A dataset unveiled today more than doubles the documented stream miles in the Chesapeake Bay Watershed, elevating the total from approximately 100,000 to over 200,000 miles. The Hyper-Resolution Hydrography Data used to generate the new stream maps stems from a collaboration between the University of Maryland, Baltimore County (UMBC), the Environmental Protection Agency’s Chesapeake Bay Program (CBP), and the Chesapeake Conservancy (CC), including UMBC alumni at CBP and CC. 

The project lays a robust foundation for sustainable management of one of North America’s most critical ecosystems, which spans six states and supports millions of residents and iconic wildlife, such as blue crabs and migrating shorebirds. The new, high-resolution dataset offers the clearest picture yet of how water moves through both pristine landscapes and altered terrain throughout the watershed. 

The novel, AI-supported mapping method the research team used also dramatically reduces costs, time, and labor required for stream mapping, making it easy to update as additional data become available or apply in other watersheds to amplify its impact. 

“The landscape is shaped by running water. Stream networks are the primary conduit between the watershed and the Bay, and now we can characterize that connection in ways that we’ve never been able to before,” says Matthew Baker, UMBC professor of geography and environmental systems, and a lead on the mapping project. In addition to locating streams and tracing their flow paths with a high degree of precision, the mapping process also allowed the team to report estimates of each channel’s width and depth along its entire length. 

A resource for restoration

“When you spend a lot of time looking at hillshade relief maps, you begin to recognize the extent of human manipulation of terrain and how dramatically we have shaped how water flows across the landscape,” Baker adds. The new data will allow individuals and organizations to improve efforts to mitigate any harms from human disruption. 

Environmental groups and government agencies, including the CC and CBP, can use the data to prioritize restoration projects, like targeted streamside tree plantings that can mitigate excessive erosion—detected as unusually steep banks or deep channels relative to a stream’s width—and filter pollutants to improve water quality. Farmers and urban planners are likely to find it useful as well, to decrease the detrimental effects of agricultural runoff or wisely manage development to avoid flooding and minimize detrimental effects on wildlife habitat, for example.  

“These maps represent over six years of hard work, and I can’t wait to see what people do with this highly anticipated dataset,” says David Saavedra ’14, environmental science. Saavedra’s role as a senior geospatial technical lead at the Chesapeake Conservancy had him intimately involved with the project from brainstorming to implementation. 

The project has been personally rewarding for Saavedra, too. “To work alongside Dr. Matt Baker all these years has been a wonderful opportunity,” he says. “I continue to learn from him every day and am proud to consider him a colleague and mentor.”

What to leave in, what to leave out?

This project is the first to harness high-resolution LiDAR data and artificial intelligence for large-scale, automated stream mapping. LiDAR, a laser-based system deployed via aircraft, captured elevation data with centimeter-level accuracy, generating a three-dimensional portrait of the terrain. AI algorithms, leveraging resources at UMBC’s High-Performance Computing Facility (HPCF), then processed the data, employing computer-vision techniques to identify channels. 

The HPCF computers mapped the entire watershed in a mere two weeks—a feat that traditional methods might take years to accomplish. The results achieved 94 percent accuracy for streams represented in existing data, and between 67 and 82 percent accuracy for previously unmapped streams, as validated by Saavedra against two other datasets, aerial imagery and LiDAR-derived topographic maps.

“I led a painstaking process of manually evaluating over 7,000 stream reaches across the watershed to conduct a thorough accuracy assessment on this novel dataset,” Saavedra says. Now that the methodology has been demonstrated effective, that level of manual validation shouldn’t be necessary if the technique is applied elsewhere.  

The algorithm needed some tweaks along the way, however. Initially, it included channel-shaped features that made less sense to include on a stream map, like detention ponds, green swales, gutters, and crop furrows. That necessitated modifications to the algorithm to remove those features.  

“Part of the challenge in interpreting the terrain was to make distinctions between those features and more natural channels,” Baker says. “So in our model, we had to eliminate some features that were mapped initially. That was unexpected.”

Eye-opening opportunities

The resulting maps offer a tenfold boost in resolution, moving from a 1:24,000 map scale to a 1:2,400 map scale with each pixel representing one square meter. The new stream maps align with recently-developed land cover maps produced at the same resolution, which are being released at the same time. 

“I think when people begin using our hyper-resolution hydrography in conjunction with the one-meter land use data, it will be eye-opening to see just how connected the landscape is to our waterways,” Saavedra says. “There are so many opportunities to improve our region’s water quality, many of which may not have been readily apparent with previous data.” 

Labeeb Ahmed ’15, environmental science, has been involved in managing the data release through his role as a geographer in the Chesapeake Bay Program at the EPA. 

“The lack of consistent high-resolution hydrography data has always been a challenge, as it is critical for numerous outcomes outlined in the Chesapeake Bay Watershed Agreement, such as mapping forest buffers, non-tidal wetlands, species habitats for brook trout and black duck, and defining stream health,” he says. “This data release will enable novel and interesting research and scientific inquiries. I’m excited to see how other researchers and stakeholders will use this data in their conservation and restoration efforts.”