Satellite-Derived Bathymetry using Adaptive Geographically Weighted Regression Model

ABSTRACT

The common practice adopted in previous attempts on Satellite-Derived Bathymetry (SDB) has been to calibrate a single set of coefficients using global regression model. In this study we propose an Adaptive-Geographically Weighted Regression (A-GWR) model that takes into account local factors in determining the regression coefficients. A-GWR model is examined as an effective solution for addressing heterogeneity and could provide better water depth estimates in near-shore region. The study has been carried out for a 30-km stretch and covers 160 km² of a complex near-shore coastal region of Puerto Rico, Northeastern Caribbean Sea. Medium-resolution (Landsat-8) and high-resolution (RapidEye) images were used to estimate water depth. Results demonstrate that the A-GWR model performs well in estimating bathymetry for shallow water depths (1–20 m), showing the correlation coefficient (R) of 0.98 and 0.99, determination coefficient (R²) of 0.95 and 0.99 and Root Mean Square Error (RMSE) of 1.14 and 0.4 m for Landsat-8 and RapidEye, respectively. The data-processing workflow has been entirely implemented in an Open-Source GIS environment and can be easily adopted in other areas.

KEYWORDS:

• A-GWR model
• Landsat-8
• RapidEye
• satellite derived bathymetry

Acknowledgements

The authors express their sincere thanks to MEXT and Osaka City University for supporting this research. They are grateful to NOAA, USA for providing high-resolution LiDAR-derived water depth data from Puerto Rico. They are also thankful to the anonymous reviewers for their valuable comments that greatly helped in improving this paper.

Funding

This work was supported by a Japanese Government (Monbukagakusho: MEXT) Scholarship awarded to the first author.