accessibility | Marcin Stępniak

Temporal resolution

Sun, 05 May 2019 00:00:00 +0100

The study of an impact of temporal resolution on precision of travel time and accessibility measurement in public transport analysis.

Publication

Stępniak, M., Pritchard, J. P., Geurs, K. T., & Goliszek, S. (2019). The impact of temporal resolution on public transport accessibility measurement: Review and case study in Poland. Journal of Transport Geography, 75, 8–24. https://doi.org/10.1016/j.jtrangeo.2019.01.007 (Open access)

Abstract

In recent years there has been a significant increase of temporally variable analyses of accessibility by public transport as a result of the increased availability of open and standardized time table information in the form of GTFS (General Transit Feed Specification) data. To date, very little attention has been paid to systematically analyze the impact of temporal resolutions on the results. Different authors have applied different standards, often in an ad-hoc manner. In this study, we address the loss of precision associated with a stepwise reduction of the temporal resolution of travel time estimations based on GTFS data for the city of Szczecin in Poland. The paper aims to provide guidance to researchers and practitioners on the selection of appropriate temporal resolutions in accessibility studies. We test four sampling methods in order to analyze four different public transport frequency scenarios, three types of accessibility measures (travel time to the nearest provider, cumulative opportunities measure and potential accessibility) and seven types of destinations ranging from high to low centrality. Additionally, the impact on spatial disparities is explored using the Gini coefficient.

We find that a reduction of temporal resolution is associated with a decrease in precision of public transport accessibility measurement. However, with up to 5-min resolutions this reduction is negligible, while computational time is reduced fivefold, compared to a 1-min resolution benchmark. Lower temporal resolutions still provide relatively precise estimations of travel times and accessibility measures. However, further resolution reductions are associated with decreasing reductions of computational time. As a result, we argue that 15-min temporal resolution provides a good balance between precision and computational time while providing very precise estimations of Gini coefficients (errors ≤0.001).

A non-linear relationship is found between the public transport frequency and the loss of precision, with lower frequencies leading to a greater loss in precision. More attention should be paid to highly centralized services, in particular when analyzed using proximity and cumulative opportunities measures. Finally, the cumulative opportunities measure is found to be highly sensitive to changes in the temporal resolution and not suited for time-sensitive accessibility analysis.

Repository

All the data used for the study, including its detailed description, can be downloaded from the open data repository ( link). The data is shared under the CC-BY-4.0 licence.

Reference: Stepniak, M.; Goliszek, S.; Pritchard, J.; Geurs, K. (2019) The impact of temporal resolution on public transport accessibility measurement. RepOD. http://dx.doi.org/10.18150/repod.7727991

All the R code used for the analysis is available from the github repository: https://github.com/stmarcin/Temporal-paper.

Additionally, a separate code which can be used to generate departure times for a given day (e.g. to conduct accessibility analysis or calculate origin-destination matrices), using user-defined sampling method and temporal resolution is stored in separated github repo: https://github.com/stmarcin/Sampling_Departure_Time.

GTFS study

Sun, 05 May 2019 00:00:00 +0100

Evaluation of a supply side of public transport using GTFS data.

Introduction

The aim of the study is to evaluate a supply side of public transport. The assumption that lies behind this study is that it should minimize the amount of data required and to guarantee its fully replicability. The result is a set of indicators which enable to evaluate public transport network in a given area and easily compare the results between different cities, functional urban areas (FUA) or metropolises, including international comparisons.

The presented example uses an example of Madrid.

Data

Three different types of data are used in the study:

GTFS (General Transit Feed Specification) which composes of a series of text files with data which contain all information about public transport schedules and associated geographic information.
Data source for an example: Open data from the Consorcio Regional de Transportes de Madrid.

