Red Tide Prediction And Decision Services By Integrating Buoy And Remote Sensing Data

Red tide is one of the most frequently occurring marine disasters in China.To fully understand the mechanism,realize its timely and accurate early-warning,verify the correctness of prediction and finally to monitor its spatio-temporal variations are the primary concerns to minimize the harm and provide scientific decision support for marine monitoring departments.In this paper,we choose the highly complex and nonlinear Zhejiang coastal waters as our study region.In general,we aim to take advantage of Geographic Information Science,Data Mining techniques and Remote Sensing technology to explore the integration application of various data and to establish a comprehensive and business-oriented framework for the prediction and decision services of red tides.The main contents of this thesis are as follows:(1)Spatio-temporal analysis and association rules mining of long-term red tide in-situ data are conducted.Facing the fact that huge amount of monitoring data are underused,we apply SOM neural network technique and association rule mining algorithm on 15 years'big data to find out the key factors of red tides as well as their changing ranges when red tides burst out.The results provide scientific criteria for red tide prediction.(2)Early-warning and occurrence prediction of red tides considering time series similarity are carried out on real time buoy data.After analyzing historical Red Tide Warning Series(RWTS),we propose a segmented similarity measurement method based on feature points,which is then combined with the data mining results to form an automatic red tide prediction pattern.It realizes separate recognition of early-warning signals and occurrence signals of red tides,and also locates the most similar red tide event in history for reference.The application of the pattern on Dachen buoy and Nanji buoy obtains accuracies of 77.8%and 88.3%.(3)A new index called RrcH is presented to identify the algae in highly-turbid coastal waters and monitor its spatial-temporal variations using GOCI images.Remote sensing images is in fact the most economic and the only way to monitor the spatio-temporal variations of red tides.Therefore,we put forward a red tide index called RrcH to efficiently extract red tide from highly-turbid coastal waters.The RrcH result is efficient and highly matched with the buoy monitoring data(r=0.9410,p<0.01).We also conduct red tide area extraction on classic red tide events from 2012-2016,the results of which can play great parts in effective emergency response.(4)The red tide service chain proposed above is integrated into a prototype system for comprehensive analysis and decision support of marine ecological environment in Zhejiang Coastal waters.We design a "Various Data Management-Data Mining-Red Tide Prediction-Red Tide Spatio-Temporal Monitoring" service chain,which combine all the research results before and the operational application of the system in various-level monitoring departments verify the effectiveness of the methods proposed in this paper.