Seagrass Sentinel: Technical Methodology

19 Feb

Overview

The Seagrass Sentinel is a cloud-native Earth Observation application designed to monitor submerged aquatic vegetation (SAV) dynamics and coastal water quality stressors. The tool uses the Google Earth Engine (GEE) API for server-side processing, using multi-sensor fusion to address common limitations in optical remote sensing of benthic habitats.

1. Data Acquisition & Pre-Processing

Optical Imagery: The system processes Sentinel-2 MSI (Level-2A) surface reflectance data.

Cloud Masking: A QA60 bitmask is used to eliminate opaque clouds and cirrus.

Water Masking: A dynamic water mask is created using the Normalised Difference Water Index (NDWI) to delineate the marine area and minimise processing load on terrestrial pixels.

Depth Correction (Lyzenga): To account for water column attenuation, we use a Lyzenga water column correction (Lyzenga, 1978; 1981). Deep-water radiometric calibration is dynamically performed based on the user's Region of Interest (ROI) to linearise the relationship between bottom reflectance and depth.

2. Classification Algorithm

Model: A Random Forest (RF) supervised classifier (Breiman, 2001) is trained in real-time.

Training Data: Ground truth training points are obtained from the UNEP-WCMC Global Distribution of Seagrasses dataset (UNEP-WCMC, Short, 2021).

Feature Space: The classifier inputs include Sentinel-2 spectral bands (B2, B3, B4), the Depth-Invariant Index (DII), and derived turbidity indices.

Class Output: The model classifies pixels into distinct categories: Seagrass Present, Seagrass Absent (Sand/Substrate), and Deep Water.

3. Environmental Stressor Detection

Turbidity: Suspended Sediment Concentration (SSC) proxies are determined using red-edge spectral algorithms to identify turbidity plumes that restrict photosynthetically active radiation (PAR).

Thermal Stress: Sea Surface Temperature (SST) is calculated using thermal bands from Landsat 8 (TIRS) and Landsat 9 (TIRS-2). A threshold of >23.5°C is used to identify potential thermal-stress areas for temperate seagrass species (USGS, 2024).

4. Change Detection

The application measures the net change in benthic cover between two user-defined time periods (Period A versus Period B). Post-classification comparison (PCC) is utilised to quantify specific transitions (e.g., Stable Seagrass, Loss, Gain).

Citation

Farrugia, J., & The Oceans Need Us. (2026). Seagrass Sentinel: Technical Methodology (1.0). Zenodo. https://doi.org/10.5281/zenodo.18690409

References

Algorithms & Methods

Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324

Lyzenga, D. R. (1978). Passive remote sensing techniques for mapping water depth and bottom features. Applied Optics, 17(3), 379–383. https://doi.org/10.1364/AO.17.000379

Lyzenga, D. R. (1981). Remote sensing of bottom reflectance and water attenuation parameters in shallow water using aircraft and Landsat data. International Journal of Remote Sensing, 2(1), 71–82.

Data Sources

UNEP-WCMC, Short F.T. (2021). Global distribution of seagrasses (version 7.1). Seventh update to the data layer used in Green and Short (2003). Cambridge (UK): UN Environment Programme World Conservation Monitoring Centre. Data DOI: https://doi.org/10.34892/x6r3-d211

Green, E.P., & Short, F.T. (2003). World Atlas of Seagrasses. Prepared by the UNEP World Conservation Monitoring Centre. University of California Press, Berkeley, USA.

Copernicus Sentinel Data (2024). Sentinel-2 MSI Level-2A Surface Reflectance. European Space Agency (ESA).

U.S. Geological Survey (2024). Landsat 8-9 OLI/TIRS Collection 2 Level-2 Science Products. USGS EROS.

Jade Farrugia