Abstract
When choosing an appropriate hydrodynamic model, there is always a compromise between accuracy and computational cost, with high-fidelity models being more expensive than low-fidelity ones. However, when assessing uncertainty, we can use a multifidelity approach to take advantage of the accuracy of high-fidelity models and the computational efficiency of low-fidelity models. Here, we apply the multilevel multifidelity Monte Carlo method (MLMF) to quantify uncertainty by computing statistical estimators of key output variables with respect to uncertain input data, using the high-fidelity hydrodynamic model XBeach and the lower-fidelity coastal flooding model SFINCS (Super-Fast INundation of CoastS). The multilevel aspect opens up the further advantageous possibility of applying each of these models at multiple resolutions. This work represents the first application of MLMF in the coastal zone and one of its first applications in any field. For both idealised and real-world test cases, MLMF can significantly reduce computational cost for the same accuracy compared to both the standard Monte Carlo method and to a multilevel approach utilising only a single model (the multilevel Monte Carlo method). In particular, here we demonstrate using the case of Myrtle Beach, South Carolina, USA, that this improvement in computational efficiency allows for in-depth uncertainty analysis to be conducted in the case of real-world coastal environments – a task that would previously have been practically unfeasible. Moreover, for the first time, we show how an inverse transform sampling technique can be used to accurately estimate the cumulative distribution function (CDF) of variables from the MLMF outputs. MLMF-based estimates of the expectations and the CDFs of the variables of interest are of significant value to decision makers when assessing uncertainty in predictions.
Introduction: The Hydrodynamicist's Dilemma
Predicting coastal flooding is a high-stakes task, but hydrodynamic models are inherently plagued by uncertainty in input data, such as wave height or bed roughness. To quantify this risk, researchers typically rely on Monte Carlo (MC) methods, which require running a model thousands of times to generate a probability distribution of outcomes.
However, researchers face a "hydrodynamicist's dilemma": high-fidelity models (like XBeach) provide necessary accuracy but are computationally grueling, often requiring days for a single simulation. Conversely, low-fidelity models (like SFINCS) are fast but may lack the physical detail required for precise decision-making. Standard MC with high-fidelity models is often practically impossible for real-world scenarios, forcing a choice between inaccurate uncertainty quantification (UQ) or overly simplified physical models.
Methodology: The MLMF Approach
This paper introduces the Multilevel Multifidelity Monte Carlo (MLMF) method to coastal flood modeling, a "best of both worlds" approach that merges two existing strategies:
- Multilevel MC (MLMC): Utilizes a hierarchy of grid resolutions (from coarse to fine) to reduce variance.
- Multifidelity MC (MFMC): Combines high-fidelity (HF) and low-fidelity (LF) models.
The core mathematical innovation is the use of a telescoping sum. Instead of estimating the expected value of a high-fidelity, high-resolution run directly, the algorithm estimates the expected value of a computationally inexpensive, low-fidelity run and then adds a series of "corrections"—the differences between LF/HF and coarse/fine levels. Because these differences have much lower variance than the raw outputs, they require significantly fewer expensive simulations to achieve convergence.
Furthermore, the authors developed a novel technique using inverse transform sampling to estimate the full Cumulative Distribution Function (CDF) from MLMF outputs. This allows practitioners to calculate the probability of extreme events, such as the likelihood of flood waters exceeding a specific threshold.
Case Studies
The method was evaluated across three distinct scenarios:
- 1D Non-breaking Wave: An idealized case with a known analytical solution used to verify algorithmic accuracy.
- 1D Carrier-Greenspan: A more complex case involving wave run-up and run-down on a sloping beach.
- 2D Myrtle Beach, SC, USA: A real-world application involving a complex dune system. This case utilized actual bed level data and simulated tidal signals to assess flooding risk from offshore water level uncertainties.
Results and Discussion
The "aha moment" of this research is the realization that while absolute water levels might vary wildly, the difference between model levels (e.g., a 5m-grid vs. a 10m-grid) is relatively stable. By sampling this narrow difference, the number of "expensive" runs required is drastically reduced.
Computational Efficiency
- Massive Speedups: In 1D test cases, MLMF was over 100 times faster than standard Monte Carlo and approximately 5 times faster than multilevel methods using only a single model.
- Real-World Feasibility: For the Myrtle Beach case, a standard high-fidelity Monte Carlo analysis would have required an estimated 40 years of compute time. MLMF achieved comparable results in just 6 years of total simulation time (distributed across cores).
- CDF Accuracy: The novel CDF estimation method closely matched results from "brute force" Monte Carlo benchmarks, proving its reliability for predicting rare, high-impact flood events.
Conclusion
MLMF shifts uncertainty quantification from a theoretical possibility to a standard engineering workflow. While the efficiency of MLMF depends on the correlation between low- and high-fidelity models, this study demonstrates that it can make complex, real-world risk assessments feasible for the first time. Future work will explore handling high-dimensional problems where multiple input parameters vary simultaneously and integrating advanced sampling techniques like Latin Hypercube Sampling or machine learning emulators.
Summary updated