SPATIO-TEMPORAL MODELS INCORPORATING NETWORK TOPOLOGY MEASURES FOR PEDESTRIAN EXPOSURE ESTIMATION

0
234
You can download this material now from our portal

SPATIO-TEMPORAL MODELS INCORPORATING NETWORK TOPOLOGY MEASURES FOR PEDESTRIAN EXPOSURE ESTIMATION

ABSTRACT

 

Analyzing the dynamic behavior of pedestrians is significant for the planning and designing of sustainable transportation networks in modern cities. A well designed and well planned network benefits social needs in terms of better security, healthier environment, and reduced economic costs due to crashes and congestion related lost productivity time. The study of pedestrian exposure analysis as it relates to pedestrian dynamics in the network, aimed at improving network security and efficiency is still at a very nascent stage. In the past, studies related to pedestrian exposure have been limited to understanding pedestrian flow on a macroscopic level. Thus, studies on the influence of network topology on the concentration of paths at a spatial-temporal level are deficient.

Space-time analysis of network topology and pedestrian exposure can provide a framework for integrated safety assessment in dense urban environments, especially in the emerging domain of connected infrastructure. This dissertation’s original contribution is in the area of spatio-temporal methods for micro-scale analysis of pedestrian exposure, with respect to network topology. The main objective of this study is to evaluate the impact of network topology on area-level, and potentially higher-level pedestrian density. Area-density is a reliable measure of space-time exposure that is an outcome of a pedestrian’s origin-destination paths. Higher-level analysis such as at the area-group level can also provide a hyper-level insight into the impact of network topology on pedestrian community density. By knowing the relative contribution of network topology accounting for pedestrian flow and destination attractiveness, insights on network design that minimizes exposure while also optimizing travel time can be gained. In this dissertation, pedestrian exposure is estimated by area-level and higher-level pedestrian densities, utilizing advanced methodologies that integrate area evaluation based on pedestrian dynamics

 

from social-force based micro simulation, and network properties from a fine resolution topological network model.

The scope of the study area is set to downtown Seattle in Washington State and the immediate surrounding areas. The network in the study area includes 11 corridors with 33 subnetworks, selected as being critical areas based on high spatial density, high pedestrian travel activity, and the potential for high pedestrian outflow onto the network in the event of an emergency. The data for the study are created using VISSIM and GEPHI, and analyzed at 5 minute intervals during the 3 peak periods (AM, MD, and PM peak).  The parameters for the area evaluation generated by the micro-simulation tool VISSIM are obtained from the dynamic behavior of pedestrians in certain areas, such as, the number of pedestrians, pedestrian area density, dwell time, and delay time. Parameters for dynamic network analysis, and metrics such as degree of node, clustering coefficients, closeness centrality, and eigenvector centrality, are produced through dynamic and hierarchical graphs using the network analysis tool GEPHI.

The objectives of this study require the integration of the spatial-temporal aspects of pedestrian exposure on the network. A spatial perspective on the impact of network topology as it relates to space-time concentration of pedestrians is developed. Spatial Autoregressive models with spatial-autoregressive disturbances and additional endogenous variables (SARAR) are developed for the study, allowing the estimation of spatial dependency, spatial heterogeneity and endogeneity in an integrated manner, while providing insights on the highest density estimates of network effects. As key parameters to estimate pedestrian exposure, pedestrian area density and time occupancy are employed, and these parameters are estimated from the variables of building volume generation, network topology measures along with area dynamics as unobserved effects. Selectivity bias being in the modeling process is captured through a recursive system, with this recursive spatial modeling process accounting for selectivity and endogenous network topologies, while incorporating spatial correlation and heteroskedasticity. Network topology measures influence density endogenously, and affect time occupancy in a recursive manner.

