Active and passive data combined depth inversion based on multi-temporal observation: comparison of model and...

Area of Science:

• * Coastal remote sensing
• * Geospatial analysis
• * Hydrographic surveying

Background:

• * Satellite observations offer valuable multi-temporal data for coastal remote sensing.
• * Improving bathymetric inversion accuracy is crucial for coastal zone management.
• * Multi-temporal data analysis can overcome limitations of single-image approaches.

Purpose of the Study:

• * To assess the effectiveness of multi-temporal satellite data for bathymetric inversion.
• * To compare machine learning models (CatBoost, Random Forest) and fusion strategies (FI, IF) for bathymetric inversion.
• * To evaluate bathymetric inversion accuracy in Culebra, Puerto Rico.

Main Methods:

• * Utilized 193 Sentinel-2 images and eight tracks of ICESat-2 ATL03 data.
• * Employed CatBoost and Random Forest (RF) machine learning models.
• * Compared Fusion followed by Inversion (FI) and Inversion followed by Fusion (IF) strategies.

Main Results:

• * Achieved R² > 97.47%, RMSE < 1.00 m, and MAE < 0.54 m using multi-temporal data.
• * The FI strategy outperformed IF, yielding an RMSE of 0.81 m.
• * CatBoost and RF models showed comparable robustness, with CatBoost achieving an average RMSE of 0.88 m.

Conclusions:

• * Multi-temporal satellite data significantly enhances bathymetric inversion accuracy and reliability.
• * The FI strategy combined with advanced machine learning models provides superior results.
• * This approach effectively mitigates environmental interferences like clouds and waves for dependable bathymetric mapping.