Studying how clouds improve climate models
associate researcher Kara Mutton read her father’s scientific american Magazine. She absorbed all she could about quantum physics, and for a while, thought she would spend the rest of her career studying it. But after Lamb completed a master’s degree in the discipline in 2008, the urgency of the climate crisis grew.The Intergovernmental Panel on Climate Change has just released its Fourth Assessment Report, which raises questions about the Earth system that are as complex as they inspire—enough that Lamb chose to turn her doctorate and subsequent work toward atmospheric science. Now at Columbia Engineering, Lamb is investigating how to use machine learning methods to better understand the microphysics of cirrus clouds.
Cirrus clouds over Federer Road in Washington state. photo: Ron Clausen via Creative Commons
Cirrus clouds form higher in the atmosphere than almost any other type of cloud — tens of thousands of feet above the ground. They form when water vapor is pushed up into the stratosphere by warm, dry air, and then freezes due to the low temperatures. The result is a thin, plume cloud of thousands to millions of ice crystals. These ice crystals, and their many different structures, shapes and sizes, affect how cirrus clouds reflect incoming sunlight and trap heat away from Earth, also known as the cloud radiative effect.
Lamb was one of the first to use machine learning to study the shape of ice crystals in cirrus clouds. In doing so, she hopes to inform the way scientists account for clouds in climate models.
“As small as it may seem, when you look at the sensitivity of climate models to cloud feedbacks, the distribution across climate models is quite significant,” Lamb said.
machine learning This uncertainty can be helped to remove this uncertainty by processing and generalizing new knowledge from large amounts of ice crystal data at unprecedented speed. Traditionally, atmospheric scientists have been required to come up with a set of mathematical equations based on physical principles to describe how a system (in this case, a cloud) works. The process can be slow, as scientists repeatedly compare the model with observations and then tweak the model’s parameters to improve performance. In contrast, machine learning can be used to learn directly from data; these algorithms build models by being “trained” on observations.
One of the many ways machine learning fits into the study of cloud microphysics is through pattern recognition. While previous research applying machine learning to ice crystal observation has focused on classifying images based on ice crystal shape, Lamb plans to go a step further, using state-of-the-art techniques to actually understand ice growth processes directly from images developed by the scientific machine learning community. method of development. Now, three months into the three-year project, Lamb is testing the algorithm. Once the model is trained, Lamb estimates she’ll feed it about 12 million different images for this project alone.
An electron microscope image of individual ice crystals captured by a weather balloon in Billings, Oklahoma, and stored in liquid nitrogen for transport back to the lab. Image credit: Nathan Magee (College of New Jersey) and Jerry Harrington (Penn State University).
“Each data source gives us a different picture,” Lamb said. “That’s where we want to use machine learning and statistical methods to try to understand how all this information is stitched together.”
Much of the data used by Lamb comes from collaborations with other universities. E.g, Carrasali A University of Albany scientist has compiled observations of ice crystals from various aircraft measurement campaigns into a massive database. at Penn State University, jerry harrington High-resolution images of the ice crystals were collected by using a balloon instrument, which were then frozen in liquid nitrogen. Other datasets come from large-scale “cloud chamber” experiments, which simulate cloud formation in the laboratory, allowing scientists to systematically study cirrus cloud formation while testing different aerosol conditions, temperatures and pressures. Lamb is using cloud chamber data from her previous research at the University of Chicago in collaboration with researchers at the Karlsruhe Institute of Technology in Germany.
Kara Lamb, an associate researcher at Columbia University’s Department of Earth and Environmental Engineering, and a colleague monitor instruments during a simulation of aerosol interactions and dynamic cirrus clouds in the Karlsruhe Institute of Technology’s Atmospheric Cloud Chamber.Photo: Kara Lamb
Yet despite all of this data, most climate models — including the ones used in the Intergovernmental Panel on Climate Change report — set the baseline assumption that ice crystals are spherical. This is partly because the models are so large, Lamb explained, that they have to rely on simplified, “computationally efficient” ways to represent complex processes.
“The model has to be something you can run fast enough to actually get an answer,” she said.
The application of machine learning can help reduce the tension between accuracy and efficiency.As Lamb made strides in using machine learning to analyze ice crystal data, she also worked with Marcus van Leer-Wolki climate school in columbia Center for Climate System Research Another project with colleagues at the National Center for Atmospheric Research focuses on how this new knowledge can be applied to community Earth system models.The project is associated with Understanding Earth with the Center for Artificial Intelligence and Physics (LEAP) And is the perfect complement to Lamb’s more in-depth study of cirrus clouds.
“If we want to understand how important the radiative effects of these clouds are, we need a more accurate representation of their ice crystal shapes in climate modeling,” Lamb said. “This is one of the biggest uncertainties that need to be addressed.”
Fortunately, the work of Lamb and colleagues provides a promising start.
