Quantification and attribution of urban fossil fuel emissions through atmospheric measurements

Background Fossil fuel combustion causes an increase in atmospheric carbon dioxide (CO2) levels and is one of the major causes of climate change. Therefore, efforts are made to reduce CO2 emissions from fossil fuel combustion through (inter)national agreements, with the most famous example being the Paris agreement. Each member state that ratified the agreement has to aim for pre-set emission reduction targets. In this collaborative effort it is important to keep track of the progress made towards these targets, but also to gain insight in which emission reduction policies are most effective to support future decision-making. Therefore, scholars have started developing atmospheric monitoring techniques, mainly focused on urban areas. Since about 70% of the anthropogenic CO2 emissions takes place in urban areas, the largest emission reductions will take place here. This causes large atmospheric signals that are relatively easy to measure. However, scholars have faced some major challenges. For example, the transport within a built-up area is complex, making the interpretation of atmospheric observations difficult. Moreover, emission reduction policies often target specific source sectors (such as road traffic or industry). Hence, these sectors should be monitored separately to understand the effectiveness of individual measures. This source attribution is impossible with only CO2 observations when source sectors are not spatially isolated. Aim The overall aim of this thesis is to improve our understanding of the monitoring requirements to constrain urban fossil fuel CO2 emissions per source sector. A key feature of a monitoring system is a network of observation sites. Therefore, the first research objective is to identify the most useful monitoring sites and network configurations. Besides CO2 we also included measurements of trace gasses that are co-emitted with CO2 during fossil fuel combustion. This happens in a ratio that is specific for a source sector and therefore these tracers have the potential to identify the source of a CO2 signal. We examined this opportunity to use co-emitted species to attribute CO2 signals to specific source sectors. Besides observations a good model representation of atmospheric transport is needed to interpret the observations. Therefore, the second research objective is to better understand the possibilities and limitations of atmospheric transport models in reproducing observed mixing ratios within/close to a city and find a useful modelling approach. The third objective is to predict high-resolution emissions in an urban area using proxy data and to gain insight in the uncertainties related to these emissions. Finally, we combine our insights related to measurements, models and emission modelling into an inversion framework to estimate how well we can constrain urban CO2 emissions per source sector (objective 4). Results and conclusions In Chapter 2 we examined the effectiveness of two observation sites close to the city border of Rotterdam, providing a gradient in the CO2 mixing ratio over the city from the upwind to the downwind site. The two sites provide one year of hourly mixing ratio gradients which are used to make a first estimate of the urban emissions. For this purpose we first examined whether the upwind site was representative for the composition of the background signal, which proved to be the case for specific wind directions. We found on average large enhancements at the downwind site compared to the upwind site for three major source areas: the city, the port and the glasshouse area. From the selected gradients we calculated emissions, accounting only for average biospheric fluxes, footprints, and boundary layer height. Although this approach is very simplified it shows reasonable flux estimates compared to the reported emissions. Nevertheless, we found that the estimates can be heavily influenced by local emissions and by transport processes that we could not take into account. For example, the presence of elevated stack emissions complicates the estimate of the emissions without detailed knowledge of the atmospheric transport. Finally, the results show that CO can potentially attribute a CO2 signal to industrial or residential source areas. We conclude that observed mixing ratio gradients can be used to make a rough estimate of the urban emissions, in which CO is of added value to identify dominant source types. In Chapter 3 we compared two atmospheric transport models: the Eulerian WRF-Chem model (1x1 km2 resolution) and the Lagrangian OPS model. Atmospheric transport models are useful to account for the impact of transport, mixing, entrainment, and biospheric fluxes on the observed mixing ratios and can help interpret the observed signals. We examined the ability of these models to reproduce the observed mixing ratios at several measurement sites along a transect from an urban (Rotterdam) to rural location. On average, WRF-Chem gives good results, reproducing meso-scale features with the correct order of magnitude for the observed CO2 mixing ratios. However, the timing of CO2 mixing ratio enhancements is often incorrect, which is mainly the result of an incorrect representation of the wind direction causing the model to sample the wrong source area. Moreover, we found that the representation of point sources is problematic. In a Eulerian model emissions get instantly mixed throughout the grid box, which causes a large underestimation