Analysis of Polar Clouds from Satellite Imagery Using Pattern Recognition and a Statistical Cloud Analysis Scheme

View More View Less
  • 1 Department of Meteorology, University of Wisconsin, Madison, Wisconsin
© Get Permissions
Full access

Abstract

The analysis of cloud cover in the polar regions from satellite data is more difficult than at other latitudes because the visible and thermal contrasts between the cloud cover and the underlying surface are frequently quite small. Pattern recognition has proven to be a useful tool in detecting and identifying several cloud types over snow and ice. Here a pattern recognition algorithm in combined with a hybrid histogram-spatial coherence (HHSC) scheme to derive cloud classification and fractional coverage, surface and cloud visible albedos and infrared brightness temperatures from multispectral AVHRR satellite imagery. The accuracy of the cloud fraction estimates were between 0.05 and 0.26, based on the mean absolute difference between the automated and manual nephanalyses of nearly 1000 training samples. The HHSC demonstrated greater accuracy at estimating cloud friction than three different threshold. methods. An important result is that the prior classification of a sample may significantly improve the accuracy of the analysis of cloud fraction, albedos and brightness temperatures over that of an unclassified sample.

The algorithm is demonstrated for a set of AVHRR imagery from the summertime Arctic. The automated classification and analysis are in good agreement with manual interpretation of the satellite imagery and with surface observations.

Abstract

The analysis of cloud cover in the polar regions from satellite data is more difficult than at other latitudes because the visible and thermal contrasts between the cloud cover and the underlying surface are frequently quite small. Pattern recognition has proven to be a useful tool in detecting and identifying several cloud types over snow and ice. Here a pattern recognition algorithm in combined with a hybrid histogram-spatial coherence (HHSC) scheme to derive cloud classification and fractional coverage, surface and cloud visible albedos and infrared brightness temperatures from multispectral AVHRR satellite imagery. The accuracy of the cloud fraction estimates were between 0.05 and 0.26, based on the mean absolute difference between the automated and manual nephanalyses of nearly 1000 training samples. The HHSC demonstrated greater accuracy at estimating cloud friction than three different threshold. methods. An important result is that the prior classification of a sample may significantly improve the accuracy of the analysis of cloud fraction, albedos and brightness temperatures over that of an unclassified sample.

The algorithm is demonstrated for a set of AVHRR imagery from the summertime Arctic. The automated classification and analysis are in good agreement with manual interpretation of the satellite imagery and with surface observations.

Save