- Ground-motion model is:
where Y is in cm∕s2, c

_{0}= 1.03, c_{1}= 0.32, c_{2}= -1.11, h = 7km and σ = 0.34. - Classify stations into four NEHRP categories: A, B, C and D (through a site coefficient, c
_{4}) but find practically no effect so neglect. - Aim to investigate scaling of ground motions for small magnitude earthquakes.
- Most earthquakes have normal mechanisms from aftershock sequences.
- Records from permanent and temporary stations of ITSAK network. Many from EuroSeisTest array.
- Records from ETNA, K2, SSA-1 and SSA-2 plus very few SMA-1 instruments.
- Filter records based on a consideration of signal-to-noise ratio. For digital records use these roll-off and
cut-off frequencies based on magnitude (after studying frequency content of records and applying different
bandpass filters): for 2 ≤ M
_{w}< 3 f_{r}= 0.95Hz and f_{c}= 1.0Hz, for 3 ≤ M_{w}< 4 f_{r}= 0.65Hz and f_{c}= 0.7Hz and for 4 ≤ M_{w}< 5 f_{r}= 0.35 and f_{c}= 0.4Hz. Find that this method adequately removes the noise from the accelerograms used. - Use source parameters computed from high-quality data from local networks. Note that because focal parameters are from different institutes who use different location techniques may mean data set is inhomogeneous.
- Note that errors in phase picking in routine location procedures may lead to less accurate locations (especially focal depths) for small earthquakes as opposed to large earthquakes due to indistinct first arrivals.
- To minimize effects of focal parameter uncertainties, fix h as 7km, which corresponds to average focal depth in Greece and also within dataset used.
- Exclude data from d
_{e}> 40km because only a few (3% of total) records exist for these distances and also to exclude far-field records that are not of interest. - Most records from d
_{e}< 20km and 2.5 ≤ M_{w}≤ 4.5. - Also derive equations using this functional form: log Y = c
_{0}+c_{1}M +c_{2}log(R+c_{3}) where c_{3}was constrained to 6km from an earlier study due to problems in deriving reliable values of c_{2}and c_{3}directly by regression. - Use singular value decomposition for regression following Skarlatoudis et al. (2003).
- Combined dataset with dataset of Skarlatoudis et al. (2003) and regress. Find significant number of
data outside the ±1σ curves. Also plot average residual at each M w.r.t. M and find systematically
underestimation of PGA for M
_{w}≥ 5. Conclude that this shows the insufficiency of a common relation to describe both datasets. - Find no trends in the residuals w.r.t. magnitude or distance.
- Find that the predominant frequencies of PGAs are < 15Hz so believe results not affected by low-pass filtering at 25–27Hz.