Measuring CO2 with Machine Learning

AJ Repinski

Artificial intelligence seems almost inescapable in today's increasingly technology driven world.

Deep learning models, such as OpenAI’s Chat GPT, have been at the forefront of public amazement and controversy since their mainstream introduction in late 2022.

Today, Fort Lewis College students are discovering new ways that artificial intelligence can be used to reduce the costs of studying the environment. 

Lincoln Scheer, a third-year computer engineering student, said he is using machine learning to measure carbon dioxide levels in areas affected by wildfires.

While one goal of this project is to map carbon dioxide levels, the project also seeks to reduce the cost necessary for environmental science, he said. 

"It's really important that we lower the costs for these sensors,” he said. “We need lower cost tools, because a lot of these communities don't have the funding.” 

So what is the price difference between these tools? Scheer says the $30,000 machines typically used in this study could eventually be replaced by inexpensive alternatives that cost $60. 

Scheer said the inexpensive sensors are less accurate than their thousand dollar counterparts, but can be calibrated with AI to match the results of high-end equipment. 

Dr. Joanna Casey, assistant professor of physics and engineering, agrees with the necessity for inexpensive alternatives. 

According to the World Health Organization, 7 million people die premature deaths due to air pollution, Casey said. 

“Having low-cost tools to measure air quality and levels of pollution can help people understand and minimize their exposure, and have lower and less health consequences,” she said. 

And for Durango, an area affected by wildfire smoke, students have a perfect testing ground, Scheer said. 

While Scheer’s project is about a year’s time from completion, he is currently working to collect wildfire data, such as at the recent Perins Peak fire, he said. 

However, this process of machine learning is slightly different from deep learning language models, such as the previously mentioned ChatGPT. 

Anders Ladow, a third year computer engineering major and recent AI collaborator with Scheer, said that machine learning models require human intervention. 

“You have to define exactly what the machine learning algorithm is doing,” he said. “What you give to it to analyze has to be really specific, and the algorithm can't make any changes to that data that you're feeding to the model.”

The main difference between deep learning models, like ChatGPT, and Scheer’s machine learning project is that deep learning models can actively change the data sets it has been fed, Ladow said. 

Despite these differences, both models are very useful for data extraction, Ladow said. 

Additionally, Casey said that air quality sensing systems using machine learning have already entered the market.

“We're standing on the shoulders of giants,” Casey said. “What we're able to do now is move into more complex problems that would be difficult to model or understand without these tools.”

Some of these problems that artificial intelligence could assist with are analyzing complex visual data, such as analyzing security footage, Ladow said. 

While tangible effects of artificial intelligence are likely a few years away, projects like Scheer’s highlight the capabilities of machine learning.