- Ground-motion model is:
where Y is in m∕s2, a = -5.984 ± 0.427, b = 2.146 ± 0.069, c = -1.772 ± 0.208, h = 2.511 ± 0.595, d = -0.023 ± 0.011, ϕ = 0.792, τ = 0.829 and σ = 1.147 for data uncorrected for site response and a = -6.514 ± 0.423, b = 1.995 ± 0.085, c = -1.468 ± 0.200, h = 2.490 ± 0.688, d = -0.029 ± 0.010, ϕ = 0.730, τ = 1.079 and σ = 1.303 for data corrected for site response.

- Correct all data in frequency domain to uniform reference site condition (generic rock profile for Switzerland and κ = 0.016s). Derive models from data with and without site correction.
- Data from: geothermal-related [Basel, Switzerland (963 records); Geysers, USA (2328 records); Soultz-sous-Foręts, France (223 records)]; gas-extraction-induced [Roswinkel, Netherlands (61 records)] and natural [Hengill, Iceland (231 records); Voerendaal, Netherlands (162 records)] seismicity. Data from 119 stations used.
- Focal mechanisms of majority of events not known.
- Select earthquakes down to about M
_{w}1 because this is about the magnitude threshold for felt events at Soultz. - Data from mixture of short-period, broadband and, in a few cases, accelerometric instruments.
- Instrument correct records and assess quality based on visual inspection and analysis of signal-to-noise ratios to retain those with ratios above 3. Note that sharp drop off in data from 0.2s upwards so restrict analysis to 0.5s. Exclude records requiring high cut-off frequency of < 10Hz because could be affecting PGA. Compute residuals w.r.t. to model of Bommer et al. (2007) and find some records, predominantly from Geysers, had very low PGAs (more than 100 times smaller) relative to predictions. Because they would hamper analysis they were removed.
- Data from different regions shows poor overlap for some magnitude-distance ranges. Overall range from
5 ≤ r
_{hypo}≤ 20km and 1 ≤ M_{w}≤ 4 well covered. - Recompute M
_{w}based on far-field source spectra to obtain mutually-consistent magnitudes. Check recomputed magnitudes against published M_{w}estimates and find that similar. Derive M_{w}-M_{L}or M_{D}conversion formulae for each region and apply them to convert M_{L}or M_{D}to M_{w}for those events that M_{w}could not be recomputed. Note that this conversion introduces uncertainty into the analysis but increases available data. - Using analysis of variance on data binned into magnitude-distance intervals conclude ground motions from induced and natural earthquakes cannot be statistically distinguished so combine.
- Compare predictions to 55 records (22 events, 13 stations) from a geothermally-active zone at Campi Flegrei (Italy) and find reasonable match, although note the large large scatter in the observations.
- Examine residuals w.r.t. M
_{w}and r_{hypo}and as histograms and find no clear trends. - Examine impact of focal depth on results by deriving models without h and with r
_{epi}rather than r_{hypo}. Find τ slightly reduced when using r_{epi}, which relate to poorly-defined focal depths. Find that focal depth has a strong impact on predictions and hence recommend model using r_{hypo}. - Study residuals w.r.t. region and find data from some regions significantly over or under-estimated by the model.
- Do not recommend model for use for M
_{w}> 3 due to limited magnitude-distance spread of data. - Argue that high values of σ associated with model are not due to inaccuracies in locations or magnitudes
because all records come from well-monitored regions where event locations are well-constrained and M
_{w}have been carefully recomputed. - Derive zone-specific estimates of τ for Basel and Soultz, where sufficient data available.
- Derive single-station ϕ (ϕ
_{SS,S}) for all 62 stations recording ≥ 10 events. Find that ϕ_{SS,S}varies considerably from one station to next. Compute mean ϕ_{SS}.