Remote sensing detection of plastic waste: recent improvements and remaining challenges

摘要

The presence of plastic litters in the environment (both onshore and offshore) has long been identified as a threat for the ecosystems. Remote sensing provides an efficient and quick access to concentration areas of plastic litters. Due to composition similarities between plastics and hydrocarbons, absorption wavelengths are expected for plastics around 1730 and 2310 nm. Kuhn's Hydrocarbon Index can be used for the detection of plastic targets on a hyperspectral aerial product. Spectral comparison algorithms (Spectral Angle Mapper and Spectral Information Divergence) as well as a spectral unmixing algorithm are used. J. Bioucas-Dias SISAL and MVSA algorithms are adapted for the automation of endmembers selection. Those previous results obtained in a controlled environment were expanded upon by using previous plastic detection algorithms as well as an index-based method (Flooding Debris Index; Biermann et al., 2020) on Sentinel 2 multispectral products. Using a cut-off value on the results of the FDI in southern Spain, onshore greenhouses and offshore plastic debris were detected. A supervised classification complemented the method, based on optical properties of five hundred greenhouses roof and five hundred non-plastic targets. In addition to plastic detection on land, we assessed temporal distribution of threadlike film presence at the sea surface on radar images in the North Atlantic. Wind conditions derived from Sentinel-1 images also help to understand the detection conditions. Spectral band configurations of free constellation of Earth observation satellites are not covering plastic absorption peaks which represent a significant limitation for their detection. The spatial resolution of commercial satellites (i.e Worldview-3) may be better adapted to plastic waste detection that have size distributions lower than open access satellite constellations. Even if offshore plastic detection with open access constellations remain challenging, image processing techniques may improve the detection.