River plastic hotspot detection from space.

Area of Science:

• Environmental Science
• Remote Sensing
• Machine Learning

Background:

• Plastic pollution is a global environmental crisis, with rivers acting as major conduits for plastic debris.
• Identifying plastic accumulation hotspots in rivers is crucial for effective mitigation strategies.
• Current methods for detecting riverine plastic are often limited in scale and accessibility.

Purpose of the Study:

• To develop and validate a scalable, semi-automated pipeline for detecting riverine plastic accumulation hotspots.
• To integrate satellite remote sensing and machine learning for enhanced plastic pollution monitoring.
• To assess the transferability of the developed method across diverse river systems.

Main Methods:

• A semi-automated, cloud-based pipeline was created using Google Earth Engine.
• High-resolution PlanetScope imagery was used for training data annotation.
• Sentinel-2 multispectral data and Random Forest classifiers were employed for hotspot detection.
• The approach was tested on the Citarum, Motagua, and Odaw river systems.

Main Results:

• The pipeline achieved high accuracy in detecting plastic hotspots within rivers (up to 99.5%).
• Optimized inter-river transfer of the model resulted in a plastic F1-score of 79%, surpassing previous benchmarks (69%).
• The method demonstrated transferability across diverse environmental conditions in the studied river systems.

Conclusions:

• The developed satellite-based pipeline offers a reproducible and scalable solution for monitoring riverine plastic pollution.
• This approach supports the development of global strategies for assessing and managing plastic pollution in rivers.
• An open-access Google Earth Engine application facilitates widespread adoption and further research.