logo image

The Source Model

The 2010 M7.1 Darfield earthquake in New Zealand started a sequence of events that propagated eastward in the Canterbury region over several years, collectively causing upward of 15 individual loss-causing events for the insurance industry. The Insurance Council of New Zealand state that the total insured loss was more than NZ$31 billion (US$19.4 billion).

With such a significant sequence of events, a lot had to be learned and reflected into earthquake risk modeling, both to be scientifically robust and to answer the new regulatory needs. GNS Science – the New Zealand Crown Research Institute, had issued its National Seismic Hazard Model (NSHM) in 2010, before the Canterbury Earthquake Sequence (CES) and before Tōhoku. The model release was a major project, and at the time, in response to the CES, GNS only had the bandwidth for a mini-update to the 2010 models, to allow M9 events on the Hikurangi Subduction Interface, New Zealand’s largest plate boundary fault, and to get a working group started on Canterbury earthquake rates.

But given the high penetration rate of earthquake insurance in New Zealand and the magnitude of the damage in the Canterbury region, the (re)insurance and regulatory position was in transition. Rather than wait for a new National Seismic Hazard Map (NSHMP) update (which is still in not available), RMS joined the national effort and started a collaboration with GNS Science as well as our own research, to build a model that would help during this difficult time, when many rebuild decisions had to be made. The RMS® New Zealand Earthquake High Definition (HD) model was released in mid-2016.

The Task Ahead

Where did we start? First, RMS experts updated the Hikurangi Subduction Interface geometry using information derived from GPS measurements (such as coupling and slow-slip event location). Examining commentary from experts in New Zealand compiled in 2005 about the 1855 Wairarapa event (near Wellington), RMS showed the likely interplay between the subduction interface and upper crust faults, in its historical event validation effort – well before this inference was made about Kaikoura (M7.8 in 2016). Providing such an event scenario for clients to run allows them to stress test their model for a type of event that now seems very likely around the Cook Strait.

RMS experts also developed and published a probabilistic method that combines geomorphology data and trench data to infer the time-dependent recurrence model for a given fault or a group of faults. The method and results for Wellington, Wairarapa and Ohariu were shared and discussed with GNS, along with their method and output for the same faults. It turns out that the local crustal faults are not dominant, as previously thought, but present the same level of risk as a M8+ on the Wellington segment of the Hikurangi subduction zone. The implied clustering of large events around the Cook Strait was also assessed and shown to compare favorably with the earthquake catalog and regional numerical models.

In the South Island, the plate boundary Alpine Fault is vital for the free movement of people, goods and electricity between the east and the west shores of the South Island. For this extensively studied fault, the RMS time-dependent model incorporated computations made using the 23 consecutive events found in trenches as well as expert knowledge on the trench environment as soon as they were published.



Figure One: Examples in yellow of the multi-segment or multi-fault events that were built and given rates, and are in the stochastic event set of the RMS New Zealand Earthquake High Definition (HD) model. The top and bottom left panels show events directly impacting Wellington, the top right panel shows events impacting the Auckland area, and the bottom right panel shows events that go up to but do not cross the Cook Strait (i.e., remain offshore from, but still impact Wellington).

When the updated NSHM for New Zealand was published in 2012, it did not feature any low frequency/high impact events around Wellington or Auckland. Such events are necessary to populate the tail of the exceedance probability curve to capture realistic financial reserve requirements. Since ruptures can “jump” from fault segment to nearby fault segment during the same earthquake, RMS complemented the stochastic event set with additional multi-fault ruptures. These were selected from more than 400,000 events that RMS had developed applying the Uniform California Earthquake Rupture Forecast (UCERF3) method to the 2010-2012 fault database, updated using data published at the end of the development. Only very large events were necessary since background events go up to M7.4.

RMS research was shared with GNS to help with their own research plan. Note that this effort happened as UCERF3 was still ongoing in the U.S. and before the multi-segment Kaikoura rupture happened in New Zealand. GNS experts helped select the most relevant events and assign rates, so that these events can populate the EP curve. RMS shared the challenges of using the UCERF3 methods in New Zealand with the researchers at the annual meeting of the Seismological Society of America in 2015.

But the biggest question at the time was: which rates should be used in Canterbury, where few faults are known and where the city of Christchurch needs to be rebuilt? In 2014, GNS Science scientists published a comprehensive statistical seismology study on short, medium and long-term rates for the region. Forecast rates for 2016 were still very high, due to the still large production of aftershocks. (Re)insurers are used to making business decisions based on mainshock-based models, and RMS felt it had to support the industry and use the re-assessed long-term rates. To further help clients, RMS innovated in terms of the characterization of background events (their orientation and length and spatial distribution). With nine years passed since the beginning of the sequence, there is now a much smaller difference between the current short-term rates and the new long-term rates. It was not the case back in 2015.

The RMS New Zealand Earthquake HD Model uses a simulation methodology over six year periods, that ensures that all events are sampled appropriately (annual rates from 10-9 to 10-2), and that time-dependent rates are updated year-by-year according to what was sampled in the previous year. This allows the users to notice what time series of mainshock events look like when preferential orientations for faulting are used in the model. This is relevant since many events in the Canterbury sequence are labeled as independent events by many de-clustering methods. Users can identify other possible ‘sequences’ over one to six years, in Canterbury and elsewhere, and use them for stress-tests.

Much science and know-how was developed during this project, and much outreach was and still is being done, both with the science community (in New Zealand and worldwide) and the industry. RMS remains active in the region, participating in many ongoing research projects. RMS users around the world have also been benefiting of this huge effort since RMS experts have been taking lessons learned in the region (and applied first to New Zealand) to other peril regions.

The story of the RMS New Zealand Earthquake HD Model shows the commitment of RMS to collect, produce and deliver the new understanding needed to ensure a more secure, long-term future to our clients and the communities they serve. For New Zealand, this happened through conducting reconnaissance work, taking stock of the local conditions and challenges and remaining committed to work alongside the local scientists, engineers, (re)insurers and regulators even as the country began recovery efforts during the Canterbury sequence.

One of the most striking features of the Canterbury sequence was the impact of liquefaction on losses. Click here for part two of this blog, which is centered around RMS innovation in this field.


Diederichs et al., Sci. Adv. 2019; 5 : eaax5703 2 October 2019

Stirling et al, BSSA 2012, 102-4, doi: 10.1785/0120110170, August 2012

You May Also Like
Delphine Map
September 27, 2018
How Risk Selection in California Can Be Affected by Earthquake Source Modeling Assumptions: A Spatial Loss Correlation Perspective
Delphine Fitzenz
Delphine Fitzenz
Senior Principal Modeler, Model Development, RMS

Delphine Fitzenz works on earthquake source modeling for risk products, with a particular emphasis on spatio-temporal patterns of large earthquakes.

Since joining RMS in 2012 after 10+ years in academia, she has strived to bring the risk and the earthquake science communities closer together through her articles and by organizing special sessions at conferences. These include the Annual Meeting of the Seismological Society of America (e.g., Earthquake Hazards and Risk: Drivers and Consumers of Earthquake Research in 2015; Risk Management Applications of Earthquake Seismology in 2016).

She gave an invited talk on “How much spatio-temporal clustering should one build into a risk model?” at the Ninth Statistical Seismology workshop in Potsdam, Germany, and was invited to "Workshop 1: Potential Uses of Operational Earthquake Forecasting (OEF) System in California."

cta image

Need Help Managing Your Portfolio?

close button
Video Title

Thank You

You’ll be contacted by an RMS specialist shortly.

RMS.com uses cookies to improve your experience and analyze site usage. Read Cookie Policy or click I understand.