Full metadata record
|Department:||Department of Computing||en_US|
|Title:||Adaptive methods for tropical cyclone prediction from time-series satellite data||en_US|
|Abstract:||A tropical cyclone is the generic term for a non-frontal synoptic scale low-pressure system over tropical or sub-tropical waters with organized convection and definite cyclonic surface wind circulation. Associated with the potential destructive winds and heavy rains, they primarily pose a threat to life and property to coastal zones. Their early identification would allow precautions to be put in place that would minimize the associated risks to people's lives and properties. In recent years, more and more research efforts have been involved in tropical cyclone forecasting, and lots of prediction models have been proposed to simulate the intensities of a variety of tropical cyclones, including statistical models, statistical-dynamical models and dynamic models. However, each approach has its own methodologies, assumptions and design goals, making it difficult to adapt to different application requirements. In this thesis, we aim at discovering new approaches in designing reformative forecasting models to predict tropical cyclones using methodologies at different levels that are more reliable for the collected data types, more efficient and effective in terms of predicted accuracy and computational costs, and more scalable to future system expansion. We propose a satellite interpretation based forecasting model, a neural network regression-forecasting model and a similarity retrieval model, each of which has its own background features. In the satellite interpretation-forecasting model, we introduce an integrated approach for tropical cyclone comparison based on typical spiral shapes using time warping technology. The position and the shape of a tropical cyclone extracted from a satellite image is the major concern. The Gradient Vector Flow (GVF) snake model is used to extract the contour points of a dominant tropical cyclone from the satellite image. The similarity of two tropical cyclones is compared using the angle features found among the successive contour points. Furthermore to achieve a better reflection of the spiral shape of tropical cyclones, we adopt a time warping approach to allow fast and accurate comparison of patterns. In the neural network regression forecasting model, we propose an integrated competitive neural network classifier to predict the maximum potential intensity of a tropical cyclone, based on a 10-year sample of western North Pacific tropical cyclones and monthly mean sea surface temperature. A large amount of feature variables are used in the network training including latitude, longitude, pressure, intensity, sea surface temperatures and so on. To deal with variety of variables, we design a variable selection procedure to choose the most important training variables to enhance the speed and accuracy of neural network training. In the similarity retrieval model, we present an approach to predict TC intensities using a feed-forward neural weight generator, which is adopted to generate a set of appropriate weights for various associated features of a tropical cyclone. We also propose the time-series similarity adjustment to measure the similarity of samples on consecutive observations of a tropical cyclone. Comparing with existing similar forecasting models, the experiments show that our proposed ones can achieve promising results.||en_US|
|Pages:||x, 104 p. : ill. ; 30 cm.||en_US|
|Subject:||Hong Kong Polytechnic University -- Dissertations.||en_US|
|Subject:||Cyclones -- Tropics -- Forecasting.||en_US|
|Subject:||Time-series analysis -- Mathematical models.||en_US|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: