A Tempered Particle Filter to Enhance the Assimilation of SAR-Derived Flood Extent Maps Into Flood Forecasting Models

Area of Science:

• Environmental science and engineering
• Hydrology and water resources
• Computational science

Background:

• Data assimilation (DA) optimally combines model simulations and observations, but particle filters (PFs) can suffer from degeneracy and sample impoverishment.
• Probabilistic flood mapping using synthetic aperture radar (SAR) data is crucial for effective flood forecasting.

Purpose of the Study:

• To introduce and evaluate a tempered particle filter (TPF) for mitigating PF issues in flood forecasting.
• To assess the TPF's ability to extend assimilation benefits over time and improve forecast accuracy.

Main Methods:

• Assimilation of probabilistic flood maps from SAR data into a flood forecasting model using an iterative TPF approach.
• Inclusion of particle mutation within the TPF to maintain ensemble diversity.
• Comparison of TPF performance against Sequential Importance Sampling (SIS) and an Open Loop (no assimilation) scenario.

Main Results:

• TPF significantly improved model forecast accuracy, reducing root mean square error (RMSE) of water levels by 80% at assimilation and 60% two days later.
• TPF-based RMSE was 20% lower than SIS-based RMSE two days post-assimilation, with benefits lasting nearly three days.
• TPF application resulted in higher critical success index values compared to SIS.

Conclusions:

• The TPF approach effectively mitigates degeneracy and sample impoverishment issues common in standard PFs.
• TPF offers more persistent benefits in flood forecasting accuracy and reliability compared to SIS.
• This study demonstrates the TPF's potential for enhancing real-time flood prediction systems.