Abstract:

On January 7, 2025, a M_S6.8 earthquake struck Dingri, Xizang, causing significant economic losses and casualties. In response, the China Earthquake Administration launched a multidisciplinary scientific investigation, among which the analysis of sequence characterization and the probability forecasting of large aftershocks is an important and meaningful part of the work. This study aims to enhance the understanding of the aftershock sequence and provide timely scientific support for field investigations. To achieve this, we employ a temporal Epidemic-Type Aftershock Sequence(ETAS)model to perform a real-time tracking analysis of the aftershock sequence over the first seven days following the mainshock. The temporal ETAS model was employed to analyze the evolving characteristics of the aftershock sequence at 0.1-day intervals, and short-term aftershock probability forecasts were generated for the subsequent one-day period. Model performance was evaluated using the Brier Score, a metric that quantifies the agreement between probabilistic forecasts and observed aftershock occurrences. The evaluation focused on different magnitude thresholds to assess the consistency and predictive skill of the model.

Key findings from our study include: 1)The fitting of ETAS model to observed aftershock activity was generally consistent with reality. The fitted model parameters suggest that the overall decay rate of aftershocks aligns closely with typical sequence decay behaviors(p=1.06). Moreover, the proportion of triggered ‘offspring’ events within the sequence is relatively low(α=1.58), indicating that off-spring events did not heavily dominate the primary aftershock activity. The model’s fitting results are consistent with the observed seismic sequence, except for a slight deviation identified around the 220th aftershock, where the observed activity exceeds the expectation based on a homogeneous Poisson process. 2)A time-tracking analysis of the model parameters across varying magnitude thresholds reveals that the parameter estimates begin to stabilize approximately 2.8 days after the mainshock. This suggests that incomplete aftershock recordings during the early phase can impact the reliability of early parameter estimation. Thus, early-stage catalog incompleteness should be carefully accounted for in operational forecasting models. 3)The model also demonstrates high sensitivity to the occurrence of strong aftershocks. When such events occur, they are quickly reflected in the intensity and frequency curves, demonstrating the model’s potential and strong applicability for short-term aftershock forecasting, particularly in a science-based emergency response context. 4)Brier score evaluation further supports the model’s forecasting effectiveness. For aftershocks above magnitude 3.5, 4.0, and 5.0, the forecasting performance consistently exceeds that of a random forecast baseline. Although the model underperforms slightly in forecasting aftershocks above magnitude 4.5 in the early stages, its performance improves over time, especially for magnitude 4.5 and 5.0 events, indicating increasing skill as more data accumulates. These findings highlight the potential of integrating Brier Score evaluation into the temporal ETAS model for assessing probabilistic aftershock forecasts.

The results demonstrate that the ETAS model provides valuable operational forecasting capabilities for guiding scientific investigations and emergency response following major earthquakes. The study also identifies key challenges for future improvements, including data completeness, parameter stability, and model adaptability to complex sequences of aftershocks. Moving forward, further refinement of hybrid forecasting approaches—integrating multiple models based on statistical and physics-based methods—could enhance the accuracy and reliability of short-term aftershock forecasting. The operational feasibility of the ETAS model, combined with rigorous evaluation metrics, underscores its role in advancing earthquake forecasting methodologies and supporting earthquake disaster risk reduction in China and beyond.

Key words: Xizang M_S6.8 Dingri earthquake, operational earthquake forecasting, epidemic-type aftershock-sequences model, probability forecasting, performance evaluation

摘要：

2025年1月7日发生的西藏定日6.8级地震对当地经济和群众生命造成了重大灾害。震后, 中国地震局围绕此次地震, 采用不同学科手段开展了跟踪式分析的科考工作, 其中序列特征分析和震后余震概率预测分析是一项重要且有意义的工作内容。为增进对此次地震序列的认识并及时为科考工作提供支撑, 文中针对震后7.1d已经积累的余震序列数据, 采用时间ETAS模型, 以0.1d为间隔进行了跟踪式分析, 对未来1d进行余震短期概率预测, 并使用描述概率预测结果与实际观测一致性的Brier评分方法对模型效能进行检验, 获得以下主要认识: 1)余震序列整体呈现贴近正常水平的衰减速率(p=1.06), 触发产生的“子事件”比例不高(α=1.58), 模型整体拟合情况与余震实际发生情况基本一致; 2)模型参数从震后第2.8d开始趋于稳定, 震后短时间内余震记录不全的问题会对模型拟合产生影响; 3)该模型的预测曲线能够快速反映第6.5d发生的5.0级余震情况, 显示出该模型对此类工作中余震短期预测较强的适应性和应用的潜在价值; 4)针对以上跟踪式分析预测结果的Brier评分结果显示, 该模型对3.5级、 4.0级和5.0级以上余震的预测优于随机预测(score<0.25), 其中对4.5级和5.0级以上余震的预测效能随时间不断提升。文中探讨了将Brier评分方法应用于时间ETAS模型概率预测效能评估中的潜力, 发现其在综合评估预测表现及预测能力随时间变化方面具有较大优势, 该模型的可操作性预测框架对支撑地震科学考察和辅助地震决策具有重要价值, 同时讨论了下一步开展此类工作需要解决的潜在问题。

关键词: 西藏定日M_S6.8地震, 可操作的地震预测, 传染型余震序列模型, 余震概率预测, 效能检验