1-meter resolution land cover mapping for U.S. cities

Keywords: very high resolution, land cover, urban
Abstract Type: Paper Abstract

Abstract

UrbanWatch is a 1-meter resolution, open-access land cover and land use (LCLU) dataset for major cities across the conterminous United States. UrbanWatch covers over 23 cities, and contains 9 LCLU classes, i.e., building, road, parking lot, tree canopy, grass/shrub, water, agriculture, barren, and others, with an overall accuracy of 91.52%. The database was produced by a novel Fine-resolution, Large-area Urban Thematic information Extraction (FLUTE) framework, which builds upon state-of-the-art semi-supervised learning and deep learning architectures and has been trained with a new benchmark dataset containing 52+ million labeled pixels to capture diverse LCLU types and spatial patterns. FLUTE addresses several challenges that frequently occur in large-area, high-resolution urban mapping, including the view-angle effect, high intraclass and low interclass variation, and multiscale land cover. The dataset is available at: https://urbanwatch.charlotte.edu.