Ground Resolved Distance Estimation of Sentinel-2 Imagery Using Edge-based Scene-Driven Approach

Abstract Sentinel‑2 satellite provides freely accessible multispectral images used in various remote sensing (RS) applications, where spatial resolution is crucial. The Ground Sampling Distance (GSD) for Sentinel’s visible and near-infrared (VNIR) bands is specified at 10 meters, but it may not accurately reflect ground resolution due to environmental effects. As a result, Ground Resolved Distance (GRD) serves as an alternative measure for actual resolution, but information about Sentinel GRD is lacking, calibration targets are not always available, and GRD may vary across different tiles. This paper estimates Sentinel’s GRD using a scene-driven approach that analyzes the edges of natural targets, reducing the challenges associated with artificial targets. The method involves selecting suitable natural targets based on their geometric and spectral characteristics, sub-pixel edge extraction, estimating the Edge Spread Function (ESF), generating the Line Spread Function (LSF), and calculating the Full-width at Half Maximum (FWHM). Two tiles of Sentinel‑2 imagery from the Shadnagar Calibration Facility, India, and Baotou, China, were analyzed. The analysis of 40 natural targets revealed average GRD values of 12.65 m, 12.40 m, 12.49 m, and 12.58 m for the red, green, blue, and NIR bands, respectively, aligning closely with results from calibration targets. The method demonstrated high accuracy and precision with a total RMSE of approximately 0.77 m and a total standard deviation of 0.19 m, respectively.