Comparative Accuracy of Satellite-Derived Bathymetry Using Random Forest, Multiple Linear Regression, and Van Hengel and Spitzer Algorithm

Fathurrahman Apriliansyah (Geomatics Engineering, Faculty of Infrastructure and Regional Technology, Institut Teknologi Sumatera, Lampung 35365, Indonesia)
Muhammad Ulin Nuha (Geomatics Engineering, Faculty of Infrastructure and Regional Technology, Institut Teknologi Sumatera, Lampung 35365, Indonesia)
Aulia Try Atmojo (Geomatics Engineering, Faculty of Infrastructure and Regional Technology, Institut Teknologi Sumatera, Lampung 35365, Indonesia)
Kuncoro Teguh Setiawan (Research Center for Geoinformatics, National Research and Innovation Agency, Cibinong 16911, Indonesia.)
Aswar Syafnur (Geophysics Department, Faculty of Mathematics and Natural Sciences, Hasanuddin University, Makassar 90245, Indonesia)

Jurnal Geocelebes
Vol. 10 No. 1: April 2026

Bathymetry mapping using conventional methods faces limitations in shallow waters. With the development of remote sensing technology, satellite-derived bathymetry (SDB) emerges as an alternative by utilizing wavelengths that penetrate water and capture depth information. This study compares the performance of three empirical SDB methods: Random Forest (RF), Multiple Linear Regression (MLR), and the Van Hengel and Spitzer (VHS) algorithm. SPOT 6 ORTHO-level satellite imagery and depth data from single beam echosounder measurements were used to construct the depth models. Model accuracy was evaluated using root mean square error (RMSE), mean absolute error (MAE), and total vertical uncertainty (TVU). Results show that the RF method achieves the highest accuracy across most depth ranges (1–5 m, 5–10 m, and 10–15 m), while the VHS algorithm performs best at 0–1 m. At depths beyond 15 meters, MLR shows relatively better performance compared to other methods, although overall uncertainty remains high. Based on the coefficient of determination (R2), RF achieves the best result with a value of 0.610, followed by MLR with 0.462, and VHS with 0.313. These findings highlight the superior adaptability of the RF method in estimating bathymetry across varying depth zones using optical satellite imagery.