TY - JOUR
T1 - UrbanWatch
T2 - A 1-meter resolution land cover and land use database for 22 major cities in the United States
AU - Zhang, Yindan
AU - Chen, Gang
AU - Myint, Soe W.
AU - Zhou, Yuyu
AU - Hay, Geoffrey J.
AU - Vukomanovic, Jelena
AU - Meentemeyer, Ross K.
N1 - Funding Information:
This work was supported by the University of North Carolina at Charlotte . The authors are grateful to the editors and four anonymous reviewers for their constructive comments, which greatly helped to improve this paper.
Publisher Copyright:
© 2022 Elsevier Inc.
PY - 2022/9/1
Y1 - 2022/9/1
N2 - Very-high-resolution (VHR) land cover and land use (LCLU) is an essential baseline data for understanding fine-scale interactions between humans and the heterogeneous landscapes of urban environments. In this study, we developed a Fine-resolution, Large-area Urban Thematic information Extraction (FLUTE) framework to address multiple challenges facing large-area, high-resolution urban mapping, including the view angle effect, high intraclass and low interclass variation, and multiscale land cover types. FLUTE builds upon a teacher-student deep learning architecture, and includes two new feature extraction modules – Scale-aware Parsing Module (SPM) and View-aware Embedding Module (VEM). Our model was trained with a new benchmark database containing 52.43 million labeled pixels (from 2014 to 2017 NAIP airborne Imagery) to capture diverse LCLU types and spatial patterns. We assessed the credibility of FLUTE by producing a 1-meter resolution database named UrbanWatch for 22 major cities across the conterminous United States. UrbanWatch contains nine LCLU classes – building, road, parking lot, tree canopy, grass/shrub, water, agriculture, barren, and others, with an overall accuracy of 91.52%. We have further made UrbanWatch freely accessible to support urban-related research, urban planning and management, and community outreach efforts: https://urbanwatch.charlotte.edu.
AB - Very-high-resolution (VHR) land cover and land use (LCLU) is an essential baseline data for understanding fine-scale interactions between humans and the heterogeneous landscapes of urban environments. In this study, we developed a Fine-resolution, Large-area Urban Thematic information Extraction (FLUTE) framework to address multiple challenges facing large-area, high-resolution urban mapping, including the view angle effect, high intraclass and low interclass variation, and multiscale land cover types. FLUTE builds upon a teacher-student deep learning architecture, and includes two new feature extraction modules – Scale-aware Parsing Module (SPM) and View-aware Embedding Module (VEM). Our model was trained with a new benchmark database containing 52.43 million labeled pixels (from 2014 to 2017 NAIP airborne Imagery) to capture diverse LCLU types and spatial patterns. We assessed the credibility of FLUTE by producing a 1-meter resolution database named UrbanWatch for 22 major cities across the conterminous United States. UrbanWatch contains nine LCLU classes – building, road, parking lot, tree canopy, grass/shrub, water, agriculture, barren, and others, with an overall accuracy of 91.52%. We have further made UrbanWatch freely accessible to support urban-related research, urban planning and management, and community outreach efforts: https://urbanwatch.charlotte.edu.
KW - Deep learning
KW - Land cover and land use (LCLU)
KW - Open access
KW - Urban
KW - UrbanWatch
KW - Very high resolution (VHR)
UR - http://www.scopus.com/inward/record.url?scp=85131365830&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85131365830&partnerID=8YFLogxK
U2 - 10.1016/j.rse.2022.113106
DO - 10.1016/j.rse.2022.113106
M3 - Article
AN - SCOPUS:85131365830
SN - 0034-4257
VL - 278
JO - Remote Sensing of Environment
JF - Remote Sensing of Environment
M1 - 113106
ER -