Research On Mechanisms Of Odd-tide Occurrence And Statistical Models For Prediction And Warning

So many accidents have occurred in North Jiangsu shoal during recentyears, mainly in the domain of Nantong. It is very surprising that theprocesses of these accidents are similar. The tide would rise untimely andcrest so soon that many people and property were lost in the accidents. Thelocal called this odd tide. In order to solve this difficult problem, NationalBureau of Oceanography set up a professional group.As a member of this project, my team initiated a brandnew solution thestatistical method. The research results are listed as below.According to the high-resolution and low-resolution remote sensingimages of the region, the changes of topography could be observed. By thetransportation of rivers, terrigenous sediment also was moved to theseashore. As high-resolution remote sensing images says, human behaviorhas been extended into marine region, which also promoted the changes oftopography.Velocity-flux model was established to find the relationship between velocity and topograph. ArcGIS was used to get the topographic data. As thefigures show, the topograph in North Jiangsu shoal is very unique. It wouldbe much more shallow and narrow off the coast. Therefore, the velocitywould be much greater. MATLAB and EXCEL was also used to verify theconclusion.Three statistical model was set up: multivariate autoregressive model,artificial neural network model and non-linear chaotic dynamic model.As for multivariate autoregressive model, the optimal model ordershould be found, which means the time scale affecting current marinefactors. Then autoregressive coefficient could be found by maximumlikelihood method. The linear relationship among the different marinefactors could be found according to the autogressive coefficient. During thestudied period, wind would impose an effect on the sea level. Besides, thechange of sea level would affect the current velocity.Time-delayed network is established to find the non-linear relationshipamong different marine factors in different time points. The character ofneural network is, that all the functions and principles in each neuron aresimple, but combined with others, a super complex network was formed.Research results show that neural network was suitable for single inputvectors, rather than multivariable ones. Non-linear chaotic dynamic model could be used for the non-linearrelationship among all marine factores. In addition, lyapunov exponents andkolmogrov entropy could be found to analyze the predictable time range.During the studied period, velocity was decomposed in north-south andeast-west directions. In east-west direction, the fitting and prediction effectsare satisfactory and predictable time range is also long enough. While in theother direction, the results are not good enough, which indicatesphenomenon of chaos would be likely to occur in this direction.