Fools, Charlatans and Liars: The truth behind earthquake prediction
By Jenny Chandler, Research Assistant, Centre for Sustainability, University of Otago

Earthquake prediction is the holy grail of seismology – it would save lives. Accordingly, a lot of time and money has been invested in it over the last 200 years, but despite best efforts it’s still not possible to predict an earthquake. Some doubt whether it ever will be, but let’s look at the current science.  

Predictions, Probabilities and Patterns

Prediction and probabilities are based on patterns – by identifying patterns in past events, people hope to predict future events. So, a more defined pattern = a more accurate prediction.  

Unfortunately, for earthquakes, the patterns are not very well defined, so scientists work in probabilities rather than predictions. What’s the difference you ask? For a prediction to be meaningful it needs to specify 3things:

1.     Where - The precise location

2.     When - The precise time and date

3.     How Big – the magnitude and duration

That level of accuracy is just not possible for earthquakes. What is possible is calculating the probability of an earthquake happening within a certain region or location within a certain timeframe. The probability of earthquakes is also called a forecast and is a lot like a weather forecast.

For example: Today’s weather forecast says: ‘70% chance of rain in Westport today’, giving us the probability of an event within a certain timeframe in a certain location. An Earthquake Forecast says: ‘75% chance of a rupture on the Alpine Fault in the next 50 years’ – event, probability, timeframe, place.

You can read more about earthquake forecasting in New Zealand at GeoNet.

Once the probability reaches 100% (or very close) then a forecast can be considered a prediction – scientists are working hard to increase the probabilities all the time.

Location Likelihoods

There are some obvious patterns for where earthquakes happen at both a global and local scale. Around the world, earthquakes often occur at tectonic plate boundaries (as you can see in the illustration below) and the fault systems associated with them. In fact, the distribution of earthquakes is what pointed scientists to plate tectonic theory in the first place.

This animation shows where earthquakes happen around the world - watch until the end to see how those patterns coincide with tectonic plates boundaries. Source: Pacific Tsunami Warning Centre

As almost all earthquakes happen on faults, having a good map of active faults really helps predict where earthquakes will happen. Below is a map of active on-land faults in New Zealand, you can check out the interactive version here.

Sitting on a tectonic plate boundary, New Zealand is fractured with faults as New Zealand Active Faults Database shows – and that’s just the ones we know about!

Unfortunately, creating an accurate map of all the faults in an area is not as simple as it sounds. The Earth’s history spans billions of years, and even on the best-preserved faults (such as the Alpine Fault) there is only evidence for the last few thousand years, which is just the blink of an eye in our planet’s lifespan.

That means that some faults may be ‘invisible’ until an earthquake happens. It may be that they rupture infrequently (like once every thousand years), or they run through an area of high erosion. This was the case for the fault that caused the destructive 2010 Darfield Earthquake that started the Canterbury Earthquake sequence – the braided rivers of the Canterbury plains had erased evidence of previous ruptures.

The Darfield Earthquake on Saturday 4th September 2010occurred on a previously unmapped fault line. It wasn’t visible until it ruptured the surface. Source: AF8 / Canterbury CDEM Group

And even on faults that scientists do know a lot about, like the Alpine Fault, there is still the question of where exactly it will rupture? The Alpine Fault is ~600km long and its next earthquake could start anywhere across that distance!

But slowly, slowly a more accurate picture of our faults is building. Each little piece of information adds something, about where they occur and how they behave, and it is these behaviour patterns which lead us on to the WHEN.

It’s all in the Timing

There are two main avenues to pursue in trying to pin down a time and date (or even a time frame) for an earthquake:

1.     The recurrence interval (the time between earthquakes) – is there a recognizable pattern to how often quakes happen on a certain fault?

2.     A precursor – is there an observable phenomenon that always precedes an earthquake?

Let’s look at each approach separately.

Recurrence Intervals – Minding the Gaps

Earthquakes are periodic and, on some faults – like our own Alpine Fault - happen at very regular intervals. On other faults the pattern is not discernible – perhaps it repeats over such a long timeframe that it is not visible to us, or perhaps there is no pattern. The average time between earthquakes on a fault is called the recurrence interval or the return period.

World-class research means scientists now have a complete record of earthquakes on the Alpine Fault going back several thousand years and this fault is the most periodic and regular fault known in the world!

