🔬 physics

A Next-Generation Nonstationary Stochastic Ground Motion Modelling of the 2001 Bhuj Earthquake: A Multi-Peak Gamma–Driven Filtered White-Noise Approach with Uncertainty Quantification and GMPE Benchmarking

This study presents a fully nonstationary stochastic ground motion model for the 2001 Bhuj intraplate earthquake that integrates a multi-peak gamma-driven modulating function, time-varying damping, and Johnson SU non-Gaussian transformations to accurately replicate observed seismic characteristics, quantify parameter uncertainty, and validate against regional ground motion prediction equations for robust seismic risk assessment.

Original authors: Jigar P. Variyavwala

Published 2026-07-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jigar P. Variyavwala

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict how a building will shake during an earthquake. To do this accurately, engineers need a "script" of the shaking—a digital recording of the ground's movement. Usually, they use real recordings from past earthquakes. But here's the problem: in places like Peninsular India (where the 2001 Bhuj earthquake happened), there are very few real recordings to choose from. It's like trying to learn how to drive a car in a snowstorm by only looking at pictures of sunny days.

This paper presents a new, high-tech way to write a perfect script for that specific type of earthquake, even without enough real recordings to copy. The author, Jigar P. Variyavwala, created a "Next-Generation" computer model that simulates the shaking of the 2001 Bhuj earthquake with incredible detail.

Here is a breakdown of how this model works, using simple analogies:

1. The Core Idea: A "Smart" Noise Generator

Think of the ground shaking as a radio playing static (white noise). A standard model just turns up the volume on this static. But real earthquakes aren't just loud static; they have a rhythm. They start with a sharp crack (the P-wave), followed by a long, rolling rumble (the S-wave).

The author's model is like a smart DJ who doesn't just turn the volume up and down randomly. Instead, the DJ uses a special "Gamma-driven" filter to shape the music.

The Two-Pulse Beat: The model recognizes that the Bhuj earthquake had two distinct "beats." It creates one pulse for the initial sharp arrival (P-wave) and a second, longer pulse for the main shaking (S-wave). This is like a song that starts with a drum hit and then transitions into a long, heavy bassline.
The Changing Filter: As the earthquake progresses, the "tone" of the shaking changes. At first, it's high-pitched and sharp. Later, it becomes low and rumbling. The model uses a "time-varying filter" that acts like a sliding equalizer, automatically shifting the sound from high frequencies (6.5 Hz) to low frequencies (1.8 Hz) as the earthquake evolves.

2. The "Heavy Tail" Problem: Not Just Average Shaking

Most earthquake models assume the shaking is "Gaussian," meaning the peaks are predictable and average. But the Bhuj earthquake was weird; it had extreme, jagged spikes that were much higher than average. In statistics, this is called "excess kurtosis."

To fix this, the author added a Johnson SU transformation.

The Analogy: Imagine a crowd of people jumping. A normal model predicts everyone jumps about 2 feet high. But in reality, a few people jumped 10 feet! The Johnson SU transformation is like a magic lens that takes the normal "average" shaking and stretches the tails of the distribution. It ensures the model produces those rare, extreme spikes (the 10-foot jumps) that actually happened in Bhuj, without messing up the average shaking.

3. Cleaning Up the Mess: The "Drift" Fix

When you simulate shaking on a computer, the math can sometimes get "sloppy," causing the ground to appear to move permanently to one side (like a car drifting off the road). This is physically impossible for an earthquake.

The Fix: The author used a Newmark-β high-pass filter. Think of this as a suspension system for the simulation. It acts like a shock absorber that instantly corrects any "drift," ensuring the ground shakes back and forth but ends up exactly where it started.

4. Checking the Work: The "Double-Check"

The author didn't just build the model; they tested it rigorously.

The "One-Record" Challenge: Since they only had one real recording (from Ahmedabad), they had to prove the model wasn't just "memorizing" that one song. They used a technique called Bootstrap Uncertainty Quantification.
- The Analogy: Imagine the model is a chef. Instead of cooking one dish, the chef cooks the same recipe 50 times with slightly different ingredients (within a safe range). If all 50 dishes taste great and are consistent, you know the recipe is solid. The study did this 50 times to ensure the model is reliable and not just a fluke.
The "Rulebook" Check: They compared their new model against two existing "rulebooks" (called GMPEs) used by engineers:
1. BSSA14: A global rulebook made for active tectonic zones (like California).
2. RKI07: A regional rulebook made specifically for stable Indian soil.
- The Result: The new model matched the RKI07 rulebook almost perfectly for longer shaking periods (which affect tall buildings). However, it completely disagreed with the global BSSA14 rulebook. This proves that using global rules for Indian earthquakes is like using a recipe for a tropical fruit salad to make a winter stew—it just doesn't work.

5. The Bottom Line

This paper delivers a validated, computer-generated script for the 2001 Bhuj earthquake. It captures the unique "two-beat" rhythm, the extreme spikes, and the changing frequency of the shaking.

By proving that this model matches regional data and handles uncertainty correctly, the author provides engineers with a powerful new tool. They can now generate hundreds of realistic "what-if" shaking scenarios for buildings in Gujarat, even though they don't have enough real earthquake recordings to do so traditionally. This helps ensure that future buildings in the region are designed to survive the specific, unique way the ground shakes in that part of India.

