Extended Abstract (2.0M)JP1.21
Using supervised learning for specific meteorological satellite applications
Richard L. Bankert, NRL, Monterey, CA; and J. D. Hawkins
Supervised learning techniques are applied to two meteorological satellite applications: cloud classification and tropical cyclone intensity estimation. Using a set of expertly labeled training data that consist of attribute records taken from the five spectral channels, a 1-nearest neighbor algorithm is employed to classify GOES-8 and GOES-10 data in terms of cloud type, clear sky, ground snow, etc. The cloud classification algorithm is currently running in real time with color-coded output images available for viewing on the web. Similarly, using a set of training data that consist of attribute records with corresponding best track intensity, a K-nearest neighbor algorithm is developed to estimate the intensity of tropical cylcones as seen in Special Sensor Microwave/Imager (SSM/I) data. Data from the 85 GHz (H-pol) channel and derived rain rate product are used to compute the attributes. Recent algorithm performance measures will be presented.
Joint Poster Session 1, Operational Applications and Artificial Intelligence (Joint between 12th Conference on Satellite Meteorology and Oceanography and Third Conference on Artificial Intelligence Applications to Environmental Science)
Monday, 10 February 2003, 2:30 PM-4:00 PM
Previous paper Next paper