GTFS feed (part)

Route network derived from OpenStreetMaps
Data on population distribution: population grids from Eurostat GEOSTAT initiative

Back to top

Evaluation

The evaluation procedure is divided into three main parts, preceded by an overview of available (or colleced) GTFS data:

stops location and their accessibility
frequency of service
accessibility

Preview

Before the evaluation, there is a need to revise a basic spatial and temporal information about all available feeds for a given case study. A map shows a spatial range of all available GTFS feeds comparing to the city and FUA limits.

The table provides information about the names and temporal coverages of all feeds in order to confirm that they cover the same period of time. It is also used to select a “typical day” for further analyses (different feeds contain a calendar data coded in a different way, thus a selection of one day is necessary in order to avoid a double-counting of particular trips and/or departures). In this example, based on the data shown in the table, we focus on a “typical working day”: 2018/09/11.

Back to top

Stops location

First part of the evaluation focus on a description of public transport network, its spatial pattern and accessibility. The table summarize basic indicators of the network using all available GTFS feeds, providing insights about existing transport modes, number of stops and total number of the departures during the selected day.
Note: due to different systems of coding of calendar, departure times and frequencies. The table provides unified information, regardless:

existence of frequencies.txt file - if this file exists, not every departure time is listed in stop_times.txt, what needs to be recalculated; this is usually the case of high frequency transport modes, like metro;
overlapping calendar data - some datasets code departure time as 25:30:00 (HH:MM:SS, i.e. 01:30:00 + 1 day) which refers to the departure realized the next day in relation to the analysed. If this is the case, there is a need to include part of the departures which are assigned to the previous day than selected for the analysis.

Finally, apart from the division by transport modes, the table summarize differences between the city within its limits, FUA and the whole GTFS dataset(s).

Then, maps visualize distribution of public transport stops in order to analyse their spatial pattern and, e.g. compare a supply of public transport during the day (peak hours) and night (low frequency, limited service).

city	FUA
stops in service during peak hours

stops in service during a night

The last part focuses on accessibility to public transport: what is a walking distance to the nearest stop in service? What is a difference between peak hours and night time? What is a difference between the city and its FUA? The next set of maps and graph address these questions.

city	FUA
walking distance to stops in service during peak hours

walking distance to stops during a night

Back to top

Frequency

The frequency of public transport differs in a course of a day, in line with peaks and valleys of density of human mobility: more departures take place during peak hours, less in out-of-peak period and much less during the night. Each city or each country has its own curve of daily changes of frequency. Nevertheless, it is important to properly identify periods different frequency, especially, when doing international comparisons. These differences are visualized by the graph presented below. Note, that y-axis shows a total number of departures and not frequency - the latter would be difficult to visualize as there are changes in e.g. number of lines. Moreover, some differences in frequency patterns may occur between a city and its FUA, so both patterns are presented simultaneously. Additionally, it enables to check what share of departures in the all FUA are realized within city limits and how it changes during a day.

Back to top

Accessibility by public transport

Regardless the number of stops, lines, departures etc., the most important function of public transport is to enable people to reach their destinations. Thus, the most important indicator which evaluates public transport network is a level of accessibility which offers this network. In this example, a potential accessibility is used, with the number of population as a proxy of destination’s attractiveness. This indicator includes relations between all pairs of origin–destination nodes in a given area and it assumes the greater importance of larger centres than smaller ones and the diminishing attractiveness of more distantly located destinations and it is expressed by the following formula.

$$A_ {i} = \sum_{j}g(M_j)*f{(t_{ij} )} $$

where $A_i$ is a potential accessibility of a zone $i$, $g(M_j)$ is the function of destination attractiveness of a zone $j$ (e.g. number of population¹), and $f{(t_{ij} )}$ is a distance decay function. In a given example, a negative exponential is used as distance decay function, with $\beta = 0.0223$ (i.e. a destination loses half-value of its attractiveness at 31 minutes travel time).

¹ we use number of population due to the fact that population data are the most broadly available in high resolution datasets.

