Bloomberg has increased its carbon emissions dataset to cover 100,000 companies.

This data consists of company reported carbon data and estimates based on either a machine-learning smart model or Bloomberg’s newly developed industry-implied model accompanied with a PCAF reliability score.

When company reported carbon emissions data is not available, Bloomberg applies estimation techniques guided by a waterfall principle to increase the scope of companies covered and thus provide a more complete picture of emissions. The first estimation technique uses Bloomberg’s machine learning-based smart model incorporating 800+ data points to estimate Scope 1 and 2 emissions with historical data going back to 2010. Last year, Bloomberg also released Scope 3 estimates for Oil and Gas, Metals and Mining, and Services industries using a methodology combining a bottom-up model with a top-down machine learning model.

When there is not enough data available to apply these machine-learning smart models, the waterfall technique automatically reverts to Bloomberg’s new industry-implied model. This method uses peer emissions and sales data to estimate a company’s emissions. These estimates are accompanied by a reliability score, using the scale that has been proposed by the Partnership for Carbon Accounting Financials (PCAF) so investors can understand the quality of the underlying datapoint.