AI Algorithm Maps Out Detailed Carbon Emissions for City Managers

It is part of China's efforts to peak its carbon dioxide emissions by 2030 and achieve carbon neutrality by 2060, the Xinhua reported.

The experimental tool of researchers from the Aerospace Information Research Institute (AIR) under the Chinese Academy of Sciences is a refitted carbon-monitoring vehicle which takes a joy-ride regularly at bustling zones in the southern Chinese city of Shenzhen.

At an experimental day, a white cylinder rose slowly on top of a van to start sensor-enabled carbon emission monitoring in the city. The carbon monitor was synchronizing the collection of carbon dioxide emission concentration on the road.

Different from ordinary carbon monitors, the vehicle was also mounted with panoramic cameras to capture street scenes. The first-hand data was transmitted to a laboratory at AIR in Beijing, where a brand-new deep learning algorithm is being trained.

Driven by the team's deep learning algorithm, those cameras could discern carbon emission and sink sources, such as moving cars, buildings and carbon-fixing vegetation, in real complex scenarios. Then the AI model can produce estimates on the carbon contributions, either positive or negative, of those sources.

"Take a single site as an example. The AI analysis revealed that small cars contribute the most to emissions, followed by trucks," said Wang Li, a researcher from AIR. "The surrounding roads and buildings also contribute to emissions."

Currently, Shenzhen has six carbon monitoring towers, but that's inadequate to make a detailed emission map and the building of more towers will be costly, making alternative and complementary monitoring methods necessary, said Zhang Yonglin, a research assistant in Wang's team.

It occurred to Zhang, who has a computer science background, that the AI could lend a hand and he developed the algorithm.

"You have at least 100 elements at one single site, and the objects keep changing," said Zhang. "Only AI can calculate viable results via an inverse modeling."

Now, a total of 650 kilometers drive-around journey has been completed, obtaining more than 100,000 records of street images and road traffic carbon emissions in Shenzhen.

After training, the AI model can produce a high-definition map within a distance of 100 meters and its accuracy for road traffic carbon emission intensity has exceeded 92 percent, according to Zhang.

The model is projected to provide a fine guide for city managers, like what causes a certain block to see higher carbon emissions and possible solutions behind it. That's an elaborate work the remote-sensing carbon satellites and monitoring towers cannot do.

"The results can remind policy-makers to take measures such as increasing efficiency of access in certain road sections, or planting more trees at a block," said Zhang.

"Sometimes, too much greenery is not all beneficial for reducing carbon concentration when it results in a pipeline effect," he added. "That's what the AI map can tell you."

4155/v