The results are presented as a set of maps which compare level of accessibility during the day (peak hours) and night and they are presented in a high resolution level of 1km² grids. Map for FUA are accompanied by zoom-in maps limited to the city area.

city	FUA
accessibility by public transport during peak hours

accessibility by public transport during a night

The last graph compares a share of population which has a particular level of accessibility (as a relation to the highest possible value measured in during a day).

Back to top

Contributors

This study was prepared in collaboration with:

Chris Jacobs-Crisioni (European Commission, Joint Research Centre)
David Sousa Vale (University of Lisbon)

Back to top

Policy support tool

Sun, 05 May 2019 00:00:00 +0100

Policy support tool for efficient transport policy: using temporal sensitive transport data to identify main accessibility restrictions.

Introduction

The main aim of the study is to prepare a methodological guidelines and describe data needs related to the evaluation of an impact of particular transport related restrictions of accessibility at the level of the whole city or metropolitan area (global restrictions) and spatial pattern of their distribution (local restrictions). The results of the analyses enable to identify main causes of the limited accessibility in a given area which in turn can be used as a base of the strategic decisions leading towards more efficient transport system.

The study extensively uses a potential of new, temporally sensitive data sources and methods which incorporate a temporal variability into the accessibility analysis. New data facilitates to compare accessibility by different transport modes (e.g. public transport and private car), as well as to identify different factors affecting spatial patterns of accessibility, including geography, quality of transport network and congestion levels and organization of public transport, including its routing, frequencies and timing.

The bottom line of the study is to apply a comparative approach in order to detect main reasons that limit accessibility level in particular areas. We compute travel times by private car using free flow and congested speeds, and we calculate travel times by public transport with a fine temporal resolution, applying a schedule-based, public transport data (GTFS format). Additionally, we prepare a pseudo-GTFS dataset, transforming the original GTFS feeds in a way that we exclude any waiting time from the model, i.e. travel time includes only in-vehicle time (or times in case of transfers) and walking to, from (i.e. from origin to stop and from the final stop to the destination) and between public transport stops (in case of transfers).

The presented document uses Madrid as a case study and focuses on accessibility to jobs during the morning peak hours (i.e. 7-10am). However, depending on particular needs and availability of data, jobs might be replaced by any other type of destination.

Back to top

Data

The main characteristic of the tool are small data requirements. In fact it requires three types of data:

road network data which permits to calculate travel time with free flow and congested speeds;
GTFS feed in order to calculate travel time by public transport (for the details consult GTFS study);
data for the origin-destinations zones

Ad. 1. In the presented example of the application of the tool the TomTom® Speed Profiles are used in order to prepare origin-destination matrices with free flow and congested speeds. However, Speed Profiles can be replaced by another (similar) data (e.g. OpenStreetMaps, HERE Traffic, TravelTime Platform or Inrix Traffic, among others).

Speed Profiles is a network dataset, where a particular edge of the network (i.e. road segment) contains a very precise information about an average travel speed, depending on: day of the week and hour with 5-minute temporal resolution. The graph below presents a relative change of speed during the course of the day (typical working day) with 100 equal to a free flow speed of a particular road segment. The graph presents a shape of curves of almost 300 different “speed profiles”.

Figure: Speed Profiles

Ad. 2. GTFS feeds used for this example are provided by Consorcio Regional de Transportes de Madrid.

Figure: Structure of GTFS feed (sample)

Apart from real GTFS feed a pseudo-GTFS is required in order to evaluate an impact of route network (routing scheme) on accessibility level. The pseudo-GTFS is a public transport network dataset with no waiting times included (i.e. door-to-door travel time is equal to in-vehicle and walking to/from/between stops).

Ad. 3. This example focuses on accessibility to jobs and it uses number of jobs as a proxy of destination attractiveness. The data are provided by Instituto de Estadistica de Comunidad de Madrid. Additionally the population number per zone is used in order to: (1) remove areas of low population density from the analysis, and (2) to calculate weighted averages of accessibility level at the city level.

