Private Vehicles Greenhouse Gas Emission Estimation at Street Level for Berlin Based on Open Data

Abstract

As one of the major greenhouse gas (GHG) emitters that has not seen significant emission reductions in the previous decades, the transportation sector requires special attention from policymakers. Policy decisions, thereby need to be supported by traffic emission assessments. Estimations of traffic emissions often rely on huge amounts of actual traffic data whose availability is limited, hampering the transferability of the estimation approaches in time and space. Here, we propose a high-resolution estimation of traffic emissions, which is based entirely on open data, such as the road network and points of interest derived from OpenStreetMap (OSM). We estimated the annual average daily GHG emissions from individual motor traffic for the OSM road network in Berlin by combining the estimated Annual Average Daily Traffic Volume (AADTV) with respective emission factors. The AADTV was calculated by simulating car trips with the open routing engine Openrouteservice, weighted by activity functions based on statistics of the German Mobility Panel. Our estimated total annual GHG emissions were 7.3 million t CO₂ equivalent. The highest emissions were estimated for the motorways and major roads connecting the city center with the outskirts. The application of the approach to Berlin showed that the method could reflect the traffic pattern. As the input data is freely available, the approach can be applied to other study areas within Germany with little additional effort.

1. Introduction

In the year 2021, the transportation sector was the third largest greenhouse gas (GHG) emitting sector in Germany [1]. Predominantly caused by road traffic, it accounted for approximately 19% of the total emissions in 2021 [1]. Decarbonization of the traffic sector is, therefore, an essential step to mitigate climate change [2]. However, despite the improvements in efficiency and consumption of motorized vehicles, recent transportation emissions were still at a similarly high level as in the year 1990 [3]. Consequently, the mitigation of traffic emissions demands special attention from policymakers, which needs to be supported by a verifiable assessment of traffic emissions on a local up to national scale.

A common approach for the estimation of GHG emissions is to combine traffic volumes of road segments with corresponding emission factors [4,5,6,7,8,9,10,11]. For example, Keuken et al. [5] estimated traffic CO₂ emissions in Rotterdam, Netherlands, and Basel, Switzerland, using traffic volume data provided by these cities. They project the traffic emissions for 2020 using one emission factor for all road types: 300 g CO₂/km. Pla et al. [8] use traffic volumes from urban traffic control and monitoring systems and a city-specific emission factor of 205.47 g CO₂-eq/km to estimate traffic GHG emissions in Valencia, Spain.

While many of these studies estimate GHG emissions at a high spatial resolution, they are often limited to certain metropolitan areas and more importantly, they do not openly provide spatially explicit data on emission estimates. Thus, high-resolution and open traffic data covering entire countries such as Germany is currently generally not available, preventing measurement-based verification of emissions on the local scale. Additionally, these studies require a vast amount of information to provide GHG emission estimates at high accuracy, as GHG emissions of traffic are influenced by many factors, such as fuel consumption, age and composition of the vehicle fleet, traveling speed, and traffic congestion. Because of differences in these parameters between different cities and countries, the emission factors applied to calculate traffic GHG emissions also vary between different cities and countries. Therefore, research has also been performed to derive city-specific emission factors based on test drives, e.g., in Bejing, Guangzhou, and Macau, where an emission factor of 257 g CO₂/km was derived for gasoline vehicles [12].

Several cities and local authorities conduct road traffic volume surveys [13,14], but their data use is often restricted. As the traffic volume data tends to be limited to smaller-scale administrative units, an application to larger regions necessitates dealing with data sets from different providers. Due to the heterogeneity of the data collection methods and the different spatial and temporal resolution of the data, such a procedure would require considerable effort. Therefore, the transferability of these approaches to other areas is limited.

An alternative to the use of traffic data can be the simulation of the traffic volume using social media data [15] or telecom data [16]. Centrality parameters such as betweenness centrality and network centrality are also considered an effective measure for analyzing network structures and are applied by a variety of studies, e.g., for analyzing urban traffic flow [15,17,18], as well as for analyzing the quality of highway tags of OpenStreetMap (OSM) road features [19]. Based on our literature survey, centrality measures have, so far, not been applied to estimate GHG emissions from road traffic.

