After the seismometers have been taken out of the ground later this year, the next phase of Lauren’s research project begins. The seismometers measure Earth’s ground motion, and the output looks similar to audio signals . “I will take the observations from the earthquake waves, and then do some modelling based on potential compositions and temperatures, and then try to connect that to the dynamics of the Earth,” Lauren says.
“Think about the hot plumes coming up in the volcanoes of Hawaii, for example,” she says. “I try to figure out how what we see physically ties into the composition and the subsequent dynamics and the motion of the Earth’s mantle. That's my current goal.”
From hand picking data to artificial intelligence
Only a couple of years ago, Lauren would have had to identify the particular signals she needed manually, and that takes time. She has done this already for previous projects. “That could get a bit boring,” she says.
However, neural networks and artificial intelligence will take over this rather tedious task. “I had a Master's student in New Mexico who came with a computational background,” Lauren says. “He developed an algorithm to train a computer to automatically identify the correct signals.”
How a cup of coffee can influence a dataset
But how do you train artificial intelligence? Only by doing the legwork first. When a human is analysing the data, they have to go through it and look for the ‘wiggles’ that correspond to the energy that the scientists are studying.
The results depend on the attention of the person who looks at the data — a cup of coffee or two in the morning or a big lunch can have an impact on the attention of the researcher. This is why Lauren prefers to use computers. “We can be sure that the computer is consistent,” she says.
The science of speed
There is more. “It took me about eight months to manually compile the data I needed,” Lauren says. “It takes about 16 hours to get a similar dataset on a supercomputer running on 12 nodes and 16 cores. It does take a bit of computational power, but to be able to have a complete data set within a day is amazing.” She adds that a computer finds it easy to identify and discard waves that are simply ‘noise’ and not useful for the researcher.
Lauren’s eight months of handpicked data weren’t made redundant by the artificial intelligence, though, because her data was used to train the code, to make it understand what the scientists were looking for.
Analysing big data with artificial intelligence
Lauren says that even though using neural networks and artificial intelligence is not completely new in seismology, the code she and her student developed is still a step forward for the study of seismic waves.
“There is so much data being generated right now, and there's only so many PhD students that you can persuade to sit in front of a computer and go for it,” she says. “A PhD student or a postdoc’s skills are much better applied to developing an exciting new code.”