Rapid analysis of large local earthquakes and their impacts (Research Aim 1.1)
Traditionally, earthquakes are often thought of happening at a point. Often, we will put a dot on the map where the earthquake occurred. Earthquakes occur along faults, though, and faults are planes. They are depicted on a (two-dimensional) map with a line. This seemingly simple change has big consequences for all of the events that follow on after the earthquake. R-CET is working to change how we depict earthquakes' and the knock on effects from that change.
How are we doing this?
One of the tools we are using is an algorithm- a complex mathematical model that has been turned into software that we can run on a computer- called FinDer. As ground shaking data comes in, all of the recorded shaking is mapped with the area of the strongest shaking noted. This is where the epicentre (where the fault first started moving) occurred and corresponds to the leftmost map in the figure below. We can then say this is where shaking is occurring or has occurred and this is where there hasn't been any shaking. Then, we can use the equations of the shaking we are observing and a database of equations of shaking we have previously experienced in other earthquakes and see which previous set of equations best matches what we are seeing this time (the "template" event). When we have found a good fit, we can then take the map of the shaking we are observing and the map of the "template" event and see how well they match. This corresponds to the second (middle) map in the figure below. Finally, after the shaking has stopped, we will have all of the data and can break the original line we plotted on the map into a set of smaller lines that correspond to the actual faults in the ground that moved. This process can only happen after the initial shaking is over, though, so it takes more time to be able to make a map like the one on the far right below.
As the figure above shows, our estimates of the shaking improve as more data is received. In the map on the left, the shaking north of Kaikoura is substantially lower than in the map on the right. Because shaking is related to how far a location is from the movement on the fault, the area north of Kaikoura is far from the star, so initial estimates of the shaking are low. In the middle map, a line has been extended from the star northwards. This means the shaking will have been worse north of the star than south of it. This is confirmed in the third map on the right where the line from the middle map has been broken into a series of overlapping rectangles. These rectangles all indicate faults that have moved in this event, the distance along the fault that moved, and the fault's orientation in space.
FinDer can only work if we have seismometers set up in a dense array. This means we need a grid of seismometers that are spaced 50 km or less apart. This is why we are installing lots of seismometers in Tai Tokerau! R-CET scientists are the first in Aotearoa New Zealand to use FinDer here, so this will benefit both our work and responding to earthquakes in Aotearoa New Zealand in general. There are many more details about FinDER in this paper if you are interested:
M. Böse, D. E. Smith, C. Felizardo, M.-A. Meier, T. H. Heaton, J. F. Cliff; FinDer ver. 2: Improved real-time ground-motion predicitons for M2-M9 with seismic finite-source characterization. Geophysical Journal International 2018; 212(1): 725-742. doi: https://doi.org/10.1093/gji/ggx430
Why does this matter?
One effect is how computer simulations estimate where the worst shaking was. Instead of focusing on a spot along the fault, the entire fault can now be considered as the origin. With that, we know that local bedrock can change how the shaking is felt in an area. Together, this can help pinpoint the worst of the shaking and help redirect first responders to the locations with the worst damage rather than discovering that locale as they respond. Knowing about the shaking can also help more accurately predict if a tsunami was generated and how large it may be.
When the earthquake's waves are well known, it becomes easier to estimate what the aftershock sequence could look like. As part of R-CET's work, we are using machine learning (that is, the computers are helping us automatically) to see if the earthquake waves bear similarities to previous earthquakes and aftershock sequences that have been recorded. We are also developing and expanding a database of previous earthquake and aftershock sequences so the computers can give us better predictions the more earthquakes we experience. Knowing what the aftershock sequence may look like gives us an idea of what we may be up against during the recovery phase after the earthquake.
To summarise this research aim, R-CET is working to make processes that take hours to weeks to conduct manually and is showing that with new computer programs, we can get the same results within minutes to hours.
Maps outputs from FinDer. This series of maps is an example of the types of outputs that FinDer creates (left) as the event is ongoing, (middle) within a few minutes of the end of the event, and (right) within a few days of the event. Each map shows more refined detail about the event that has taken place and can provide first responders a better idea about where strong shaking has taken place and, therefore, who might need the most help. Warmer colours indicate more intense shaking.
Figure created by: J. Andrews (GNS Science)
Maps becoming more refined as more data was processed after the 2016 Kaikōura earthquake. Moving from left to right, the bottom maps show the peak ground acceleration (the maximum rate at which the speed of the ground movement changed) based on the data at 1.2 hours, 8.7 hours, 5 days, and 3 months after the earthquake. The top maps show the likelihood of an earthquake in a given location. Landslides are a process that commonly occurs during or immediately after an earthquake because the ground has been disturbed, potentially leading to instabilities. Similarly, moving from left to right, the predictions become more precise as more data processing is undertaken. Research Aim 1.1 in R-CET will create new processes for this data to be processed as it comes it, expediting how quickly we get the results in the far right.
Figure adapted from:
Kate E. Allstadt, Randall W. Jibson, Eric M. Thompson, Chris I. Massey, David J. Wald, Jonathan W. Godt, Francis K. Rengers; Improving Near‐Real‐Time Coseismic Landslide Models: Lessons Learned from the 2016 Kaikōura, New Zealand, Earthquake. Bulletin of the Seismological Society of America 2018; 108 (3B): 1649–1664. doi: https://doi.org/10.1785/0120170297
Team members involved:
Bill Fry (GNS Science)
Anna Kaiser (GNS Science)
Yannik Behr (GNS Science)
Jen Andrews (GNS Science)
Elisabetta D'Anastasio (GNS Science)
Sigrun Hreinsdottir (GNS Science)
Adrian Benson (Victoria University of Wellington)
Simon McClusky (Australian National University)
Brendan Cowell (University of Washington)
Diego Melgar (University of Oregon)