlation and optical flow methods in tests

Abstract

An algorithm to estimate motion from satellite imagery is presented. Dense displacement fields are computed from time-separated images of of significant convective activity using a Bayesian formulation of the motion estimation problem. Ordinarily this motion estimation problem is ill-posed; there are far too many degrees of freedom than necessary to represent the motion. Therefore, some form of regularization becomes necessary and by imposing smoothness and non-divergence as desirable properties of the estimated displacement vector field, excellent solutions are obtained. Our approach provides a marked improvement over other methods in conventional use. In contrast to correlation based approaches, the displacement fields produced by our method are dense, spatial consistency of the displacement vector field is implicit, and higher-order and small-scale deformations can be easily handled. In contrast with optic-flow algorithms, we can produce solutions at large separations of mesoscale features between large time-steps or where the deformation is rapidly evolving.