Submitted to: IEEE Transactions on Geoscience and Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/31/2002
Publication Date: N/A
Citation: Interpretive Summary: We tested a new atmospheric correction algorithm for Landsat ETM+ satellite imagery by comparing surface-based reflectance measurements with reflectances calculated from the satellite imagery. The surface reflectance measurements of landcover types within the Beltsville Agricultural Research Center were made coincident with a satellite overpass using a hand-held spectroradiometer. Results showed that the new algorithm retrieved surface reflectances from the satellite imagery accurately. Further analysis showed how accurate satellite reflectances improves the accuracy of land cover classification, land cover change detection, and broadband albedo calculation. These results demonstrate the usefulness of the new algorithm and the importance of accurate satellite data corrections before many types of uses.
Technical Abstract: Accurate retrieval of surface reflectance from satellite imagery requires a correction for atmospheric effects. A new atmospheric correction algorithm was previously developed that uses histogram matching of land cover types under clear and hazy conditions. This approach corrects for both atmospheric water vapor and aerosols, and assumes fairly homogeneous distribution of these elements over the scene. Validation of the algorithm was conducted by comparing surface reflectances calculated from Landsat ETM+ satellite imagery with surface reflectances measured with a hand-held spectroradiometer. The hand-held measurements were made of land cover types on the Beltsville Agricultural Research Center coincident with a Landsat satellite overpass. Comparison of the surface-based reflectances with the satellite reflectances showed excellent agreement. Analysis of the effects of more accurate satellite reflectance retrieval on land cover classification, NDVI change detection, and nadir view-based broadband albedo mapping demonstrated significant improvements. Land cover classification was checked with site visits. The NDVI change detection used 1999 and 2000 image dates and a logistic regression model to calculate the probability of change. The results showed that atmospherically corrected imagery had a greater probability of detecting NDVI changes. Differences between broadband albedo calculated from corrected and uncorrected imagery were significant, thus demonstrating that neglecting bidirectional effects, calculations using corrected values are closer to expected values. This study is of interest to users of satellite imagery for all types of applications.