Tests of Spectral Cloud Classification Using DMSP Fine Mode Satellite Data.
Abstract
A computer-based processor for satellite imagery was tested on samples of DMSP visible and IR imagery data smoothed to 0.6 n mi resolution. The data were displayed on the AFGL Man-computer Interactive Data Access System so that meteorologists could label small areas (25 x 25 n mi) with one of nine possible cloud categories from the AF 3D Nephanalysis Program (3DNEPH). The computer-based processor labeled the same areas by computing a two-dimensional fast Fourier transform (FFT) and comparing the results to average wavenumber spectra for the cloud categories. Classification accuracies were 65% for visible, 65% for IR and 81% for combined data. The classification accuracies were appreciably better than chance and a simplified processor which used only the averaged values of satellite data over an area. Accuracies improved is some categories were merged. The results were also compared to a cloud typing procedure in the 3DNEPH and to some earlier studies. The results were generally good for categories with small-scale cloud features such as cumulus or cirrus clouds, but the overall accuracy of classification for all cloud categories was not significantly better than verifications cited in earlier studies. Two potential refinements to the spectral processors, namely, removing the effects of backgrounds such as land, water, and snow cover and minimizing sensitivity to varying fractional cloud cover from case to case, are also discussed. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 02, 1980
- Accession Number
- ADA094119
Entities
People
- James T. Bunting
- Ronald F. Fournier
Organizations
- Air Force Research Laboratory