Semi-Automated Seagrass Classification Using Earth Engine Python API

Description: This script classify dense seagrass beds in satellite images (from Sentinel and Landsat sensors) using machine learning (Support Vector Machine). The outputs can be exported to EE Assets. All the training and validation matrices and accuracies can be saved as an Excel file in your working directory.

NOTE: The classifier will use only the aerosol (if available), blue, green, red and Blue/Green (from Depth Invariant Index) bands.

By Luis Lizcano-Sandoval
College of Marine Science, University of South Florida
[email protected]
Updated: 09/03/2021

Sugested citation:

Lizcano-Sandoval, L., Anastasiou, C., Montes, E., Raulerson, G., Sherwood, E. & Muller-Karger, F. (2022) Seagrass distribution, areal cover, and changes (1990–2021) in coastal waters off West-Central Florida, USA. Estuarine, Coastal and Shelf Science 279: 108134. DOI: 10.1016/j.ecss.2022.108134

Workflow:

1. Import required images, collections, data, etc.
2. Mask clouds, land, and deep areas >20m
3. Apply Depth-Invariant Index (band-ratios)
4. Sample bands: B1, B2, B3, B4, B/G
5. Train models and classify (SVM)
6. Get confusion matrices and accuracies
7. Export output to EE Assets (.tiff)
8. Save matrices in local computer (.xlxs)

Demo:

Sentinel-2 L2A Image:

Cloud Mask:

Land Mask:

Depth Mask:

Classified Image: