Did you miss a session from the Future of Work Summit? Head over to our Future of Work Summit on-demand library to stream.


In 2015, representatives from more than 196 countries met in Le Bourget, France to sign the Paris Agreement. The legally binding treaty limits global warming to a rise of well below 2 degrees Celsius compared to preindustrial levels, preferably capping warming at 1.5 degrees. While the Paris Agreement doesn’t spell out how the undersigned are expected to achieve this goal, some countries have pledged to cut their net climate emissions to zero by 2050.

For these and other steps to be successful, reliable data is key. While the ability to evaluate companies’ carbon footprints will be critical for countries seeking to comply with the measures, only a fraction of companies currently disclose their greenhouse gas emissions. But researchers at Bloomberg Quant Research and Amazon Web Services claim to have successfully trained a machine learning model to estimate the emissions of businesses that don’t disclose their emissions.

The researchers say investors could use this model to align their investments with international regulatory measures and achieve net-zero goals. Some regions, including the European Union, require investors to apply a “precautionary principle” that penalizes non-disclosing companies by overestimating their emissions.

“The project was born out of necessity, really,” Bloomberg global head of enterprise data science at Bloomberg Naz Quadri told VentureBeat via email. “Climate risk is a key sustainability topic and in order to manage it effectively, financial market participants need to better understand the carbon footprint of companies in their investment or loan portfolios. Unfortunately, current disclosure by companies is low, so the market needs to use estimates. The main driver for this project was to fill in the gaps in reported carbon data.”

Training the model

Prior work attempted to estimate companies’ carbon emissions using a combination of conventional statistical approaches and machine learning. But according to the researchers, these approaches relied on assumptions that don’t always hold true in reality, like the idea that companies in the same industry emit roughly the same level of emissions.

To train their model, the researchers identified more than 1,000 features and 24,052 rows of disclosed emissions from datasets containing company financials (like balance sheets and income statements), corporate locations, and ESG records. The ESG records had over 500 metrics alone, covering areas like carbon emissions and resource and energy use; human rights and diversity and inclusion; and criteria based on management structure, executive compensation, and employee relations.

In an experiment designed to evaluate the model’s accuracy, the researchers say the model closely estimated the emissions of companies in industries including health care, technology, financial, materials, real estate, utilities, energy, communications, and more. In future work, the team plans to add more features from datasets across areas like corporate policy, supply chain, and factory data.

“We are currently working on expanding our product. Phase 1 of the project focused on “scope 1 and 2″ emissions,” Quadri said. “We are now focused on adding scope 3 emissions which relate to emissions across the supply chain of companies, requiring a completely new approach to modelling. In addition, we’re studying the possibility of creating estimates for other … metrics, such as water consumption.”

While studies suggest some forms of machine learning contribute significantly to greenhouse gas emissions, the technology has also been proposed as a tool to combat climate change. For example, an IBM project delivers farm cultivation recommendations from digital farm “twins” that simulate the future weather and soil conditions of real-world crops. Other researchers are using AI-generated images to help visualize climate change, and nonprofits like WattTime are working to reduce households’ carbon footprint by automating when electric vehicles, thermostats, and appliances are active based on where renewable energy is available.

Facebook chief AI scientist Yann LeCun and Google Brain cofounder Andrew Ng, among others, have argued that mitigating climate change and promoting energy efficiency are worthy challenges for AI researchers.

“Similar to how AI and machine learning techniques help tackle the complex problems in self-driving cars and news content understanding, they can be utilized to learn the complex relationships between the thousands of attributes of a company and its greenhouse gas emissions alongside the data challenges mentioned above,” Quadri added. “Our estimations can help regulators as well as investors to make more effective decisions to mitigate greenhouse gas emissions.”

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn More