Fossil fuel consumption projection and emission extrapolation

This page describes a simple, purely data-based method to project fossil fuel consumption and related CO$_2$ emissions within a year with available monthly data from eurostat. The basic idea of the approach is to extrapolate trends of fuel consumption of past months within a year to total yearly values.

As first step, the latest datasets from Eurostat are retrieved using the monthly commodity balances of natural gas, petroleum products and solid fossil fuels:

In case not all months of a year are available, yearly data are extrapolated based on projection factors that are calculated using past data. For each fuel $f$, each month $m$ and each past year $y$ (starting 2013), a scaling factor $F_{f,m,y}^{cons}$ is calculated. This scaling factor describes the ratio between the actual final consumption at the end of the year $C_{f,y}$ and the consumption up to the respective month $C_{f,m,y}$:

The monthly scaling factors of all years are assumed to be normally distributed. Their averag value $\overline{F^{cons}_{m,f}}$ is used to scale the actual data:

For each scaling factor, also a standard deviation estimator with rule-of thumb bias correction is calculated:

This deviation allows to estimate the uncertainty of the fuel consumption projection. The figure below shows results of the approach to predict data of 2022, with scaling factors calculated from the years 2013-2021. Although the prediction is not bad, there is still some systematic error that cannont be explained by pure statistical features. The unstable situation of energy markets in the last years might have added additional uncertainties. For example, the full year natural gas consumption is overestimated taking data until summer 2022. High prices during the second half of the year probably led to a consumption decrease compared to previous years that could not have been projected. Extrapolation results must therefore be interpreted with great caution. Data displayed on the tracker start page always use training data up to the last year.

In order to estimate CO₂ emissions from these data, fuel-specific emissions of the following commodities are calculated using emission factors taken from the National Inventory Report and net calorific values (NCV) taken from Statistics Austria Energy Balances:

With the yearly consumption data $C_{f,y}$ of fuel $f$, the $NCV_f$, and emission factor $e_f$, an estimate for the actual emissions $E^{estimated}_{y}$ can be calculated: $$E^{estimated}_{y} = \sum_f C_{f,y} \cdot NCV_f \cdot e_f$$ Note that Biofuel related consumptions (Biodiesel and Biogasoline) must be subtracted from the Diesel and Gasoline consumption data.
The figure to the right (upper chart) compares these estimated emissions with the actual emissions from the energy-related sectors (Transport, Energy&Industry, Buildings). It can be seen that there is a gap, probably caused by several systematic errors:

• The calculated data do not consider process emissions from non fossil-fuel based industry secors (mainly cement and lime production, approximately 2.5 Mt_CO2 in 2021).
• Calculated data focus on CO₂ emissions and do not consider CH₄ and N₂O emissions from, e.g., fugitive emissions.
• Energy-related emissions from the agriculture sector are included in the calculated emissions from the fuel balances, but not in the energy-sector emissions.
• Not all fuels are considered.
• Integrated process routes in the industry sector (especially steel and chemical industry) might be oversimplified.

To account for these systematic effects, an approach similar to the consumption projection above is used: For each year's estimated emissions $E^{estimated}_{y}$, a scaling factor $F^{em}_y$ is calculated, leading to an average $\overline{F^{em}_y}$ over all years with standard deviation $\hat{\sigma_{F^{em}_m}}$ that is used to project the total emissions of the energy sectors $E^{scaled}_{y}$: $$E^{scaled}_{y} = E^{estimated}_{y} \cdot \overline{F^{em}_y}$$ The figure to the right (lower chart) shows upscaled emissions (using $\overline{F^{em}_y}$) of all years. It can be seen that the approach estimates the actual emissions reasonably well.

Thje combination of fuel consumption extrapolation and emission estimation allows to estimate CO₂ emissions based on non-fully available yearly data. In the first step, yearly fuel consumption is extrapolated from monthly data and in the second step, emissions are estimated based on the described scaling: $$ E^{scaled}_{y} = (\sum_f C_{f,m,y} \cdot \overline{F^{cons}_{f,m}} \cdot NCV_f \cdot e_f) \cdot \overline{F^{em}_y}$$ The statistical uncertainties of both approaches add up (uncertainty propagation): $$\sigma^{E,total}_{y} = \sum_f (C_{f,m,y} \cdot NCV_f \cdot e_f \cdot \overline{F^{em}_y} \cdot \hat{\sigma_{F^{cons}_{f,m}}}) + (\sum_f C_{f,m,y} \cdot \overline{F^{cons}_{f,m}} \cdot NCV_f \cdot e_f) \cdot \hat{\sigma_{F^{em}_y}}$$ The first term describes the uncertainty from fossil fuel consumption extrapolation, the second term describes the uncertainty from emission scaling.
Finally, the emissions from other, not-energy sectors (Agriculture, Waste, Fluroinated Gases) are added. For those sectors, a more simple approach is chosen: the average emission trend of the past three years is extrapolated to the projection year.

The figure on the left shows the total projected emissions for the year 2022. Emissions are projected using data up to the respective month on the x-axis. As can be seen, projected yearly emissions are overestimated when using data of only the first few months, but fit better the more months are used. Again, energy market disruptions in 2022 may have caused stronger consumption decreases during the second half of the year, leading to systematic projection errors. The lower chart on the left shows the two uncertainties. As expected, the uncertainty from consumption extrapolation decreases the more monthly data is available, and the emission scaling uncertainty is more or less constant.