Application Of Pattern Informatics Method To Earthquake Prediction Based On Ionospheric Data From Zhangheng (01) Satellite

As a worldwide scientific issue,earthquake forecast is still being unsolved,and the use of " Non-seismic precursors " with large variations in time scale and anomaly amplitude for universal earthquake prediction is an important research direction.The China Seismo-Electromagnetic Satellite(CSES)is a microsatellite launched by China on February 2,2018,using for monitoring seismic electromagnetic activity,also known as the Zhangheng-1 electromagnetic satellite.It has completely recorded information on more than 500 strong earthquakes of magnitude 6 or higher and nearly 60 strong earthquakes of magnitude 7 or higher around the world,and has stored a large amount of observation data.Extract ionospheric seismic precursory anomalies and capture "deterministic" precursory anomalies requires in-depth development and research on ionospheric disturbance anomaly extraction methods.The Pattern Informatics(PI)method,as a statistical seismology method,has high spatial and temporal resolution and fusion capability,and previous studies have shown that the method can meet the demand for information extraction of spatial and temporal dynamic changes of electromagnetic satellite data.Therefore,the PI method was applied to the ionospheric data processing of CSES in this paper,and we study the influence of parameter selection on the prediction performance of the PI method based on the basic process of processing seismic activity data by PI method,and improve the PI method to obtain the Modified Pattern Informatics(MPI)method after understanding the degree of influence of each parameter on the prediction performance of PI method.After pre-processing the ionospheric data from CSES,the MPI method is used to study the application of seismic prediction using CSES ionospheric data.The main study contents and results are as follows:1.PI method for seismic catalog data(1)When studying the effect of parameter selection on the prediction performance of the PI method,it was found that a large difference in seismicity in the study area would lead to a decrease in the prediction performance of the PI algorithm.This is due to the fact that the calculation process of the PI method involves the normalization of all grid parameters in the study area,and the more seismically active areas would mask the anomalies in the less active areas.(2)Exploring the activity pattern of small earthquakes before strong earthquakes after selecting the optimal parameters and study area reveals that the main activity pattern of small earthquakes before corresponding to large earthquakes shows a spatial tendency to gradually move closer to the epicenter;temporally,the occurrence of small earthquakes gradually increases and shows a distinct period of calm before the occurrence of large earthquakes.2.MPI method for processing ionospheric data of CSES(1)The effects of ascending-track(night)and descending-track(day)data on the calculation results of the MPI method are analyzed through seismic case studies.It was found that the distributions and trends of MPI hotspots calculated using ascending track data and descending track data are approximate,but the evolution process of MPI hotspots calculated using descending track data and the temporal and spatial locations of seismic anomalies are more stable.This may be due to the fact that the ionosphere is affected by solar radiation during the day,ionized particles are denser,and CSES instruments can better monitor changes in plasma parameters.(2)The effects of different window lengths of anomaly learning periods and prediction periods on the MPI calculation results were analyzed by seismic case studies.It is found that the window length selection of anomaly learning time period and prediction time period has a significant effect on the extraction of seismic ionospheric anomaly information by MPI method.This is due to the different smoothing effects of different time window sizes on the data,where a shorter time window suppresses the changes of high-frequency data and makes it difficult to extract a wide range of evolutionary trends.A longer time window suppresses the transformation of high-frequency data,but also over-smoothes the low-frequency data.(3)The retrospective study of some global Mw≥7.0 earthquakes and some Mw≥6.0 strong earthquakes in mainland China in 2019-2021 by the MPI method found that: 1)75% of the earthquake cases have anomalies that last up to 3 days before the earthquake,while 60% of the cases have anomalies one month before the earthquake,and the duration of abnormalities in 80% of cases was less than one month.2)The orientation and distance of the anomalies relative to the epicenter migrate with time.3)Most earthquake examples exist anomaly,which near the direction of the equatorial.4)The MPI method can better identify and extract the ionospheric disturbances caused by the earthquake without other strong seismic disturbances;5)The evolution of the anomalies is as follows: anomalies appear-probability increases-probability decreases-gradually disappear.(4)By comparing and analyzing the calculation results of others anomaly extraction methods and MPI method.It is found that,1)Most of the other anomaly extraction methods adopt single track or partial track data for anomaly extraction,and the location of the tracks involved in the calculation is fixed,which leads to the inability to extract the location and orientation of the most relevant anomaly.The MPI method calculates the anomaly of each grid parameter by constructing a two-dimensional distribution of ionospheric parameters,based on which the MPI method can extract the most relevant anomalies.2)Most of the other anomaly extraction methods select the part of the track near the epicenter in a fixed time interval for calculation,which leads to a poor understanding of the anomaly change process.The MPI method can obtain the continuous anomaly evolution process through calculation,which helps to understand the spatial and temporal patterns of earthquake impact on the ionosphere.Compared with the previous MPI method for DEMETER satellite data,this paper has made some improvements,including 1)Introducing the concept of Moore’s neighborhood from meta-cellular automata and using Moore’s nearest neighbor method instead of kriging interpolation to complete the missing data.2)In order to remove the influence of the "noise" generated by the lower electron density variation on the calculation results,the minimum electron density value 0)(8 and the predicted electron density value 0) are defined based on the continuous variation of electron density in the ionospheric environment,,and the intensity of the electron density variation in the range of 0)(8< 0)< 0) is using to predict the situation of 0)> 0).3)In order to obtain negative anomalies of ionospheric parameters,the calculation process of the exponential function in Equation(4.7)is modified,and the absolute value of the average change of the exponential parameter-electron density is obtained and involved in the probability calculation.4)Normalize the overall MPI results to better capture the evolution of the seismicionospheric disturbances,and select the top 10% hotspots of PI as the anomalies for "noise" eliminating.The above improved MPI method can better respond to the spatial and temporal variation of seismic ionospheric anomalies and help to use the seismicionospheric perturbation information for short-range prediction practice.