Background in EU context on accessibility indicators
Accessibility is a major issue in the EU narratives since a long time and the Maastricht Treaty of 1992, when the concept of Trans-European transport Networks (TEN) was introduced for the first time (Spiekermann et al. 2015). The idea behind is the assumption that transport infrastructures and accessibility are necessary conditions for economic growth in the Union, having a direct impact on the attractiveness of regions for business and people. This led to important researches and indicators creation at regional scale (NUTS2, NUTS3) on potential accessibility to economy (GDP, in general) or labour market (population). Since the economic crisis of 2010’s, global discourse based on economic growth has slightly moved: The Territorial Agenda of the European Union, a policy document agreed at the Informal Ministerial Meeting of Ministers responsible for Spatial Planning and Territorial Development on 2011, stated that a fair and affordable accessibility to services of general interest, information, knowledge and mobility are essential for territorial cohesion. Providing services and minimising infrastructure barriers can improve competitiveness, and the sustainable and harmonious territorial development of the European Union (European Union 2020a). Still today, in the Territorial Agenda 2030 of the EU (European Union 2020b) remains the need to enable more equal opportunities, including accessibility and affordability to public services for people and enterprises, whatever they are located (art.6, 26, 27, 30, 47, 50, 62). It underlines also that this especially concerns remote areas that lack access to public services and economic and social opportunities (art.28).
Numerous projects and researches have been carried out during the last decades to support these policy perspectives, namely in term of territorial observation within the ESPON Programme (Mathis et al. 2004; Spiekermann et al. 2015; Kluge and Spiekermann 2017). These projects have provided concepts, indicators, tools and recommendations to monitor accessibility measures in a European context. From a methodological point of view, the indicators created within these analytical and conceptual projects combine generally three approaches:
Travel cost: these indicators measures the accumulated travel cost to reach a destination. Travel costs are popular because easy to interpret (average travel cost, travel time).
Cumulated opportunities: number of activities, population, economic activities, education or tourist amenities that can be reached in a given time (How many intercontinental flights can be reached from European regions within a maximum travel time of three hours ? How many people can I reach within a day round trip ?) The choice of the time threshold is crucial for the calculation of these indicators and depends ideally on the traveller realities (business traveller, population living in a rural area) and the service this person is looking for (an intermodal connection, a job, a service).
Potential accessibility: it is based on the assumption that the attraction of a destination increases with size, and declines with distance, travel time or cost. Destination size is generally represented by population, but it can also other indicators, such as GDP. Thus, potential accessibility can be seen as an indicator useful for evaluating the size of a market for suppliers of goods and services (Kluge and Spiekermann 2017). It allows to identify the relative competitive position of a territory (generally a region) towards given destinations.
The choice of appropriate origins and destinations depends largely on the scale of analysis: the ESPON TRACC project (Kluge and Spiekermann 2017) identified several spatial contexts (Global, European, Regional) for selecting relevant destinations: at global level access to global cities and major intercontinental hubs; at European level to main European metropolitan areas; at local scale to basic facilities (Kompil et al. 2019) or services of general interest (Noguera et al. 2017; Kompil et al. 2022). In that domain, interesting is the initiative led by L. Dijkstra and Poelman (2008), who used a travel time indicator to measure remoteness of rural NUTS 3 regions. A NUTS 3 was considered as peripheral if more than half of its population have a travel time by car of more than 45 minutes to the nearest city (more than 50 000 inhabitants). It classified regions as Predominantly Urban, Intermediate Close to a City, Intermediate Remote, Predominantly Rural Close to a City and Predominantly Rural Remote.
All these valuable inputs have provided - and continue to provide - important conceptual, theoretical and methodological basis to follow-up the accessibility of European territories at several territorial scales, and helping policy makers to have a better view on transport and service investments needs. However, three limits can be considered in this existing framework and deserve further investigations:
Rural-oriented approach: Most indicators focus the accessibility of territories to urban functions (transport hubs, employment areas, population centers or services concentrations). It would be interesting to adapt the scope to centrality markers also existing in rural areas.
