[Paper Review] Earthquake Forecasting Based on Data Assimilation: Sequential Monte Carlo Methods for Renewal Processes

Maximilian J. Werner, Kayo Ide+1 more|ArXiv.org|Aug 11, 2009

earthquake and tectonic studies37 references4 citations

TL;DR

This paper proposes a data assimilation framework for earthquake forecasting using sequential Monte Carlo methods applied to renewal point processes, specifically the Optimal Sampling Importance Resampling (OSIR) particle filter. It demonstrates that accounting for noisy observed event times via OSIR yields significantly more accurate forecasts and parameter estimates than benchmark methods ignoring data uncertainty, with a 98.3% likelihood advantage in synthetic tests.

ABSTRACT

In meteorology, engineering and computer sciences, data assimilation is routinely employed as the optimal way to combine noisy observations with prior model information for obtaining better estimates of a state, and thus better forecasts, than can be achieved by ignoring data uncertainties. Earthquake forecasting, too, suffers from measurement errors and partial model information and may thus gain significantly from data assimilation. We present perhaps the first fully implementable data assimilation method for earthquake forecasts generated by a point-process model of seismicity. We test the method on a synthetic and pedagogical example of a renewal process observed in noise, which is relevant to the seismic gap hypothesis, models of characteristic earthquakes and to recurrence statistics of large quakes inferred from paleoseismic data records. To address the non-Gaussian statistics of earthquakes, we use sequential Monte Carlo methods, a set of flexible simulation-based methods for recursively estimating arbitrary posterior distributions. We perform extensive numerical simulations to demonstrate the feasibility and benefits of forecasting earthquakes based on data assimilation. In particular, we show that the forecasts based on the Optimal Sampling Importance Resampling (OSIR) particle filter are significantly better than those of a benchmark forecast that ignores uncertainties in the observed event times. We use the marginal data likelihood, a measure of the explanatory power of a model in the presence of data errors, to estimate parameters and compare models.

Motivation & Objective

• To address the challenge of forecasting earthquakes when observed event times are corrupted by measurement noise.
• To develop a fully implementable data assimilation method for point-process models of seismicity, particularly renewal processes.
• To evaluate whether sequential Monte Carlo methods can improve forecast accuracy and parameter estimation in the presence of data uncertainty.
• To compare the performance of data assimilation-based forecasts against a benchmark that ignores uncertainties in observed event times.
• To demonstrate the feasibility and benefits of using marginal data likelihood for model comparison and parameter estimation in seismic forecasting.

Proposed method

• The study employs Bayesian data assimilation to combine noisy observations with a prior model of seismicity, treating the problem as a state-space or hidden Markov model (HMM).
• It models seismicity using a lognormal renewal process as the state process, with observed event times subject to measurement errors.
• Sequential Monte Carlo (SMC) methods are used to recursively estimate the posterior distribution of unobserved true event times given noisy observations.
• The OSIR particle filter is implemented using an optimal importance density and resampling to maintain particle diversity and reduce degeneracy.
• The marginal data likelihood is computed to evaluate model performance and estimate parameters via maximum likelihood.
• The method is tested via extensive numerical simulations on synthetic 100-event renewal processes with known true and noisy observed event times.

Experimental results

Research questions

• RQ1 Can data assimilation based on sequential Monte Carlo methods improve earthquake forecasts when observed event times are noisy?
• RQ2 How does the performance of the OSIR particle filter compare to a benchmark forecast that ignores data uncertainty in event timing?
• RQ3 To what extent can the marginal data likelihood be used to estimate parameters and compare models in the presence of observational noise?
• RQ4 Does the OSIR filter produce less biased parameter estimates than the benchmark method when fitting renewal processes to noisy data?
• RQ5 Can the proposed method be generalized to real-world earthquake forecasting and paleoseismic data analysis?

Key findings

• The OSIR particle filter produces forecasts that are significantly more accurate than the benchmark method, which ignores data uncertainty in observed event times.
• In 98.3% of 1,000 synthetic replicas, the OSIR-based forecast achieved a higher log-likelihood score than the benchmark, indicating superior explanatory power.
• Parameter estimates from the OSIR filter were substantially less biased than those from the benchmark, with reduced variance and improved consistency.
• The marginal data likelihood effectively quantifies model performance and enables robust parameter estimation under data uncertainty.
• The OSIR filter successfully handles non-Gaussian statistics of earthquakes and maintains stable performance across multiple event sequences.
• The method demonstrates feasibility for real-world applications, including paleoseismic data analysis and models of Coulomb stress changes with location errors.