Further investigations predict and simulate traffic emissions based on a wide range of existing emission calculation models, such as the Calculation of Air Pollutant Emissions from Road Transport (COPERT) model [20] and the DEMO model [21]. However, these models are very complex and, therefore, not well suited for local governmental institutions and especially smaller communities.

The research gap can be summarized as the following: Overall, scalable approaches to estimate traffic GHG emissions that do not rely on traffic data with limited availability are missing. We contribute to filling this research gap by proposing a high-resolution estimation of GHG emissions by motorized private traffic based on open data that should be transferable to most regions in Germany with little additional effort. The approach does not require detailed traffic count data. Instead, it is based entirely on open data.

2. Methods and Data

2.1. Data and Study Site

Berlin, the capital of Germany, is located in eastern Germany and covers an area of about 890 km², of which 15% is traffic area [22]. In 2021, Berlin had approximately 3.66 million inhabitants. The population density was higher in the central part of the city within the inner commuter train circle [23]. Berlin is crossed by the motorways A100/A111/A113 in the NW-SE direction and the motorway A115 in the SW-NE direction. These motorways all connect to the motorway ring A10 surrounding Berlin and leading long-distance traffic around the city (Figure 1).

We used road network data, points of interest (POI) data, survey data on travel behavior, data of the spatial population distribution, the total population of Berlin, data on the stock of motorized vehicles, and emission factors. The data was accessible without any charge. We extracted the road network from OSM based on a bounding box (13.014° E to 13.834° E and 52.293° N to 52.719° N), which covered the city of Berlin and an additional buffer of a 5 km minimum to the city borders.

For the information on travel behavior, we used the German Mobility Panel (MOP), a longitudinal and nationwide survey performed annually over the last 25 years [24]. In our study, we referred to the MOP time series data, which was collected from September 2019 to February 2020 and is considered the most recent data not affected by the 2020 COVID-19 pandemic [24]. Population density information was obtained from the Global Human Settlement Layer (GHSL) [25]. The current population estimation of the Office for Statistics of Berlin-Brandenburg [23] was used to calibrate the population of our research area calculated based on the GHSL. The stock of motorized vehicles by the federal state was provided by the Federal Motor Transport Authority [26]. To calculate the GHG emissions, we used emission factors provided by Transport Emission Model (TREMOD) [27], which are based on the Handbook Emission Factors for Road Transport (HBEFA) measured vehicle statistics [28].

2.2. Data Processing

Our approach consisted of three key steps: First, we generated the betweenness centrality of each OSM road segment in Berlin based on an OpenRouteService (ORS)-based simulation. Second, we estimated the Annual Average Daily Traffic Volume (AADTV) of privately owned, motorized vehicles based on the results of the ORS centrality simulation and on statistical information about the total number of trips in the research area. Finally, we estimated the GHG emissions for the road segments by combining the AADTV and the respective emission factors provided by TREMOD.

2.2.1. ORS Centrality Simulation

In our approach, betweenness centrality was expressed by the number of simulated trips on a certain road segment. We calculated centrality by simulating 20,000 trips themed and weighted by human action functions (

Table 1. Human action functions based on [15], computed weights of the human action functions as well as trip groups by purpose and respective percentage of the traffic volume according to the German Mobility Panel (MOP). Since the human action functions and the MOP grouping are not identical, we harmonized the classifications by combining the themes of the human action functions and the MOP trip purposes. We interpreted the MOP group “back home […]” as the respective return trip. Therefore, the percentage of the traffic volume for the group “back home […]” was split relatively among the percentages of the other trip groups. The weights of the human action functions were computed as the cumulative percentage of the traffic volumes, which were assigned to the respective human action function. The percentages of the trips back home are lower because one can, e.g., go to work, then to recreation and then home. * Differences in the totals are due to rounding. ** Designations translated from German.