The key findings of this study can be summarized as follows. 1) Block level building generation has a significant size effect, with high average total impact and unit elasticity. 2) Community structure appears to matter as being affected by attributes of adjacent areal units; and centrality and connectivity within a network are also significant. 3) The SARAR model appears to be a very plausible end-user model for efficient pedestrian exposure prediction and it can be claimed that constructing the network topology of a grid, and developing a facility level trip generation inventory, is sufficient to estimated pedestrian exposure.

The spatial-temporal perspective on the impact of network topology and area dynamics measures for pedestrian exposure analysis makes this study unique. This proposed approach could offer a more robust understanding of the complexities of pedestrian behavior and network dynamics, and can be expected to provide improved insights relating to security along the network, and contingency planning. Furthermore, the identification of areas with high pedestrian exposure or high area-level density in the network can aid in the planning and development of a more efficient walking environment, transit systems (bus routes or operation interval alteration); and possibly the installation of eco-friendly transportation facilities such as bicycle or Segway rentals, and walker or bicycle paths.

 

Key words: Pedestrian, Network, Exposure, Topology, Spatial-Temporal, Density,

Occupancy, Vulnerability, Crosswalk, Micro-simulation

 

TABLE OF CONTENTS

List of Figures…………………………………………………………………………………………………….. viii

List of Tables……………………………………………………………………………………………………….. ix

 

Acknowledgements …………………………………………………………………………………………………… xi

Chapter 1 Introduction ………………………………………………………………………………………………. 1

1.1Background …………………………………………………………………………………………………. 11.2Objective …………………………………………………………………………………………………….. 4

1.3Analytical framework …………………………………………………………………………………… 5

Chapter 2 Literature review ……………………………………………………………………………………….. 12

2.1Pedestrian exposure research …………………………………………………………………………. 12

2.2 Spatial analysis research ……………………………………………………………………………….. 15

2.3 Network topology research ……………………………………………………………………………. 18

Chapter 3 Microscopic simulation for pedestrian area evaluation ……………………………………. 20

3.1 Overview of study area …………………………………………………………………………………. 20

3.2Simulation of area evaluation ………………………………………………………………………… 24

3.3 Simulation output and analysis ………………………………………………………………………. 263.4 Pedestrian dynamics analysis …………………………………………………………………………. 32

3.5 Time-distance matrices …………………………………………………………………………………. 35

3.6 Crosswalk analysis ……………………………………………………………………………………….. 39

Chapter 4 Network topology analysis ………………………………………………………………………….. 44

4.1 Network topology measures …………………………………………………………………………… 46

4.2Node-Link construction ………………………………………………………………………………… 52

4.3Adjacency matrix …………………………………………………………………………………………. 54

4.4 Network analysis ………………………………………………………………………………………….. 57

Chapter 5 Methodology …………………………………………………………………………………………….. 68

5.1 Spatial auto-regressive model structure …………………………………………………………… 715.2Accommodating spatial dependence of area density measures …………………………… 74

5.3SARAR Model estimation …………………………………………………………………………….. 75

5.3.1 Estimation by maximum likelihood (ML) ………………………………………………. 755.3.2 Estimation by IV/GMM ………………………………………………………………………. 76

5.3.3 SAR and SARE Model estimation ………………………………………………………… 80

Chapter 6 Modeling results for pedestrian exposure ……………………………………………………… 81

6.1 Descriptive statistics of data…………………………………………………………………………… 83

6.2 Spatial dependency, heteroskedasticity, and endogeneity test …………………………….. 86

6.2.1 Spatial dependency test ……………………………………………………………………….. 86

6.2.2 Heteroskedasticity test …………………………………………………………………………. 88

6.2.3 Endogeneity test …………………………………………………………………………………. 90

6.3 Base models ………………………………………………………………………………………………… 94

6.4 SARAR model development ………………………………………………………………………….. 100

6.5 Recursive system model development …………………………………………………………….. 110

6.5.1 Selectivity correction …………………………………………………………………………… 110

6.5.2 Time occupancy models ………………………………………………………………………. 112

6.6 Model post-estimations …………………………………………………………………………………. 115

6.6.1 Prediction accuracy …………………………………………………………………………….. 115

6.6.2 Marginal effects and elasticities ……………………………………………………………. 117

6.6.3 Validation with observed pedestrian volumes …………………………………………. 119

Chapter 7 Conclusions and recommendations ………………………………………………………………. 122

7.1 Local and global findings ………………………………………………………………………………. 122

7.2 Directions for future research …………………………………………………………………………. 127

 

