A Spatiotemporally Variable Model for Nowcasting Storm Motion Vectors using Remotely Sensed Raster Data

Abstract

This paper presents a tracking and forecasting model that performs both pattern tracking and forecasting of the motion fields of the rainfall patterns that are detected using raster-based remote sensors (weather satellites and radars). The technique uses a distributed version of the cross correlation idea on subsets of the subsequent images to determine the velocity field at any time step. The velocity field obtained from the subsets is spatially interpolated to the pixel level to determine a high resolution version of the velocity field. The forecasting part implements a novel idea of using an exponential filter with parameter updating to adaptively fit the temporal evolution of the velocity vectors at every pixel. The effectiveness of the model is illustrated using Meteosat Images. We were able to effectively track and forecast the velocity vectors of the cloud patterns at every pixel of the Meteosat extent. Our initial experience indicates that the developed model shall benefit many application domains.