What question did this study set out to answer?

This technical note aims to outline the methodology for monitoring seagrass and environmental stressors using advanced algorithms.

February 22, 2026Open Access

Seagrass Sentinel: Technical Methodology

Key Points

This technical note aims to outline the methodology for monitoring seagrass and environmental stressors using advanced algorithms.
Cloud-native application designed with Google Earth Engine (GEE) API.
Utilizes multi-sensor fusion of Sentinel-2 optical imagery and Landsat thermal data.
Employs Random Forest machine learning for detecting seagrass presence and environmental threats.
Implements algorithms for turbidity and thermal stress detection, plus change detection methods.
Real-time detection of seagrass and aquatic vegetation dynamics is achieved.
Accuracy assessments indicate reliable algorithm performance with real-time error evaluation.
Automated processes effectively isolate study areas from noise in the data.

Abstract

The Seagrass Sentinel is a cloud-native Earth Observation application designed to monitor submerged aquatic vegetation (SAV) dynamics and coastal water quality stressors globally. This technical note describes the algorithmic methodology for detecting seagrass presence and absence, as well as environmental threats, using the Google Earth Engine (GEE) API. The methodology leverages multi-sensor fusion, combining optical imagery from Sentinel-2 MSI (Level-2A) with thermal data from Landsat 8 and 9 (TIRS/TIRS-2). A Random Forest (RF) machine learning classifier is trained dynamically using ground truth data from the UNEP-WCMC Global Distribution of Seagrasses dataset. Key technical features described in this document include: Water Column Correction: Implementation of the Lyzenga depth-invariant index algorithm to account for light attenuation in the water column. Dynamic Masking: Automated masking of deep water, clouds (QA60), and land to isolate the benthic study area. Automated Accuracy Assessment: Real-time 70/30 train-test split of in situ coordinates to generate independent error matrices, evaluating Overall Accuracy (OA) and Cohen’s Kappa coefficient. Stressor Detection: Algorithms for detecting turbidity plumes (suspended sediment) and thermal stress anomalies (>23.5°C) in real-time. Change Detection: Post-classification comparison (PCC) methods for quantifying habitat loss and gain over user-defined temporal baselines. This document serves as the primary technical reference for the Seagrass Sentinel tool, developed by The Oceans Need Us to democratise access to satellite monitoring for marine conservation.

Seagrass Sentinel: Technical Methodology

Key Points

Abstract

Cite This Study