The discussed paper presents a scaling model for Absolute Acceleration Spectra in Europe, to be used in seismic hazard analyses. The discusser will comment on the following aspects of this paper: (1) uniformity of the data set used, (2) applied data processing techniques, (3) consistency in classification of the local site conditions, (4) adequacy of local soil classification used, (5) magnitude scale used, (6) assumed magnitude dependence, (7) some trends in the distribution of the residuals for different types of sites, (8) agreement with results by other authors for a region in Europe (territory of former Yugoslavia) and ( 9) some aspects related to use of scaling laws for response spectra for seismic hazard calculations (e.g. need for a wider period range of the spectra and for analytic approximation of the distribution of the residuals).
(1) The authors state that they use 'large and uniform data set' (42 per cent of their data is from Italy, 19 per cent from Greece, 13 per cent from the former Soviet Union, 12 per cent from the former Yugoslavia, and the remaining 14 per cent from Algeria, Azores, Bulgaria, Iran, Portugal and Turkey). They use a total of 422 records from 157 earthquakes. Trifunac and Todorovska and Trifunac et al. described the differences in attenuation of intensity scales for natural seismological zones of the Balkan Peninsula (Southeast Europe), proposed by Shebalin. The observed differences for these regions are significant, but as yet not fully researched and understood. Thus, it is not likely that attenuation of strong motion in Italy, Greece, former Yugoslavia, and former Soviet Union can be described by one common law. Furthermore, it is difficult to describe the attenuation of strong motion by only a frequency-dependent exponent of source to station distance. Other important factors which influence the attenuation function are e.g. the 'size' of the seismogenic zone (distributions of fault width, length and depth and orientation of the fault plane) and the stress in the source region, which could be modelled via the source dimension S and the coherence length S
. The attenuation law of the form rL, where n is a frequency-dependent exponent, is inadequate for small and for large r (e.g. see Figures 6 and7 in the discussed paper) and cannot be used to describe near-field, intermediate field, far-field and surface wave attenuation, all by a single value of n. Thus, the data used by Ambraseys et al. is not uniform with respect to the different attenuation laws which apply in the regions contributing the data. Strong motion data recorded in former Yugoslavia (325 three-component records from 183 earthquakes) has been used to develop a frequency dependent attenuation function and scaling laws of response spectra specific for that region. Correlation of peak amplitudes of strong motion with one of the intensity scales for that region (Mercalli-Cancani-Sieberg scale) and frequency-dependent duration of strong ground motion have also been studied using the same database. All of these studies -discuss the differences between attenuation of strong ground motion in California and in former Yugoslavia. Some of the differences have been interpreted in terms of different soil and geologic settings of the recording stations.
(2) The acceleration data used by the authors should have been corrected for the instrument response (using nominal values for the instrument constants). The natural frequency of AR-240 accelerographs is