Global optimization for data assimilation in landslide tsunamis models

TL;DR

This work addresses the challenge of calibrating a coupled submarine landslide-tsunami model using only free-surface measurements. It casts data assimilation as a global, bound-constrained optimization problem and introduces a parallel hybrid framework based on multi-path Simulated Annealing and Basin Hopping, with local searches via bounded $\text{L-BFGS-B}$ and gradients computed by finite differences. Through synthetic tests and a laboratory dataset, the authors demonstrate identifiability of the key parameters $(r,\theta,n)$ and show that the parallel multi-path hybrids robustly locate the global optimum, outperforming purely local or multi-start approaches. The study provides a practical methodology for calibrating complex two-phase models and offers a foundation for comparing competing landslide-tsunami representations against measured data, with implications for forecasting and hazard assessment.

Abstract

The goal of this article is to make automatic data assimilation for a landslide tsunami model, given by the coupling between a non-hydrostatic multi-layer shallow-water and a Savage-Hutter granular landslide model for submarine avalanches. The coupled model is discretized using a positivity-preserving second-order path-conservative finite volume scheme. The data assimilation problem is posed in a global optimization framework and we develop and compare parallel metaheuristic stochastic global optimization algorithms, more precisely multi-path versions of the Simulated Annealing algorithm, with hybrid global optimization algorithms based on hybridizing Simulated Annealing with gradient local searchers, like L-BGFS-B.

Figures (13)

• Figure 1: Sketch of the model.
• Figure 2: Schematic visualization of the BH$\text{M}$ algorithm (with $\text{M}=4$).
• Figure 3: Sketch of the channel, initial condition and position of the tide-gauges.
• Figure 4: Synthetic generated series vs calibrated ones with the multi-start L-BFGS-B. Target series in red, simulated series in blue.
• Figure 5: Synthetic generated series vs calibrated ones. Target series in red, simulated series in blue.