Characterization of the Landsat-7 ETM+ Automated Cloud-Cover Assessment (ACCA) Algorithm

 

Richard Irish, Science Systems & Applications, Inc.

John Barker, NASA Goddard Space Flight Center

Samuel Goward, University of Maryland College Park

Terry Arvidson, Lockheed Martin

 

Photogrammetric Engineering & Remote Sensing, vol. 72, no. 10, pp: 1179–1188.

In 1999, NASA launched the latest in a series of Landsat satellites designed to image the Earth’s land areas for mapping and research surveys at moderate spatial resolution of about the size of football fields. The Landsat-7 digital camera is a scanner called the Enhanced Thematic Mapper Plus (ETM+) and records eight separate images in the visual, infrared and thermal portions of the electromagnetic spectrum.  The goal of the current Landsat mission is to collect cloud-free imagery of the Earth for each season of the year to permit trending for changes.  Since the launch of the first satellite in 1972 the imagers on the Landsat satellites have provided a continuous record of changes over most of the land areas of the Earth.  This archive of images is the basis for creation and maintenance of land use and land cover change by satellite remote sensing as a scientific discipline.

Clouds interfere with the mission objective.  Nothing can be done about the actual distribution of clouds over the Earth, however, by measuring the cloud cover of acquired imagery mission planners know whether or not they have to take additional imagery.  The Automatic Cloud Cover Assessment (ACCA) software was created to accomplish this goal. The ACCA algorithm has been used to automatically calculate the amount of cloud cover for more than 400,000 180-by-180 Km “scenes” in the Landsat archive.  The objective of this paper was to assess the precision and accuracy of the ACCA algorithm. 

To validate ACCA, visual assessment of clouds from “Browse” imagery was used for comparison to the automated ACCA scores for a stratified sample of 212 ETM+ 2001 scenes. This comparison of independent cloud-cover estimators produced a 1:1 correlation between the ACCA scores and visual assessments with no bias to a standard error of ±5 %.  The largest misidentification errors were at high altitudes and at low solar illumination where snow was misclassified as clouds. The largest errors where clouds were missed occurred over water where undetected optically thin cirrus clouds were present.  After analyzing ACCA scores by latitude, seasonality and solar elevation angle it was determined the algorithm works equally well under all conditions.  The standard error of ±5% for the cloud cover of a scene was sufficiently precise for mission planning purposes, especially given the absence of any significant systematic error.  The cloud cover scores generated by the Landsat-7 ACCA algorithm have been reliably assisting mission controllers in scheduling and confirming cloud-free ETM+ acquisitions for its entire mission.