Elements: A GeoAI Toolset for Identifying Locations and Drivers of Pond Collapse in Salt Marshes from High-Resolution Earth Imagery

Salt marshes provide important coastal services,for example, habitats for fishery species, buffers against damage due to rising sea levels and storm surges, and pools of soil organic carbon. Restoring coastal marshes has been recently listed as a “fundamental pillar of fighting the climate crisis” by the White House Nature-based Solutions Roadmap.

One significant threat to coastal salt marshes is the collapse of ponds, which can have cascading effects on the overall health and resilience of the marsh. The detection of pond collapse is essential for effective conservation and management. Coastal research communities rely on a rapidly growing amount of high-resolution remote sensing imagery from drones, aerial planes, and satellites.

Due to the sheer data volume, scientists urgently need automatic tools to analyze the Earth imagery data. This project aims to develop an integrated GeoAI toolset to train deep learning models for marsh feature mapping (channel networks and ponds). The tool also integrates spatial topological analysis and temporal change analysis of mapped marsh features.

Results produced from the tool will be applied to geomorphology research through interdisciplinary collaboration. Educational activities include curriculum development, mentoring a broad group of high school students in data science seminars, as well as a year-long project for a selected number of high school students for the regional Science Fair competition.

The main goal of the project is to build GeoAI toolset to generate a high-resolution (0.5m to 1m) marsh feature database (channel networks and ponds) for the most vulnerable estuaries across the U.S. coast. First, such a high-resolution marsh feature database enables the detection of locations and drivers of pond collapse. Second, it enables new geomorphology research opportunities by capturing the ability of small-scale (few meters) tidal networks to drain ponds, feed sediments, and restore vegetation.

Third, the GeoAI toolset provides a weakly supervised learning framework to automatically train deep learning models from imperfect vector labels. Finally, the open-source tool will be coded in a general and transferrable fashion so that images from other estuarine regions can be used as input for the tool, enabling the same analysis to be performed.

This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Geosciences Directorate’s Division of Research, Innovation, Synergies, and Education and Division of Earth Sciences.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.