Automated computational delimitation of SST upwelling areas using fuzzy clustering

In our previous work we applied fuzzy clustering to the problem of identification of upwelling areas from Sea Surface Temperature (SST) images, and showed that the approach was promising. However, the approach required a user-supplied information for annotation of the upwelling area on the map in order to fine-tune parameters of the method.In this paper, we modify the method to apply it in a fully automated manner without any pre-specified expert knowledge. We describe a computational system, FuzzyUPWELL, that provides a framework needed for a totally unsupervised segmentation and delimitation of upwelling areas on SST images. The FuzzyUPWELL system integrates an unsupervised fuzzy clustering algorithm, a threshold procedure combining a set of features extracted from clusters to determine the upwelling fronts, a mechanism to delimitate the upwelling areas by fuzzy boundaries defined from measures of classification uncertainty, and a Graphical User Interface (GUI).The system has been successfully applied to a collection of 113 images obtained at the coastal ocean of Portugal during the upwelling seasons of 1998 and 1999. The collection covers much diverse upwelling situations. The system is shown to be robust to false positives when analysing its response on SST images without upwelling.

► Fuzzy segmentation of SST upwelling images by unsupervised fuzzy clustering.
► Automatic detection of upwelling frontlines by features extraction from clustering.
► Definition of upwelling fuzzy boundaries by measures of classification uncertainty.
► Automated system shows accurate results on analysing much diverse upwelling SST's.
► Robustness of the system to deal with noise and with false positives.