Automatic satellite-based flood mapping for disaster response

As flood events are becoming more frequent and more severe, the impacts of flooding on a growing and urbanizing domestic population have intensified. The last four decades have brought significant increases in flood related deaths, size of flood-affected population, and flood associated economic damages. To mitigate the societal impacts of flooding, access to reliable, real-time information on flood-affected areas has proven critical for coordinating efficient inter-agency disaster response. Multiple challenges have limited the adoption of satellite-based flood mapping in disaster response, including: lack of available imagery at exact time/location of flood event; computational limitations to storage/processing of large spatial data sets; and challenges associated with automatic classification of flooded pixels (i.e. lack of available ground-truthing for model training and limitations to quantifying classification uncertainty). This presentation will provide an overview to a joint USGS-NASA-NGA-Univ. Alabama initiative to develop a platform for automatic near-real time flood detection, using multiple sources of satellite imagery for use in disaster response. The platform will be based on an open-source library for supervised classification of satellite radar and optical imagery using deep neural networks currently under development at the NASA Ames and USGS Innovation Center at Moffett Field, CA.