Estimation of earthquake magnitudes from epicentral intensities and other focal parameters in Central and Southern Europe

• Published in: Geophysical journal international Vol. 151; no. 3; pp. 824 - 834
• Main Authors: Gutdeutsch, Rolf, Kaiser, Diethelm, Jentzsch, Gerhard
• Format: Journal Article
• Language: English
• Published: Oxford, UK Blackwell Publishing Ltd 01-12-2002; Blackwell Science Ltd
• Subjects: correlation coefficient; earthquake catalogues; Europe; local magnitude; magnitude–intensity relations; orthogonal regression; surface wave magnitude
• ISSN: 0956-540X; 1365-246X
• DOI: 10.1046/j.1365-246X.2002.01804.x

The publication of an earthquake catalogue by Kárník in 1996 (a continuation and revision of an earlier one (1969)) makes important data available covering one century of the seismic history of Central and Southern Europe. It allows us to study in detail empirical relations between the magnitude and other focal parameters. In this study well-known relations combining two or three focal parameters, M=A+BI0+C log(H), Ms=D+EI0+F log(H), ML=G+OI0+P log(H), ML=Q+RMs+Slog(H), are investigated (M, Kárník's magnitude; ML, local magnitude; Ms, surface wave magnitude; I0, epicentral intensity; H, focal depth in kilometres). The data show a considerable scatter with respect to the relations above. The relations are considered useful, if the following significance criteria are fulfilled. (1) The data sets comprise a minimum of 20 entries.(2) The partial correlation between the two most important parameters is greater than 70 per cent. (3) The parameter of least importance still influences the correlation of the others by more than 5 per cent. The partial correlation coefficients help to decide whether the data are to be rejected as insufficient for the regression analysis or to determine the level beyond which it is useful to perform a regression analysis excluding the parameter of lowest importance. Two kinds of regression are carried out: (1) standard linear regression assumes that only M or ML, respectively, are in error, while the remaining two parameters are error-free. (2) Orthogonal regression assumes that all three parameters have errors. This is the case for the data in the catalogue used here. The orthogonal regression M=−1.682 + 0.654 I0+ 1.868 log (H), with a standard deviation of ±0.284, differs considerably from Kárník's empirical relation M= 0.5 I0+ log (H) + 0.35 for shallow foci, but agrees well with the results of earlier studies by the authors for earthquakes in SE Europe. The data set M, I0, H (for H < 50 km) fails criterion (3). The orthogonal least-squares fit without log (H) has been found as follows: Ms= 0.550 I0+ 1.260, with a standard deviation of ±0.412. We observe systematic regional deviations from this relationship, which need further investigation. The correlation analysis shows that ML and Ms are weakly linked with log (H), but the correlation between ML and Ms is very high (93 per cent). Therefore, the orthogonal relation between ML and Ms without the log (H) term was chosen: ML= 0.664 + 0.893 Ms, with a standard deviation of ±0.163. The correlations between ML, I0 and log (H) do not fulfil the significance criteria. For the purpose of earthquake hazard analysis the orthogonal regression visualizes simultaneously the errors of all input data, i.e. δMLi, δMi and δ log (Hi). Our new relationships result from orthogonal regression analysis using a large high-quality data set. They should be applicable in Central and Southern Europe unless there are regional relationships available that fit the data better.