Task 5 - Towards the European Catalogue
of Damaging Earthquakes
The goal of Task 5 is to review the procedures according to which earthquake parameters are determined from macroseismic data in Europe, to assess common procedures and to use them to determine the parameters of the earthquakes listed in the upgraded root file of damaging earthquakes (URfDE).
Parameter determination in the input PEC
As mentioned before, the quality of the supporting data sets (roots) heavily influences the parameters. However, the determination procedures represent the next main reason of inhomogeneity of parameters found in the PEC, which warns against the current procedures of merging inhomogeneous data sets. Some examples of different parameters derived from the same root taken from Camassi et al. (1998), are shown below:
same root, different Io, similar location
Or Ye Mo Da Ho Ax Rt Rc Ix Io Lat Lon GRU88 1553 08 17 19 TORGAU SIE 2C 0 70 51.600 13.000 LEY86 1553 08 17 19 TORGAU S40 2C 0 80 51.580 13.000The root for both entries is Sieberg, 1940:
Or Ye Mo Da Ho Ax Rt Rc Ix Io Lat Lon LAB95 1703 07 28 Gelnica RET52 2B 0 80 I 48.860 20.970 ZSA88 1703 07 28 GOLNICBANYA 1 2C 0 65 L 48.850 20.930 LAB95 1830 07 11 09 CENTR. SLOVAKIA RET52 2B 0 70 H 48.750 19.350 ZSA88 1830 07 11 10 LIBETBANYA 1 2C 0 45 L 48.750 19.370 LAB95 1855 01 31 12 CENTR. SLOVAKIA RET52 2A 0 65 H 48.460 18.960 ZSA88 1855 01 31 13 SELMECBANYA 1 2B 0 50 L 48.450 18.900
The root for both entries in the three cases is Rethly, 1952.
same root, similar Io, different location
Or Ye Mo Da Ho Ax Rt Rc Ix Io Lat Lon LAB95 1652 03 07 Kosice RET52 2B 0 50 48.720 21.270 ZSA88 1652 03 07 N HUNGARY 1 2A 0 60 K 48.800 20.000 LAB95 1852 11 15 22 PERNEK - MODRA RET52 2A 0 60 H 48.640 17.160 ZSA88 1852 11 15 23 SASVAR 1 2A 0 55 L 48.630 19.150The root for both entries is Rethly, 1952.
This topic is widely analysed in the paper by Cecic et al., 1997. Most PEC say that the epicentral co-ordinates are determined as the barycentre of the highest degree isoseismal; however many doubts arise because isoseismals are available only for a limited percentage of the events.
The main problems obviously arise for earthquakes with a limited number of data or with uneven data distributions (offshore source, frontier areas events studied by one side only, and so on). In some of the first cases it appears that the authors were tempted of "calibrating" the location of the event against other, presumed similar, better known ones, locating therefore the earthquake "where it should be located" according to the known seismicity. Obviously this is a biased procedure which might lead to inconsistent determinations.
However, even in the case that data are available in a significant number, unformalized procedures lead to different determinations.
Location accuracy is also determined according to varied definitions and given in the 55% of the cases over the total number of the WF entries. Tab. 5.1 summarises the situation of the location accuracy estimates in the input PEC.
Depth determination is a controversial problem, as many authors do not trust in the current (few) available procedures, which are all based on the Kovesligethy-Blake-Sponheuer method.
In the PEC the depth is determined in most catalogues: the exceptions are Lambert et al. (1996), Mezcua and Martinez Solarez (1993), Sulstarova and Kociu (1975), while Papazachos and Papazachou (1989) report the values N (normal) or I (intermediate) only.
However, depth determinations are given for the 16% of the total number of entries only, with percentages in the single PEC varying from 100 to 1 (Tab. 5.1); in many cases the depth corresponds to some fixed values adopted by the compilers (see for instance Fig. 5.1).
Some information on the situation of depth in the current PEC, including accuracy codes, their values and number of occurrences, are given in Tab. 5.1.
Definition and determination procedures of Io are of varied type. From Stucchi and Bonnin (1995) it can be inferred that for some PEC Io always represents the maximum observed intensity Ix, while for others the relation between Io and Ix is more flexible; in these cases Io represents a potential value of I extrapolated to the epicentre according to hardly reported procedures.
Tab. 5.2 shows that only 5 PEC give the two parameters Io and Ix; that both figures are reported for a limited number of cases; and that the differences among them ranges between +3 and -2/3. The PEC by Musson (1994) gives only Ix.
