Extract Surface Water Features from Elevation Data
Contributed by the Center of Excellence for Geospatial Information Science (CEGIS) of USGS and CyberGIS Center for Advanced Digital and Spatial Studies (CyberGIS Center) at UIUC
Up-to-date topographic map data are critical for land planning and resource management, particularly in assessing risks from and responding to natural hazards such as severe weather, flooding, and landslides. National mapping databases typically provide only a single snapshot of surface water features (also known as hydrography) at the time of data collection. Traditional hydrography mapping methods involve flow-routing techniques on elevation models, followed by manual editing to align features with high-resolution imagery.
Recent advancements in AI, particularly deep learning, show promise in automating the extraction of hydrography from elevation data using existing datasets for training. These AI approaches can reduce manual tasks, resulting in cost-effective workflows that enable easy updates as changes occur to terrain and landscape. The objective of this challenge is to apply AI techniques to extract hydrography from elevation data based on existing training datasets while also providing uncertainty estimates for the extracted features. Participants should focus on quantifying uncertainty in relation to data quality and the precision of the applied methods.
See Data and Problem Description
Integrate Surface Water Data for The National Map
Contributed by the Center of Excellence for Geospatial Information Science (CEGIS) of USGS and CyberGIS Center for Advanced Digital and Spatial Studies (CyberGIS Center) at UIUC
The USGS National Map is vital for land management, research, environmental agencies, and hazards mitigation and recovery. To improve the usability of National Map data, we seek research focused on data integration and model development that leverages extensive open-data sources. Current hydrography data in the National Map includes waterbodies and a nationwide routable network, capturing a snapshot of surface water features at the time of collection. However, detailed temporal trends of surface water extent are not well represented.
We invite research to develop standardized methods for aligning high-resolution satellite images with vector hydrographic networks and waterbody features. Strategies should import and use coincident 1-meter multispectral images and digital line hydrographic data. Successful solutions should automate the alignment of these datasets, using techniques to adjust images to match line data.
Identifying network and waterbody features within the image data will enable time-series analysis of surface water trends and potentially aid in characterizing water conditions. Additionally, researchers should explore how to integrate these alignment strategies in deep-learning workflows currently being developed by the USGS for topographic mapping and hydrologic modeling. Finally, a critical component of this challenge is to measure the relative performance and error of various data alignment methods.
See Data and Problem Description