@article{epos1518,
          volume = {102},
          number = {B11},
           month = {November},
          author = {Mariana Eneva and Yehuda Ben-Zion},
           title = {Application of pattern recognition techniques to earthquake catalogs generated by model of segmented fault systems in three-dimensional elastic solids},
       publisher = {American Geophysical Union},
            year = {1997},
         journal = {Journal of Geophysical Research},
           pages = {24513--24528},
             url = {https://episodesplatform.eu/eprints/1518/},
        abstract = {Techniques and seismicity parameters described by Eneva and Ben-Zion [ 1997] 
are used to examine synthetic earthquake catalogs generated by Ben-Zion [ 1996] for precur- 
sory patterns of large model events. Different model realizations  represent various levels of 
fault zone disorder. These include models with uniform properties (U), a Parkfield-type as- 
perity (A), fractal brittle properties (F), and multi-size-scale heterogeneities  (M). The seis- 
micity parameters used are based on information contained  in typical earthquake catalogs 
reflecting earthquake distribution  in space, time, and size. The analysis highlights the com- 
plexity of the information content of the synthetic earthquake catalogs. Simple repetitive 
precursory signals have not been found. However, local extrema in the examined parameters 
are found to have significant association  in time with large events. Thus our techniques and 
parameters may be useful for intermediate-term  earthquake prediction, especially when pa- 
rameters are used in combinations. Some analysis results are the same for all model realiza- 
tions and some depend on the model. Features characterizing all catalogs are as follows: (1) 
Large model events are statistically predictable on the basis of patterns in the distribution of 
smaller events. (2) For a given parameter, the type of precursory extrema (maxima or min- 
ima) is the same for all models. (3) The interparameter correlation for any parameter pair has 
the same sign (positiv e or negative) in all models. (4) The large events are neither slip- nor 
time-predictable based on previous large events. Results that differ from model to model 
include the following: (1) The degree of predictability of large events correlates with the 
degree of regularity in the assumed  fault properties, following the order U, F, A, and M. (2) 
There is no one-to-one correlation between type of precursory extrema (maxima or minima) 
and type of precursory  trends (increase or decrease);  this produces great variations in observ- 
able trends for any given parameter, both from model to model and for different events in the 
same model. (3) The interparameter  correlations  vary among models, with the highest corre- 
lations in model F. Most discussed patterns are in agreement with observations  from seismi- 
cally active zones, laboratory models, and mining-induced seismicity.}
}