Advanced models for hourly marginal CO2 emission factor estimation: A synergy between fundamental and statistical approaches

Global warming is caused by increasing concentrations of greenhouse gases, particularly carbon dioxide (CO 2 ). The reduction of carbon dioxide emissions is thus an energy policy priority. A metric to quantify the change in CO 2 emissions is the marginal emission factor. Marginal emission factors are needed for various purposes, for example to analyze the emission impact of electric vehicle charging. This paper presents two methodologies to estimate the marginal emission factor in electricity systems with high temporal resolution. The first is an energy systems model that incrementally calculates the marginal emission factors. This calculation is computationally intensive when the time resolution is high, but is very accurate because it considers relevant market factors on both the supply and demand sides and emulates the electricity market dynamics. The second is a Markov Switching Dynamic Regression model, a statistical model designed to estimate marginal emission factors faster, and it is benchmarked against the dynamic linear regression model widely used in the marginal emission factor estimation literature. For the German electricity market, we estimate the marginal emission factor time series both historically (2019, 2020) using Agora Energiewende and for the future (2025, 2030, and 2040) using estimated energy system data. The results indicate that the Markov Switching Dynamic Regression model outperforms benchmark models. Hence, the Markov Switching Dynamic Regression model is a simpler alternative to the computationally intensive incremental marginal emission factor, especially when short-term marginal emission factor estimation is needed. The results of the marginal emission factor estimation are applied to an exemplary low-emission vehicle charging scenario to estimate CO 2 savings by shifting the charge hours to those corresponding to the lower marginal emission factor. We implemented the emission-minimized charging approach using both marginal emission factors. Over a 5-year period, the Markov Switching Dynamic Regression model appears to save 47.9 % of emissions on average, compared to 6.5 % real-world saving. The maximal value possible with incremental MEFs would be 37.6 %. • We compare a complex energy system model with a faster MSDR model for MEF estimation. • The energy system model is accurate but too computationally intensive for real-time use. • The MSDR model provides fast and efficient MEF estimation with high accuracy. • Benchmarking confirms that MSDR outperforms traditional statistical MEF estimation methods. • Emission-minimized electric vehicle charging using the MSDR model yields an apparent 47.9 % CO 2 reduction and incremental MEFs show real-world savings of 37.6 %.