Automatic Cloud Classification from Multi-Spectral Satellite Data.
Abstract
This project was aimed at developing an operational 'expert' system to perform the classification of satellite images into cloud types. The approach we have used is based on a number of assumptions. The first one is that such a classification is possible with satellite images of 1 km (or more) resolution. A second assumption, which lays the foundations for all classifications, is that there exists a parameter space wherein some clustering of the data occurs, so the task is to identify this parameter space from the data. An additional assumption necessary to physically interpret the results, but not necessary for the classification itself, is that the clusters found in this parameter space can be related to cloud types or physical features. We chose a Bayesian classifier for our classification. We believe that this type of classifier is best suited for the task because clouds are fuzzy objects which have overlapping characteristics. Also, with a Bayesian classifier, each point in the parameter space has a probability to belong to each class, although this probability may be anywhere between zero and one.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 1991
- Accession Number
- ADA255125
Entities
People
- Catherine Gautier
- D. Lavallee
- D. Schweizer
- Peter A. Petérson
Organizations
- University of California, San Diego