Keywords: Coastal Monitoring, Earth Observation, Machine Learning, Shoreline Change
This case study demonstrates an Artificial Neural Network (ANN) approach to extract shorelines from Sentinel-2 and Pléiades Neo imagery via the Earth Observation Data Hub. By automating land-water classification and shoreline boundary detection, the workflow enables efficient environmental monitoring and analysis of coastal change.
Figure 1: Comparison of satellite imagery and classification results at Connah's Quay. (a) Pléiades Neo true-colour image ANN-classified landcover map and binary image. (b) Sentinel-2 true-colour image, classified landcover map and binary image. These images illustrate differences in spatial detail and classification precision between medium-resolution (10 m) and very high-resolution (30 cm) data.
Automated Shoreline Mapping Using Satellite Imagery
Coastal shorelines are dynamic environments shaped by erosion, sediment transport, and sea-level changes. Traditional shoreline mapping relies on manual surveys, which can be time consuming and limited in coverage. This case study explores how machine learning and Earth Observation (EO) data can automate shoreline extraction, offering a scalable approach for efficient coastal monitoring.
The study focuses on Connah’s Quay, aiming to evaluate how satellite data resolution affects shoreline delineation and classification accuracy. Connah’s Quay is prone to flooding and dynamic shoreline changes, making it an ideal site to test automated extraction methods and assess their ability to capture fine-scale land–water boundaries. By comparing medium- and very high-resolution imagery, the potential for automated approaches to detect fine-scale land-water boundaries was assessed.
Open and Commercial Imagery Acquisition via EODH
For this study, two near-coincident satellite datasets were used to compare how resolution affects shoreline detection:
Sentinel-2: 10 Oct 2024 | 10 m resolution | Open (ESA)
Pléiades Neo: 12 Oct 2024 | 0.3 m resolution | Commercial (Airbus)
These dates allowed comparison under similar environmental conditions, minimising the effects of tides or weather on shoreline position.
Figure 2: Pléiades Neo acquisition of Connah's Quay, Wales © Airbus DS (2024)
Figure 3: Difference map between Pléiades Neo and Sentinel-2 classified outputs. Blue areas indicate where Sentinel-2 underestimates the land-water class relative to PNEO, red areas indicate overestimation, and white indicates agreement. This map highlights the spatial discrepancies introduced by varying satellite resolutions.
Python-based Machine Learning in Jupyter Notebooks
Processing and analysis were conducted in Python, using a combination of geospatial and machine learning libraries. The workflow used the cloud-based infrastructure of the Earth Observation Data Hub, allowing multi-source data integration and efficient computational scaling.
Key tools and platforms included:
- Python libraries: for geospatial processing, data manipulation, and machine learning
- Jupyter Notebook: enabled iterative development and visualization in a single integrated environment
- Earth Observation Data Hub: provided access to Sentinel-2 and Pléiades Neo imagery, metadata filtering, and cloud-hosted computation
By integrating data access, preprocessing, and machine learning within the EODH, the workflow allowed a seamless transition from raw imagery to classification results.
Shoreline Extraction Methodology
To enable automated shoreline extraction and comparison across datasets, a multi-step workflow was developed:
- Preprocessing. Cloud masking was applied to remove unusable data, and regions of interest were clipped to focus the analysis on the study areas. Metadata such as cloud coverage was also used to filter out scenes with excessive interference.
- Training Data Creation. Landcover polygons representing water, sand, agriculture, and artificial surfaces were manually digitised and rasterised to align with the satellite imagery. Random pixel samples were then extracted across multiple images to capture spectral variability, with selected bands normalised to prepare the dataset and improve ANN generalisation.
- Classification. A pixel-based Artificial Neural Network (ANN) model was trained on the labelled data, learning spectral and spatial patterns to classify new imagery into the defined landcover types.
- Postprocessing. The land-water boundary was created by combining sand, agriculture, and artificial surfaces into land, leaving water as the separate class. The resulting binary land-water masks were refined to remove noise and ensure smooth boundaries. Morphological operations were applied to further clean the classification and improve shoreline delineation.
- Dataset Comparison. The classified outputs from Sentinel-2 and Pléiades Neo were compared to assess the impact of satellite resolution on shoreline detection, highlighting discrepancies between medium- and very high-resolution imagery.
- Shoreline Extraction. Final shorelines were derived by extracting contours from the refined masks. These contours were then converted into georeferenced vectors suitable for GIS analysis and further spatial interpretation.
Figure 4: Shorelines extracted from ANN-classified imagery overlaid on the Pléiades Neo image. Green lines represent shorelines derived from Pléiades Neo classification, while red lines represent shorelines from Sentinel-2 classification. The visualization demonstrates how higher-resolution imagery allows for more detailed and precise shoreline delineation.
What the EODH provided
☑️ Immediate access to EO data. Sentinel-2 imagery available instantly without the need for download
☑️ Cloud-based processing. No reliance on local computing power
☑️ Integrated workspace. Scripts, data, and outputs stored in one organised environment
☑️ Single point of access. Centralised platform for discovering and working with EO datasets
Why Earth Observation Is Important For Coastal Monitoring
By combining AI-driven classification with the EODH platform, this workflow provides a scalable, repeatable approach for coastal monitoring. Key benefits include:
Accurate detection of land-water boundaries at multiple spatial scales
Reduced reliance on field surveys and manual digitisation
Support for long-term environmental monitoring and coastal management decisions
Framework for multi-site and multi-temporal analysis, enabling consistent assessments of shoreline change over time
Future Work
This case study demonstrates the potential of automated shoreline extraction using cloud-hosted EO data. The workflow can be applied to additional imagery, allowing shoreline features to be captured across multiple sites over longer temporal scales. By integrating both high-resolution and open-access datasets, it can support ongoing monitoring of shoreline change over time.
