Khaniya et al. (2025) Using ensemble optimal interpolation with dynamic covariance matrices for assimilation of water level observations in a distributed rainfall-runoff-inundation model

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-12-09
• Authors: Manoj Khaniya, Yasuto Tachikawa, Kodai Yamamoto, Takahiro Sayama, Sunmin Kim
• DOI: 10.1016/j.jhydrol.2025.134733

Research Groups

• Graduate School of Engineering, Kyoto University, Kyoto, Japan
• Disaster Prevention Research Institute, Kyoto University, Kyoto, Japan

Short Summary

This study investigates two ensemble generation strategies for the computationally efficient ensemble optimal interpolation (EnOI) scheme to produce dynamic covariance matrices for assimilating water level observations into a distributed rainfall-runoff-inundation model, demonstrating that EnOI can provide improved state estimates compared to deterministic simulations, particularly with an adaptive error parameter estimation approach.

Objective

• To investigate two ensemble generation strategies for producing dynamic covariance matrices within the ensemble optimal interpolation (EnOI) scheme for assimilating water level observations into a distributed rainfall-runoff-inundation model, through both synthetic and real experiments.

Study Configuration

• Spatial Scale: Distributed rainfall-runoff-inundation model; spatially correlated Gaussian noise applied. Specific geographical area not detailed in the provided text.
• Temporal Scale: Focus on operational assimilation applications and real-time flood forecasting. Specific simulation durations not detailed in the provided text.

Methodology and Data

• Models used: Distributed rainfall-runoff-inundation model, Ensemble Optimal Interpolation (EnOI), Ensemble Kalman Filter (EnKF).
• Data sources: Water level observations (assimilated), synthetic data (for synthetic experiments), real-world data (for real experiments).

Main Results

• The EnOI scheme performs worse than the EnKF algorithm when true system and observation error characteristics are perfectly known.
• EnOI may provide better results than EnKF in real applications where identifying an optimal EnKF is challenging.
• Both proposed EnOI approaches (fixed error parameter and adaptive) improve state estimates compared to deterministic simulations, at some additional computational cost.
• The fixed error parameter approach works well when the background ensemble spread is commensurate with the system error, but degrades significantly if the spread is too large.
• The adaptive approach reduces sensitivity to the error parameter, particularly at assimilation locations, but requires further work to improve state estimation across the entire domain.

Contributions

• Proposes and evaluates two novel ensemble generation strategies for the EnOI scheme to enable dynamic covariance matrix estimation, addressing a key limitation of traditional EnOI.
• Demonstrates the potential of EnOI as a computationally efficient alternative to EnKF for operational flood forecasting, especially in scenarios with imperfect knowledge of error characteristics.
• Provides insights into the performance and sensitivity of these EnOI strategies, highlighting the benefits of an adaptive approach for error parameter estimation.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Khaniya2025Using,
  author = {Khaniya, Manoj and Tachikawa, Yasuto and Yamamoto, Kodai and Sayama, Takahiro and Kim, Sunmin},
  title = {Using ensemble optimal interpolation with dynamic covariance matrices for assimilation of water level observations in a distributed rainfall-runoff-inundation model},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134733},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134733}
}