GNS Science Operational Earthquake Forecasting in New Zealand: Advances and Challenges Annemarie Christophersen, David A. Rhoades, David Harte & Matt C. Gerstenberger Presentation at 9 th International Workshop on Statistical Seismology, 17 th June 2015

GNS Science Outline Background: Some definitions and activities prior to Canterbury Canterbury sequence Advances during recent earthquakes Challenges Catalogue transition Incomplete data Magnitude scales Example Wilberforce Conclusions and recommendations

GNS Science Some definitions Operational Earthquake Forecasting (OEF) ‘procedures for gathering and disseminating authoritative information about the time dependence of seismic hazard to help communities prepare for potentially destructive earthquakes’ GeoNet ‘official source of geological hazard information for New Zealand’

GNS Science Prior to Canterbury No regular forecasts The Short Term Earthquake Probability (STEP) model automated and expected numbers and probabilities distributed by internal NZ-wide STEP hazard maps calculated within GeoNet, but not used internally or externally The medium term earthquake likelihood model EEPAS used internally and in published client reports.

GNS Science Canterbury response Initially Omori decay; forecast tables online compared to observations. STEP model forecast maps Ensemble model for the rebuild of Christchurch: 50 year model –Expert elicitation for weighting of different model components –Monthly and yearly forecast up-dated monthly –Tweaks along the way

GNS Science

Updated Forecast for Christchurch region starting 1 st June 2015 Screenshot from:

GNS Science Recent earthquakes and advances Three M≥6.5 earthquakes in the Wellington region in ● Probabilities of M>4, 5, 6 & 7 calculated with STEP model ● Comparison of probabilities to National Seismic Hazard model ● Scenarios of sequence development ● ETAS model (Harte, 2013) set up to calculate forecasts

GNS Science Scenario One - Very Likely (over 98% within 30 days)Very Likely Aftershocks will continue to decrease in frequency as expected (and in line with forecasts), remaining relatively deep (i.e. deeper than 25 km) in the Pacific plate. Scenario Two - Very Unlikely (1% within 30 days)Very Unlikely Another earthquake occurs, with a similar size - magnitude 6 to 7 - to the Eketahuna quake. The earthquake could be at the same depth as the current aftershocks, in a nearby part of the Pacific plate. Or the quake could be centred at a shallower depth (i.e. less than 25 km) in the overlying Australian plate. If a shallower earthquake happens, there may be stronger shaking at the surface than the Eketahuna earthquake. There are two examples in the east of the North Island where a large earthquake was followed by an earthquake(s) of similar or greater magnitude. In the first example, a magnitude 6.5 earthquake in June 1942 near Masterton was followed by two subsequent events, a magnitude 6.8 in August and a magnitude 6.0 in December the same year. In the second example, a magnitude 5.9 earthquake in February 1990 near Weber and Porangahau was followed by a magnitude 6.2 earthquake in May 1990 in the same area, which damaged buildings near Weber. Scenario Three - Extremely Unlikely (less than 1% within 30 days)Extremely Unlikely A larger magnitude quake - greater than magnitude 7 - on the 'plate interface' (the transition between the Pacific and Australian plates) is extremely unlikely, however, as with many places in New Zealand an earthquake of this size can occur at any time. The chances of one occurring are temporarily increased by the Eketahuna earthquake, in the area outlined in the box in the map below. The expected intensity of shaking from these earthquakes has already been built into the New Zealand Seismic Hazard Model, which is used by the engineering community to design buildings that can withstand intense shaking in New Zealand. Note: These probabilities will be updated as new forecasts are produced. Scenarios for the January 2014 Eketahuna earthquake available at

GNS Science Challenges: Catalogue transition Change of data acquisition system from manual to automated in 2012 Short overlap of both systems Initially problems with locating off-shore events Different magnitude corrections for calculation local magnitude M L Magnitude calculation still changing xml format; very difficult to retrieve data.

GNS Science Incomplete data: 50 hours after Darfield

GNS Science Incomplete data: 3 days after Christchurch

GNS Science Incomplete data: Canterbury Number comparison

GNS Science Incomplete data: Omori decay

GNS Science Magnitude scales All our earthquake likelihood models use catalogue with M L Ground motion prediction equation expect M w Twice as many M L events compared to M w at M=5.0 Correcting M L by regression equation changes b-value Regression equation applied to correct Canterbury forecast Still working out what to do with other earthquake likelihood models.

GNS Science Magnitude scales

GNS Science Magnitude scales

GNS Science Magnitude scales

GNS Science Example “Wilberforce earthquake” Lake Coleridge Wilberforce earthquake Canterbury model forecast area Initially M=6.4 at the edge of Lake Coleridge Within the area of Canterbury model Got the wheels going Location moved Magnitude downgraded to 6.0 M w = 5.6 Not many aftershocks at all, nor people at risk

GNS Science Getting the wheels going 45 min instructions from David Rhoades on the phone Download catalogue; reformat; patch with older catalogue Run EEPAS models; STEP model; combine models In the mean time: Doubts about the choice of model Canterbury long-term models had large component that were not declustered – over-cook rates Not enough weights to aftershock models – under-cook in short time. What would be a better model? STEP or STEP EEPAS combination? What can actually be done in a useful time that is not already set up? Example “Wilberforce earthquake”

GNS Science Example “Wilberforce earthquake” All forecasts start 6 th January 2015 and are for the region from degrees East and degrees South. 1 Month MagExpected # EE model Expected # STEP Expected # EEPAS1F Expected # NSHM

GNS Science Example “Wilberforce earthquake” All forecasts start 6 th January 2015 and are for the region from degrees East and degrees South. 1 Year MagExpected # EE model Expected # STEP Expected # EEPAS1F Expected # NSHM

GNS Science Conclusions Numerous opportunities to practice OEF in recent years. Advances in communications, in particular scenarios. Procedures still somewhat adhoc but efforts underway to automate forecasts. On-going work –Understanding and resolving catalogue issues –Stakeholder consultation to establish what information to provide –Hybrid models for different time scales

GNS Science Recommendations Decide on procedures (including models to use) before the event Engage with stakeholders about their information need Communicate what you know – as well as what you don’t know.