References ………………………………………………………………………………………………………… 130

Appendix A  Distributions of area dynamics measures …………………………………………… 137

Appendix B  Network topology graphs ………………………………………………………………… 139

  1. Corridor 1 …………………………………………………………………………………………… 139
  2. Corridor 2 …………………………………………………………………………………………… 140
  3. Corridor 3 …………………………………………………………………………………………… 141Corridor 4 …………………………………………………………………………………………… 142
  4. Corridor 5 …………………………………………………………………………………………… 143
  5. Corridor 6 …………………………………………………………………………………………… 144
  6. Corridor 7 …………………………………………………………………………………………… 145
  7. Corridor 8 …………………………………………………………………………………………… 146
  8. Corridor 9 …………………………………………………………………………………………… 147
  9. Corridor 10 …………………………………………………………………………………………. 148
  10. Corridor 11 …………………………………………………………………………………………. 149

Appendix C Network topology measures on a sub-network scale level …………………….. 150

Appendix D Comparison of observed and predicted pedestrian volumes ………………….. 151

Appendix E Word clouds for this dissertation ……………………………………………………….. 152

Chapter 1

Introduction

1.1  Background

 

Transportation infrastructure is increasingly focusing on improving the general quality of life for people in modern cities. This is largely due to people pursuing safer and more sustainable traveling environments, with a shift towards non-motorized transportation as a viable alternative to automobiles for both economic and health reasons. Recently, with these changes in social needs, many major cities around the world are developing themselves as walkable cities, and making an effort to eliminate the hindrances to walkability with their urban networks. Therefore, improvements to ensure a safe walking environment, with efficiently connected walkable areas that are both comfortable and enjoyable to use, have become serious concerns for cities, planners, policy decision makers and transportation researchers.

The efficiency of urban networks and safety of walking environment are significant factors for making a walkable city. A well-designed and planned network benefits social needs in terms of better safety and security, healthier environment, and reduced economic costs due to crashes and congestion related lost productivity time. These benefits serve as an important motivation to make users of the network prefer non-motorized modes of transportation such as walking or using a bicycle over motorized modes. Safety is one of the main targets of the field of transportation, and numerous studies are being conducted with aim of improving safety across all modes of transportation. With regards to walking, people are more likely to avoid walking if they are exposed to harmful environments on pedestrian paths. Given these trends, pedestrian exposure should be discussed as a main measure for explaining relationships with sustainable transportation networks, a goal that forms the core of this dissertation.

Pedestrian exposure is a multidimensional term.  It can mean exposure to crash conflicts, exposure to adverse weather, exposure to security threats, or exposure to non-vulnerability related phenomena such as media advertisements. This dissertation focuses on the vulnerability related aspects of pedestrian exposure. Vulnerability of pedestrians is increasingly becoming a concern in the United States due to the continued aging of the US population. It puts certain components of the population at an enhanced level of risk despite well-intentioned strategies for security management. For example, the services provided by the National Terrorism Advisory System

(NTAS), or other similar systems based on the NTAS like the New York City Government Terror Threat Alert System, are limited in effectiveness due to the reactive nature of the threat alert system itself. The announcements of “Imminent Threat Alert” and “Elevated Threat Alert” go into effect after credible information is received (NTAS Guide 2011). When the threat classifications are assigned, the information can then be disseminated through map-based national security warnings. This service provides geographic information in a basic topographic alert manner that highlights affected areas after an alert has been issued. Similar services include publically available mapping tools, such as Google, NAVTEQ, and other similar state-run Department of Transportation services, that provide real time up-to-the-minute traffic condition information integrated into a map interface. Also, services, such as the National Weather Service, provide up-to-date weather information through geographic imagery. A dynamic threat alert system that provides real time information on a public domain does not currently exist or available.

