Using Landsat 8 and 9 operational land imager (OLI) data to characterize geometric distortion and improve geometric correction of Landsat Multispectral Scanner (MSS) imagery

Abstract

The Landsat 1–5 Multispectral Scanner (MSS) acquired images in 1972–1992, but their usage is limited, particularly by low geometric accuracy. In the latest USGS Landsat Collection 2 processing, <1 % MSS archive could be geometrically processed to the highest-level Tier 1 category. We present a novel methodology to characterize and correct MSS geometric distortions using Landsat 8 and 9 OLI L1TP images as references, and using a new many-to-many matching strategy applied to MSS and OLI time series to provide large quantities of tie-points needed for effective MSS image geometric characterization and correction. The method is demonstrated at four U.S. sites (Landsat path/rows), including a site over Las Vegas that had significant surface change, and considering 182 MSS and 38 OLI Collection 2 images. Significant and spatially-variable MSS geometric distortions were found, with mean RMSE distortions of 43.2, 25.2, 24.0, 22.2, and 21.0 m for Landsat 1, 2, 3, 4, and 5 MSS L1TP images, respectively, and 327.0 m for Landsat 1 MSS L1GS images. After geometric correction, these mean RMSE values were reduced to 12.0, 12.0, 13.8, 12.0 and 11.4 m for the Landsat 1–5 MSS L1TP images, and 15.0 m for the Landsat 1 MSS L1GS images. The geometrically corrected MSS L1TP images acquired over each site were mutually well registered with site mean RMSE values from 9.6 to 20.4 m. The results indicate the methodology can be applied to the MSS archive in support of Landsat time series applications extending back to the 1970s, and to include a greater proportion of the MSS archive in the USGS Landsat analysis ready data (ARD) data suite that requires Tier 1 processing.