Technical Summary: Next-Generation Nonstationary Stochastic Ground Motion Modelling of the 2001 Bhuj Earthquake

Problem Statement
The stochastic simulation of earthquake ground motions in intraplate tectonic environments, such as Peninsular India, faces significant challenges due to sparse strong-motion databases, complex wave-phase structures, prolonged strong-motion durations, and distinctly non-Gaussian amplitude statistics. Conventional Gaussian or single-pulse models fail to capture these characteristics. The 2001 Bhuj earthquake ( $M_w$ 7.6), which caused extensive damage in Ahmedabad ( $R_{rup} = 230$ km), serves as a critical benchmark for this intraplate scenario. However, existing Stochastic Ground Motion Models (SGMMs) have not simultaneously addressed multi-peak P/S-wave envelope physics, piecewise spectral bandwidth evolution, non-Gaussian amplitude statistics, and formal parameter identifiability within a single, rigorously validated framework. Furthermore, calibration uncertainty in SGMM parameters is rarely quantified, and GMPE comparisons for intraplate scenarios often rely on inappropriate active-tectonic models.

Methodology
The study develops a fully nonstationary, physically calibrated SGMM based on the Modulated Filtered White-Noise (MFWN) framework (Rezaeian and Der Kiureghian, 2010), augmented by three targeted extensions:

Multi-Peak Gamma-Driven Modulation (Extension 3A): Replaces the single-pulse envelope with a superposition of two independent gamma pulses to independently model the P-wave arrival (0–15 s) and the dominant S-wave/coda phase (10–85 s). The amplitude parameters are analytically derived to ensure the peak envelope matches the target PGA for each phase.
Piecewise Time-Varying SDOF Filter (Extension 2A): Implements a time-varying damping scheme ( $\zeta_f$ ) and a linear frequency sweep ( $f_0 = 6.5$ Hz to $f_n = 1.8$ Hz). The damping ratio is segmented into three phases: 0.08 for narrow-band P-waves, 0.30 for moderate S-waves, and 0.70 for broadband coda.
Johnson SU Non-Gaussian Transformation (Extension 1A): Applies a Johnson SU transformation to reproduce the heavy-tailed amplitude statistics (excess kurtosis $k_u = 45.03$ ) observed in the Ahmedabad record. Crucially, the Johnson SU parameters ( $\delta, \lambda$ ) are recalibrated to ensure the ensemble mean PGA matches the recorded value of 0.106 g, resolving a previous factor-of-11 overestimation.

Parameter Identifiability and Uncertainty Quantification
The study conducts a formal identifiability analysis, demonstrating that 13 of the 17 model parameters are uniquely constrained by physically interpretable target metrics (e.g., peak times, decay rates, spectral bandwidth, kurtosis), while amplitude parameters are analytically fixed. Calibration uncertainty is quantified using a bootstrap resampling procedure over 50 parameter-perturbed realisations. This yields spectral confidence bands with a lognormal standard deviation ( $\sigma_{ln}$ ) of approximately 0.62–0.69, consistent with NGA-West2 inter- and intra-event variability.

Results and Validation

Simulation Fidelity: The calibrated model successfully reproduces the dual-peak envelope structure, the 85-second effective duration ( $D_{5-95}$ ), and the high-frequency impulsive content. The ensemble mean PGA is 0.108 g (within 2% of the recorded 0.106 g), and the ensemble mean Arias intensity is within 10% of the target.
GMPE Benchmarking: The simulated response spectrum is validated against two Ground Motion Prediction Equations (GMPEs):
- BSSA14 (NGA-West2): The model shows systematic divergence, with the SGMM predicting substantially higher spectral accelerations than BSSA14 at most periods. This confirms the inapplicability of active-tectonic models to stable-continent intraplate events at large distances.
- RKI07 (Peninsular India): The SGMM spectrum falls within $\pm 1\sigma$ of the RKI07 regional model at all 11 tested periods ( $T \ge 0.4$ s). At $T = 1.0$ s, the residual is nearly zero ( $\epsilon = -0.007$ ), indicating near-perfect agreement with the regional stable-continent attenuation model.
Spectral Statistics: The coefficient of variation (CoV) for the spectral acceleration across the ensemble ranges from 65% to 76%, reflecting physically realistic intra-event variability.

Significance and Claims
The paper claims to advance the state of the art in intraplate seismic risk quantification by providing a validated, computationally efficient ensemble generator that does not rely on sparse recorded databases. Key contributions include:

Resolution of Calibration Ambiguity: The derivation of analytical normalisation conditions for multi-pulse gamma functions and the recalibration of Johnson SU parameters to ensure physical consistency with recorded PGA.
Formal Identifiability: Establishing that the model parameters are not degenerate for single-record calibration, provided specific physical targets are met.
Regional Validation: Demonstrating that a physics-based SGMM calibrated to a single record can reproduce the statistical expectations of regional empirical models (RKI07) for periods relevant to mid- to high-rise building response ( $T \ge 0.4$ s).
Integration with PBEE: Positioning the calibrated model within the Stochastic Polynomial Chaos Expansions (SPCE)-based Performance-Based Earthquake Engineering (PBEE) fragility framework, offering a pathway for robust seismic risk assessment in regions lacking extensive strong-motion databases.

The study concludes that while the model overpredicts short-period spectral accelerations ( $T < 0.2$ s) due to the heavy-tail nature of the Bhuj record, this is a physically interpretable feature of intraplate rupture rather than a calibration error. The model is presented as a robust tool for generating synthetic ground motions for structural risk assessment in the Gujarat region and similar intraplate environments.