Nandita Basu studies how human activity can impact water quality, specifically how nutrient runoff can impact large areas. Think of the Mississippi River basin or the Chesapeake Bay watershed. Much of the work Basu, a professor of water sustainability and ecohydrology at the University of Waterloo in Canada, does looks at nitrogen and phosphorus concentrations in streams and rivers and then links them to sources in the landscape, such as agricultural land use.
It’s work that necessarily depends on physical sampling of water in the field, but as Basu notes, researchers quickly find fundamental limits in this type of work.
“When you work with these water quality data, one thing that immediately becomes really evident is the lack of data. There are millions of streams, and there are only so many that we can go take samples from all the time,” she said.
By necessity, researchers who study water quality end up using models input with what information they have, “but often, those models are not really grounded in data, you can’t really trust them,” she said.
That’s why Basu is so excited about AquaSat, a new data set from researchers at Colorado State University, the University of North Carolina, and others that correlates water quality samples from U.S. rivers, streams, and lakes with more than 30 years of remote sensing images taken by Landsat satellites operated by NASA and the U.S. Geological Survey.
“The AquaSat data set is absolutely amazing,” she said. “I can imagine using it quite extensively.”
Remote Eyes on Water Quality
Matthew Ross, an assistant professor of ecosystem science and sustainability at Colorado State University, is the lead author on a 2019 paper in Water Resources Research detailing the AquaSat project and started his career taking water quality field samples. As a postdoctoral researcher in Tamlin Pavelsky’s lab at the University of North Carolina at Chapel Hill, however, Ross became interested in using satellites for larger-scale measurements. “I was sort of surprised that more people weren’t using remote estimates of water quality,” he said.
The eight Landsat satellites have provided continuous and global imaging of terrain since 1972, and although those missions have focused on land, Ross and his colleagues realized there should be “optically relevant” parameters in images of water too. “That’s things that should change the color of water,” he said. For AquaSat, they were interested in chlorophyll a, a measure of algae in water that turns it green; sediment, which can yield a tan color; dissolved carbon, which can darken waters and is a measure of carbon leached from the landscape; and Secchi disk depth, a measure of total water clarity.
Ross and his colleagues then correlated images taken by Landsat 5, 7, and 8 between 1984 and 2019 with on-the-ground samples of the imaged bodies of water that measured the optically relevant parameters. Researchers pulled sample data from the U.S. Water Quality Portal and the Lake Multi-scaled Geospatial and Temporal Database (LAGOS) data set, both of which record water quality measurements in U.S. streams, rivers, and lakes. The resulting 600,000 matchups of remote sensing and sample data allow for more reliable predictions of water quality based on future Landsat imaging alone.
“It gives you a ground truth. It’s basically a way to calibrate models that are using Landsat to estimate water quality parameters,” Ross said. “We can use these more data-rich, empirically driven ways of prediction that previously weren’t available because no data set like this existed before we made it.”
Applications and Accessibility
“With this data set we can look at all of these lakes and rivers and look at the water quality trajectories over time,” Basu said. For instance, researchers can track the water quality in a particular river over a 30-year period and correlate it with land use and farming practices in the surrounding landscape to estimate their impact. “Maybe,” she noted, “the farming practices have not changed that much, but maybe it’s climate that’s changing the conditions.”
Ross hopes to do more than just provide a new and useful data set for other water quality researchers. “Our goal is to make it a lot easier for anyone to use [the AquaSat data set] to build models that predict water quality,” he said.
He has already seen some evidence that is happening. The AquaSat data set has been shared openly on Figshare (an open-access repository where researchers can preserve and share figures, data sets, images, and videos), where it has attracted some amateur attention.
“I’ve gotten a bunch of high school and early college computer science folks emailing me about how to train neural nets on our data,” Ross said. “Those emails are always exciting because of the idea of there’s a bigger community that can engage with the data in an easier way.”
Right now, building models and making water quality predictions require some coding skills, but Ross said the ultimate goal is to create a user-friendly interface that could be used by water quality and environmental professionals to make decisions about water resources, such as reservoirs. “Getting these data and ideas into the hands of municipalities is certainly one of my long-term goals,” he said.
Beyond creating more user-friendly access to AquaSat going forward, Ross says he hopes to extend the data set with additional satellite imagery, such as the NASA Moderate Resolution Imaging Spectroradiometer (MODIS), satellites, and future missions.
“I’d say the biggest game changer for doing full stack hydrologic sciences from space is the SWOT mission, which is launching 2022,” he said. The Surface Water and Ocean Topography satellite will provide the height of large rivers and lakes. These data, according to Ross, could be combined with Landsat color information to allow researchers to do things like estimate the discharge and sediment volume in an ungauged river.
But the future projects Ross is most excited about involve getting enough on-the-ground data to validate satellite imagery in parts of the world that have little water quality data available to begin with. “In places that are changing rapidly, like in Honduras or Brazil, South Africa or other places, going back in time with Landsat satellites there is incredibly valuable,” he said. “To me, that’s one of the biggest value adds and why it’s so important to make this data set global, so we can validate a more global model.”
—Jon Kelvey (@jonkelvey), Science Writer