Back to top

Data preparation and preliminary analysis

Preparation of Origin-Destination matrices

The input data for the the analysis are several origin-destination matrices (OD matrices) for a particular transport mode. As travel time by public transport highly depends on the departure time (as it affects waiting times, transfer times etc.) in case of public transport there are several OD matrices, one for each of the departure times. Based on the previous study we use a hybrid sampling method and 5-minute temporal resolution, in order to reduce an impact of temporal resolution on travel time measurement. In result, we use 36 OD matrices which are then aggregated in order to obtain:

average travel time (using a hybrid-based average, for the details see: Stępniak and Jacobs-Crisoni, 2017)
minimum travel time in order to prepare a “best-case scenario” (the highest possible accessibility level of a given area during the morning peak hours).
maximum travel time in order to prepare a “worst-case scenario”
The last two are used to o evaluate an impact of public transport schedule on accessibility level, i.e. to what extent the fact that one can (or cannot) freely select their departure time affects the level of accessibility they experience.

In case of private car accessibility, two matrices are required: one for flee flow speeds and one for congested speeds. In case of availability of more temporally detailed data (i.e. several OD matrices for different departure times, e.g. every 15 minutes), the minimum and maximum travel time may be prepared, as it is done in case of public transport. Nevertheless, the variability of travel time by car is more regular and smooth, thus this information is only a supplementary one.

In result the number of OD matrices is limited to eight: four for public transport and four for private car travel times.

Accessibility calculation

In the second step, the accessibility values for each of the scenarios are calculated. In this case a potential accessibility measure is used:

`$$A_ {i} = \sum_{j}g(M_j)*f{(t_{ij} )} $$

where $ A_i $ is a potential accessibility of a zone $ i $, $ g(M_j) $ is the function of destination attractiveness of a zone $ j $ (e.g. number of job), and $ f{(t_{ij} )} $ is a distance decay function. In a given example, a negative exponential is used as distance decay function, with $ \beta = 0.0223 $ (i.e. a destination loses half-value of its attractiveness at 31 minutes travel time).

Additionally, the coefficient of variation is calculated as it facilitates to evaluate a variability of travel time on accessibility level (only for public transport accessibility).

Scenario Abbreviation	Description	Comment
Car accessibility
FreeFlow	Free flow speed	Benchmark scenario
Car_Best	Car best-case scenario	Congestion
Car_Avg	Average car	Congestion
Car_Worst	Car worst-case scenario	Congestion
Public transport accessibility
FullFreq	PT - no waiting time	PT route network
PT_Best	PT best-case scenario	Frequency of PT
PT_Avg	Average PT	Frequency of PT
PT_Worst	PT worst-case scenario	Frequency of PT
PT_VarCoeff	Coefficient of variation of PT accessibility	variability of PT accessibility

The dataset which contains all the data above can be found in the project open data repository.

Spatial pattern of accessibility

Once accessibility values are calculated, we can focus on preliminary analysis of spatial patterns of accessibility in particular scenarios. Four of them are particularly important from the policy perspective: accessibility by car with free flow speeds, accessibility by car with congested speeds, accessibility by public transport in no-waiting-times scenario and average accessibility by public transport. In order to facilitate comparison of maps, all of them follow the same color palette and breaks.

accessibility by car	accessibility by public transport

The preliminary analysis of spatial patterns of accessibility enable to identify main regularities in spatial distribution of accessibility level. First, it seems that all of them follow the same central-periphery division. However, some irregularities as well as differences between scenarios can be found. In case of congestion, there is a difference between south and north (and south-west in particular), while in case of public transport we can find a star-shape pattern, along the rail links. Moreover, one can find a strong difference in accessibility level between car and public transport scenarios. The next section focuses on these general differences.

