两种短期概率预测模型在2017年九寨沟7.0级地震中的应用和比较研究

Application and comparison of two short-term probabilistic forecasting models for the 2017 Jiuzhaigou,Sichuan,M_s7.0 earthquake

本文选用"传染型余震序列"(ETAS)模型和Reasenberg-Jones(R-J)模型,分别对九寨沟M_s7.0地震序列的模型参数稳定性、余震发生率预测和余震概率预测进行了比较研究,并利用"地震信息增益"(IGPE)、N-test和Ttest检验方法对预测效果进行了评价.研究结果表明,ETAS模型和R-J模型的序列参数分别在震后t_2=2.0天和t_2=1.50天后趋于稳定,此次九寨沟M_s7.0地震序列的衰减较为正常;对未来1天的余震发生率预测和余震概率连续滑动预测表明,ETAS模型给出的余震发生率和余震概率数值均低于R-J模型预测结果;IGPE结果显示,ETAS模型在95%的置信区间上预测效果明显优于R-J模型;统计检验结果表明,在序列参数较不稳定的震后早期阶段,ETAS模型预测失效而R-J模型预测效果较好,在序列参数稳定阶段,ETAS模型预测效果较好而R-J模型预测失效.根据上述分析,在与此次九寨沟M_s7.0地震类型相同的地震的余震预测策略上,如可在序列参数不稳定的震后早期阶段使用R-J模型、在此后使用ETAS模型,或可取得较好的预测效果.

In this paper, we use the ETAS model and the Reasenberg-Jones （RJ） model to compare the model parameter stability, aftershock occurrence rate and aftershock probability forecasted of the Jiuzhaigou MS7.0 earthquake sequence, and the information gain per earthquake （IGPE）, N-test and T-test were used to evaluate the forecast results also. The results show that the sequence parameters of ETAS model and R-J model are stabilized after t2=2.0 days and t2=1.50 days respectively. The attenuation rate of MS7.0 earthquake sequence is normal when compare to other sequences in continental China. The forecast results of the aftershock occurrence rate and aftershock probability for the next one day show that the values of forecast results of the ETAS model always lower than the R-J model. The results of IGPE show that the ETAS model has better forecast effect than R-J model in the 95% confidence interval. The results of N-test and T-test show that, the ETAS model forecasts failure and R-J model forecast better in the early stage after mainshock occurred with unstable sequence parameters, however, in the sequence parameter stability stage, the above results are the opposite. According to the above analysis, it is possible to obtain a better forecast effect if the R-J model is used in the early stage after mainshock occurred with unstable sequence parameters in the Jiushagou MS7.0 earthquake sequence, and the ETAS model is used after this time.