Human Action Function [15]		MOP Statistics
Theme	Weight (%) *	Trip Group by Purpose **	Traffic Volume (%) *
1. Work	24.8	Work, official or business back home […]	14.2 10.6
2. Education	2.4	Education back home […]	1.4 1.0
3. Shopping	49.7	Procurement and service Other private errands back home […]	19.9 8.5 21.3
4. Recreation	22.9	Recreation back home […]	13.1 9.8

) using the ORS. Table 1 outlines the human action functions “work”, “education”, “shopping” and “recreation” [15]. We weighted the functions based on statistical information of the traffic volumes obtained from the MOP (Table 1). Since the human action functions and the MOP grouping are not identical, we harmonized the classifications by combining the themes of the human action functions and the MOP trip purposes. We interpreted the MOP group “back home […]” as the respective return trip. Therefore, the percentage of the traffic volume for the group “back home […]” was split relatively among the percentages of the other trip groups. The weights of the human action functions were computed as the cumulative percentage of the traffic volumes, which were assigned to the respective human action function. The workflow of the ORS centrality simulation is described in detail below.

A general overview of the inputs, trip requirements and output of the ORS Centrality simulation is given in Figure 2. We extracted topological information as well as the road classes in the research area from OSM [29]. To avoid translation misunderstandings, we have used the term “roads” for all types of streets, as opposed to the OSM tag “highway = *” which also includes footpaths and objects on roads, such as crossings and rest areas. Trips were simulated using the profile “driving car”, which considers only roads with the tags highway = [motorway, motorway_link, motorroad, trunk, trunk_link, primary, primary_link, secondary, secondary_link, tertiary, tertiary_link, unclassified, residential, living_street, service, road, track]. Roads with the tag “track” were only included if they were accessible for cars. We downloaded the selected POIs in Berlin using the Ohsome API [30].

Every single trip was characterized by its starting point, its destination point and their respective distance. The starting points were generated by random sampling inside the bounding box described above, using population density as a weight. The destination points were sampled from OSM POIs weighted by the importance of the human action functions the POIs belonged to (Table 1). The assignment of the POI to the human action functions was determined by the respective main activity performed at the POI (see Appendix A for a list of POIs assigned to each human action function).

In addition to the placement of the starting points and the destination points, we integrated two requirements for the length of the trips. First, assuming that the car is not used for trips shorter than 1 km, only trips longer than 1 km were accepted. Second, based on empirical findings [31], the simulation generated more short-distance trips than long-distance trips using the distance-decay function from Zia et al. [15]. The routes generated by the simulation of the trips were then matched to the OSM road network using the fast map matching developed by Canfast [32].

Finally, we computed centrality as the cumulative number of trips passing through a road segment. We normalized the centrality values by the number of simulated trips. Using a representative number of 20,000 trips, this simulation could reflect the traffic pattern within the research area. Therefore, the normalized centrality served as a ratio value for the following calculation of the AADTV.

2.2.2. Calculation of the AADTV

The AADTV is defined as the number of vehicles passing a road segment per day on an annual average [33]. We calculated the AADTV for each road segment as a proportion of the total number of trips by privately owned, motorized vehicles within the study area, using the normalized centrality as a ratio (c.f. Figure 3). First and foremost, we computed the population of our research area, the Berlin bounding box, in order to calculate the total number of trips by privately owned, motorized vehicles of all residents. The population of the bounding box of Berlin was calculated by summing up the raster values of the GHSL within the bounding box. Equation (1) shows the computation. To calibrate the calculated population, an extrapolation factor f was generated according to Equation (2) and the official population size of 3.6641 million for the city of Berlin [23].

$$po{p}_{BB}=f\times {\displaystyle \sum _{n=1}^{n_{BB}}}valu{e}_{BB}$$

(1)

$$f=\frac{po{p}_{Berlin}}{{\sum }_{n=1}^{n_{Berlin}}valu{e}_{Berlin}}$$

(2)

where:

• $po{p}_{BB}/po{p}_{Berlin}$: population size within the Berlin bounding box/within the city limits of Berlin;
• $valu{e}_{BB}/valu{e}_{Berlin}$: raster value of the population distribution raster layer extracted from the GHSL for the entire Berlin bounding box/for the city of Berlin;
• $n_{BB}/n_{Berlin}$: total number of raster cells of the population density raster for the entire Berlin bounding box/for the city of Berlin;
• f: extrapolation factor.