Back to top

Accessibility restrictions at city scale

The aim of this section is to facilitate a general comparison accessibility values in different scenarios and detect the main factors which limits accessibility at the city (metropolitan) level. The box-plot shows a general distribution of accessibility level, while table focuses on relative differences, but it additionally uses population weighted averages in order to reflect the extend to which observed differences affect inhabitants.

Speed Profiles

First, we can see that accessibility in all car scenarios are higher than in public transport ones. The exception is made only for not realistic scenario with no-waiting times, which is slightly higher than the most congested scenario. Second, the impact of congestion is less significant (less than 10%) than impact of change of transport mode (~25%). Finally, selection of a particular departure time is more important in case of travel by public transport than in case of car trips (difference between best and worst case scenarios: 90.7% vs 78.1%). This shows that those who travel by public transport have to larger extent adapt their daily schedule to public transport schedule than car traveler to the traffic conditions (congestion).

Table 1: Population weighted relative differences between all accessibility scenarios
	Car accessibility	Public transport accessibility
scenario	Free flow	Best	Average	Worst	no waiting times	Best	Average
Car Best case	95.4
Car Average	90.3	94.6
Car Worst case	86.5	90.7	95.8
PT no waiting times	87.1	91.3	96.6	100.9
PT Best case	76.5	80.2	84.8	88.5	87.7
PT Average	68.0	71.2	75.3	78.6	77.8	88.7
PT Worst case	59.9	62.7	66.3	69.2	68.5	78.1	87.9

Back to top

Accessibility restrictions at local scale

This section focuses on spatial pattern of impact of particular accessibility restrictions. The title of a particular map indicates which of the two scenarios are compared against each other. The in-depth analysis enable to evaluate the extent to which a particular transport-related accessibility restriction affects inhabitants of a given zone.

Spatial pattern of accessibility restrictions

Back to top

Bivariate maps

The next set of maps focuses on an evaluation of impacts of pairs of factors. As the maps presented above show a spatial pattern of impact of a particular restriction, bivariate maps puts these results in the broader context. Apart of the evaluation of a given factor, they inform what these restrictions means for inhabitants, e.g. congestion affects inhabitants in the areas with low level of accessibility by public transport to the larger extent, as people have limited alternatives other than travel in traffic jams). Using this kind of illustration facilitates to identify areas of intervention which should be addressed in the first order.

First map confronts impact of congestion and the level of accessibility by public transport. If high level of congestion affects areas with relatively efficient public transport, it may stimulate inhabitants to use public transport instead of private car, as it becomes to be more competitive. On the contrary, high level of congestion and low accessibility by public transport means that inhabitants are stuck in traffic jams because they don´t have any reasonable alternative. In this case policy makers (or transport planners) should focus on this area first.

Second map facilitates to identify how public transport can be improved: to what extent accessibility in a given area is limited due to low (insufficient) frequency or in order to improve accessibility, there is a need to reorganize a route network of public transport.

The third map analyse to what extent inhabitants of particular zones are double affected by limited accessibility by public transport, i.e. not only by low average level of accessibility by public transport, but also by the high variation in accessibility level (so to larger extent their daily schedule need to focus a public transport schedule).

Back to top

Concluding remarks

The presented decision support tool facilitates an identification of the most severe restrictions of accessibility level, so transport planners and policy makers will know what they should focus on in order to improve quality of life of inhabitants of a city. It provides a general information about the most important accessibility restriction at the city level and provides a set of very detailed information about the spatial pattern of impacts of particular restrictions.
In the future, a technical note will be shared through a github repository and relevant information will be shared through project repository. At the moment, an example dataset of accessibility values used for the presented study is available throught the open data repository.

Back to top

Contributors

This study was prepared in collaboration with:

Borja Moya-Gómez (tGIS Research Group, Complutense University of Madrid)
Juan Carlos Garcia Palomares (tGIS Research Group, Complutense University of Madrid)
Amparo Moyano (Department of Civil Engineering, Universidad de Castilla-La Mancha)

Back to top