Super resolution enhancement of Landsat imagery and detections of high-latitude lakes

• Main Author: Ethan D. Kyzivat
• Format: Other/Unknown Material
• Language: English
• Published: Zenodo 2022
• Subjects: Super resolution; SISR; Landsat; Planet labs; size distribution; SR; object detection; lake; ESRGAN; cubesat; SmallSat; upscaling; downscaling; geoAI; Old Crow; Yukon
• Online Access: https://doi.org/10.5281/zenodo.7306219

This archive contains native resolution and super resolution (SR) Landsat imagery, derivative lake shorelines, and previously-published lake shorelines derived airborne remote sensing, used here for comparison. Landsat images are from 1985 (Landsat 5) and 2017 (Landsat 8) and are cropped to study areas used in the corresponding paper and converted to 8-bit format. SR images were created using the model of Lezine et al (2021a, 2021b), which outputs imagery at 10x-finer resolution, and they have the same extent and bit depth as the native resolution scenes included. Reference shoreline datasets are from Kyzivat et al. (2019a and 2019b) for the year 2017 and Walter Anthony et al. (2021a, 2021b) for Fairbanks, AK, USA in 1985. All derived and comparison shoreline datasets are cropped to the same extent, filtered to a common minimum lake size (40 m 2 for 2017; 13 m 2 for 1985), and smoothed via 10 m morphological closing. The SR-derived lakes were determined to have F-1 scores of 0.75 (2017 data) and 0.60 (1985 data) as compared to reference lakes for lakes larger than 500 m2, and accuracy is worse for smaller lakes. More details are in the forthcoming accompanying publication. All raster imagesare in cloud-optimized geotiff (COG) format (.tif) with file naming shown in Table 1 . Vector shoreline datasets are in ESRI shapefile format (.shp, .dbf, etc.), andfile names usethe abbreviations LR for low resolution, SR for high resolution, and GT for “ground truth” comparison airborne-derived datasets. Landsat-5 and Landsat-8 images courtesy of the U.S. Geological Survey For an interactive map demo of these datasets via Google Earth Engine Apps, visit: https://ekyzivat.users.earthengine.app/view/super-resolution-demo Table 1 : File naming scheme based on region, with some regions requiring two-scene mosaics. Region Landsat ID Mosaic name Yukon Flats Basin LC08_L2SP_068014_20170708_20200903_02_T1 LC08_20170708_yflats_cog.tif “ LC08_L2SP_068013_20170708_20201015_02_T1 “ Old Crow Flats LC08_L2SP_067012_20170903_20200903_02_T1 ...