Researchers have used deep learning to model more accurately than ever how ice crystals form in the atmosphere. Of them paperpublished this week in the journal PNAS, hints at the potential to dramatically increase the accuracy of weather and climate forecasts.
Researchers have used deep learning to predict how atoms and molecules behave. First, the models were trained on small-scale simulations of 64 water molecules to help them predict how the electrons in the atom would interact. Then the models reproduce those interactions on a larger scale, with more atoms and molecules. It is the ability to accurately simulate electron interactions that allows the team to accurately predict physical and chemical behavior.
“The properties of matter emerge from the way electrons behave,” said Pablo Piaggi, a researcher at Princeton University and lead author of the study. “Clearly simulating what happens at that level is one way to capture much richer physical phenomena.”
This is the first time this method has been used to model something as complex as the formation of ice crystals, also known as ice nucleation. This is one of the first steps in cloud formation, which is where all precipitation originates.
Xiaohong Liu, a professor of atmospheric science at Texas A&M University who was not involved in the study, said half of the precipitation – whether it’s snow or rain or hail – starts off as ice crystals, then goes on grows and produces precipitation. If researchers can model ice nuclei more accurately, it could provide a huge boost to weather prediction in general.
The nuclear formation of ice is currently predicted on the basis of laboratory experiments. The researchers collect data on ice formation under various laboratory conditions, and that data is fed into models that predict weather under similar real-world conditions. This method sometimes works well enough, but often it is not accurate because of the large number of variables related to actual weather conditions. If even a few factors differ between the lab and the real world, the results can be quite different.
“Your data is only valid for a certain area, temperature, or type of laboratory,” says Liu.
Predicting ice nucleation from how electrons interact is much more accurate, but it is also computationally expensive. It requires researchers to model at least 4,000 to 100,000 water molecules, and even on a supercomputer, such a simulation can take years to run. Even that would only be able to model interactions for 100 picoseconds or 10-ten seconds — not long enough to observe ice nucleation.
However, using deep learning, the researchers were able to run the calculations in just 10 days. The interval is also 1,000 times longer — still a fraction of a second, but just enough to see nucleation.
Of course, more precise models of ice nucleation alone will not make the forecast perfect, says Liu, since it is only a small, though important, component of the weather model. details. Other aspects are also important – understanding how water droplets and ice crystals grow, for example, and how they move and interact under different conditions.
However, the ability to more accurately model how ice crystals form in the atmosphere will greatly improve weather predictions, especially those related to whether it will rain or snow. or not. It could also aid climate forecasting by improving its ability to model clouds, which affect the planet’s temperature in complex ways.
Piaggi said future research could create nuclear models of ice in the presence of substances such as smoke in the air, potentially improving the accuracy of the models even further. Because of deep learning techniques, it is now possible to use electron interactions to model larger systems over longer periods of time.
“That opened up a fundamentally new field,” says Piaggi. “It has and will play an even bigger role in our chemistry simulations and in our materials simulations.”
