Exploring plastic detectability on riverbanks using remote sensing

Plastic pollution is an emerging environmental challenge, threatening terrestrial, freshwater and marine ecosystems. Rivers are major pathways and storage systems, and large-scale plastic monitoring is necessary to effectively reduce plastic pollution. This study investigates the detectability of plastics on riverbanks across spatial scales, ranging from in-situ hand-held spectrometers to large-scale satellites. We designed an experiment using two artificial plastic targets placed on the riverbanks of the Nederrijn, the Netherlands. The first target was a white polyester sheet of four different sizes (0.5x30 m², 1x30 m², 2x30 m², 3x30 m²), and the second target consisted of transparent PET bottles with two different sizes and surface concentrations (3x30 m² with 4 items/m², 15x30 m² with 8 items/m²). Data were collected with several sensors, covering a range of spatial, spectral, and temporal resolutions: the ASD Handheld 2 Spectroradiometer, the MAIA S2 multispectral camera, Sentinel-2, PlanetScope SuperDove, and EnMAP. We analyzed the reflectance spectra, developed a new index (SI-13), and applied a Naïve Bayes detection model to test the detectability of the plastic targets. Sentinel-2 images were successfully used to detect the three largest polyester targets. The PET targets were however not detected. In addition, we found high correlations (-0.93) between polyester target size and several spectral indices. Our results suggest that plastic detection satellite remote sensing is limited by both spatial resolution and plastic concentration. This paper serves as a proof of concept to show that plastic detection in riverbank environments using satellite and camera imagery is feasible and should be investigated further.