EventsThe 6th International Electronic Conference on Remote Sensing
Published
This submission belongs to the session S6. Remote Sensing of the Environment of the event The 6th International Electronic Conference on Remote Sensing
Published date
14 Oct, 2026
Academic Editor
author-avatarYanbing Bai
Citation
Paris Antonakis, Stavros Kolios, Evaluating the sensitivity of Sentinel-2 combinations of spectral bands and Indices for Floating Marine Plastic Detection, in Proceedings of The 6th International Electronic Conference on Remote Sensing, 19 October–21 October 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Evaluating the sensitivity of Sentinel-2 combinations of spectral bands and Indices for Floating Marine Plastic Detection

Paris Antonakis 1
image
1. Department of Aerospace Science and Technology, National and Kapodistrian University of Athens, Athens 10679, Greece
Abstract

Floating marine plastic debris is a growing threat to marine ecosystems, but its detection using multispectral satellite imagery remains difficult because target responses vary with material, fractional pixel coverage, and environmental conditions. This study systematically compares five Sentinel-2 spectral indices—the Floating Debris Index (FDI), Plastic Index (PI), sign-preserving kernel Normalized Difference Vegetation Index (kNDVI), Index-1, and Index-5—using observations from nine Greek Plastic Litter Project campaigns conducted between 2018 and 2023. Although the experimental targets were documented in earlier field studies, the original contribution here is their unified cross-campaign and cross-material evaluation under a common water-baseline calibration and statistical separability framework. The dataset comprises 22 target/control combinations, including 18 genuine plastic targets ranging from approximately 616 to 1 m², organic-material controls, and near-target water zones.

For each campaign, index values over targets were compared with the local water mean and expressed as deviations normalized by water-background variability (Δ/σ_water). FDI produced a positive deviation exceeding 2σ (standard deviation values) for all 18 plastic targets, including the smallest sub-pixel targets. Across all 22 target/control zones, its response ranged from 2.3σ to 82σ, with a median near 10σ. However, similarly significant responses over reeds, wood, and near-target water confirmed that FDI detects floating or anomalous surface material rather than plastic specifically. Index-1 was positive for all 18 plastic targets but exceeded 2σ for only 14; all three 2023 sub-pixel targets fell below this threshold (1.1–1.9σ). PI was less consistent, producing negative deviations in 9 of 22 combinations and near-background responses for some plastics (−0.2σ for bottles and 0.1σ for HDPE). Signed kNDVI also lacked directional consistency, with 13 positive and 9 negative responses. Index-5 showed narrower but potentially useful material-discrimination capability: under matched 2021 conditions, HDPE mesh and wood differed by approximately 0.298 index units (−0.297 versus +0.001). Zonal analysis additionally showed monotonic FDI signal decay from target core to surrounding water.

These findings identify FDI as the most scale-resilient detector of floating anomalies while confirming its limited specificity to plastic. A potential practical implication is that FDI could be used for initial candidate identification, with water-related masks and complementary spectral indices subsequently explored for background characterization and material discrimination. However, this hierarchical multi-index framework remains a hypothesis arising from the comparative results and requires independent validation to determine whether it can be efficient for the detection of genuine plastic targets.

Keywords
Marine plastics
Sentinel-2
spectral bands
remote sensing indices
evaluation
SHORELINE CHANGE ANALYSIS OF THE CHENNAI–MAHABALIPURAM–MARAKKANAM COAST USING REMOTE SENSING, GIS AND MACHINE LEARNING TECHNIQUES
Integrating Remote Sensing and Crop Physiological Processes for Large-Scale Winter Wheat Yield Estimation in the Indo-Gangetic Plain