Most PEC use uncertain values of Io, such as 6/7, 9/10, etc., usually interpreted as intermediate figures (6.5, 9.5); some do not (see for instance Fig. 5.2). The situation of the accuracy estimates is also given in Tab. 5.2.
For the time-window 1400-1899 all magnitudes are obviously determined from macroseismic data. Therefore, all magnitude figures from all PEC were recompiled in the BEECD WF in the macroseismic magnitude (Mm) parameter, no matter whether they were declared as Ms calibrated (Costantinescu and Marza, 1982) or whether their calibration was not declared at all. The only exception is the PEC by Musson (1994), which gives ML calibrated magnitudes. The PEC by Shebalin et al. (1974), Sulstarova and Kociu (1975), Mezcua and Martinez Solares (1993), Leydecker (1986), Lambert et al. (1996) do not supply magnitude values.
Tab 5.3 summarises the procedures according to which M is determined, as reported from the PEC and/or as inferred from the analysis of the data. Determination procedures belong to two main families: a) correlating M with the area of some low degree (such as III or IV) isoseismal (see for instance Musson, 1994; Ahjos and Uski, 1992 - partly); b) correlating M with Io (the others). Depth is considered by relations of type M/Io only.
In Fig. 5.3 three families of the relations presented in Tab. 5.3 are plotted together (without depth; with depth fixed at two values). The relations of the first type give similar values of M for Io = 5, while they diverge for higher values of Io; only the relation used by Costantinescu and Marza (1992) gives definitely higher values. Relations of the second type clearly show that for events up to 15 km deep the contribution of depth is moderate, while the behaviour of relations for events with depth of 35 km noticeable rises the value of corresponding M.
In the reality, the application of the relations does not appear so straightforward. Fig. 5.4 and Tab. 5.3 shows that figures found in some PEC actually come from the declared procedures, while they do not in some others. In Fig. 5.4 the data obtained using relations with depth are reported in green (open squares); for two catalogues (COM82 and ZSA88) depth values are also reported. Magnitudes calculated without depth are given in red (open circles); in some cases they follow the relations declared by the corresponding catalogue (as ZSA88) in some others they do not (as in COM82 and RIB82). Some of these misfits could be related to the fact that, for some events, earthquake parameters, including M, come directly from other parametric catalogue.
Once again, the situation of accuracy estimates are given in Tab. 5.2, which also shows that only two PEC (Kondorskaya and Shebalin, 1982; Labak, 1995) give accuracy estimates for M.
Proposed procedures of parameter determination
After analysing the current procedures, the matter was discussed in the Milano meeting (April 1997): here follows a short review of the main parameters and the proposed solutions.
It is the time given by the study (root), in the sense that it is the author's decision to assess the time of the earthquake, to which all earthquake records are to be referred; this holds for roots of type 1 to 3. Though conversion to UTM can be useful, original time can be of some importance as well, provided that it is clear in which time-system it comes; therefore, both original and UTM times can be reported.
The first issue concerning epicentre determination is the choice between attempting to distinguish between macroseismic epicentre (sensu stricto) and "barycentre" (sensu Cecic et al., 1996). In most cases of historical earthquakes these are likely to be the same, and in those cases where they are not, probably the distinguishing information is not likely to be something that is found in the intensity database (e.g. information on surface rupture, knowledge about anisotropic attenuation).
Type 1 roots. The following methods could be used for determining co-ordinates:
It appears that most people had used method (i) in some form, e.g. arithmetic
mean of data points of highest intensity; this is not surprising given that
this is the simplest approach.
A new solution consists in the algorythm proposed by Gasperini and Ferrari (1995), with some post-computation review in the most critical cases (poor data sets, offshore earthquakes, etc.). This review can include also comparison with the centroid of isoseismals, because the previous result may be adversely affected by clustering of datapoints due to population distribution.
In the case of events with very poor data sets, e.g. one datapoint, it can be suggested that, if the epicentre is assigned to the single datapoint automatically, such cases needed to be flagged in a very clear way.
Type 2 and 3 roots. In these cases it is necessary to use the centroid of isoseismals, provided that it is possible to draw them without affecting the raw data by inserting too large personal judgements. In the case this is impossible to avoid it seems better to produce a very rough and preliminary intensity distribution ant to follow the procedure proposed for type 1 roots.
Type 4 and 5 roots. There is no way other than to adopt the location proposed by the input PEC.
There are not so many options: the Kovesligethy-Blake-Sponheuer method is well established, though the limits are known. One issue is whether Kovesligethy-type methods should be applied to raw data points or isoseismal radii/areas. Some studies had adopted the first method to avoid the possible problem of subjectivity in drawing isoseismals, though someone argues that irregularities in the distribution of data points would distort the results.