The total number of trips per day was calculated by multiplying the total population within the bounding box of Berlin $po{p}_{BB}$ by the annual average daily number of 1.71 car trips per person provided by the MOP statistics [24]. Finally, we estimated the AADTV for each road segment as the product of the normalized centrality value and the computed total number of trips (Equation (3)).

$$AADT{V}_{rs}=\frac{c_{rs}}{20000}\times 1.71\times po{p}_{BB}$$

(3)

where:

• $AADT{V}_{rs}$: annual average daily traffic volume of the given road segment;
• $c_{rs}$: centrality (cumulative number of trips passing through the given road segment);
• $po{p}_{BB}$: population size within the Berlin bounding box.

2.2.3. Estimation of the GHG Emissions from Motorized Vehicles

Figure 4 provides an overview of the GHG emission estimation from the AADTV. We classified the OSM road segments into three categories based on the road types of TREMOD shown in

Table 2. Weighted average emission factors of vehicles (year 2020) provided by Transport Emission Model (TREMOD).

Road Types	CO2 Equivalents per Vehicle Kilometer (g)
motorways	189
roads outside of built-up areas	148
roads inside of built-up areas	210

: OSM “motorways”, “roads outside of built up areas” and “roads inside of built up areas”. OSM roads tagged as “highway = motorway” were categorized as motorways. All remaining roads were categorized as “roads outside of built-up areas” if they were outside of the administrative boundary of Berlin and as “roads inside of built-up areas” if they were inside the administrative boundary of Berlin. The emission factors ef were calculated as weighted averages between the emission factors for cars and the emission factors for two-wheelers provided by TREMOD [27] (Equation (4)). The weights w were based on the number of motorcycles and cars in Berlin provided by the Federal Motor Transport Authority [26].

$$ef=e{f}_{cars}\times w_{cars}+e{f}_{2wheelers}\times w_{2wheelers}$$

(4)

The weighted average was calculated under the following two assumptions. First, privately owned motorized vehicles were mainly cars and motorcycles. Second, the emission factor of two-wheelers served as the average emission factor of all types of motorcycles.

Finally, we estimated the annual average daily GHG emissions for each OSM road segment by multiplying the AADTV by emission factors for privately owned motorized vehicles for the corresponding road type.

3. Results

The total annual GHG emissions from privately owned motorized vehicles of residents within the administrative boundary of Berlin were estimated as 7.3 million t CO₂ equivalents per year (

Table 3. GHG emissions from privately owned motorized vehicles of residents in the city of Berlin aggregated by human action functions.

Human Action Function	CO2 Equivalents per Year (kt)
1. Work	1787.69
2. Education	182.45
3. Shopping	3646.96
4. Recreation	1717.03
Total	7334.11

).

The spatial pattern of emissions of privately owned motorized vehicles in Berlin was heterogeneous (c.f. Figure 5). Motorways and primary roads directly connected to the city center showed the highest emissions (>30 t CO₂ equivalent per road kilometer per day). The two roads with the highest emissions were motorways A100 and A115, exhibiting emissions > 50 t/day on most of their segments. Connecting the suburbs and the motorway ring around Berlin (A10) with the city center, they serve as important commuting routes. The A115 also connects Berlin with Potsdam, the capital of the federal state Brandenburg. The road segments with the highest emissions (>100 t/day) were on motorway A100 between exit Kaiserdamm Süd and Dreieck Funkturm, and between exits Gradestraße and Oberlandstraße. These also had the highest estimated AADTV, which was roughly 500,000 private motorized vehicles/day. The motorway ring around Berlin (A10) had lower emissions (20 to 30 t/day). Residential and minor roads generally had low emissions (<10 t/day).

