Matching opportunistic column measurements of CO2 with pixel-wise scaled emission tracers from a forward model for the urban area of Thessaloniki – Can we detect strong sources?

A rapid reduction in greenhouse gas emissions is critical to limit global warming. In order to act effectively, it is important to be able to monitor emissions and their changes over time. A special interest lies in the monitoring of urban areas due to their significant contribution to the amount of global anthropogenic emissions. However, the heterogeneous structure of urban areas makes it difficult to attribute observed changes in atmospheric concentrations of e.g. CO₂ to localized emission sources.

Here, we present results from an opportunistic measurement campaign in the framework of the COllaborative Carbon Column Observing Network (COCCON) in Thessaloniki, Greece. During the campaign period in October 2021 and summer 2022, XCO2 amounts were observed with two EM27/SUN FTIR spectrometers located at different places in the city. A total of 20 days of co-observations at different locations were recorded, with differences between the measurement locations of up to 2 ppm.

These observations are compared to regional hindcasts generated with the NWP forward model ICON-ART using emissions from the high resolution ODIAC inventory. The agreement between pairs of observed and simulated XCO2 columns obtained in this way is often limited, while other meteorological quantities are well represented in the model. Assuming that the largest source of the XCO2 discrepancies is originating from the inventory, we fragment the urban area of the inventory into different pixels, simulating the contribution of each individual pixel as a separate tracer within the model. The pixel-wise emitted tracers are scaled after run time to optimize the agreement with the observations. In this linear superposition the re-weighting of tracers imply which pixels need to be assigned with higher emissions than stated by the ODIAC inventory. As expected, the agreement between measured and modeled XCO2 columns can be significantly improved with this method, while regions with potentially high emissions (e.g. the harbor area) receive an upscaling.

This demonstrates that even smaller datasets without strong emission signatures can contain extractable emission information when processed carefully in conjunction with a good meteorological forward model.