GNS Earthquake Geologist Dr Ursula Cochran explains what we know about past earthquakes on the Alpine Fault, and what that means for the future.

The Alpine Fault is currently as good as it gets anywhere in the world for using recurrence intervals to predict timing of earthquakes. But it’s not good enough to form a prediction, geological time is just too big and human timescales too small. However, this evidence and other research has enabled scientists to work out the probability of the next Alpine Fault earthquake and recent calculations indicate there is a 75% chance of an Alpine Fault rupture in the next 50 years (read more here)

Precursors – Looking for a Sign

If we are unlikely to ever get to 100% probability using recurrence intervals, then what about the other option – precursors?

People have been looking for earthquake precursors for thousands of years (and many have declared that they have found one). Weather, dreams, psychic visions, clouds, and strange animal behaviour have all been proposed over the centuries.

For a precursor to be useful it must fulfil two criteria:

1.     It ALWAYS precedes an event

2.     It ONLY precedes an event

No precursor has been found that fits both these criteria.

The idea is not off the table though – it is possible that the tension within the earth that precedes a large earthquake could be observable in some way. Scientists have observed some promising correlations with compression of the rock, release of gases, electromagnetic changes and small tremors. But these things can happen without an earthquake occurring too, so without more information we can’t rely on them as warning signs yet.

Looking for precursors is a lot like hindsight – it involves capturing lots of background data and then after an earthquake happens, looking back through that data for patterns and signals.

Machines and Models

So, we are back to looking for patterns. And data. Which leads us to computers – machine learning, Artificial Intelligence (AI) and neural networking technologies and expertise are being applied to large sets of data to create more and more accurate predictive models.

A predictive model uses data from past events to look for patterns that can be expressed as a mathematical equation. In general, more data makes a more accurate model, and we have more data than ever before (as well as greater processing power). Neural networking and AI are also being applied to seismic data with some hopeful experiments in progress (like this one).

New Zealand is right on this bandwagon. A set of sensors is being deployed right along the Alpine Fault. This new project - SALSA – will give more detailed data on the Alpine Fault than ever before, including all the background ‘seismic noise’ and small tremors. Who knows what secrets and patterns that ‘noise’ might reveal? (Read more here)

John Townend, Geophysics professor at Victoria University, introduces the SALSA project and explains how it will help scientists improve modelling for the Alpine Fault. Source: Out There Learning]

Added to real-life data from faults is data from lab-induced earthquakes. In this video, Marie Denolle, a seismologist from Harvard University, explains how her team has managed to predict aftershock sequences in the laboratory using computer modelling and AI.  

More Good News

While scientists aren’t yet at the point where they can predict future earthquakes in real-life, they can let you know very quickly once it’s started. Introducing…. Earthquake Early Warning systems.

Earthquake Early Warning (EEW) systems are for after an earthquake has started and rely on the fact that telephone and internet communications travel faster than seismic waves.

Obviously, if you are very close to the epicentre the warning won’t be much use, but it could give people anything from a few seconds to a few minutes of warning to prepare. Japan and the USA currently have EEW systems in place and researchers at Massey University have been investigating how they could be implemented in New Zealand.

A few seconds or even minutes doesn’t sound like much preparation time, but there are a few scenarios where it could make a real difference. It’s enough time: to turn off the gas stoves in a restaurant, for the surgeon to put down her scalpel, to do an emergency shut down at the nuclear power plant or for high-rise lifts to stop and let people out.

Google is also currently trialing an early-warning system in New Zealand using the accelerometers in people’s smartphones. The idea is that the phones will sense a strong motion and send out an automated alert to others on the network (read more here).

The Quest Continues...

Earthquake prediction is elusive. Perhaps we will never get there, but we can be inspired by the ground-breaking projects underway and the exciting applications of computer technology. If all those tools keep being applied in collaborative and creative ways, it is possible that one day, not only fools and charlatans, but credible scientists will be able to predict earthquakes and prove Charles Richter wrong once and for all.  

This article is part of a series developed in collaboration with AF8[Alpine Fault Magnitude 8] team aimed at explaining earthquake science and increasing understanding of earthquake risk and resilience. With thanks to our science partners for their contributions: QuakeCoRE: New Zealand Centre for Earthquake Resilience, Resilience to Nature's Challenges, University of Otago and GNS Science.