Tackling the challenge between lab- and field-based detection of floating plastics using hyperspectral remote sensing

Plastic pollution in aquatic ecosystems has increased dramatically in the last five decades, with strong impacts on human and aquatic life. Recent studies endorse the need for innovative approaches to monitor the presence, abundance, and types of plastic in these ecosystems. One such approach gaining rapid traction in the remote sensing community is the use of hyperspectral cameras to identify floating plastic litter. However, most experiments using this approach have been conducted in controlled laboratory environments, making findings exceptionally challenging to apply in natural environments. We present a method that links lab- and field-based identification of macroplastics using hyperspectral data (1150-1675 nm). Two experiments using riverbank-harvested macroplastics were set up in (1) a laboratory environment, and (2) at the Rhine River. The reflectance characteristics of the sample items were analysed to understand the influences of the two environmental settings. Eleven lab-based images (n = 786.264 pixels) and two field-based images (n = 40.289 pixels) were used for these analyses. Next, multiple classifier algorithms such as support vector machines (SVM), spectral angle mappers (SAM) and spectral information divergence (SID) techniques were applied, because of their robustness to varying light intensities and high accuracies in mapping spectral similarities. Our results showed that SAM classifiers are most robust in separating plastic debris from natural or human-made background elements, such as vegetation, quay walls, and sand. By applying lab-based spectral data for plastic detection in our field-based images, we were able to attain user accuracies up to 93.6% (n = 8.370 plastic pixels) using SAM. The same approach resulted in accuracies of 50.2% and 65.4% for SID and SIDSAM, respectively. The limitations of this study concern the low number of images used for training and classification, hardware issues related to the hyperspectral camera, and the unforeseen dynamic nature of environmental conditions outside a laboratory. Nevertheless, this study provides key fundamental insights in linking lab-based data to plastic detection in the field. In doing so, a contribution to the development of future spectral missions to monitor plastic pollution in aquatic ecosystems is made.