Study On The Rapid Estimation Of The Potential Damage Zone By Integrating Onsite-and Regional-warning

Earthquake early warning technology is designed to provide early warning information before destructive seismic waves reach the target area,so as to gain enough time for earthquake emergency measures to achieve disaster mitigation benefits.In the early stage of the earthquake,the rapid prediction for area that may be damaged after the earthquake(potential damage area)has always been the premise and difficulty for the earthquake early warning system to issue effective warnings accurately.Part of the existing research predicted the earthquake intensity of the epicenter and surrounding area based on the simplified fault model and the law of attenuation of ground motion.However,it is affected by the problems such as the underestimation of magnitude for large earthquakes and the time-consuming determination of the faut model;the other part is mainly to establish an on-site ground motion prediction model based on the correlation between the P-wave early warning parameters and the subsequent peak ground motions,which directly estimates the extent of the potential damage area according to the distribution of damage to the station and its surroundings.However,the accuracy of the on-site ground motion prediction model still needs to be improved.In response to the above problems,based on the Chinese strong motion data in the period of 2007-2017,this article optimized the earthquake magnitude estimation from the perspectives of all P-wave time window parameter characteristics,source theory,multi-parameter decision-making combination,optimized the on-site ground motion prediction models by applying the machine learning methods,developed a new grid interpolation method to generate real-time prediction PGV distribution maps,estimated the potential damage area based on the rupture characteristics of large earthquakes and small-moderate earthquakes,and finally established a real-time,multi-step optimization estimation method for the potential damage zone of earthquakes in China that combine regional and on-site early warning.The research results obtained are as follows:1)This study proposed a real-time magnitude estimation method using all P-wave time windows,showing that this method can ensure the stability of the estimation of small-moderate earthquakes while improving the underestimation of large earthquakes.From the perspective of the P-wave time window length closely related to the magnitude estimation parameters,the statistical magnitude relationships for the five parameters P_d,P_v,IV2,?_c and?_p^max under all P-wave time windows were established.Based on the characteristics of the statistical relationship,a real-time magnitude estimation method was proposed by selecting the amplitude parameters P_d and P_v under the all P-wave time windows.Through offline simulation,the advantages and feasibility of this method has been verified with comparing to the results of using a fixed 3s time window.2)Based on the P-wave displacement peak time history curve,a source parameter estimation method was proposed by select a suitable source model according to the magnitude of the earthquake.Moreover,the real-time application of this method was explored.Based on the generated P-wave displacement peak time history curve,the characteristics of the curve corresponding to the source time function were extracted through model fitting.The attenuation relationship was selected properly for the magnitude estimation based on the height of the platform section of the curve.The source time function generated by the theoretical source model was proposed to be used for exploring the relationship between the model parameter and the arrival time at the platform.The point source model and the Haskell rectangular model were selected properly according to the magnitude of the earthquake for estimating the length of the fault and the stress drop based on the predicted time of reaching the platform.With comparing to the research results of other scholars,the accuracy of the method in estimating the source parameters was verified.Finally,the real-time application of the method in earthquake early warning was explored.3)By prejudging the combination of multiple early warning parameters,a threshold-based method for continuously estimating the magnitude was proposed.Considering the timeliness of real-time magnitude estimation,the magnitude statistical relationships for?_c and P_d under an extended time window were established by using the near-field data set with epicenter distance less than 60km in the Sichuan-Yunnan region.Based on the characteristics of?_c and P_d in estimating magnitude within different magnitude segments and different time window lengths,a threshold-based method was proposed to judge whether to introduce?_c parameter for magnitude estimation.This method used the weighted average approach based on the time window length to optimize the final magnitude estimation under multiple triggers.Finally,the accuracy and timeliness of the method has been verified.4)After training and optimizing the multi-parameter Support Vector Machine(SVM)models for predicting magnitude and PGV,the on-site ground motion threshold-based prediction model was reconstructed using the magnitude and PGV predicted from the SVM models.With the help of machine learning in terms of better performance in predicting magnitude and PGV comparing to the traditional single parameter,the best feature parameters were selected for training the SVM models of magnitude and PGV.The magnitude and PGV values obtained from the SVM models were used to replacing?_c and P_d in the existing threshold-based model.Finally,the simulation shows that the accuracy of determining potential damage at the stations and the estimation of the potential damage area has been greatly improved after combining the threshold-based on-site ground motion prediction model with the machine learning approach.5)A prediction method for real-time generation of PGV distribution maps that combines the laws of ground motion attenuation and station correction was proposed.The real-time magnitude estimation results of multi-method fusion were added into the ground motion prediction equation,and the constructed SVM-PGV prediction model was used to predict PGV at the stations.By defining the weights that are corresponding to the station spacing,the influence of the distance between the grid points and the stations,the correction of the grid points by the stations,and the predicted value of the source attenuation to the grid points were fused to generate a smooth and accurate PGV distribution map in real time.6)Based on the accurate real-time prediction of the PGV distribution map,a rapid prediction method of potential damage zones suitable for large earthquakes and small-moderate earthquakes adapted to their respective rupture characteristics was proposed.Based on the point source characteristics of the rupture of small-moderate earthquakes,the potential damage area was obtained directly according to the corresponding relationship between PGV and instrument intensity;considering the linear rupture characteristics of large earthquakes and the time-consuming determination of the fault model,it is proposed to firstly determine the rupture length according to the magnitude,and then to determine the fault-related parameters and potential damage areas in real time based on the minimum error between each intensity circle determined by the fault model attenuation and the predicted PGV distribution.As the results of the off-line application show,the potential damage area obtained from this method is consistent with the actual intensity circle of the investigation.