4. Discussion

4.1. Annual Average Daily Traffic Volume (AADTV)

Overall, our approach could reflect the traffic pattern in Berlin, with the highest AADTV estimated for motorways and main roads connecting the city center with the outskirts and lower AADTV on residential and minor roads. This traffic pattern is in line with studies of other cities, as found by Keuken et al., McDonald et al., Li et al. and Wen et al. [5,6,11,34]. However, the estimated AADTVs on the motorways of Berlin, with a mean AADTV of 94,200 vehicles per day, were unrealistically high. Comparable studies found traffic volumes between 1000 and 2000 vehicles per hour on the motorways of Chengdu [11], up to 4400 vehicles per hour on the motorways of Bejing [6], and around 40,000 vehicles per day on the motorways of Basel and Rotterdam [5].

Our approach has several limitations, which may lead to an over- or underestimation of AADTV on individual road segments. For example, some assumptions in our model are based on average values for motorized individual traffic in Germany. First of all, the average trip length in Berlin is only 6 km [35], while it is 11 km in German metropolises overall [24]. This has a significant influence on the traffic volume. Further, the number of trips per person with motorized individual traffic in Berlin is presumably smaller than the German average of 1.71 [24], as the share of trips traveled with motorized individual traffic in Berlin is only 34%, while the German average is 57% [36]. Another limitation is that the actual shares of the trips according to the different human action functions in Berlin might differ from the numbers in the MOP that were used in our study. Moreover, our method generates trips based only on population density and POIs and, therefore, does not account for socioeconomic factors such as income or unemployment rate. Due to the reasons stated above, the absolute AADTV estimates should be treated with care. However, the relative proportions of the AADTV values across space are presumably realistic.

Accounting for local peculiarities of mobility behavior, such as the average trip length or the motorization rate in Berlin, could presumably also improve the accuracy of the AADTV estimation. However, the required mobility data for this is not always available. Therefore, we deliberately chose average values for Germany in order to facilitate the nationwide applicability of the approach, accepting that this would cause inaccuracies. If local mobility statistics are available, the parameters of our method can still be adapted to a specific local use case.

It has been projected that despite a declining population, the number of registered cars in Germany will probably increase until 2030 [37]. This implies that the factors used for calculating AADTV, such as the average daily number of trips per person, should be updated regularly, as the relationship between population and traveled car mileage is subject to change. Moreover, the effects of the COVID-19 pandemic on mobility have elucidated that travel behavior can change rapidly [38].

Relying entirely on open data, our approach is flexible and can be adapted to different time periods and spatial extents, thus contributing significantly to the prospect of estimating historic emissions and serving as a baseline. Moreover, it can reflect changes in travel behavior because the shares of the activity functions of the total number of simulated trips can be adapted, e.g., to have a higher share of work and education trips on weekdays than on weekends.

4.2. Emission Factors

The highest emission factor was used for roads inside Berlin, which is sensible considering that city traffic is usually characterized by low speeds with frequent starting and stopping. GHG emissions of a car are usually highest at the startup stage of the vehicle and at low speeds [6]. This is reflected by the high emissions of primary roads inside Berlin. However, there is considerable variance among the emissions of the roads inside Berlin, with major roads in the city center and roads connecting the center to the outskirts having higher emissions than roads in the outer districts. This shows that the emissions on the main roads in the city center are caused by much higher traffic flows than in the outskirts. Motorways in the city have among the highest emissions despite the lower emission factor used for motorways, showing that the motorways inside the city have especially high traffic flows. The lowest emission factor used for roads outside of Berlin is reflected by low emissions on all roads outside the city limits.

Our emission factors (Table 2) were similar to the 205.47 g CO₂-eq/km that Pla et al. [8] derived for Valencia. The emissions per road kilometer can vary between different cities, e.g., because of differences in the vehicle fleets [5]. This is shown by the emission factors used in comparable studies, which range from 205.47 [8] to 300 g CO₂/km [5]. Because of these differences in the emissions per road kilometer, it may be advisable to derive local emission factors using methods as in Pla et al. [8] or Zhang et al. [12] when applying our approach to a different study area.

