Comparative study of different methods for bathymetric mapping based on multi-source remote sensing data.

Area of Science:

• Geospatial analysis
• Coastal zone management
• Remote sensing technology

Background:

• Satellite-derived bathymetry (SDB) is vital for mapping shallow coastal waters.
• SDB accuracy is influenced by satellite data and bathymetric model selection.
• Effective SDB supports sustainable coastal resource management.

Purpose of the Study:

• To evaluate five SDB inversion methods using Landsat-8, Sentinel-2, and WorldView-2 data.
• To compare the performance of empirical and machine learning models for SDB.
• To determine optimal satellite imagery and model combinations for shallow water bathymetry.

Main Methods:

• Utilized five SDB inversion methods: log-transformed linear, log-transformed band-ratio, k-nearest neighbor regression, random forest, and support vector regression.
• Employed Landsat-8, Sentinel-2, and WorldView-2 satellite imagery with measured bathymetric data.
• Compared bathymetric results from different spatial resolutions across empirical and machine learning models.

Main Results:

• All tested satellite data and models reflected general water depth trends.
• Machine learning models generally outperformed empirical models in accuracy.
• Random forest model demonstrated the best performance across various data types.
• Sentinel-2 data showed stability in empirical models; WorldView-2 excelled in machine learning models.

Conclusions:

• Machine learning models, especially random forest, provide higher accuracy for SDB.
• WorldView-2 imagery is optimal for machine learning-based SDB, while Sentinel-2 is stable for empirical models.
• This research guides the selection of satellite data and models for improved SDB accuracy and application.