Local scale: Most indicators and analysis are aggregated at NUTS levels (NUTS2, NUTS3). It provides a valuable picture of the European area. But for some aspects, finer geographical grain is required to discuss on intra-regional disparities, to overcome MAUP effects or more basically to better identify challenges that population have to face locally. According to our knowledge, no harmonized database on accessibility at EU local scale have been fully implemented.
Reproducibility and open source solutions: This is probably the most important. All the existing indicators produced in a European context use at the moment proprietary data. In general, the project coordinator is also the one who build and maintain (or pay for) the transport network and its attributes (maximum speed, transit, traffic congestion, etc.) required to compute accessibility indicators. This implies breaks in the reproducibility of the workflow and, as a consequence, difficulties to maintain or update core accessibility indicators. Taking the example of the ESPON Program, accessibility indicators updates required several contracts to the owner of the methodology, and the transport network behind. Even if the work done in this context is undoubtedly of high quality, we argue that exploring open source solutions have to be considered to conduct a more in depth democratization of these indicators, their use, their possible updates or, more important, possible criticism on the methodology behind.
General analyticial framework
This section shows the overall analytical framework behind the calculation of accessibility indicators. This allows to introduce the challenges in term of data collection and quality assessment developed in the next sections of the document.
Input data
Several data are required to build accessibility and travel time indicators (figure below).
Origins are required for travel time calculations. In the context of territorial observation in the field of accessibility, territorial divisions can be used (A). In this case, it requires to define geographic coordinates for the origins. Geometric centroids of the territorial units can be considered, or a specific location for each territorial unit of the nomenclature (demographic peak, city hall, etc.).
Two families of potential destinations for travel time indicators can be distinguished: First, destinations to a specific amenity or service, such as hospitals (C). Some attributes are ideally linked to these points of interest (capacity, opening date, private-public operator, etc.). The more accurate the attributes of the points of interest are, the greater the diversity of accessibility indicators will be. Another approach consists in considering as a destination the location that the origins (symmetric matrix). The interest here consists in using masses associated to the territorial units (D), such as population, GDP or jobs to compute specific accessibility indicators afterwards.
The network infrastructures (B) are obviously a central component in this framework. It corresponds to a structured network of lines (edge-expended graph) with specific attributes (maximum speed, traffic lights, service penalties, stops, directions, etc.). Routing engines use shortest path algorithms. One of the most known shortest-path algorithms was proposed by E. W. Dijkstra (1959). It finds shortest path between two vertices of a graph. Several improvements have been applied to this early proposal, to speed-up computing and tackle diverse problems of a general and complex graph (Schweimer et al. 2021). Transport modes (road, rail, air, bike, walk) require to be analyzed separately since they obviously do not follow the same itineraries and characteristics. Moreover, the transport network and its specificities can be affected by natural aspects (lakes, mountains) or by human amenities (a border, a road restricted to specific uses, etc). Routing engines allows to take into account most of the time these characteristics to compute accurate travel time indicators between origin-destination pairs.
Travel time calculation through routing engines
When origin destination pairs are identified and gathered, routing engines allow to get the associated travel geometries, its associated length, duration and sometimes costs. Several providers deliver this service, based on proprietary and commercial (Google routing engines, …) or free and open source (OSRM, …) solutions.
As an example, below is presented outputs coming from two routing engines for the travel Montpellier train station - CIHEAM premises location, by car and by bike. Results are slightly different: 10.8 km and 21 minutes for OSRM ; 7.4 km and 20 minutes for Google. This is explained by travel time algorithms which are not the same according these data sources and the fact that Google takes into consideration traffic congestion.
By bike the travel corresponds to 6.5 km and 34 minutes for OSRM and 5.9 km for 25 minutes for Google. These examples shows interestingly that in both cases the travel is adapted to the transport mode : shortest path and little roads by bike ; higher speed roads with more distance by car.
The challenge here consists here in calculating these travel time for a large amount of origin destination pairs. This is indeed not possible to using demo API for OSRM or Google Map API without specific installation or paying the access to large matrixes.
