- Ground-motion model is (MOD1 and MOD3):
where: Y is in m∕s2, a = -2.666 ± 0.072, b = 1.158 ± 0.020, c = -2.312 ± 0.055, h = 1.734 ± 0.197, σ

_{inter-event}= 0.142, σ_{intra-event}= 0.358 and σ = 0.385 for MOD1 without site term and a = -2.710 ± 0.064, b = 1.165 ± 0.021, c = -2.244 ± 0.044, h = 1.779 ± 0.158, e = 0.225 ± 0.004, σ_{inter-event}= 0.151, σ_{intra-event}= 0.276 and σ = 0.315 for MOD3 with site term. - Use 3 site classes using approach of Emolo et al. (2011) (see Section 2.341):
- s = 1
- Site with significant positive mean residual.
- s = 0
- Site with mean residual not significantly different from zero.
- s = -1
- Site with significant negative mean residual.

Derive site terms for all stations in second step, based on residuals w.r.t. to overall model derived without considering site effects. Only statistically-significant terms (using Z-test and 5% level) are retained. Assign site corrections of either -1, 0 or 1 based on residuals. Find reduction in ϕ but not in τ when applying site correction.

- Use data from 29 stations (with 104 to 211 records each) of dense (20 × 10km
^{2}) Lawrence Berkeley National Laboratory/Calpine-Geysers network (01/09/2007–15/11/2010) of induced (by water injection and steam extraction) seismicity at The Geysers vapour-dominated geothermal field (N. California, USA). Instruments I/O Sensor SM-6 geophones (f_{0}= 14Hz) then Oyo GS-11D 4.5Hz sensors. - Select earthquakes with focal depths ≤ 5km since deep events assumed to be natural earthquakes.
- Select only waveforms with signal-to-noise ratio greater than 10 in frequency range 0.5–35Hz.
- Instrument correct data. Remove mean and trend baselines. Filter using zero-phase 4-pole Butterworth filter with passband 0.7–35Hz. Extract parameters in time window starting at origin time and ending at time corresponding to 98% of total energy in waveform, which were tapered with 0.1 taper width with a cosine window. Finally differentiated and filtered again to obtain accelerations.
- Test another functional form including term for anelastic attenuation, for which report coefficients. Choose
final model due to: its simplicity, σ and R
^{2}. - Examine inter-event and intra-event residuals (as scatter plots and histograms) w.r.t. M
_{w}, r_{hypo}and for each station separately. Find no significant trends based on fitting lines to residuals. - Examine standard deviations at each station before and after site correction (single-station σ, ϕ
_{SS}, analysis). Compute average value weighted by number of records per station, ϕ_{SS}. Compare to overall total σ and find single-station σ is lower (0.2410 compared to 0.3849) and that the different between them is smaller when using MOD3 (0.2377 compared to 0.3152). For one station (CLV) find ϕ_{SS,S}is higher than σ, which relate to local geological conditions. - Find slight positive trend in inter-event residuals w.r.t. focal depth, which relate to possible decrease in stress with depth or variation in rigidity modulus due to heterogeneities in highly-fractured medium.