- Ground-motion model is [based on base model of Abrahamson and Silva (1997, 2008)]:
_{1}= 8.92418, a_{2}= -0.513, a_{3}= -0.695, a_{4}= -0.18555, a_{5}= -1.25594, a_{6}= 0.18105, a_{7}= 7.33617, a_{8}= -0.02125, a_{9}= 0.01851, σ = 0.6527 (intra-event), τ = 0.5163 (inter-event) and σ_{Tot}= = 0.8322 and b_{lin}= -0.36, b_{1}= -0.64 and b_{2}= -0.14 [taken from Boore and Atkinson (2008)]. Fix c_{1}= 6.5. pga4nl is predicted PGA in g for V_{s,30}= 760m∕s. See Boore and Atkinson (2008) for b_{nl}, c and d [not repeated by Akkar and Çağnan (2010)]. - Characterise sites using V
_{s,30}and use the site response terms of Boore and Atkinson (2008) because of their simplicity and fairly good performance for data (demonstrated by intra-event residual plots and their distributions that do not show clear trends, except perhaps for V_{s,30}> 720m∕s). Majority of records from NEHRP C (360 ≤ V_{s,30}≤ 760m∕s) and D (180 ≤ V_{s,30}< 360m∕s) sites with very few from sites with V_{S30}≥ 760m∕s. All sites have measured V_{s,30}values. - Use three faulting mechanisms:
- Normal
- F
_{N}= 1, F_{R}= 0. 28% of records. - Strike-slip
- F
_{N}= 0, F_{R}= 0. 70% of records. - Reverse/thrust
- F
_{N}= 0, F_{R}= 1. 2% of records.

- Focal depths between about 0 and 50km with most between 5 and 20km.
- Use data from the recently compiled Turkish strong-motion database (Akkar et al., 2010), for which the independent parameters were carefully reassessed.
- Note that there are many singly-recorded earthquakes.
- Vast majority of data from M
_{w}< 6 and r_{jb}> 10km. - Explore several functional forms (not shown). Try to keep balance between rigorous model (for meaningful and reliable estimations) and a robust expression (for wider implementation in engineering applications).
- Data from 102 mainshocks (346 records) and 35 aftershocks (88 records).
- Bandpass filter records using method of Akkar and Bommer (2006).
- Compare PGAs from unfiltered and filter records and find negligible differences.
- Note that aim of study is not to promote the use of poorly-constrained local models.
- Use pure error analysis (Douglas and Smit, 2001) to investigate magnitude-dependence of σ. Find strong dependence of results on binning strategy (including some bins that suggest increase in σ with magnitude) and, therefore, disregard magnitude dependency.
- Derive GMPEs using data with minimum thresholds of M
_{w}3.5, M_{w}4.0, M_{w}4.5 and M_{w}5.0 to study influence of small-magnitude data on predictions. Find that equation using M_{w}5.0 threshold overestimates PGAs derived using lower thresholds; however, ranking of predictions from GMPEs using thresholds of M_{w}3.5, M_{w}4.0 and M_{w}4.5 is not systematic. - Note that due to limited records from reverse-faulting earthquakes, the coefficient a
_{9}needs refining using additional data. - Examine inter-event residuals for PGA, 0.2s and 1s w.r.t. M
_{w}and intra-event residuals w.r.t. r_{jb}and V_{s,30}. Fit straight lines to residuals and also compute bias over ranges of independent variables. Test significance of trends at 5% level. Find no significant bias w.r.t. M_{w}nor w.r.t. r_{jb}. For V_{s,30}for 1s find significant overestimation for V_{s,30}> 450m∕s, which relate to linear site term. Suggest linear site term needs adjustment using Turkish data. - Compute inter-station residuals and identify 9 outlier stations, which are those with residuals mainly outside range generally observed.
- Examine bias of residuals for mainshock and aftershock records. Find weak evidence for overestimation of aftershock motions but this is not significant at the 5% level.
- Combine Turkish and Italian data from ITACA (1004 records) and derive GMPEs using same functional
form, except using site classes rather than V
_{s,30}directly, to test observed differences between local and global GMPEs. - Compare focal depth distributions, using histograms with intervals of 5km, of the datasets for various
GMPEs. Compute mean and standard deviations of M
_{w}for each depth bin. Find that records from Turkey and Italian are on average deeper than those for other GMPEs, which seems to explain lower observed motions. Conclude that focal depth can be important in explaining regional differences.