Scholarlink

<p>Identifying background seismic events and incorporating them into a suitable model is the stepping stone in earthquake forecasting. In this study, we apply both deterministic and probabilistic approaches to decluster the seismic catalog, employing widely used deterministic methods such as Gardner–Knopoff (G-K), Grünthal, Reasenberg, and Uhrhammer algorithms, as well as advanced probabilistic techniques including the Nearest Neighbour (NN) and Stochastic Declustering (SD) methods. The primary objective of this process is to separate background (independent) events from aftershocks and other triggered seismicity. Following the declustering process, we assess the assumption of the adherence of the resulting background seismicity to the Poissonian model. However, our findings indicate that the background seismicity derived from the applied declustering method does not conform to a homogeneous (single-rate) Poisson model. This discrepancy suggests the necessity of adopting more flexible or advanced statistical models, such as switched rate Poisson processes, to more accurately characterize the background seismicity in the study region. The present study also effectively modeled the background seismicity using a Markov Modulated Poisson Process (MMPP) with multiple states, where the number of states was optimized based on the selected declustering method. For Southern California background seismicity modeling, a three-state MMPP model was found to be most suitable for the Gardner–Knopoff, Grünthal, Uhrhammer, and Nearest Neighbour methods, while the Reasenberg method performed best with a four-state model. The Stochastic Declustering approach was however best represented by a two-state MMPP model. In addition, a forecasting analysis was conducted by leveraging historical seismicity data (1981-2019) to evaluate the potential of statistical models in predicting future earthquake occurrences (2020-21). Specifically, we compared the performance of the MMPP model—capable of capturing regime-switching behavior and temporal variability in seismic rates—with the AutoRegressive Integrated Moving Average (ARIMA) model, which is widely used for time-series forecasting. Therefore, this comparative study provides a clear evaluation of the advantages and drawbacks of each method in predicting background seismic activity. The result yields important insight into how each approach performs in forecasting earthquake events within the study region.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Modeling background events across Southern California using the Markov-modulated Poisson process

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • Aditi Seal,
  • Subhajyoti Sarkar,
  • Niptika Jana

摘要

Identifying background seismic events and incorporating them into a suitable model is the stepping stone in earthquake forecasting. In this study, we apply both deterministic and probabilistic approaches to decluster the seismic catalog, employing widely used deterministic methods such as Gardner–Knopoff (G-K), Grünthal, Reasenberg, and Uhrhammer algorithms, as well as advanced probabilistic techniques including the Nearest Neighbour (NN) and Stochastic Declustering (SD) methods. The primary objective of this process is to separate background (independent) events from aftershocks and other triggered seismicity. Following the declustering process, we assess the assumption of the adherence of the resulting background seismicity to the Poissonian model. However, our findings indicate that the background seismicity derived from the applied declustering method does not conform to a homogeneous (single-rate) Poisson model. This discrepancy suggests the necessity of adopting more flexible or advanced statistical models, such as switched rate Poisson processes, to more accurately characterize the background seismicity in the study region. The present study also effectively modeled the background seismicity using a Markov Modulated Poisson Process (MMPP) with multiple states, where the number of states was optimized based on the selected declustering method. For Southern California background seismicity modeling, a three-state MMPP model was found to be most suitable for the Gardner–Knopoff, Grünthal, Uhrhammer, and Nearest Neighbour methods, while the Reasenberg method performed best with a four-state model. The Stochastic Declustering approach was however best represented by a two-state MMPP model. In addition, a forecasting analysis was conducted by leveraging historical seismicity data (1981-2019) to evaluate the potential of statistical models in predicting future earthquake occurrences (2020-21). Specifically, we compared the performance of the MMPP model—capable of capturing regime-switching behavior and temporal variability in seismic rates—with the AutoRegressive Integrated Moving Average (ARIMA) model, which is widely used for time-series forecasting. Therefore, this comparative study provides a clear evaluation of the advantages and drawbacks of each method in predicting background seismic activity. The result yields important insight into how each approach performs in forecasting earthquake events within the study region.