link: https://isprs-archives.copernicus.org/articles/XLIII-B3-2022/1011/2022/isprs-archives-XLIII-B3-2022-1011-2022.pdf

Very simple paper. Nothing novel done.

What have they done?

Built a Random Forest model to predict seaweed presence along the west coast of Ireland.

Introduction Section

Introduction is full of justifying the importance of seaweed and in introducing remote sensing as an important way for monitoring. Largely irrelevant for our purposes. We already know that these things are super important and super useful.

Study Area and Acquiring Dataset

Study area is the entire western coast of Ireland. 3000km of coast with everything ≤ 750m outside and inside the shoreline included.

Sentinel-2 data was chosen for satellite imagery:

• 10m resolution data
• 13 different bands from visible to far infrared
• High temporal resolution of 5 days for their regions

Date of images: 28/05/2021 - Good meteorological conditions and cloud absence over a large part of the study area

Processing of Dataset

• Pixels in satellite images/tiles outside study area were cropped out
• Resampling of all bands to 10m was performed
• Random indices were created from bands: 9 features in total at the end eg: NDVI, BNDVI and NDWI
• Target classes were defined and polygons were created for each class: Seaweed, water, Rocks, Fine Sediment, Vegetation, Clouds (to mask out areas in the final outputs)
• Creation of polygons was based on Sentinel-2 true colour images, ground surveys information and spectral indexes calculated.