Deep Learning based Surface Water Mapping with SAR leveraging Google Earth Engine
Abstract
Throughout South and Southeast Asia local communities are continually impacted due toflooding events and water-related disasters. SERVIR, a NASA-USAID initiative, works inpartnership with regional hubs around the world building capacity through promoting the use ofEarth observation and geospatial service co-develop. The Hydrologic Remote Sensing Analysisfor Floods (HYDRAFloods) service is an open source Python application for downloading,processing, and delivering near-real time surface water maps derived from remote sensing data.Building upon this service, new approaches such as automatic data labeling techniques andsurface water modeling leveraging Deep Learning were investigated leveraging Google EarthEngine and the Google AI Platform. This talk focuses on new methodologies incorporated intothe HYDRAFloods system through a comparison of data labeling approaches, hyperparametertuning for a surface water U-Net convolutional neural network, and independent validation effortutilizing high resolution imagery. Through this comparison or new methodologies these resultscontinue to contribute to improving the rapid and automatic operational surface water mappingeffort, potentially increasing the impact to beneficiaries, end-users, and stakeholders duringhumanitarian assistance events.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMGC43D..04T