4.3. Emissions of Privately Owned Motorized Vehicles

The total estimated GHG emissions of road traffic in Berlin of 7.3 million t CO₂ equivalent per year were in line with the road traffic CO₂ emissions estimated for Rotterdam, which were 1.1 million t CO₂ per year [5]. Since the population of Berlin is roughly six times the population of Rotterdam, the per capita road traffic emissions estimated for both cities are similar. However, the road traffic CO₂ emissions of Basel, which were estimated as 0.2 million t CO₂ per year [5], were much lower, even when accounting for the difference in city size, considering that the population of Berlin is roughly 21 times the population of Basel. This difference in the emission estimates can be attributed to a lower number of per capita vehicle kilometers in Basel [5], highlighting that travel behavior can vary significantly across different cities, even within Europe.

The spatially explicit information on traffic emissions in the results of our study can be used to identify the location of high-emission road segments and to analyze the causes of the emissions in-depth. It is not surprising that the highest emissions were estimated for the motorways connecting the city center with the suburbs and the remaining motorway system since these motorways do not only serve as important commuter routes but also have to handle the traffic between Berlin and other places in Germany and beyond. Primary roads connecting the city center with the suburbs also had high emissions because they were important commuting routes as well. Conversely, residential and minor roads probably had generally low emissions because they are widely distributed throughout the city and, therefore, disperse traffic flow, which is in line with Li et al. [6].

The spatial emission pattern estimated for Berlin in our study was similar to studies of other cities around the world, e.g., Beijing, Los Angeles, Dallas-Fort Worth, or Chengdu [6,11,34]. However, in Beijing, ring roads have a high share of traffic emissions [6]. Conversely, the motorway ring A10 around Berlin had lower emissions than the motorways running through the city because, unlike the ring roads in Beijing, it circles the city outside the city borders and presumably mostly carries traffic going from places outside of Berlin to places outside of Berlin. Much of this traffic was not captured by our approach because trips were only simulated within the Berlin bounding box and not to or from other cities that are further away. Conversely to Berlin, the cities of Los Angeles and Dallas-Fort Worth are characterized by a lower population density and a more extensive highway network, leading to a more even distribution of emissions over the highway network [34].

Since our study lacks a comparison with local GHG measurements, this would be an important step for future research. Moreover, the method could be enhanced in the future to also include pollutants such as carbon monoxide, nitrous oxides and particulate matter. Spatially explicit information on these pollutants is relevant as well because these pollutants are associated with an increased respiratory risk in spatial proximity to roadways [7].

5. Conclusions

Our study estimated the total annual GHG emissions from privately owned motorized vehicles in the city of Berlin as 7.3 million t CO₂ equivalent per year. The emissions ranged from <10 t/day on most residential and minor roads up to 100 t/day and more on the motorway segments with the highest emissions. The study showed that a betweenness centrality-based approach relying on open data could be applied to estimate GHG emissions of motorized individual traffic. As it depends entirely on open data, the approach provides the opportunity to estimate spatially explicit GHG emissions of road traffic for other areas in Germany. The spatially explicit information allows for in-depth analyses of the emissions of the road system, such as identifying the locations of especially high-emission road segments. While the general spatial patterns of the estimated emissions can be trusted, the actual emission values of the road segments should be treated with care. Since local travel behavior can deviate from the German averages used in our approach, the estimated emission values are subject to uncertainty. Further limitations to the validity of our emission estimates are that the generation of trips does not account for socioeconomic factors and that we do not compare the estimated emissions to local measurements. The accuracy of the emission estimates could be improved by adapting the parameters to the mobility behavior in the respective study area.

Despite its limitations, our centrality-based approach has potential for future applications because it is scalable, both spatially and temporally. Its potential applications range from analyses of GHG emissions for specific traffic situations of a city, such as morning rush hours, up to a nationwide estimation of GHG emissions of motorized individual traffic in Germany. The results of our study can also be used as prior emissions for a measurement-based validation on a local scale in an inverse framework such as [39].