In the domain of safety related vulnerability analysis, pedestrian exposure has emerged as an area of concern with the increase in mode shifts observed in the traveling US population.

Coupled with land use planning that promotes high density housing choices in dense urban environments, US households have now started to consider car-less choices as a lifestyle strategy, while relying increasingly on access to public transit, Uber cars, taxis and bicycles as alternative modes of transportation. Additionally, there is the desire of the US household to increase engagement in physical activities outside the home, with a goal to improve personal health and quality of life indicators. Limitations currently exist in the analytical tools available for the estimation of mode shifts, and the analysis of the spatial distribution of these shifts, especially in the pedestrian mode. This creates limitations in the analysis of exposure of pedestrians to vehicle conflicts at intersections, conflicts in multimodal interfaces, and also in the development of network design concepts that can help reduce barriers to pedestrian travel in cities. For example, without accurate estimation of pedestrian density, pedestrian volumes cannot be estimated for a given spatial area. This in turn leads to inaccuracies and biases in the estimation of key parameters such as pedestrian crash occurrence and severity likelihoods. Furthermore, the lack of a reliable spatial pedestrian estimate creates a gap in terms of our understanding of spatial energy distributions from human activity in the non-motorized time-space prism.

This dissertation proposes a robust computational methodology for a network-level, dynamic, topography-and-built-environment-based pedestrian exposure framework.  This proposed system will be built on a spatial econometric framework that allows statistical inference on key network, pedestrian area and pedestrian flow measures.  The final outcome is expected to result in a topographical and built-environment-intensive prediction system, aimed to capture high-resolution dynamic variations in pedestrian activities, highlighting areas of vulnerability through spatial and time-of-day dynamics.

The rationale for this dissertation is therefore, to address in a multidimensional manner:

  1. a) how topography enables or disrupts pedestrian vulnerability in urban environments, and b) how the built environment through the extensive set of organized component complexities enables or disrupts security threats to non-motorized traveling populations. The extant literature bears no evidence that a multidimensional effort to tackle topography and built environment interactions simultaneously has been undertaken. Therefore, this dissertation is aimed at studying in a dynamic real-time manner, the methodological standpoint of addressing endogenous pedestrian and network topology variables through the topography-built-environment-integrated lens.

1.2  Objective

Analyzing the dynamic behavior of pedestrians along with area evaluations is significant for the planning and designing of sustainable transportation networks. The understanding of travel patterns or the walking behavior of pedestrians and identifying key attributes of high density pedestrian areas provide insights into how pedestrians travel in the network. Network topology is a term that depicts the general physical structure of a network along with representations of spatial connectivity between all key areas within the network. The study of pedestrian behavior combined with pedestrian area evaluations is essential when using network topology to improve security and efficiency. However, current available literature relates pedestrian exposure to pedestrian flow on a macroscopic level, and studies on the contribution of network topology to the concentration of paths on a spatial-temporal level are lacking. Space-time analysis of pedestrian flow in pedestrian areas can provide a framework for integrated safety assessment as well as sustainable transportation planning in dense urban environments in the emerging domain of connected infrastructure. Furthermore, the methodology this dissertation aims to develop can also be extended to energy footprint analysis of large covered infrastructure spaces, thereby providing guidance for the sustainable design of such facilities.

