Objective Method Research Of Tropical Cyclone Intensity Estimation Based On Satellite Cloud Image And Relevance Vector Machine

A Tropical Cyclone(TC)is one of the most influential natural hazards in the world.Therefore,the forecasting accuracy of intensity and track is crucial to reduce the impacts of this disaster.In recent years,remote sensing satellite cloud image gradually becomes the main means of monitoring TC.In this paper,geostationary satellite cloud image and Relevance Vector Machine(RVM)are used to establish TC intensity objective estimation model combined with the TC kernel scale and central latitude,compared with traditional Linear Regress(LR)model.Three main research works are completed as followed:(1)Construct eyed TC intensity objective estimation model based on geostationary satellite cloud image and RVM.Firstly,the infrared satellite cloud image is denoised by gaussian smoothing.Then,the TC eye area is segmented by a Partial Differential Equation(PDE)based on a Geodesic Active Contour(GAC)model.The brightness temperature gradient data of the eyewall are obtained.And the maximum of gradients and different mean gradient in different probability are calculated.In this paper,TC intensity objective estimation model based on RVM is established(single characterization factor and multiple characterization factors,respectively).Experimental results show that the intensity estimation error of RVM model is smaller than LR model both in the single factor model and multiple factors model.With the increase of characterization factor’s dimension,the errors of both models reduce.However,the errors of RVM model reduce more than LR model.(2)The TC center is taken as reference point,construct TC intensity objective estimation model based on fused satellite cloud image and RVM.Firstly,Laplacian pyramid fusion algorithm is used to fuse the infrared and water vapor data,a fused satellite cloud image is obtained.Then,the TC center is taken as reference point,and the deviation angle-gradient co-occurrence matrix is constructed.In this paper,the relationship between multiple statistical parameters of co-occurrence matrix,TC kernel scale,central latitude and TC intensity is studied,RVM is used to establish TC intensity objective estimation model.Experimental results show that the optimal radial kernel scale for intensity estimation is 200 km after comparing the intensity estimation error of every scale.The intensity estimation error of RVM model is smaller than LR model when the 9 best statistical parameters of co-occurrence matrix are used to construct model.The errors of both models reduce when the TC central latitude is combined into the model.RVM model performs better than LR model.(3)Every point is taken in turn as reference point,construct TC intensity objective estimation model based on fused satellite cloud image and RVM.Every point is taken in turn as reference point,the fused satellite cloud image is used to construct deviation angle-gradient co-occurrence matrix.Then,the minimum,median and mean of statistical parameter matrix of co-occurrence matrix are calculated.In this paper,the relationship between multiple statistical parameters(minimum,median and mean)of statistical parameter matrix,TC kernel scale,central latitude and TC intensity is studied,RVM is used to establish TC intensity objective estimation model.Experimental results show that the optimal radial kernel scale for intensity estimation is 200 km after comparing the intensity estimation error of every scale.The mean of statistical parameter matrix is more suitable for TC intensity estimation.When every point is taken in turn as reference point,the intensity estimation error of model is smaller than the model based on data of taking TC center as reference point.The intensity estimation error of RVM model is smaller than LR model when the 9 best statistical parameters of co-occurrence matrix are used to construct model.The errors of both models reduce when the TC central latitude is combined into the model.RVM model performs better than LR model.Throughout the three TC intensity objective estimation methods based on RVM,RVM model has better high-dimensional nonlinear processing ability,intensity estimation ability and better algorithmic stability,compared with LR model.TC intensity can be estimated by RVM model effectively.