The earthquake, second by second

What is real here, and what is simplified

• The fault and the rupture. The Main Marmara Fault as mapped in GEM's Global Active Faults database. Rupture length and width grow with magnitude (Wells & Coppersmith 1994: 29 km at Mw 6.5, 120 km at 7.5, 160 km at 7.7), centred on the Kumburgaz segment south of Istanbul, breaking both ways at 2.8 km/s.
• The waves. P waves at 6 km/s and S waves at 3.5 km/s from a 10 km deep hypocentre, so the first jolt reaches central Istanbul 5–9 s after the fault starts to break and the strong shaking 3–7 s later. That gap is what an earthquake early-warning system has to work with. Shaking lasts longer for bigger ruptures and further away (an approximation: source duration plus 0.1 s per km).
• How hard it shakes. Two ground-motion models (Akkar et al. 2014, Boore et al. 2014) through OpenQuake, 80 simulated earthquakes per magnitude with spatially correlated variability, site conditions from topographic slope. Colours show the mean intensity per kilometre square.
• Which buildings fail. 1.2 million real building footprints with building types from ESRM20 and İBB's age and height data. Each building's chance of complete or extensive damage is averaged over the 80 earthquakes, and one plausible outcome is drawn from those chances; each drawn building fails at a random moment while it shakes. Run it again with another magnitude and different buildings fail.
• Streets and hospitals. 20,700 km of OpenStreetMap roads. A collapsing building throws debris about half its height into the street; a segment counts as blocked when less than 3 m stay open, from the moment its first neighbour collapses. Travel times to the nearest of 417 hospitals use typical speeds per road type on what is left of the network, without traffic. "Cut off" marks squares normally within 30 minutes of a hospital that are more than 30 minutes away afterwards. Hospitals themselves are not damaged in this model.
• Calibration. Uncalibrated, ESRM20's fragility curves predict about four times İBB's damage estimate. Here all curves are stretched by one factor (2.3×) so that Mw 7.5 matches İBB's 46,600 very heavily or heavily damaged buildings; the other magnitudes keep that factor. Deaths follow from the same models and are not calibrated; they come out below İBB's 13,900 for Mw 7.5.
• Not shown: tsunami, landslides, liquefaction, fires, aftershocks, bridges and tunnels. The seismogram is an illustration scaled to the modelled peak acceleration, not a simulated record.