Coastal cliffs make up over half of the world’s shorelines and are susceptible to irreversible erosion. Despite their global prevalence and increasing vulnerability due to sea level rise and coastal development, large-scale and long-term monitoring of cliff retreat remains limited. Traditional measurement methods such as LiDAR and GPS surveys are accurate but costly, labor-intensive, and geographically constrained. Satellite imagery offers a scalable and cost-effective alternative, yet existing algorithms are primarily designed for sandy coastlines and are not suitable for cliff environments.

This study investigates the potential of satellite imagery to monitor cliff erosion by adapting an existing Satellite-Derived Shoreline (SDS) algorithm, CoastSat, into a Satellite-Derived Cliff (SDC) algorithm, CliffSat, answering the main research question: ’How can satellite-derived shoreline detection methods be adapted and applied to extract coastal cliff erosion, and how do these satellite-based measurements compare to in-situ erosion data?’

To address this, CoastSat was modified by incorporating a combined spectral index (NDVI and SwiRed) to distinguish cliff tops from sandy foreshores and other land types. Additionally, composite imagery was used to reduce noise, leveraging the relative stability of cliff lines compared to dynamic shorelines. The algorithm was validated along the Holderness Coast in England — a 60 km stretch of rapidly eroding clay cliffs with biannual in-situ measurements.

Validation was conducted using both single satellite images and yearly composite images. The algorithm’s performance was assessed by comparing satellite-derived erosion trends (via linear regression) and total erosion amounts against in-situ data.
• Erosion Trends: Both methods showed similar performance, with a bias of -0.1 meters/year and a standard deviation of 0.7 meters/year — well within acceptable limits (bias < 0.3 meters/year, standard deviation < 0.7 meters/year).
• Total Erosion: The composite method had a slightly higher bias (-0.6 meters) than the single image method (-0.3 meters), but a notably lower standard deviation (4.7 meters vs. 5.5 meters), especially when recorded erosion was below 10 meters. These results fall within acceptable thresholds derived from other SDS algorithms (bias < 3 meters, standard deviation < 7 meters).

These findings demonstrate that satellite imagery can be effectively used to monitor cliff erosion, with the adapted algorithm performing as well as — or better than — established shoreline detection models. Future research should focus on validating the algorithm across diverse cliff types, exploring alternative spectral indices, and integrating elevation data to enhance accuracy. Additionally, the use of higher resolution satellite imagery may further improve performance, particularly in areas with complex land cover or high erosion rates.