We discuss the challenging task of forecasting the occurrence and magnitude of earthquakes.
On 6 February 2023, the world was aghast to learn the news that a 7.8-magnitude earthquake had hit Turkey and Syria. The quake was followed by a 7.5-magnitude event and more than 200 aftershocks, resulting in a catastrophic scale of destruction. As of the time of writing, more than 35,000 people are known to have died in what has become the world’s deadliest quake in more than a decade — and sadly, the death toll is still expected to rise.
Turkey has a long history of earthquakes as most of the region sits on the Anatolian tectonic plate between two major faults — the North Anatolian Fault and the East Anatolian Fault1. The former is known to be particularly devastating and has caused disastrous quakes in the past, although the most recent quake is believed to have struck along the East Anatolia fault zone. Overall, according to the Disaster and Emergency Management Authority, from 1900 through 2019, more than 1,700 earthquakes with a magnitude of at least 5.0 have been recorded in Turkey; in 2019 alone, there were more than 23,000 quakes recorded, with 20 of them having a magnitude of at least 5.0.
The 6 February tragedy has left many wondering if, in a region substantially prone to earthquakes, such as Turkey, it would be possible to forecast their occurrence in order to plan appropriate evacuation strategies and thus avoid the potentially catastrophic consequences.
As it turns out, quake prediction is notoriously difficult. The United States Geological Survey (USGS) FAQ page even states that “neither the USGS nor any other scientists have ever predicted a major earthquake” and that they do not expect to know how to predict earthquakes in the foreseeable future.
There are multiple reasons why predicting quakes is so hard. Even though scientists do have a good understanding regarding how they occur, determining which early warning signs translate into when, where, and with what magnitude an earthquake will hit is not a trivial task. Some natural factors that have been thought to be connected with quakes include, for instance, increased amounts of radon in water sources (smaller breaks in the subsurface rock may form prior to the main rupture, making rocks more permeable and potentially leading to the escape of radon) and odd animal behavior. However, there is little empirical evidence that these factors are indeed connected with quakes: some of these events happen for other reasons, and some quakes occur without having any of these precursor factors. Forecasting the magnitude of an earthquake also remains challenging, as small and large quakes are thought to start in similar ways: ruptures often occur suddenly, creating earthquakes that wreak havoc with little notice.
Scientists have also attempted to create sophisticated mathematical models of the movement of known tectonic plates, but precisely predicting earthquakes would require extensive mapping and analysis of the Earth’s crust. More recently, researchers have turned to machine learning2,3 to develop nowcasting models by taking advantage of large amounts of data, such as seismology readings and data on Earth’s surface deformation. Nevertheless, while promising, these models haven’t yet achieved the required precision. Challenges include a lack of data on the early warning signs to feed into the machine learning pipelines — given that these warning signs are not yet entirely understood — and the fact that historical quakes occurring more than 25–30 years ago lacked automated and digitized data collection at that time, resulting in an incomplete data source.
It goes without saying that earthquake prediction is a high-stakes task. According to the World Health Organization, earthquakes have accounted for nearly half of all deaths from natural disasters in recent decades. Failure to provide warning about strong quakes is obviously harmful, as the primary goal is to save as many lives as possible. However, false alarms have downsides too, as they can incur unnecessary emergency costs and undermine the credibility of these systems, which will ultimately affect how people will respond to alarms in the future. Ensuring that the predictions are as accurate as possible while minimizing these risks is daunting.
Still, even though some scientists do not expect earthquake prediction to be possible in the foreseeable future, we certainly hope, for the sake of our communities, that they are mistaken and that we can continue to improve our earthquake early-warning systems and save countless lives.
Naddaf, M. Turkey–Syria earthquake: what scientists know. Nature (6 February 2023); https://doi.org/10.1038/d41586-023-00364-y
Rouet-Leduc, B. et al. Geophys. Res. Lett. 44, 9276–9282 (2017).
Hulbert, C. et al. Nat. Geosci. 12, 69–74 (2019).
Rights and permissions
About this article
Cite this article
Why can’t we predict earthquakes?. Nat Comput Sci 3, 115 (2023). https://doi.org/10.1038/s43588-023-00418-1