Karimi et al. (2025) Remote sensing-based bathymetry mapping in shallow lakes: comparative analysis of Sentinel-2 and Landsat-8 imagery integrated with machine learning techniques

Identification

• Journal: Advances in Space Research
• Year: 2025
• Date: 2025-10-10
• Authors: Neamat Karimi, Omid Torabi
• DOI: 10.1016/j.asr.2025.10.028

Research Groups

• Department of Water Resources Study and Research, Water Research Institute, Tehran, Iran

Short Summary

This study evaluates the efficacy of Sentinel-2 and Landsat-8 satellite imagery combined with machine learning (MLP, RF, SVR, XGBoost) for bathymetric mapping in shallow inland waters. It demonstrates that the Multi-Layer Perceptron (MLP) model, particularly when integrated with Landsat-8's optical and thermal infrared bands, can produce sub-metre accuracy bathymetric maps, highlighting the significant contribution of thermal data.

Objective

• To assess the potential of medium-resolution multispectral satellite imagery (Sentinel-2 and Landsat-8) integrated with machine learning techniques to generate high-quality bathymetry for shallow inland waters.
• To conduct a comparative analysis of Sentinel-2 and Landsat-8 imagery performance in bathymetric mapping.
• To quantify the contribution of Landsat-8's thermal infrared bands to bathymetry estimation.

Study Configuration

• Spatial Scale: Chah Nimeh No. 3 reservoir in southeastern Iran.
• Temporal Scale: Satellite imagery dates are not explicitly specified, but the study utilizes contemporary Sentinel-2 and Landsat-8 data. In-situ measurements were collected via a remotely operated hydrographic survey vehicle.

Methodology and Data

• Models used: Multi-Layer Perceptron (MLP), Random Forest (RF), Support Vector Regression (SVR), XGBoost. Hyperparameters for MLP were optimized via systematic grid search.
• Data sources:
  • Satellite imagery: Sentinel-2 and Landsat-8 (medium-resolution multispectral, including thermal infrared bands for Landsat-8).
  • Observation: 52,000 in-situ water depth measurements collected via a remotely operated hydrographic survey vehicle.

Main Results

• The Multi-Layer Perceptron (MLP) model achieved the highest accuracy among the tested algorithms.
• Landsat-8 inputs with MLP yielded an RMSE of approximately 0.82 m (R² ≈ 0.97), while Sentinel-2 inputs yielded an RMSE of approximately 1.10 m (R² ≈ 0.95) during initial evaluation.
• For the final comparison using optimized MLP models and test data (30% of in-situ points):
  • Landsat-8 (all optical and thermal bands): Mean Absolute Deviation (MAD) = 0.50 m, Root Mean Square Error (RMSE) = 0.78 m, Mean Absolute Percentage Error (MAPE) = 9.6 %.
  • Sentinel-2: MAD = 0.58 m, RMSE = 0.88 m, MAPE = 10.4 %.
• Direct spatial comparison of the final bathymetric products from both sensors showed strong agreement (R² = 0.90, mean difference = +0.11 m, standard deviation = 1.7 m).
• Landsat-8 thermal bands (Band 10 and Band 11) showed unexpectedly strong correlations with depth (R² = 0.52 and R² = 0.53, respectively).
• The inclusion of thermal bands in the Landsat-8 MLP model substantially improved performance (R² = 0.98, RMSE = 0.78 m) compared to a model excluding thermal bands (R² = 0.93, RMSE = 1.30 m), representing a 67 % increase in RMSE without thermal data.
• The study achieved sub-metre bathymetric products, with errors less than 5% relative to the 18 m depth range of the Chah Nimeh reservoir.

Contributions

• Demonstrates the effectiveness of integrating freely available medium-resolution multispectral and thermal imagery with machine learning (specifically MLP) for generating sub-metre accuracy bathymetric maps in optically clear shallow lakes.
• Provides a comprehensive comparative analysis of Sentinel-2 and Landsat-8 for bathymetric mapping, establishing Landsat-8's superior performance when thermal bands are included.
• Highlights the significant and previously underappreciated contribution of thermal infrared bands from Landsat-8 to accurate bathymetry estimation, attributing it to depth-linked thermal patterns.
• Offers a cost-effective and operationally valuable tool for monitoring and managing data-sparse inland reservoirs, reducing reliance on expensive and impractical in-situ surveys.

Funding

Not specified in the provided text.

Citation

@article{Karimi2025Remote,
  author = {Karimi, Neamat and Torabi, Omid},
  title = {Remote sensing-based bathymetry mapping in shallow lakes: comparative analysis of Sentinel-2 and Landsat-8 imagery integrated with machine learning techniques},
  journal = {Advances in Space Research},
  year = {2025},
  doi = {10.1016/j.asr.2025.10.028},
  url = {https://doi.org/10.1016/j.asr.2025.10.028}
}