Movement and deformation prediction method of nonstationary meteorological radar patterns based on a spatio-temporal texture element model

摘要

This paper discusses a Dynamics Texture Method that predicts futuristic movements and deformations of meteorological patterns from past time-sequential patterns. All shapes of patterns appeared in typhoons and intensive heavy rainfalls can be assumed to a pattern model blended from many small patterns with differencings. Thus this paper applies several spatio -temporal texture patterns so as to approximate a wide spectrum of patterns. Prediction is made by solving partial difference equations. Pattern velocities are estimated in two stages: roughly by optical flow and precisely by the SVD. Futuristic velocities are modified by the Navier-Stokes equation. Experimental results show that predicted vertical patterns with strong nonstationarity and nonlinearity become similarity with actual patterns in the leading time and that this method is useful for heavy rainfall disaster prevention.