Mason Lee is currently a sophomore at Brown studying Computer Science-Applied Math and Earth, Environmental, and Planetary Sciences. He is especially interested in machine learning and planetary climate science. I sat down with Mason to learn more about his interest in ecology, mycology, and his experience as a volunteer firefighter in his home state of California.

Laura: What are you currently doing at Brown?

Mason: The stuff that I've been working on recently is the intersection of deep learning and climate science, so a lot of physics-informed neural networks to improve climate models to understand natural processes. Like taking a lot of big remote sensing data, whether that be satellite imagery or weather monitors, and then imbuing them with physics and learning new insights about the environment. I work in the Northern Change Lab, which is a lab here at Brown that studies glacier hydrology. The work I'm doing there is developing a deep-learning computer vision technique to map glacial rivers.

L: Could you tell me more about this research?

M: Okay, so in Northwest Greenland, there is an issue where the climate models that they have—the regional ones for Northwest Greenland— are not actually accurate, and they're overpredicting the amount of water that's entering into the ocean. And this is because we don't have an understanding of how water actually flows off the ice sheet and the equations that we're using to understand how the water flows are not actually accurate. So, I developed a method that can take satellite imagery and then map the locations of the rivers. And then there were multiple parts to that as well, where there were these highly discontinuous rivers, so we used deep learning to create these continuous river networks. With that, we think we will able to more accurate climate models where we can route water over hundreds of square miles on the ice sheet in a way that we never were able to before because of those discontinuities. That is what we're working on to improve those regional climate models. I'm also involved in the Population Studies department here, which is kind of a new thing. We're looking at the dynamics of population migration in response to different natural disasters. We're looking at how we can understand where people go, how adaptive people are—in their beliefs, to natural disasters—and what effects correlate to people staying in their house or moving and how long it takes them to come back. That's in partnership with the IRS. 

L: Yeah, that sounds really interesting. So, how did you become interested in ecology and climate modeling?

M: So before I came to Brown, I took a gap year in 2020. And I'm from Los Angeles, where are you from?

L: I'm from Boston.

M: Okay, I'm sure you know that we have crazy wildfires out on the west coast. I was a wildland firefighter in California, for CALFire, which is the state agency there. We were in this little town with a bunch of wildfires [during the August Complex Fire]. It was one of the largest fires in California and saw a million acres burn. We had to do structure protection, and help evacuate people. It was a very jarring experience, you know. We worked 20 to 48 hours every day for five months straight just putting out these huge fires. You'd see crazy stuff. You'd see animals on fire running through the forest, you'd see people's houses burning down. It was crazy. 

So after that, you know, I got in touch with this nonprofit that I was working with before, and we started this citizen science research group. I've been into mycology for a while, and there was this really cool collection of people across California like this mycology professor, Mia Maltz at UC Irvine, and she kind of started this group. After a fire, FEMA would come in, and they would like try and take whatever debris they can off of the house, these houses that were burned down. But there's still so much ash around. When the water comes, the rainy season is right after the fire season in California, and it all washes into the waterways. So we're thinking, how can we stop this point-source pollution before enters? All these heavy metals, this is nasty stuff. We came together and created this biofilter using oyster mushrooms, which can bioaccumulate a lot of toxins, especially these heavy metals. We were working with some native tribes in the region, and we did a test study in Santa Cruz, in the remnants of a neighborhood where there were over 400 houses that burned down during the CCU fire. We did around 20 houses that we put these BioWattles around the perimeter. We inoculated them with oyster mushrooms. But unfortunately, we didn't have the resources to actually get that research funded. We did all the soil sampling, but we were never able to actually get those samples processed, unfortunately. So that's kind of where that stands right now. There are a lot of other groups in California that are working on the remediation part of the problem.

L: Do you see yourself getting back involved in that project soon? Or are you shifting your focus toward ML/AI-centered work?

M: I think they go hand in hand with each other. I think the modeling side is cool, but I don't like it in some sense because a lot of the climate intelligence sector serves the purpose of underwriting predatory loans from insurance companies that actually price people out of these neighborhoods. Even if you're gonna make an accurate fire model, you're gonna say, okay, I can really understand the effects of this, but you have to be really careful with that information, too. It's often used by insurance companies to make insurance insanely expensive, and people will never be able to pay for that stuff. They just won't be able to live in that area anymore. I think, you know, there's this whole section of machine learning and physics-informed neural networks, which is really the intersection of engineering and data science where we can use sensory information to build really robust systems—really robust biological systems—especially biofilters. So that's the stuff I'm interested in, and I think I'm gonna explore a lot in the future. 

L: I was wondering if you could say more about the science behind biofiltering. You described your experience with forest fires and how the mushrooms can filter the smoky air but are there any other use cases?

M: 100%. The first paper that I published was the research I'd been doing in high school for four years, which was around this biofilter. We were using these bacteria called Methanotrophic bacteria, and they just have the ability to degrade the carbon-hydrogen bond in atoms and a lot of pollutants. Specifically, chlorinated hydrocarbons have these really long chains, so they're able to break that down. What I mean to say by that is there's a large range of pollutants that can be degraded naturally by microbes given the right environment, and I think there's a ton of work that needs to be done in that field to do that. All this soil is polluted, and you can't just ship it off to some factory and then clean it. There has to be a way of applying these methods in the soil that's really cheap and fast, natural and non-invasive. There's an incredible amount of work that needs to be done in that sector. The amount of pollutants and the range of pollutants that can be degraded by these bacteria is almost all of them, except the heavy metals, which can be bioaccumulated by mushrooms. We have to find a way to make these systems scalable and robust, and I think that's where machine learning comes in. We can learn the really complicated dynamics of these fungi and bacteria to do that.

L: That is really cool. So, keeping in mind our biweekly theme of Decomposition, is there anything that you wanted to add, or anything else that you wanted to talk about?

M: Yeah, I think in terms of decomposition, just listening to nature, and trying to understand what it's telling us will be really, really important in the future. Something that I was always interested in was how fungi, you know, there's a theory that fungi connect. Ectomycorrhizae fungi the roots connect with each other in between the trees. That is what connected all the individual trees in the forest together. It is like the driver of the ecosystem, you could like move resources around like that. And it's so powerful to think of it that way. We're so far removed from that sort of information, it's in a completely different language. The natural world, in terms of data, is hard to understand. I think it's impossible to understand the mathematics, it's so complex. I think machine learning is going to be the way where we can frame the problem in the correct way where we're not trying to solve it, but we can really learn from what the state of the environment is. And then from there, I think we can really leverage that in certain ways. All the tools are already there. It's just like, you know, how can we elevate the ones that are the most useful?