Improved Co-Registration of Sentinel-2 and Landsat-8 Imagery for Earth Surface Motion Measurements

Abstract

Abstract:The constellation of Landsat-8 and Sentinel-2 optical satellites offers opportunities for a wide range of Earth Observation (EO) applications and scientific studies in Earth sciences mainly related to geohazards. The multi-temporal co-registration accuracy of images provided by both missions is, however, currently not fully satisfactory for change detection, time-series analysis and in particular Earth surface motion measurements. The objective of this work is the development, implementation and test of an automatic processing chain for correcting co-registration artefacts targeting accurate alignment of Sentinel-2 and Landsat-8 imagery for time series analysis. The method relies on dense sub-pixel offset measurements and robust statistics to correct for systematic offsets and striping artefacts. Experimental evaluation at sites with diverse environmental settings is conducted to evaluate the efficiency of the processing chain in comparison with previously proposed routines. The experimental evaluation suggests lower residual offsets than existing methods ranging between RMSE xy = 2.30 and 2.91 m remaining stable for longer time series. A first case study demonstrates the utility of the processor for the monitoring of continuously active landslides. A second case study demonstrates the use of the processor for measuring co-seismic surface displacements indicating an accuracy of 1/5 th of a pixel after corrections and 1/10th of a pixel after calibration with ground measurements. The implemented processing chain is available as an open source tool to support a better exploitation of the growing archives of Sentinel-2 and Landsat-8.

Exploitation of Sentinel-2 Time Series to Map Burned Areas at the National Level: A Case Study on the 2017 Italy Wildfires

“…Since S2 data acquired during the year 2017 belong to different processing baselines, exhibiting variable geolocation spatial accuracy, each pixel revealed a shaky multitemporal profile as a result of the weak spatial coherence [55]. For this reason, an image co-registration step, performed separately for each S2 granule, was included in the processing chain in order to identify the horizontal linear shift for each observation date and to co-register the multitemporal S2 observations.…”

“…The S2 weak geolocation spatial accuracy, determined in a recent study to be around 12 m [55], currently requires an additional preprocessing step in order to improve the consistency of multitemporal pixel profiles. The new S2 archive release from the forthcoming S2 reprocessing campaign, foreseeing a geometric refinement making use of a large set of global ground control points obtained via automatic matching, should overcome this limitation and allow for improved geolocation spatial accuracy.…”

The 2016 M7 Kumamoto, Japan, Earthquake Slip Field Derived From a Joint Inversion of Differential Lidar Topography, Optical Correlation, and InSAR Surface Displacements

“…First, image-parallel stripes are introduced by the sensor from a push-broom effect (e.g., Barnhart et al, 2014;Stumpf et al, 2018). First, image-parallel stripes are introduced by the sensor from a push-broom effect (e.g., Barnhart et al, 2014;Stumpf et al, 2018).…”

High-Resolution Spaceborne, Airborne and In Situ Landslide Kinematic Measurements of the Slumgullion Landslide in Southwest Colorado

“…Varying methods utilizing precise image registration and pixel offset processing techniques on both optical and SAR datasets have shown their worth with respect to monitoring surface motion (e.g., Landslides) [46][47][48][49][50]. Dense pixel offset techniques have other advantages over DInSAR processing methods in that offsets are derived in both the range (across track) and azimuth (along track) directions.…”