Huy et al. (2026) Physics-aware feature engineering improves spatial transferability of satellite estuarine salinity models: a Mekong Delta evaluation

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2026
• Date: 2026-09-28
• Authors: Nguyen Dam Quoc Huy, An Nguyen, Tran Thi Kim
• DOI: 10.1016/j.rsase.2026.102297

Research Groups

• Ho Chi Minh City University of Natural Resources and Environment (Vietnam)
• Universit´e Grenoble Alpes (France)
• University of Science (Vietnam)
• Vietnam National University Ho Chi Minh City (Vietnam)

Short Summary

This study evaluates the use of physics-aware feature engineering to improve the spatial transferability of satellite-based estuarine salinity models in the Mekong Delta. The results show that incorporating physical processes and interactions improves model performance, particularly when using Random Forest models.

Objective

• Investigate the effectiveness of physics-aware feature engineering in improving the spatial transferability of satellite-based estuarine salinity models.

Study Configuration

• Spatial Scale: Regional (Mekong Delta)
• Temporal Scale: 2016–2023 dry seasons (November–June)

Methodology and Data

• Models used: Random Forest
• Data sources: Harmonized Landsat–Sentinel-2 reflectance, Sentinel-1 backscatter, discharge, tidal forcing, and hydrodynamic pseudo-observations from a 3D ocean model (CROCO)

Main Results

• The incorporation of physical processes and interactions improves model performance, with R2 values increasing from 0.26 to 0.78.
• Discharge-position interactions are identified as the dominant predictors using SHAP attribution.

Contributions

• This study provides original insights into the use of physics-aware feature engineering for improving spatial transferability in satellite-based estuarine salinity models.
• The results highlight the importance of incorporating physical processes and interactions to improve model performance.

Funding

• Not specified in the paper.

Citation

@article{Huy2026Physicsaware,
  author = {Huy, Nguyen Dam Quoc and Nguyen, An and Kim, Tran Thi},
  title = {Physics-aware feature engineering improves spatial transferability of satellite estuarine salinity models: a Mekong Delta evaluation},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2026},
  doi = {10.1016/j.rsase.2026.102297},
  url = {https://doi.org/10.1016/j.rsase.2026.102297}
}