Houses crumbled and buildings flattened within seconds of the 7.8-magnitude earthquake that hit Turkey and Syria early Monday morning, leaving rescue workers scrambling to find survivors among the dead.
The horror of the near-instantaneous destruction raises a question: Why didn't anyone know the earthquake was coming?
The answer is tricky. The ability to predict where and when an earthquake will occur has eluded earth scientists for years, though the stakes couldn't be higher. Earthquakes account for nearly half of all deaths from natural disasters over almost two decades, according to the World Health Organization.
Many geologists say it's nearly impossible to perfectly predict an earthquake, due to the sheer complexity of analysing the entirety of the planet's crust. Others say that a slew of new technology - including artificial intelligence, which may help make predictions faster and more precise, and smartphones, which can instantly send alerts and warn people to find shelter - can help save lives.
- Turkey-Syria earthquake: Expats in UAE fear for safety of loved ones amid widespread damage back home
- Turkey-Syria earthquake: Death toll in the thousands after massive quake and aftershocks
- UAE leaders condole Syrian, Turkish Presidents over earthquake victims
- Videos show buildings collapsing and rescuers in action in Turkey and Syria
But even the most promising efforts offer only seconds, or in some rare cases minutes, of advanced notice - making it hard to evacuate in time. A future where technology more precisely predicts the location, time and severity of an earthquake seems years away, earth science experts said, while inaccurate estimates can do more harm than good.
"An earthquake happens very, very quickly," said Christine Goulet, director of the US Geological Survey (USGS) Earthquake Science Center. "It's fair [to say] that at this point, we are not capable of predicting earthquakes at all."
The plate movements that underpin earthquakes happen slowly, and ruptures often occur suddenly, creating earthquakes that wreak havoc with little notice.
Major earthquakes, such as the 2010 temblor in Haiti, came as a surprise. To prevent incorrect guesses, geologists have begun to focus on the odds of an earthquake happening rather than attempting to forecast individual events.
Scientists use geological measurements, data from seismology machines and historical records to highlight areas that are at risk for an earthquake - and then use statistical models to assess the likelihood of one happening in the future.
But unlike weather forecasting - which has been improved by computing power, mathematical models and the rise of drones and satellites - the quality of earthquake prediction has lagged.
Can animals predict earthquakes?
Over the past half-century, scientists tried to predict earthquakes using several methods and found little success.
In the 1970s and 1980s, researchers set out to find signals that might precede earthquakes, looking at a hodgepodge of cues like animal behavior, radon emissions and electromagnetic signals. At times, the results showed patterns, but none were reliable enough to meet scientific muster, said John Rundle, a professor of physics and geology at the University of California at Davis.
In the 1980s, earthquake scientists said a segment of the San Andreas Fault near Parkfield, Calif., was overdue for an earthquake and analyzed reams of historical data to predict it. They decided an earthquake would hit by 1993, but it didn't happen until 2004 - when it barreled through central California without warning.
That was a "kind of death knell" for earthquake prediction, Rundle said, prompting many scientists to focus more on statistical models and probability assessments rather than a weather-like forecast.
But as technology continued to advance, earthquake early-warning systems developed. These networks use seismology machines to detect and analyze tremors - and plug into a system that sends notifications to people a few seconds before an earthquake hits.
ShakeAlert, a system built by the USGS, can send a notification to a person's phone giving them roughly 20 seconds to a minute advanced notice before an earthquake.
The technology culls data from USGS field station sensors, which measure the intensity of ground shaking. When a station detects an earthquake, computers can calculate station data and predict within five seconds where the shaking will go.
Cell carriers can then issue warnings to users in the prospective area. The system works because internet and cellular signals travel at the speed of light, which is far quicker than the slow pace at which earthquake waves travel through rock.
But providing a heads-up of more than a few seconds is very hard to do, multiple earthquake experts said. Precisely predicting earthquakes would require extensive mapping and analysis of Earth's crust, including marking every stress point to carefully track which ones might be close to rupturing.
There's also an element of randomness to when an earthquake happens, which can sometimes occur without any warning signs, the experts added. Even if the technology shows promise, many scientists fear a product gets put out early without rigorous testing and fails, it makes people less trustworthy of the technology.
"False alarms are almost worse than correct predictions," Rundle said. "Because [then], people lose faith in the system."
Researchers are also turning to artificial intelligence, using machine learning software, which ingests large amounts of data and spots patterns. The hope, experts said, is to have software quickly analyze more data than humans can to help them better understand what precedes earthquakes to spot more warning signs.
For example, some are developing "nowcasting models," inspired by how the Federal Reserve predicts the health of America's economy, Rundle said.
Scientists are feeding machine-learning models troves of data, from seismology readings to radar data on how Earth's surface is deforming, to get better at predicting the time and location of future earthquakes, he added.
But even if the technology is mastered it is unlikely to be incredibly precise. In a best-case scenario, scientists could likely predict the location of an earthquake within a roughly a 600-by-600-mile range, and still over the course of a few years.
Anything more detailed would be unlikely, because the amount of data available on past earthquakes is still lacking, he said.
"Earthquake data only became automated and digital in the last 25 or 30 years," he said. "So we are working with data that was rather incomplete before that time."
Alternative earthquake prediction theories have also arisen, but many are to be viewed skeptically, earthquake scientists said.
One controversial method relies on studying the alignment of the planets. A researcher from the Netherlands went viral on Twitter on Monday after seemingly using this method to accurately predict the details of the Turkish earthquake several days in advance.
Goulet said there are all kinds of unsubstantiated and un-researched earthquake prediction methods, adding that if these results are analyzed over a long period, there's no real element of predictability that holds up.
"I don't know of anyone who made reliable predictions repeatedly ever," Goulet said. "If this was this easy, we would do it."