Sleeter, R., Carter, E., Jones, J.W., Eggleston, J., Kroeker, S., Ganuza , J., Dobbs, K., Coltin, B., McMichael, S., Shastry, A., Longhenry, R., Ellis, B., Jiang, Z., Phillips, J., and Furlong, P. M., 2020, Satellite-Derived Training Data for Automated Flood Detection in the Continental U.S.: U.S. Geological Survey data release, https://doi.org/10.5066/P9C7HYRV.
When captured at the appropriate time, satellite imagery can reveal the extent of flood inundation and consequent damage to human infrastructure. Emergency managers rely on flood extent maps derived from government and commercial satellites to aid first responders and to inform the allocation of disaster response and relief resources. However, no single federal agency currently generates derived data layers of flooded areas consistently across the U.S. from the range of available satellite platforms. Through funding from the National Geospatial-Intelligence Agency (NGA), the U.S. Geological Survey (USGS) is facilitating the development and integration of machine-learning (ML) algorithms developed at National Aeronautics and Space Administration (NASA) Ames Research Center (National Aeronautics and Space Administration 2020), in tandem with terrain and statistical modeling approaches developed by NGA, The University of Illinois, and The University of Alabama, to create a workflow that rapidly generates improved flood map products. Project goals include (1) embedding the resulting tools and procedures in USGS operational capabilities to automatically generate flood extent products that support hazard response and (2) building foundational open source tools that stakeholders can use for hydrologic monitoring and science-based assessments. This initial project will “train” ML algorithms to use inputs from at least three satellite sensors: Synthetic Aperture Radar (i.e., Sentinel-1) and two optical sensors (i.e., WorldView-2 and 3, and one other that has yet to be determined). This data release includes the geospatial training data developed for the Worldview-2 and 3 sensors that are used as input for the ML process.
