Cloud detection is critical step for improving our scientific understanding of ongoing natural and human-induced global climate change. However, clouds above snow- and ice-covered surfaces over polar regions are especially difficult to detect because their temperature and reflectivity are similar to that of the surface. In this talk, we provide efficient algorithms to detect polar clouds using data from NASA's Earth Observing System. We first developed a fast algorithm to distinguish cloud from snow and ice, using the data provided by the Multi-angle Imaging SpectroRadiometer (MISR). We also proposed another method fusing the data from MISR and the Moderate Resolution Imaging Spectroradiometer (MODIS) to improve the polar cloud detection for both sensors.
Motivated by the satellite data fusion problem, we studied the machine learning algorithms using a Gaussian kernel. We showed that the leading eigen-vectors of a Gaussian kernel represent the clusters of the data. Based on this relationship, we provided a unified view of the Gaussian kernel related machine learning algorithms including Semi-supervised learning algorithms, Kernel Principle Component Analysis, Spectral Clustering, and Support Vector Machines.
Refreshments will be served at 3:45 p.m. in Room 008 of the Statistics Building