Using Remote Sensing and AI to Monitor Plastic Debris in our Rivers

We hear frequently these days about the problem of plastics in our environment, from trash left along our streets and highways to the Great Pacific Garbage Patch in the North Pacific. In fact, the United Nations Environment Programme has listed plastic pollution in the oceans as one of the foremost emerging pollution challenges. But there is a plastic problem closer to home in our rivers and much of that pollution will eventually get to the ocean.

Current methods for detecting plastic debris in aquatic environments are time-consuming and expensive because they require labor-intensive sampling. A recent study conducted at the St. Anthony Falls Lab (SAFL) by then-graduate student Mohammadali Olyaei (Department of Civil, Environmental, & Geo- Engineering) and Associate Professor and CSE DSI affiliate Ardeshir Ebtehaj (CEGE) explored a new way to monitor plastics in freshwater environments. 

This first-of-its-kind study used remote sensing and machine learning to monitor plastic debris in a test environment at SAFL. This research was published recently in the peer-reviewed journal Nature: Scientific Data and indicates that the methods developed in this study show promise for addressing the problem of plastic debris in our freshwater environments like the Mississippi River.

The researchers had to set up a data acquisition system above the hydraulic flume at SAFL. The water running through the flume came directly from the Mississippi. The spectroradiometer used to measure diffuse reflectance of the floating debris was mounted on a mobile carriage system above the flume, along with a digital single-lens reflex camera synchronized with it. Spectral measurements were made from above the tilted flume, where the researchers could simulate various conditions (e.g., foamy, turbid).

The study included both virgin plastics and weathered plastics, the latter collected from the Mississippi River and Lake Hiawatha in Minneapolis. These plastics were released upstream and when they came into the field of view, the researchers recorded with both the spectrometer and the camera multiple times until all the plastic had passed beyond the field of view. 

According to Olyaei, “We could use this technology to identify different types of plastics in the water simultaneously.” By identifying the spectral signatures of different plastics, they could filter out materials that are found naturally in freshwater, such as seaweed, sediments, driftwood, and water foams.

The researchers used machine learning in the technical validation phase of the study. They split the dataset from single virgin plastics into training (70%) and testing (30%). Precision for all plastic types exceeded 90%.

Since this study was carried out under controlled conditions at SAFL, the next step will be to test the system in a natural environment by designing a remote sensing system that is mounted on bridges over waterways like the Mississippi River and use machine learning to help distinguish between the different kinds of plastics.

Research like this is possible here at the University of Minnesota because of the extensive resources available at SAFL for simulating a wide variety of conditions in the indoor flumes as well as the outdoor stream. Christopher R. Ellis, a senior research associate at SAFL, was also on the research team.

This work was funded by the Minnesota Environment and Natural Resources Trust Fund (ENRTF) as the Legislative-Citizen Commission on Minnesota Resources (LCCMR) recommended. The ENRTF is a permanent fund in Minnesota that provides funding for the protection and conservation of Minnesota’s natural resources.

The entire research paper is available on Nature’s website: A Hyperspectral Reflectance Database of Plastic Debris with Different Fractional Abundance in River Systems.