The main objective of this study is to evaluate the impact of network topology on arealevel (and potentially higher-level) pedestrian density and time occupancy. Area-density and time occupancy are reliable measures of space-time exposure that is an outcome of pedestrian origindestination paths. Higher level analysis such as at the area-group level can also provide a hyperlevel insight into the impact of network topology on pedestrian community density.  By knowing the relative contribution of network topology accounting for pedestrian flow and destination attractiveness, insights on network design that minimize exposure while also optimizing travel time can be gained. In this dissertation, pedestrian exposure will be estimated by area-level (and higher-level) pedestrian densities and time occupancies, utilizing advanced methodologies that integrate area evaluation based on pedestrian dynamics from social-force model based microsimulation, and through network properties obtained from a fine resolution topological network model. A study of the extant literature shows that this potentially fruitful area of research has not been explored along the lines of network topology, insight into which this dissertation hopes to provide to the transportation research community.

1.3  Analytical framework

The background and objective of this dissertation are described in the introductory part of chapter 1. Development of walkable cities with improved levels of pedestrian safety has driven the movement towards sustainable transportation. Towards understanding the spread of pedestrian risk over a wide spatial-temporal setting, aimed at predicting and improving future walking environments along urban networks, it is necessary that pedestrian exposure is estimated along with network topology. Therefore, the purpose of this study is to evaluate the impact of network topology on areal level and potentially higher levels for pedestrian density and time occupancy. Therefore, spatial models were developed incorporating network topology measures for pedestrian exposure estimation.

The overall framework and flow of ideas to accomplish the objectives of this dissertation is presented in figure 1-1.

The spatial scope of the study area was set to downtown Seattle in Washington State, and the immediate surrounding areas. The temporal scope was limited to three peak hour periods (AM, MD, and PM peak), and the micro-simulation part of the analysis was conducted at 5minute intervals resulting in 12 time panels for each peak period.

In chapter 2, literature reviews are discussed with respect to three topics; 1) pedestrian exposure research, 2) spatial modeling analysis, and 3) network topology related research.

The database for pedestrian exposure estimation is evaluated through several steps. The data development and analytical process is divided into four main steps; 1) simulations of pedestrian area evaluation, 2) pedestrian dynamics records and area dynamics evaluation analysis, 3) network graph and relevant matrices constructions, and 4) network topology analysis.

In chapter 3, simulations of pedestrian area dynamics evaluations are discussed. The first stage of this simulation process deals with area evaluation wherein area units are defined. There are definitions for three types of area units; 1) an areal unit comprised of a structure such as a building, terminal or parking garage, 2) an areal unit comprised of a sidewalk between two crosswalk endpoints, and 3) an areal unit of crosswalk. Secondly, a large scale network with 33 sub-networks is constructed for micro-simulation of pedestrian area evaluations. Each subnetwork includes facilities, sidewalks and crosswalks, and then the defined areal units are set up as walkable areas that pedestrians can access. In order to construct the network for the study area, “Areal units” are designated as map shapes on micro-simulation, and pedestrian in-and-out flows produced by trip generation calculations are assigned for each facility areal unit. Area evaluation is conducted by dynamic micro-simulation VISSIM, and parameters obtained from the simulation such as a walking speed, number of pedestrian, total walking distance, total delay, total gain time and total dwell time are utilized as the measures to estimate pedestrian exposure in an areal unit.

Figure 1-2 shows the relevant measures for pedestrian exposure estimation from the micro-simulations of area dynamics along with pedestrian dynamics, and topology measures from network analysis. The generating process of these measures will be discussed going forward.

