AI based method for studying changes on the Swedish coast
A novel Artificial Intelligence (AI) method based on Copernicus data was developed to identify physical changes along the Swedish coast, especially physical constructions such as piers and jetties. Using Sentinel-2 data in an Open DataCube (ODC) environment, the coastline is first identified using advanced convolutional (U-Net) models, then the rate of change (and whether the change is permanent or temporary), and lastly small constructions along the shoreline are detected.
The long-term goal is to transform the methodology into a permanent monitoring service that can help municipalities to combat environmental crime, for example, by identifying illegal dredging and excavation activities that affect the marine environment and ecosystem. In addition, there is an added value for municipalities and regions in a Copernicus-based tool. With the continuous supply of Sentinel data, it will become relatively easy to define dredging activities in time and to reduce costs of on-site monitoring and control, especially costs of aerial photography. This will support marine coastal planning regarding the dynamics of the coastal zone and demonstrate the robustness of AI-based technology for coastal and marine research.
The scripts of the workflow are publicly available in a Git repository: https://gitlab.ice.ri.se/sdl/shallow-water-sdl
The following codes can be found there: 0-pixel-alignment.ipynb, 1-mock-mechanistic-ts.ipynb, 2particle-filter-discrete-obs.ipynb, 3particle-filter-continuous-obs.ipynb (Prepare time series, create mock data, get MNDWI and NDWI distributions).
You can find more information in the project report.