To manage risks, health authorities regularly inspect waters and issue warnings and beach closures when necessary. The National Oceanic and Atmospheric Administration (NOAA) works with states and other local partners to issue harmful algae bloom forecasts, as well as weather forecasts, during harmful algae blooms.
On-site testing requires hours of boat rides to manually collect water samples and send them to a lab for analysis, taking more than a day and requiring multiple tests. It’s even harder to know where to test before a bloom starts to spread.
NASA’s Earth-orbiting satellites are already tracking harmful algae blooms with a unique global view. By integrating diverse datasets, new AI tools can serve as powerful tools to help communities decide where to focus their efforts.
“At the very least, tools like this can help us know when and where to collect water samples when an algal bloom outbreak is starting,” said Michelle Gierach, a scientist at NASA’s Jet Propulsion Laboratory in Southern California and one of the paper’s co-authors. “It can also foster collaboration between experts and encourage new ways to conduct science and deliver decision support products.”
Currently, satellites can detect a variety of cues that signal algal blooms. For example, hyperspectral sensors aboard NASA’s Plankton, Aerosols, Clouds, and Ocean Ecosystems (PACE) satellite can identify algal communities by size, shape, and pigment. Other instruments, like TROPOMI (Troposphere Monitoring Instrument), capture the faint red light emitted by species such as K. brevis as they photosynthesize.
The research team, consisting of Gierach, Kelly Luis of NASA JPL, and research data scientist Nick LaHaye of the Spatial Informatics Group, compiled findings from five space missions and instruments, including PACE and TROPOMI.
The challenge for them was the amount of raw data involved. How will AI differentiate between the deep ocean and the coastline? Will it be able to recognize blooms across different data streams? Will it be able to process input from both satellites and underwater sensors?
The team developed a self-supervised machine learning system designed to learn patterns from multiple types of satellite data and compare them with field observations. This approach allows AI to recognize relationships between different data sources without the need for prior labeling.
The system was trained on satellite data collected in 2018 and 2019. Field and laboratory measurements were then used to add real-world context to the patterns the system was recognizing. The scientists evaluated the tool’s performance over subsequent time periods in the same geographic region. Initial results show that it can accurately identify and map harmful blooms containing certain species, such as K. brevis, and that it works well in complex coastal waters with swirling sediments, vegetation, and runoff.
“Applying self-monitoring AI to large streams of satellite data is becoming a powerful tool for generating actionable ocean intelligence,” said Nadia Vinogradova Schiffer, chief program scientist at NASA Headquarters in Washington.
The team now aims to refine the tool with more data from more coastlines, expand testing to other types of water bodies, including lakes, and make the tool available to decision makers in the next few years.
“The aim of this initiative is to begin bridging technologies to better serve end users and their needs, from aquaculture to tourism,” Lewis said. “We’re going to use all of NASA’s assets to do that.”
