TL;DR
This paper presents the results of the PRENOLIN benchmark's verification phase, comparing 23 numerical codes for 1D nonlinear seismic site response. The study highlights that while linear cases converge well, nonlinear simulations show significant code-to-code variability, particularly for low-frequency motions and high strains, underscoring the importance of precise input parameter definitions and constitutive model choices.
What Problem Does This Paper Address?
Accurately predicting how seismic waves are amplified or attenuated as they travel through soil layers (site response) is crucial for seismic hazard assessment. This is especially true for strong ground motions, where soil behavior becomes nonlinear, meaning its stiffness and damping change with strain. However, simulating this nonlinear behavior is complex, and different numerical codes and constitutive models can produce widely varying results. This variability, known as epistemic uncertainty, needs to be quantified and understood to improve the reliability of seismic hazard predictions. The PRENOLIN project aims to address this by benchmarking various numerical codes under controlled, idealized conditions before validating them against real-world data.
Methods
The PRENOLIN project's verification phase involved 21 teams and 23 computational codes. They were tasked with simulating seismic wave propagation through three idealized 1D soil profiles under various conditions:
- Linear Cases (Elastic and Viscoelastic): Initially, participants compared their results for linear elastic and viscoelastic soil behavior. This step was crucial for ensuring a common understanding of input parameters, units, and basic numerical implementation, leading to convergence with analytical solutions (Haskell-Thomson).
- Nonlinear Cases: For nonlinear simulations, participants used various constitutive models to represent soil behavior under a wide range of shear strains. They were provided with simple, idealized soil profiles and different input motions (pulse-like and real earthquake records scaled to different peak ground accelerations). Boundary conditions at the sediment-bedrock interface (rigid vs. elastic) were also varied.
- Analysis: The results were compared based on surface acceleration time histories, Fourier transfer functions, pseudoresponse spectra, peak shear strain profiles, and stress-strain curves. Variability was quantified using standard deviations in logarithmic units.
Datasets
The study utilized three idealized soil profiles (P1, P2, P3) designed to represent common geotechnical scenarios, ranging from shallow homogeneous layers to thicker profiles with velocity gradients or distinct layers.
For input motions, the benchmark used:
- A simple Ricker pulse: A synthetic waveform designed to test basic nonlinear response characteristics.
- Two real earthquake accelerograms: One with high-frequency (HF) content and another with low-frequency (LF) content, scaled to three different peak ground acceleration (PGA) levels (0.5, 1, and 5 m/s²).
These inputs were applied at the bedrock level under two boundary conditions: rigid base (representing a downhole sensor) and elastic base (representing an outcropping rock motion).
Key Insights and Ideas
- Linearity is Key for Baseline: The initial comparison of linear elastic and viscoelastic cases was vital. It revealed discrepancies due to misunderstandings of "input motion" definitions and numerical implementation issues, highlighting that even simple cases require careful specification and can expose fundamental code differences.
- Nonlinearity Amplifies Uncertainty: The real challenge emerged in nonlinear simulations. The variability between code predictions increased significantly with strain level, especially for low-frequency input motions.
- Constitutive Models Matter (A Lot): The choice of nonlinear constitutive model and its specific implementation (e.g., shear strength, damping curves) proved to be a major driver of code-to-code variability.
- Input Motion Definition is Critical: Ambiguities in defining "input motion" (outcrop vs. within motion) caused initial discrepancies. Clarifying these terms was essential for achieving convergence in later stages.
- Shear Strength and Damping Control are Crucial: The study found that specifying the soil shear strength profile accurately and the use (or not) of damping control (Masing rules) significantly impacted the results, with damping control showing a particularly large influence on stress-strain curves.
- LF Motions are More Challenging: Low-frequency input motions, which tend to induce higher strains, led to greater variability in predictions compared to high-frequency motions.
Key Results
- Linear Convergence: After initial iterations and clarification of definitions, all teams converged to the Haskell-Thomson analytical solution for elastic and viscoelastic cases.
- Nonlinear Variability: For nonlinear computations, epistemic uncertainties (code-to-code variability) increased with strain level. For a PGA of 5 m/s² at the base, this variability reached around 0.2 (log₁₀ scale).
- Reduction of Uncertainty: The study showed that variability could be reduced by almost 50% when constitutive models used the same shear strength and damping implementation.
- Parameter Sensitivity: Peak shear strain values were highly sensitive to the input motion's frequency content and intensity. For LF motions at the highest PGA, peak strains were significantly larger than for HF motions, even with the same PGA.
- Groupings for Reduced Scatter: Analyzing results based on different code characteristics (attenuation implementation, numerical scheme, constitutive model, shear strength, damping control) revealed that groupings related to shear strength and damping control (G4) showed the most significant reduction in variability.
Limitations and Future Work
- Idealized Cases: The verification phase used simplified, idealized soil profiles and boundary conditions. Real-world sites are more complex, with 2D/3D geometries and more intricate soil layering.
- Limited Input Motions: While two real earthquake records were used, they represent only a fraction of possible seismic scenarios.
- Numerical Issues: The study acknowledges that timestep choices, while necessary for stability, could also contribute to discrepancies, especially in strongly nonlinear cases.
- Future Work: The paper sets the stage for the validation phase, which will compare these numerical predictions against actual strong-motion recordings from well-characterized sites. This will be crucial for assessing the practical reliability of the codes and models. Further research is suggested on the impact of damping control and the precise definition of shear strength profiles. The authors emphasize the need for multiple codes and careful parameter specification for realistic site response analyses.
Summary updated