 

Figure 1-1 Relevant measures for pedestrian exposure analysis

 

Micro-simulation is a critical component of the analysis process, and it permits a large scale network or sub-network level simulation, as well as various scenarios or situational planning. The micro-simulation program chosen for this study is the PTV VISSIM platform, which maintains the capability of generating fine-level high-resolution data from a multitude of simulation parameters. VISSIM also serves as the analytical engine for integrating the macrolevel flow data with the micro-level travel and activity information of individuals. Of particular interest is the high-resolution data output generated from the network level simulations. The data output consists of various measures of effectiveness (MOE) that capture network-level metrics. These network MOEs can be tied to vulnerability measures in which areas of dense occupancy and high-activity, given time-of-day dynamics, will establish a vulnerability index based on dynamic time-sliced measures of human travel such as speed, delays, group compositions by topography measures such as gradients, built environment measures such as land use densities, road network densities, and multimodal interface indicators such as terminals for ferry, bus, rail, and bicycle hubs. These measures can be used to develop a complementary map-based visualization methodology, simulation output and topography/built environment measures can be converted to multi/hyper spectral data to provide interactive spectral overlays as described by Ware (2005).

The next stage is to analyze pedestrian records and area evaluation results. The measures of pedestrian records such as walking time, walking distance and current destination are used for tracing each pedestrian’s trajectory with a one-second resolution. Tracing pedestrian trajectories is a critical basis for this study since the measures with spatio-temporal dimension are set by accumulating the travel times and travel distances of each pedestrian at an areal level. Time-space measures (time  distance matrices) may quantify pedestrian exposure. Quantifications of pedestrian exposure in terms of continuous space-time measures (CST measures) are more insightful for security and safety planning purposes than those limited to pedestrian flow

(volume/hour) alone.

Chapter 4 discusses network graph theory, network construction, and analysis of network topology measures. The two dimensional network constructed in chapter 3 consists of areal units based on Google maps, which is simplified to a diagram consisting of nodes (areal unit) and links (connections between areal units). With these node-link diagrams, adjacency matrices are set at a corridor level for evaluating connectivity, centrality and accessibility among the nodes

(pedestrian areal units) in the network. Node-link diagrams and adjacency matrices form a stepping stone for analyzing network topology. Network topology measures for each areal unit are estimated by the network visualization and analysis software GEPHI that provides measures for degree, closeness centrality, betweenness centrality, eigenvector centrality, eccentricity, hub, pagerank, modularity, number of triangle and clustering coefficient.

In chapters 5 and 6, the methodological components of this dissertation and the resulting outputs are discussed. The various spatial modeling techniques used to meet the goals of this study included a spatial autoregressive model (SAR); a spatial error model (SARE); and a spatialautoregressive model with spatial autoregressive disturbances, exogenous regressors, and additional endogenous regressors (SARAR).

For capturing selection bias of non-zero density areas in spatial models of density, a recursive system is applied. A recursive spatial model accounts for selectivity and endogenous network topologies while incorporating spatial correlation and heteroskedasticity. A logit model is developed to capture selection bias, and then time occupancy models of SARAR structure are developed as adding the probability of the non-zero density from the logit model.

In chapter 6 of this dissertation, network topology measures are evaluated with respect to their effect on area density endogenously, and their influence on time occupancy in a recursive manner. Therefore, network measures and trip generation volumes are used as endogenous and exogenous variables respectively in the spatial models, and the measures obtained from the micro-simulation of area evaluation are treated as instrument variables. In order to validate the developed models in the dissertation, prediction comparisons with observed pedestrian volume data, marginal effects and elasticity estimations are conducted.

Lastly, chapter 7 describes conclusions and recommendations of this dissertation. Overall, this section was aimed at providing a brief overview of the goals, major findings, and viability of the proposed methodology in this dissertation.  The ultimate research vision for this body of work is to provide for informed decisions pertaining to important policy implications and sustainability applications. Potential impacts of this study would be using accurate predictions of pedestrian exposure to provide real-time public guidance in the event of emergencies, to target areas of significant conflict between motorized travel and pedestrians, or to enable environmental agencies to flag areas of potentially higher foot-traffic.

 

 

 

I
 

II
 

III
 

IV
 

V

SPATIO-TEMPORAL MODELS INCORPORATING NETWORK TOPOLOGY MEASURES FOR PEDESTRIAN EXPOSURE ESTIMATION

Leave a Reply