A. Gangopadhyay, Kevin Lydon, Jeffrey A. Rezendes, R. Balasubramanian, I. Valova
{"title":"Towards an automated detection of the Gulf Stream North Wall from concurrent satellite images","authors":"A. Gangopadhyay, Kevin Lydon, Jeffrey A. Rezendes, R. Balasubramanian, I. Valova","doi":"10.23919/OCEANS.2015.7404566","DOIUrl":null,"url":null,"abstract":"Developing computational methods to automatically identify the Gulf Stream North Wall (GSNW) and similar currents in the ocean is a long-standing need for many types of operational ocean models. Specifically, the Feature-Oriented regional modeling system requires an accurate digitization of the GSNW and Rings (eddies) on a regular basis. Typical methods to determine its position and boundaries require skilled human operators to do a time-consuming manual extraction of visualized features. These experts are performing a feature extraction task that can be automated to save time, guarantee objectivity, and potentially increase precision.In this paper we present first-results from two independent approaches of addressing this issue. In one of the approaches, the dynamical approach, the methodology begins by finding the most-likely bounds of iso-sea-surface-height contours within which the Gulf Stream north wall might fall. Other features, such as eddies, which are also captured, will be set aside after a round of shape analysis. Any gap in the isoheight contours is filled with segments that are generated by combining the slopes from different heights.The second, a machine-learning approach uses an artificial neural network over a GSNW dataset, which has been generated weekly over past six years (2009-2015) by analysts. An artificial neural network is a type of learning algorithm designed as a system of neurons with connections among them. This neural network will first use the analyst-designated GSNW paths to determine the neural weights of the radial basis functions. Then the network will use the concurrent sea-surface height and temperature data that were used to identify those lines, and train itself to develop a smart network which will be able to identify GSNW paths from the concurrent satellite images on its own, with little to no human intervention.In the long-term, we expect to merge the two techniques in a unique and unifying construct to be used operationally. A general approach of this methodology has the potential of being used for other similar operational modeling, reanalysis and skill assessment of numerical model system with data assimilation.","PeriodicalId":403976,"journal":{"name":"OCEANS 2015 - MTS/IEEE Washington","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2015-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"OCEANS 2015 - MTS/IEEE Washington","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.23919/OCEANS.2015.7404566","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 3
Abstract
Developing computational methods to automatically identify the Gulf Stream North Wall (GSNW) and similar currents in the ocean is a long-standing need for many types of operational ocean models. Specifically, the Feature-Oriented regional modeling system requires an accurate digitization of the GSNW and Rings (eddies) on a regular basis. Typical methods to determine its position and boundaries require skilled human operators to do a time-consuming manual extraction of visualized features. These experts are performing a feature extraction task that can be automated to save time, guarantee objectivity, and potentially increase precision.In this paper we present first-results from two independent approaches of addressing this issue. In one of the approaches, the dynamical approach, the methodology begins by finding the most-likely bounds of iso-sea-surface-height contours within which the Gulf Stream north wall might fall. Other features, such as eddies, which are also captured, will be set aside after a round of shape analysis. Any gap in the isoheight contours is filled with segments that are generated by combining the slopes from different heights.The second, a machine-learning approach uses an artificial neural network over a GSNW dataset, which has been generated weekly over past six years (2009-2015) by analysts. An artificial neural network is a type of learning algorithm designed as a system of neurons with connections among them. This neural network will first use the analyst-designated GSNW paths to determine the neural weights of the radial basis functions. Then the network will use the concurrent sea-surface height and temperature data that were used to identify those lines, and train itself to develop a smart network which will be able to identify GSNW paths from the concurrent satellite images on its own, with little to no human intervention.In the long-term, we expect to merge the two techniques in a unique and unifying construct to be used operationally. A general approach of this methodology has the potential of being used for other similar operational modeling, reanalysis and skill assessment of numerical model system with data assimilation.