Urban Flood Dataset: A globally representative satellite-based labeled dataset of flooding in urban settings
Abstract
Development and rapid urbanization exacerbates the occurrence of urban inundation resulting in an increased exposure of people and infrastructure to flood hazards. For the remote sensing community, documentation of flood extent and duration is a persistent challenge given the ephemerality and spatial complexity of urban inundation. Yet, advancements in the spatial and temporal resolution of satellite imagery, data fusion, and deep learning techniques offer opportunities to improve urban flood detection. These state of the art methods are data intensive, requiring abundant and high-quality labels to train algorithms and externally validate pre-trained flood detection models. To meet this demand, we present a novel dataset named Urban Flood Dataset, a globally representative dataset of urban flood events labeled on 3-meter PlanetScope imagery from 2017-2022. Label classes include high and low confidence flooding and no flooding. A total of 20 unique events were selected to represent globally diverse geographies, urban built-up form, and variable flood types including pluvial, fluvial, storm surge, and dam failure. For locations with repeat flood events, labels are provided for multiple events. In addition to the labels, co-occuring public satellite imagery from Landsat 8, Sentinel-1 and Sentinel-2 are included if acquired within 2 days after the date of the labeled image. For each label, supplementary data layers of permanent surface water and percent impervious surface are also provided. While other satellite-based labeled flood datasets are publicly available for model training and validation, the urban focus of the Urban Flood Dataset provides opportunities for others to contribute to the advancements in satellite image-based modeling of flood detection in urban settings. This dataset adds to the small but growing suite of labeled datasets that can be leveraged to train machine learning and deep learning algorithms to detect surface water inundation inclusive of urban environments.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H46D..03F