This repository is a testbed for shoreline mapping algorithms using satellite imagery. It contains all benchmark datasets, input files, evaluation codes and output results (under /evaluation).
Different algorithms can be used to map the position of the shoreline on satellite imagery like Landsat and Sentinel-2, and extract long-term time-series of coastal change. Satellite-derived shoreline workflows differ on many aspects including:
- the way images are pre-processed (pan-sharpening, compositing, co-registration)
- the spectral indices used for detecting the water edge (NDWI, MNDWI, AWEI)
- the contouring method (at-pixel scale, sub-pixel, hard/soft classification)
- the water level correction that is applied to the shorelines (tide, beach slope, wave setup)
With this project, we want to showcase the diversity of algorithms that can be applied to the satellite imagery and create a platform that can be used to:
- evaluate the accuracy of established satellite-derived shoreline (SDS) algorithms against benchmark datasets with a set methodology.
- test new algorithms, future developments and enhancements of existing SDS workflows.
The following notebooks are available in this repo:
- 1_preprocess_datasets.ipynb: download and preprocesse the grountruth data at the four benchmark sites.
- 2_check_shoreline_accuracy.ipynb: compare your satellite-derived shoreline time-series against the groundtruth (adjust to read your submission files).
- 3_evaluate_submissions_Landsat_MSL.ipynb: evaluate all the submissions using Landsat against the MSL contour.
- 4_evaluate_Landsat_vs_Sentinel2.ipynb: evaluate the time-series obtained from Landsat and Sentinel-2 for the submissions that use individual images (no composites).
- 5_evaluate_wave_correction.ipynb: evaluate the effect of adding a wave-setup correction.
Participants can run their shoreline mapping algorithm at each of the sites using the input files provided.
Everybody is welcome to submit new benchmark datasets, or complement the existing ones, as long as the data they upload is publicly available.
Currently there are 4 validation sites available, which are downloaded from their respective sources and processed into time-series of shoreline change along cross-shore transects:
- Narrabeen, Australia ref
- Duck, North Carolina, USA ref
- Truc Vert, France ref
- Torrey Pines, California, USA ref
The first Jupyter notebook, 1_preprocess_datasets.ipynb, provides the code to download and process the publicly available shoreline datasets into time-series of shoreline change along cross-shore transects. Everyone is encouraged to run this notebook to get familiar with the benchmark sites.
The following inputs are provided for each site:
Region of Interest (ROI): to download/crop the satellite imagery.Cross-shore transects: to extract time-series of cross-shore shoreline change and apply a water level correct.Modelled tides: time-series of tide levels every 5 min from the FES2014 global tide model.Wave parameters: time-series of deep-water wave parameters from the ERA5 reanalysis.(Beach slopes): not available yet, will be estimated from the topographic surveys and can be used to apply a water level correction.
The geospatial layers can be visualised by opening the QGIS file qgis_overview.qgz include in the repository.
Currently, additional information for applying satellite-derived shorelines at each site is available in the sites_info.txt.
| Location | Average Beach Slope (tanBeta) | EPSG code | Groundtruth Elevation Datum | Offset NAVD88 to MSL | MHWS contour (above MSL) |
|---|---|---|---|---|---|
| NARRABEEN | 0.1 | 28356 | MSL | NA | 0.7 |
| DUCK | 0.1 | 32119 | NAVD88 | -0.128 m | 0.585 |
| TRUCVERT | 0.05 | 32630 | MSL | NA | 1.5 |
| TORREYPINES | 0.045 | 26946 | NAVD88 | 0.774 m | 0.792 |
Please use these inputs to ensure that differences between algorithms are not a result of differences in the inputs.
The time-series of shoreline change should be submitted for each transect in a consistent format to make the evaluation easier.
The submission folder will contain the shoreline change time-series from the different participating teams. An example of submission is provided in the example_submission folder. There is one folder per site and in that folder the shoreline time-series are separated in two subfolders:
raw_timeseries: contains the raw time-series of shoreline change along the transects, as extracted from the images (no water level correction).tidally_timeseries: contains the tidally-corrected shoreline change time-series along the transects (can also include a wave correction).
The format of the time-series is one .csv file per transect, named with the following convention PROFILENAME_timeseries_raw.csv and PROFILENAME_timeseries_tidally_corrected.csv where:
- the first column contains the date in UTC time
- the second column contains the cross-shore distance of the shoreline from the origin of the transect
- the third column is optional and contains the name of the satellite mission from which the datapoint was extracted
An example of csv file is provided for Narrabeen under /submissions/example_submission/NARRABEEN.
To submit your results, please:
- fork this repository;
- Copy the
submission_examplefolder and rename (e.g., 'team_CoastSat'); - Change the files in the folder;
- Create a Pull request to the original repository to submit your results.
If you need any help with this submission, please post in the GitHub Issues page.
The deadline for this first round of analysis is the end of the year (01/12/2022).
Please put any questions on the GitHub Issues page so that everybody can read/comment.
We acknowledge the creators of this repository which was used as a template: https://github.com/gwmodeling/challenge