CEDA Sentinel

Process overview

This repository is designed to use input features (shapefile or geopackage) to extract Sentinel 2 imagery over the UK from the analysis ready Sentinel 2 images available on the CEDA archive.

The Python package in this repository src/ceda_s2 contains classes to search for images, to plot them for visual inspection and to download them as geotiff files.

The CEDA images are cloud optimised geotiffs and this allows the processes created in this repository to download subsections of image tiles intersecting input features directly from their online storage, rather than downloading large whole image tiles. This saves space and can be particularly useful if extracting images to prepare a training dataset for deep learning methods such as UNET.

The process initially extracts links to suitable images for each input feature. These links are recorded in a column in the in a copy of the input search shapefile or geopackage. The process checks the image is not mostly nodata or cloud covered to ensure only usable images are recorded against the search features.

Once the search is complete, images covering each feature can be plotted (an interactive plot window allows to scroll through each search result as RGB true colour images), and/or downloaded as geotiffs.

Setup

Either use Docker as described below or install packages from the requirements.txt:

pip install --requirement requirements.txt

How to run

The easiest way to extract images is by running src/main.py.
For example with a geopackage of search features stored in the inputs directory of this repository:

python src/main.py --search-features inputs/test_search.gpkg \
--start-date 2024-04-01 --end-date 2024-05-31 --plot --download

The above will:

• Search for suitable cloud free images for each feature within the specified date range.
• Add the link to each suitable image as an attribute to the features and extract the image date.
• Pop-up a plot window so can scroll through each image/feature and view as RGB true colour image.
• Download the images to the outputs directory.
• Save a copy of the input shapefile or geopackage with a new field of image links and dates.

Note the search is one to many, so the number of features will often be greater in the output layer if the search finds multiple suitable images for a feature within the specified date range.

Various other arguments can be used to control the behaviour of the src/main.py process. For example, you can just search initially without any plotting or downloading (by including --start-date, --end-date, but not --plot and --download arguments). You can then plot or download later by specifying the search result shapefile (or geopackage) as --search-features (and this time can omit the start and end dates as it will use existing image links found by default).

To see all the options:

python src/main.py -h

Docker

Can use the docker image. In a terminal cd to the repository and then run:

docker build . --no-cache --file .devcontainer/Dockerfile -t ceda

The Docker image built from the devcontainer/Dockerfile includes tk for plotting from within the docker image using X11 forwarding. When start the container need to include this:

 docker run --rm -i -t -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix \
 -p 127.0.0.1:8888:8888 -w /app --mount type=bind,src="$(pwd)",target=/app ceda