Having highly accurate and reliable Digital Elevation Models (DEMs) of the Earth’s surface is critical to orthorectify Landsat imagery. Without such accuracy, pixel locations reported in the data are difficult to assure as accurate, especially in more mountainous landscapes, where the orthorectification process is the most challenging. To this end, the Landsat Calibration and Validation Team (Cal/Val) compared the Copernicus DEM (CopDEM) to the DEM that is currently used in Collection-2 processing (called “Collection-2 DEM”). NGS ground-surveyed and lidar-based ICESat-2 points were used, and the CopDEM shows improvement to be less than 1 m globally, except in Asia where the accuracy and resolution of the DEM were greater for the CopDEM compared to the Collection-2 DEM. Along with slightly improved accuracy, the CopDEM showed more consistent results globally due to its virtually seamless source and consistent creation methods throughout the dataset. While CopDEM is virtually seamless, having greater than 99% of their data coming from a single source (Tandem-X), there are significantly more voids in the higher elevations which were mostly filled with SRTM derivatives. The accuracy of the CopDEM fill imagery was also compared to the Collection-2 DEM and the results were very similar, showing that the choice of fill imagery used by CopDEM was appropriate. A qualitative assessment using terrain-corrected products processed with different DEMs and viewing them as anaglyphs to evaluate the DEMs proved useful for assessing orbital path co-registration. While the superiority of the CopDEM was not shown to be definitive by the qualitative method for many of the regions assessed, the CopDEM showed a clear advantage in Northern Russia, where the Collection-2 DEM uses some of the oldest and least accurate datasets in the compilation of the Collection-2 DEM. This paper presents results from the comparison study, along with the justification for proceeding with using the Copernicus DEM in future Landsat processing. As of this writing, the Copernicus DEM is planned to be used in Collection-3 processing, which is anticipated to be released no earlier than 2025.

There is an ever-increasing need to use an accurate and consistent geometric ground reference in the processing of remotely sensed data products, as this reduces the burden on the end-users to account for the differences between the data products from different missions. In this regard, the U.S. Geological Survey (USGS) initiated an effort to harmonize the Landsat ground reference with the Sentinel-2 Global Reference Image (GRI) to improve the co-registration between the data products of the two global medium-resolution missions. In this paper, we discuss the process, results, and the improvements expected from this harmonization of two ground references using space-triangulation-based bundle adjustment techniques. The ground coordinates of the Landsat reference library, consisting of five million Ground Control Points (GCPs) were adjusted in a series of four simultaneous bundle block adjustments using thousands of Landsat-8 (L8) scenes anchored with more than 300,000 control points extracted from the GRI dataset. The net adjustments to each of the four blocks, namely, Australia, Americas, Eurasia, and Islands, varied anywhere from 1 to 13 m, depending on the accuracy of the GCPs in these blocks. The use of the GRI dataset in our bundle adjustment not only improved the absolute accuracy of the Landsat ground reference, but will also improve the co-registration between Sentinel-2 and Landsat terrain corrected products, as the European Space Agency plans to process the Sentinel-2 products using the GRI dataset. Independent validation of the Landsat products processed using harmonized GCPs with the GRI dataset indicated a global misregistration error of less than 8 m Circular Error Probable at 90 % (CE90), an improvement from the 25 m prior to harmonization. The improvements to the Landsat products using the harmonized GCPs are expected to be available to the public as part of Landsat Collection-2 processing by the end of 2020.

The Landsat Collection-2 distribution introduces a new global Digital Elevation Model (DEM) for scene orthorectification. The new global DEM is a composite of the latest and most accurate freely available DEM sources and will include reprocessed Shuttle Radar Topographic Mission (SRTM) data (called NASADEM), high-resolution stereo optical data (ArcticDEM), a new National Elevation Dataset (NED) and various publicly available national datasets including the Canadian Digital Elevation Model (CDEM) and DEMs for Sweden, Norway and Finland (SNF). The new DEM will be available world-wide with few exceptions. It is anticipated that the transition from the Collection-1 DEM at 3 arcsecond to the new DEM will be seamless because processing methods to maintain a seamless transition were employed, void filling techniques were used, where persistent gaps were found, and the pixel spacing is the same between the two collections. Improvements to the vertical accuracy were realized by differencing accuracies of other elevation datasets to the new DEM. The greatest improvement occurred where ArcticDEM data were used, where an improvement of 35 m was measured. By using theses improved vertical values in a line of sight algorithm, horizontal improvements were noted in some of the most mountainous regions over multiple 30-m Landsat pixels. This new DEM will be used to process all of the scenes from Landsat 1-8 in Collection-2 processing and will be made available to the public by the end of 2020.