Monitoring plastic accumulations in a river environment using machine learning on Sentinel-1 SAR data

Abstract

Plastic litter is a growing environmental issue. Microwave (MW) technology has the potential for an all-time and weather remote detection of floating macroplastics via satellites or airborne platforms. However, none of the early studies in the existing literature has addressed MW sensor requirements or the application of Machine Learning (ML) for automated monitoring with readily available data. In this study, we employ a supervised learning workflow to monitor, in a preliminary phase, floating macroplastic accumulation in a river environment utilizing polarimetric Sentinel-1 (S-1) SAR data. The combination of co-polarization (VV) and cross-polarization (VH), specifically VV-VH and VV+VH, resulted in detection accuracies exceeding 90%, without overfitting. Analysis of scattering behavior revealed that both VV and VH backscatter were sensitive to plastic patch size, with a linear relationship. This differs from the low-intensity scattering of river water. The findings highlight the importance of dual polarization for effective MW-based plastic monitoring missions.