Floating litter detection at the estuary of Puducherry using Sentinel‐2 data and machine learning model

Abstract

Listen

Translate article

Marine Floating Litter (MFL) comprises synthetic polymers or plastic that are purposefully tossed out or inadvertently mislaid along rivers, drains, beaches, or other coastal environments that connect the marine ecosystem. Given the adverse impact of the chemicals involved in plastic manufacturing on the marine environment, including marine biota, the need to locate, track, and remove the floating plastic has become demanding. Increasing focus has been placed on remote sensing for its multispectral imaging with high resolution and its capability to monitor large areas systematically. Compared to old-fashioned in-situ observation, satellite imaging products are pecuniary and non-chronophagous. Therefore, the current study exploited the full functionality of open-source Sentinel-2 satellite-retrieved data to detect MFL. The Union Territory of Puducherry (Southern region of India) is the study location. Field measurements were collected from the Ariyankuppam River and the Chunnambar River. A spectral signature profile was created to distinguish other floating debris from floating plastic. The Floating Debris Index (FDI) and Plastic Index (PI) were modeled to validate the performance. It has been proven that FDI performs better at detecting MFL than PI. A Random Forest machine learning model was applied to classify water, plastic debris, and slurry debris. The overall observation of the model details the distribution percentage and accuracy assessment of the proposed model. The accuracy of classifying water, slurry debris, and plastic debris are 64.91 %, 61.6 %, and 96.71 %, respectively, indicating room for improvement in classifying water and slurry debris. The proposed model is used for the first time at the Union Territory of Puducherry and is significantly more efficient compared to in-situ observation in terms of time, cost, and workforce.