Unprecedented Data On Global River Quality, Quantity Now Gathered From Space, Powered By UMass Amherst-Built Software

The river data can help with drought and flood prediction, water resource management for drinking water and irrigation, infrastructure planning, and environmental monitoring

Software created at the University of Massachusetts Amherst is turning data from international satellite missions into usable information on both the volume of and quality of water flowing through all of Earth’s rivers wider than 50 meters. The open-source framework, Confluence, is the first platform to estimate flow volume together with river sediment on one platform that is freely available to a global community of water resource managers, planners, policy makers, climate scientists and hydrologists.

“From a societal standpoint, it’s important to know how much water’s in the river, but really, total river knowledge comes from knowing both its quantity and quality,” says Colin Gleason, Armstrong Professor of civil and environmental engineering at the UMass Amherst Riccio College of Engineering and the primary investigator leading the software’s creation for NASA.

“We don’t have all the water quality pieces we might want in Confluence, but it is unique in that it’s observationally driven software that is producing both estimates of river quantity and quality at the same time globally—and that’s never been done,” added Gleason, a hydrologist who led the international effort to incorporate various river discharge algorithms into the framework.

Confluence marks the first time hydrologists can create timely models of river flow and water quality at a global scale, leveraging data from three satellites: Surface Water and Ocean Topography (SWOT), a $1.2B satellite mission launched by NASA and France’s Centre National d'Études Spatiales (CNES) to calculate river discharge, or how much water flows through a particular point in a river at a specific time, as well as LANDSAT and Sentinel-2, which provide data on suspended sediment.

We don’t have all the water quality pieces we might want in Confluence, but it is unique in that it’s observationally driven software that is producing both estimates of river quantity and quality at the same time globally—and that’s never been done. - Colin Gleason, Armstrong Professor of civil and environmental engineering at the UMass Amherst Riccio College of Engineering

Producing data on river discharge is a top mission requirement and NASA recently announced meeting this milestone via the successful inaugural run of Confluence. River discharge data can be used for managing water resources for drinking and agriculture; planning infrastructure, like dams; anticipating flood and drought risks; and modeling water patterns as they relate to climate change.

Gleason has studied changes in rivers, both globally and in areas heavily dependent on rivers for agriculture and hydropower (namely, high-mountain Asia), as well as the regulatory implications of what it means to have enough flow to be categorized as a river. Now that NASA has adopted and runs Confluence with regularity, “everyone with a computer can access these river data and understand all global rivers regardless of access to ground-based data: we’re all starting from 2nd base,” says Gleason. “Without Confluence, you need a staggering amount of specialized knowledge and computing power to even step in the batter’s box.”

Confluence also uses AI and computer vision to “read” images from LANDSAT and Sentinel-2 to estimate how much suspended sediment is flowing in the river. River sediment influences the efficiency of dams and shapes coastal erosion.

Importantly, Confluence is the first platform to pull river discharge and sediment data together automatically. This feature was developed in collaboration with Subhransu Maji, a professor in UMass Amherst’s Manning College of Information and Computer Science, and his Ph.D. student, Rangel Daroya, working alongside hydrologists from the University of Pittsburgh.

Previous river sediment models have relied on external data inputs, such as elevation maps, and were heavily influenced by factors like cloud cover and surrounding terrain. “The current algorithm doesn’t need any of that,” says Maji. Instead, their computer vision algorithm was trained to detect rivers and discern between impeding factors such as clouds, snow and terrain shadow, simply by analyzing the image. “This allows us to isolate the pixels where sediment concentration can be reliably estimated. The approach removes significant computational and data bottlenecks and produces a unified, higher-fidelity product than previous methods.”

“These are independent observations that we make from space,” says Gleason. “Confluence is unencumbered by what you think the river should be. It’s simply direct measurements of the river inverted to discharge and sediment, so it’s true to the actual physical state of the planet.”

Daroya developed the algorithm to speed up the image processing that made sediment estimation possible. “With the standard pipelines, it would take about 20 seconds per image, which doesn’t sound a lot, but if you scale it up to a global estimation pipeline of 400,000 images a day, that would take 90 days to process the whole planet—and that’s just one pass,” she says. Her algorithm shrank the process down to around 0.8 seconds per image.

“Being able to work on this kind of problem that is not just computer science, but actually working together with a real-world problem that has impacts in society is actually quite important,” says Daroya. "There is so much more coming along, and with these new tools and advanced satellites, we are seeing the future of interdisciplinary science unfold."