The chief issues is to decide if certain parameters in the equation should be modelled by region or by earthquake, and whether the calculations should be done in a mechanistic way or in a way that allowed some choices to the investigator.
As mentioned above, epicentral intensity Io values is determined as observed maximum values or extrapolated values, mostly according to personal judgement.
An approach for type 1 roots where Io is derived critically from Ix is found in Camassi and Stucchi (1997). Here Ix represent the maximum observed intensity (with the exception of the values assigned to territories or single buildings, not taken into account for assessing this parameter), while Io is determined with the purpose of giving a value suitable for size determination. The result is that 267 earthquakes, out of a set of 950 with intensity data, got Io different from Ix. For 28 events Io was given a value > Ix; they correspond to earthquakes whose data distribution is highly asymmetric (mostly offshore or border areas events). For 239 events Io was given < Ix; they correspond to cases where the number of datapoints with Ix was not statistically significant and the next lower intensities were below Ix-1.
For type 2 and 3 it is suggested that Ix is assessed in a preliminary way and Io thereafter.
Mm can be:
i) derived from Io
ii) derived from Io and isoseismal radii
iii) derived from isoseismal radii or felt area
iv) derived from processing directly the intensity datapoints.
Further than the equations mentioned above
(Tab. 5.3) and
other similar, an example of a method of type i) was adopted recently
in Italy (Rebez e Stucchi, 1997), showing that it was possible to calculate
mean magnitude (Ms) and standard deviation for each value of Io. This could
be applied as a look-up table rather than as a linear correlation, since
there did not seem to be a linear relation for the whole range of I values.
Obviously, that correlation was established between Io and Ms data of the
The question is, whether to try and go for one European correlation or for regional ones. Fig. 5.5 gives the plot of all couples Io/Ms in the WF (1400-1899), the mean value and the error. Also the errors are not that large, the dispersion of the data (notice that each point in the plot can correspond to tens of couples with the same values !) suggests that it might be better to fit regional correlations.
Method of type iii) has many references: part of them is referenced in Tab. 5.3. Ambraseys derived, performing regression analysis from 520 isoseismals of about 9600 intensity datapoints associated with 163 shallow (h<26 km) earthquakes in the Balkans, the following relation
Ms = -1.41 + 0.65 (Ii) + 0.0033 (Ri) + 2.03 log (Ri) + 0.32 P
where Ri is the hypocentral distance, i.e. R = r2 + h2,
and h is the focal depth in km. P is = 0 for mean values and 1 for 84
As the use of this equation requires prior knowledge of h, a mean depth can be calculated as a part of the regression. Ambraseys obtained ho = 9.7 km, which gives the following final relation:
Ms = -1.54 + 0.65 (Ii) + 0.0029 (Ri) + 2.14 log (Ri) + 0.32 P
Method of type iii) also carries the already mentioned problems related
to isoseismals and is likely to be applied in about 30% of cases only. To
overcome this, R. Musson suggests that, once correlations between magnitude
and various isoseismal radii (or attenuation equations) had been established
for the events where the data were good, a series of templates could be prepared
for expected isoseismal radii of events of magnitude 4, 5, 6 and 7. For any
data set where it was impossible to draw isoseismals, the data could be compared
to the templates and the lowest value that provided a credible fit could
be selected to give a minimum credible magnitude (MCM). This would then provide
a classification of events into those at least magnitude 4, those at least
magnitude 5, and so on.
Such a classification could be entered using integer values so that magnitude 4 would indicate an MCM of 4, while 4.0 would indicate a calculated magnitude from a good data set. In some cases it would be possible to estimate an upper bound as well as a lower one, and that magnitudes could be given as 4-5 or 4-6.
An example of the method of type iv) was proposed and used by Gasperini and Ferrari (1995); it is based on a regional calibration against Ms and it also requires some post-computation review in the most critical cases. This method can be adopted for BEECD roots of type 1, with critical comparison with the results obtained from the other methods, provided that the problem of calibration is solved.
A problem common to all methods discussed is to which type of magnitude (Ms, ML, Mw) the calibration should be done.
It is a common practice in different countries to set some parameter first and then derive others. Some examples:
A new approach should try to compute all the parameters simultaneously. Since
non-linear inversion are involved, the quality/quantity of the macrosesimic
data might not be sufficient, and thus some kind of sorting of methodologies
according to data quality has to be envisaged.
Thus, for the parameterisation of BEECD data, the following possible approaches, sorted in decreasing order of preference, are suggested:
Although updated methods are available, as shown above, some of them have been scarcely tested. To proceed a step forward, two tests were performed on earthquakes with datapoints.
