projects
Detail
Publication date: 1 de June, 2021Learning Spatio-Temporal Oceanographic Patterns (LSTOP)
This project is an approach to develop fuzzy clustering methods and their extension to dynamic versions, for the automatic identification, analysis and tracking of oceanic mesoscale phenomena in the Iberian Coastal Ocean. Knowledge aware neural networks will also be studied for classification and prediction of these phenomena by incorporating domain knowledge. Specifically, we are going to study the application of these methodologies to the presence of eddies, coastal upwelling and a coastal counter-current off Iberia, including their spatio-temporal evolution, from satellite remote sensing images.
Eddies are a fundamental feature of the ocean circulation as they provide a mean of mixing of ocean waters as a consequence of turbulent motions associated with them. Thus, several physical, chemical and biological parameters are likely to be affected by the dynamics of eddies. Coastal upwelling is a mass of deep, cold, and nutrient-rich seawater that rises close to the coast. The upwelling occurs when winds parallel to the coast induce a net mass transport of surface seawater in a 90º direction, away from the coast, due to the Coriolis force. Deep waters rise in order to compensate the mass deficiency that develops along the coastal area.Thus, eddies and upwelling are very energetic phenomena and their knowledge is important, for studying oceanic circulation, as well as related applications, which include fisheries exploitation, coastal ocean monitoring and oceanic pollution detection. The identification, tracking and prediction of these oceanographic phenomena are crucial in the studies of ocean dynamics, which involves the analysis of large data sets, as in the use of infrared imagery. Replacing the intensive time-consuming and subjective human-made interpretation by automatic analysis tools that allow the recognition, tracking, and prediction of these phenomena is a timely need.
Remote sensing images are inherently imprecise, not because of randomness in the data but because the observational scale of the phenomenon generally implies that its description bears some degree of imprecision. Therefore, the techniques of fuzzy set theory are particularly adequate in image segmentation for the recognition and characterization of heterogeneous patterns with boundaries not sharply defined, as is the case of eddies and upwelling patterns. Visualization of fuzzy membership maps honours the observation and interpretation of the fuzziness of the natural phenomena.
Artificial Neural Network (ANN) classifiers are one of the most robust classification systems used in image processing that provides good skills in dealing with noise and for prediction purposes.
We will prepare and analyse remote sensing and numerical simulations imagery. Knowledge of the oceanographic phenomena and their expression in the imagery will be summarized in the form of working definitions to be used by the fuzzy and neural network methods.
We will develop fuzzy clustering methods and extend them to dynamic versions. A dynamic fuzzy clustering is defined by time-dependent cluster prototypes and by degrees of membership of entities to the clusters. In the course of time these two measures will change in virtue of the dynamics of the phenomena being described. The application of these models to deal with oceanographic phenomena will be studied. An ANN classifier already developed for detecting eddies (RENA project: PDTCE/CTA/49945/2003) will be conjugated with logic programming for improving its performance and embedding domain knowledge. In order to accomplish these goals, it will be explored the simplicity of algorithms that allow the conversion of a propositional logic programs to one hidden layer feed-forward ANN. Dynamic fuzzy methods and knowledge aware neural networks will provide new tools for image understanding.
The validation of the proposed methods applied to remote sensing images will be done, whenever possible, using simulated Sea Surface Temperature (SST) data and surface velocity results from an ocean circulation model.
Sname | Eddy |
---|---|
Reference | PTDC/EIA/68183/2006 |
Funding Total | 160.000,00 Euros |
Results | Identifying mesoscale eddy-related structures from remote sensing (RS) images of the sea surface temperature (SST) off the Portugal coast is a complex task due to the Ocean dynamics of this region. Here, upwelling currents and bathymetry effects produce countless and highly heterogeneous SST patterns, features of interest may have smooth boundaries, as well as edges associated to strong temperature gradients may not correspond to any eddy. All this limits the effectiveness of an image processing based on edge-features (which can be successfully applied to automatically detect eddies in other oceanographic areas, for instance close to the Gulf Stream). The scope of the present report is documenting the work progress performed in 2008 (LSTOP project under Grant PTDC/EIA/68183/2006) to implement a new Model for identifying Eddy-Related Structure from Iso-SST pattern (MERSI). The novelty of the MERSI scheme is the exploitation of iso-SST patterns associated to the eddy-related structure to code with a rule-based definition the process that allows for their visual identification (knowledge-based approach). In practice, this enables revealing various morphological parameters of the eddy-related structure (i.e., the location, scale, symmetry and rotation). A software application called SEAEDDY has been build on top of the MERSI scheme to enable interactive functionalities. SEAEDDY provides access to reference information valuable to improve the exploitation of SST data allowing for annotating the RS image and benchmarking the subjectivity of the visual survey. |
URL | http://www.fct.mctes.pt/projectos/pub/2006/Painel_Result/vglobal_projecto.asp?idProjecto=68183&idElemConcurso=895 |
State | Concluded |
Startdate | 01/01/2008 |
Enddate | 01/12/2011 |