Multi-fault earthquakes are increasingly recognised as a common occurrence but remain poorly understood. For example, the 2016 Kaikōura earthquake in New Zealand ruptured a series of >20 faults or fault segments, raising questions whether empirical ‘rules’ for fault rupture complexity, established from global earthquake catalogues, apply to structurally complex regions. In this thesis, I modelled empirically-derived relative likelihoods of different surface rupturing earthquakes in the Kaikōura region to test how much of an outlier the 2016 event was compared to historical earthquakes globally.

The major contribution of this study is a new method for generating multi-fault rupture scenarios. Five different empirical co-rupture parameters from historical surface-rupturing earthquakes were used to model the relative likelihood of different rupture pathways: angular change, jump distance, fault type change, step over type, and the cumulative number of steps. The parameters were combined to quantify the passing ratios between adjacent fault sections and run through a Monte Carlo-type algorithm called an ‘empirical rupture simulator’ (ERS). The ERS starts on a predefined fault section or ‘seed fault’ and tests the passing ratios at active ends of the rupture in each iteration against a randomly generated number. If a connection passes, the rupture progresses to a new section and repeats until all possible adjacent sections have failed, stopping that rupture scenario. The ERS looped 250,000 times to build a complete rupture set before counting duplicate ruptures to determine relative likelihoods (occurrences per number of trials) of unique scenarios and classes of scenarios.

The ERS was tested on three seed sections: the epicentral section of The Humps fault in the 2016 Kaikōura earthquake, the western section of the Conway segment of the Hope fault, and the epicentral section of the San Jacinto fault in the 1812 San Juan Capistrano earthquake (Southern California). Each of the seed section cases tests an aspect of the ERS compared to characteristics of the fault systems derived from physical modelling and/or paleoseismology. For example, The Humps fault simulation tests if the model can replicate the 2016 Kaikōura earthquake. The Hope fault simulation tests the rupture segmentation and magnitude ranges against paleoseismic studies. Finally, the Southern California simulation tests the empirically derived passing ratios against a dense network of paleoseismic studies and physical models.

Most simulated ruptures originating on The Humps fault remained on The Humps, with the most common connections being to the Leader and Mt. Culverden faults. The expected magnitude range (based on length alone) of a Humps fault rupture is a MW 6.3-7.7. The most common magnitudes were MW 6.3 (single section) and MW 6.7. The ruptures along the Conway segment of the Hope fault indicate that the most likely scenario is a full-length segment rupture. The Hanmer Basin inhibited 99% of ruptures, and the eastern splay faults (Jordan, Kowhai and Fyffe faults) allowed continued propagation of ~30% of ruptures from the Conway segment. The most likely magnitude was MW 7.0 ± 0.2, and MW ranged between 6.3 and 7.8. The ERS magnitude distribution closely matched the paleoseismic magnitude estimates of MW 7.0-7.4 based on single-event displacements. In the Southern California simulations, the ERS predictions of magnitude (MW 6.5-8.0) closely resembled the distribution of magnitudes for paleoearthquakes in the San Jacinto-San Andreas region (MW 6.7-7.8). The passing ratios through the Cajon Pass (the connection between the San Jacinto and San Andreas faults) were 0.2407 and 0.345 (24% and 34.5% of ruptures pass the connection), which is slightly higher than the 20-23% that paleoseismic evidence would suggest.

While the results replicate the rupture complexity of the historical earthquake catalogue, the ERS did not replicate the 2016 Kaikōura earthquake using empirical relationships. Instead, results from the ERS indicate that the 2016 Kaikōura earthquake was either (i) exceedingly unlikely based on current understandings of historical ruptures, (ii) involved complex 3D fault dynamics not captured in a 2D empirical approach, (iii) involved the induced rupture of secondary faults at rotating block boundaries, or (iv) a combination of these factors. Another interpretation could be that the ERS more accurately represents mature fault systems in structurally ‘simple’ settings due to bias in the underlying data.

Overall, the results suggest that empirical approaches in weighting or down-weighting rupture scenarios in seismic hazard models could be beneficial, but perhaps only in specific settings. The ERS also has potential applications in paleoseismology for determining rupture extents and expected magnitudes. There is some future potential for the ERS in early warning systems and post-earthquake disaster responses.