Test 1: three earthquakes from varied regions, "weighted mean"
A sample study was performed in order to check the stability of the most common methods for the estimate of epicentral parameters on earthquakes coming from different areas. The selected events are:
1. the mainshock of the Calabrian earthquake sequence of 1783
2. the Atalanti, Greece earthquake, Greece, 1894
3. the Cagli (Central Italy) earthquake of 1781
4. the Colchester, UK, earthquake of 1884
They represent very different tectonic and geographic setting, and the
distribution of intensity points is varying accordingly. Io ranges from IX
(case 1) to VIII (case 4).
To determine earthquake parameters two different methodologies have been adopted.
The first one determines the baricenter separately for each intensity class. The second one determines the epicentral co-ordinates as the weighted mean of all the points. The weight is given by the CRAM attenuation relationship (Magri et al., 1994) in the form
Is = Io + a - b*D (1/3)
where Is is the site intensity, Io is the epicentral intensity, D is the epicentral distance and a, b are parameters. The parameter a has the effect of a correction of the epicentral intensity, while b modulates the decay with distance. Two possibility are given: either a and b are fixed to a known value or they become two more parameters to be determined beside the epicentral co-ordinates (2 or 4 free parameters model). Fig. 5.6 to 5.8 show some examples of parameterisation of the studied events.
First of all it is clear that if the earthquake occurs near to the coastline,
the baricentres of the different intensity distributions are not coincident;
therefore, the first method does not seem to be very robust.
On the contrary, the weighted average provides a much more "reasonable" result (i.e. it gives an epicentre where most of the expert would put it in a visual localisation). It is interesting to note that the area comprised by the 95% variation of the parameters for the CRAM model is few km. This area is obviously smaller when also the a and b parameters are set free. The problem is that these parameters appears to be quite different for the four cases considered.
5.9 shows a graph comprising the four couples of values
with their 95% confidence interval and superimposed the values derived for
about 40.000 data points of the Italian macroseismic data base (Monachesi
and Stucchi, 1997) and a regression line. It can be noted the striking
correlation between a and b values. This is a similar phenomenon
to the one pointed out for the M/I relationship by Mucciarelli (1998), indicating
a possible statistical bias due to a pivotal behaviour.
Unfortunately, the trade-off between the two parameters masks the possibility of discriminating between the two factors that could explain the diversity between values for earthquakes coming from different zones:
Test 2: main earthquakes in Italy 1400 - 1599, Gasperini and Ferrari (1997)
The methodology proposed by Gasperini and Ferrari (1995 & 1997) was used to determine the parameters of the strong earthquakes of the Italian area, 1400-1599, taken from the R roots listed in the "upgraded root file of strong earthquakes" (URfSE): in all, 18 earthquakes with Ix ranging from 5 to 11 and a number of datapoints from 2 to 200.
The parameters determined are Lat, Lon, Io, Mm; determinations are given in Tab. 5.4, compared with the initial parameters.
Some major changes are visible, mostly on Io (earthquakes of 1403, 1410, 1473) where the new value is almost always greater than the old one, with the exception of the events of 1501, 1542 and 1564. The location changes in a few cases (mostly 1403, 1410, 1414, 1511), while M also undergoes some drastic changes. It has to be stressed, however, that the main changes are due to both the improvement of the roots and to the use of the new methodology.
Updated methods largely agreed are available in the European scientific community at the end of the project, yet. Therefore, the project partners could not reach a steady consensus among themselves on which methods should be adopted.
Moreover, it must be considered that most procedures proposed recently (e.g. Gasperini and Ferrari, 1995 & 1997) adopt formalised methods which apply to roots of type 1 only, while the compilation of the final catalogue must necessarily adopt solutions also for roots of the other types.
Nevertheless, the project has shown some possible ways. The compilation of the final catalogue will be accomplished in a few months, putting the data on the project web site (http://emidius.mi.cnr.it/BEECD/home.html), formulating possible solutions and asking former partners and other compilers to give their agreement or alternatives.
Part of this task, with main references to the strong earthquakes, will be achieved in the frame of the recently started ENVIRONMENT project namedFAUST (FAUlts as a Seismologists' Tool), already active at the Internet address http://faust.ismes.it. One of the aim of this project is to provide an association with seismogenic fault for all the European events with M > 6. This will be done using very recent and more advanced techniques to derive information about the geometry of active faults from intensity data points (Gasperini et al., 1999; Mendez et al., 1996) overcoming the concept of catalogues of point sources.