of local and downwind mixing ratios for sources with a small horizontal extent. Using the OPS model improves the representation of point sources, because it has no spatial discretization. The difference between OPS and WRF-Chem is only visible up to approximately 15 km from major stack emissions, such that point sources further away from observation sites can be represented by WRF-Chem as well. An additional advantage of the OPS model is that it can be driven by locally observed meteorological data, such that it overcomes the wind direction issue from WRF-Chem. However, the OPS model is sub-optimal for area source emissions over a large domain and therefore we conclude that a combination of both models is the best option in Rotterdam. Finally, the results in Chapter 3 show that urban sites are well-exposed to urban fossil fuel fluxes and can be used to separate between different source areas (such as the residential and industrial area), especially if besides CO2 also CO is included. Sites that are further removed from the city (semi-urban) provide a better constraint on the total flux. Chapter 4 explored the potential of several data streams to predict high-resolution emissions. These data were combined in a dynamic fossil fuel emission model that estimates emissions based on additional knowledge about the emission landscape. First, we calculated the total yearly emissions for the Netherlands per source sector using activity data (such as Gross Domestic Product), emission factors (the amount of CO2 emitted per amount of fuel consumed) and energy efficiency (amount of fuel consumed per amount of activity). Then the total yearly emissions were disaggregated to hourly and 1x1 km2 scale using proxies and hourly activity data. In this way we created a dynamic emission map based on a wide range of parameters that are specified per source sector. One major advantage is that we can estimate the (unknown) uncertainty in the high-resolution emissions from the (better-known) uncertainty in the model parameters. We find that we can estimate the yearly emissions for the Netherlands with a 15% uncertainty when using generalized proxies (i.e. based on general, large-scale activity data and emission factors). Using more specific knowledge about the region (e.g. about technological advancement) and local activity data reduces this uncertainty. We can also use the emission model to calculate emissions of co-emitted species by multiplying the CO2 emissions with the typical emission ratios for each source sector. These emission ratios are variable and uncertain and the emissions of co-emitted species have a larger uncertainty than the CO2 emission. Finally, the model parameters have a physical meaning and can be linked to emission reduction policies, making it a useful tool for policy-makers. With the dynamic emission model we identified the most important and uncertain parameters affecting the emissions (CO2 emission factors, emission ratios and time profiles). In Chapter 5 we tried to optimize these parameters using a newly developed inverse modelling framework. The inversion system uses the multi-model framework described in Chapter 3 to translate the emissions calculated by the dynamic emission model into mixing ratios of CO2, CO, NOx (nitrogen oxides) and SO2 (sulphur dioxide). We used the same modelling framework to create pseudo-observations, which are used to validate the model. The only difference is the values appointed to the parameters in the emission model (generalized data for the prior, local data for the pseudo-observations). We performed an experiment to explore the difference between an urban and a rural observation network, which shows that the CO2 signals captured by the rural network are too small to contain relevant information. The urban network performs well and gives a good estimate of the total yearly emissions for the Rotterdam area (5% error). When we included observations of the co-emitted tracers the emission estimate per source sector generally improves. Some sectors remain difficult to constrain, for example due to the lack of large enhancements or the lack of a clear emission ratio signature. The time profiles can also be constrained relatively well, at least the day-to-day variability. However, for households the error in the time profile gets aliased into the emission factor, causing the emission factor to be less well constrained. When we introduced erroneous atmospheric transport the results deteriorate drastically, especially for power plants and industry (i.e. point sources) which suffer most from the transport errors. We conclude that an inversion system with a dynamic emission model as a prior has great potential for monitoring urban emissions, but transport errors currently hamper its applicability to real observations. This work contributed to a better understanding of the complexity of the urban fossil fuel emissions and what is needed to monitor this. Urban observations provide useful information and, depending on the size and shape of the monitoring network, can be used to constrain urban emissions in more or less detail. Observations of co-emitted species have the potential to attribute CO2 emissions to specific source sectors and are an important addition to our inversion framework. The dynamic fossil fuel emission model has several major advantages over a regular emission map, being flexible and physically meaningful. Although several challenges remain, the work described in this thesis is an important step in the development of urban monitoring capacities