For example, certain models use reported sector intensity averages to estimate Scope 3 emissions. The limitation with this approach is that it fails to account for differences between similar companies in the same sector, for example, Tesla versus Volkswagen when it comes to fuel source. Additionally, this approach does not solve for the inconsistent nature of the data on which the estimates are based, further perpetuating inaccurate Scope 3 emissions data.
Using Models to More Accurately Estimate Illusive Scope 3 Emissions
Since Bloomberg’s Scope 1 and 2 estimation models are trained on reported emissions data, the lack of quality Scope 3 emissions data means Bloomberg could not employ the same machine learning techniques. Instead, Bloomberg continues to refine its approach for estimating Scope 3 emissions for specific sectors using a combination of a bottom-up model with a top-down machine learning model. This approach works best for industries such as Oil & Gas and Metals & Mining since it requires the use and processing of sold products as a key input for estimating Scope 3 emissions.
To break this down further, the bottom-up model is composed of an operating metric multiplied by a GHG emissions factor. The best possible operating metric is the amount of product sold in the fiscal year, in units of production. When this detail is not reported by the company, Bloomberg uses production data as a proxy. The operating or production metric serves to represent the amount of a hydrocarbon or metal produced, extracted, or sold. For the carbon emissions factor, this is sourced from official government tables that show the number of emissions per unit that come from using or processing those materials.