Estimating the Carbon Dioxide Emissions of Thessaloniki by Combining Observations and Simulations Based on an Inventory Rescaling Approach

Our planet is warming at a rapid pace. The increase in global surface temperature is driven by the growing concentration of greenhouse gases in our atmosphere, primarily due to the extensive use of fossil fuels. Currently, the most important greenhouse gas that is emitted from human activities is carbon dioxide (CO2). Climate change is already having a dramatic impact on our environment, and the effects are projected to intensify as emissions continue. To mitigate the worst effects of climate change, targeted action is needed to achieve a rapid reduction in the emissions of all greenhouse gases, not only CO2. Precise monitoring methods are needed to improve current emission inventories for informed decision-making.

However, the currently available emission inventories contain considerable discrepancies. Especially at smaller scales, such as urban areas, the uncertainties are high. Concurrently, urban areas are major sources of anthropogenic greenhouse gases. Existing inventories are constructed using a bottom-up approach based on reported emission activities. Top-down estimation of emissions is an alternative approach, where emissions are calculated from atmospheric observations. ... mehrThe method can be used to validate and improve bottom-up estimates. It has been applied in several cases, mostly covering global to regional scales. Top-down estimation of urban emissions is challenging because of the dense and heterogeneous emission structure, and the difficulty to perform accurate simulations of atmospheric transport at small scales.

In this work, I perform a measurement-based estimation of the CO2 emissions of the urban area of Thessaloniki, Greece. For Thessaloniki, the discrepancies between the bottom-up inventories are substantial: For 2019, the EDGAR inventory reports urban emissions of 3.1 Mt, which is significantly higher than the estimates by the CAMS (1.7 Mt) and ODIAC (1.8 Mt) inventories.

For the emission estimation, a measurement campaign was conducted in collaboration with partners from the Aristotle University of Thessaloniki. Two solar Fourier Transform Infrared spectrometers of the type EM27/SUN were operated over a period of three months, covering three weeks in October 2021 and ten weeks between May and July 2022. The EM27/SUN can record time series of the column-averaged dry-air molar fraction of CO2 (𝑋CO2) with high precision and temporal resolution. One spectrometer was positioned centrally within the city. The second spectrometer was equipped with solar power supply for portability, enabling positioning at various locations. The collected dataset contains 179 hours of observation, excluding the observations for calibration.

To interpret the recorded dataset, a corresponding simulation of the emission sources and transport of CO2 is required. The ICON-ART model – an operational model at the German Weather Service (DWD) – was used for this purpose. The ODIAC inventory was used as a starting point, and within the city it was separated into different sub-areas to allow scaling of the emissions during post-processing. An estimate of the net ecosystem exchange was constructed from different available datasets. The simulations show a good agreement when compared to wind and pressure observations in the city and water vapor columns co-observed by the EM27/SUN. However, the agreement is poor for 𝑋CO2. The Pearson correlation coefficient between the simulated and observed time series of 𝑋CO2 was only 0.1.

The agreement is enhanced significantly by rescaling the anthropogenic emissions inside the city area. To find an optimal scaling, a least-square approach is applied. Two different methods are compared: Firstly, the whole city is uniformly scaled, leaving just one degree of freedom. Secondly, the previously separated source regions of the city center are scaled individually, giving 30 degrees of freedom. Furthermore, two subsamples with good prior agreement are selected to evaluate the robustness of the results. The optimization leads to a significantly improved agreement between simulated and observed time series for both subsamples. The largest improvement is found in the smaller sample, where 5 days were selected from the full time series. Here, the correlation coefficient improves from 0.34 to 0.77. Still, discrepancies remain in the time series for all configurations. Possible reasons for this include the inaccurate representation of the biogenic sinks and sources, imperfect simulation of tracer transport and limitations of the model setup such as short simulation time and limited spatial resolution.

Despite the remaining discrepancies, the different scaling configurations show robust results: When looking at the configurations where all source regions are scaled individually, expected emission hot spots consistently receive higher weights, supporting the correct attribution of emissions by the optimization approach. For every configuration, the rescaling results in a distinct increase of Thessaloniki’s total emissions. The estimates range from 2.9 to 4.4 Mt/yr. This indicates an underestimation of the actual emission in the ODIAC and CAMS inventories and supports the higher estimate from the EDGAR inventory. The results demonstrate the potential of measurement-based methods to enhance our knowledge about urban scale emissions.