Operational Coregistration of the Sentinel-2A/B Image Archive Using Multitemporal Landsat Spectral Averages
Philippe Rufin, David Frantz, Lin Yan, Patrick Hostert
Abstract
Geometric misalignment between Landsat and Sentinel-2 data sets as well as multitemporal inconsistency of Sentinel-2A and -2B data sets currently complicate multitemporal analyses. Operational coregistration of Sentinel-2A and -2B imagery is thus required. We present a modification of the established Landsat Sentinel Registration (LSReg) algorithm. The modifications enabled LSReg to be included in an operational preprocessing workflow to automatically coregister large volumes of Sentinel-2 imagery with Landsat base images that represent multiannual monthly spectral average values. The modified LSReg was tested for the complete Sentinel-2 archive covering Crete, Greece, which is a particularly challenging region due to steep topographic gradients and high shares of water in Sentinel-2 tiles. A coregistration success rate of 87.5% of all images was obtained with a mean coregistration precision of 4.4 m. The mean shifts of 14.0 m in the x-direction and 13.4 m in the y-direction before coregistration were found, with maxima exceeding four pixels. Time series noise in locations with land cover transitions ( n = 585) was effectively reduced by 43% using the presented approach. The multitemporal geometric consistency of the Sentinel-2 data set was substantially improved, thus enabling time series analyses within the Sentinel-2 data record, as well as integrated Landsat and Sentinel-2A and -2B data sets. The modified algorithm is implemented in the Framework for Operational Radiometric Correction for Environmental monitoring (FORCE) version 3.0 (https://github.com/davidfrantz/force).