Urban and Smart : Prospects and Challenges with New Forms of

Piyushimita (Vonu) Thakuriah Dean, Bloustein School of Planning and Public Policy Distinguished Professor of Transportation and Urban Informatics

NTTS 2019 Please do not distribute without permission BlousteinUrban Big School/RutgersData Centre University Personal and Wearable Tech Trends An explosion of ICT solutions and data Integrated Systems

Smart Smart Transportation Buildings

Connected Infrastructure

Courtesy ETSISmart, collaborative, self-organizing systems Bloustein School/Rutgers University

Generations of “Smart Cities” Critical Ingredients:

. ICT infrastructure; Version 1: Smart Infrastructure . Effective resource management; . Cost reduction and accountability;

. Performance monitoring. Smart

Version 2: Smart . Business-led development; Innovations . Strengthened civic leadership;

. ICT-based urban innovations. City Innovative Version 3: Smart . Well-informed and engaged citizens; Citizenry . Addressing problem causes in addition to service delivery; . Social innovations – innovative solutions for urban problems;

. Social , education and social capital; Future City . Citizen choices and wellbeing. Bloustein School/Rutgers University

One example - Connected, Cooperative and Anticipatory Transport

Systems

Existing Information Information Existing Environment

. Intelligent Transportation Systems . Structural Health Monitoring for asset management . Connected systems V2X: . Vehicle-to-Vehicle (V2V)

. Vehicle-to-Infrastructure (V2I) New of Elements

. Vehicle-to-Grid (V2G) Information Environment Information Bloustein School/Rutgers University

Emerging Forms of “” for Urban Applications Infrastructure A wide spectrum of naturally-occurring data: . Physical – low to high-tech (multi- Generated through transactional, operational, modal transport, planning and social activities not all of which connected were specifically designed for research or the vehicles, smart linkage of such data to purposefully designed buildings, V2G) data . ICT – Complexities associated with which (e.g. systems, sensor voluminous, heterogeneous, unstructured, hard- networks to-access) require special considerations: . Data . Technological . Methodological . Theoretical/epistemological . Political economy UrbanBloustein Big School/Rutgers Data Centre University Urban Informatics Data-intensive approaches to analyzing, visualizing, simulating, understanding, interpreting structured and unstructured data on cities and urban areas to address complex urban challenges.

Edited volume of NSF workshop: “Big Data and Urban Informatics” UrbanBloustein Big School/Rutgers Data Centre University Big Data and Better Urban Living Actions and Analytics . Urban infrastructure development and monitoring – building and monitoring transport, energy, ICT, water and other infrastructure . Detection systems . Understanding links, causality and supporting . Urban operations management – transport operations and traffic processes flow management, energy management and optimisation, crime . Forecasting and detection and prevention understanding the future . Citizen engagement/civic participation – involvement in plan- . Evaluation of actions or making, and idea-generation; crowdsourcing travel and other potential actions information . Engagement . Timeliness . - create and maintain well-designed, good quality . Fit for purpose places and sites . Value-for-money . – large-scale: urban land-use planning, mega- . Understanding biases, infrastructure planning; small-scale: site design, brownfield planning uncertainty, robustness of and regeneration projects findings . Keeping up with the . Urban knowledge discovery – understanding emerging issues, rapidly changing data behaviours, public mood, critical concerns landscape – including . Urban policy analysis and evaluation – impact of proposed high- privacy, citizen awareness speed rail construction, crime prevention strategies and Bloustein School/Rutgers University Grand Challenges for Urban Management . How to operate cities effectively and efficiently . How to build and manage robust and resilient infrastructure . How to evaluate potential consequences of complex social policy change on urban areas . What makes the economy resilient and strong – how to develop shock- proof cities . How to drive economic growth and revenue . How to support business innovation and economic competitiveness . How cities can recover from man-made or natural disasters . What interventions are needed for healthy behavior . What strategies are needed for lifelong learning, civic engagement and community participation . How does one address challenges of social exclusion UrbanBloustein Big Data School/Rutgers Centre University

Social Hazards and Trust in Data - A need to balance the Good, the Bad and the Ugly

New and data has many benefits in the urban space but also has the potential to lead to unfair practices and unintended consequences Bloustein School/Rutgers University Joining up crime detection and safe transport

. About 1.25 million people died in 2013 in road crashes worldwide (World Health Organization, 2013) – many in urban areas . Many types of traffic deviance leading to crashes are not random, but has a root cause in the same social conditions that result in concentrations of crime. . Crime and traffic crashes often spatio-temporally overlap in cities and are responsible for decreased accessibility and quality of life in cities. . Determine a more unifying approach and integrate operational and policy strategies. . BUT variable levels of reporting – incidents in some areas, especially poor, deprived areas tend to be underreported in official records UrbanThe Sensing Big Data City: Centre Real-time Monitoring of Cities Context-Awareness and Semantic Enrichment Using Twitter to Understand Local Concerns and Events Can we use language patterns detected in different parts of the city to understand underlying uses, activities, and concerns? System to help identify social and functional concerns and issues potentially for planning or operational action, eg, where people are not happy with services

Known incident from transportation sensor data from highways agency Detecting Road Negative tweets – tweets posted Incidents from when there is no Twitter data Positiveincident tweets – tweets posted when there is an incident Bloustein School/Rutgers University Crime – a huge societal issue Study Area . City of Chicago . 758 homicides in 2016 . 98 people killed, 2028 seriously injured in 2014 (latest figures) Complexity of the problem . Significant concentrations of crime and crashes in micro-places, but also spread throughout city . Deep distrust of authority and contested relationships . Limited English speaking capacity in some areas and limited knowledge of social, medical and legal options . Problem with underreporting of crashes and crimes in some areas Bloustein School/Rutgers University Predictive Analytics of . Generally, crimes increase with crashes. Relationship is more evident at points less Traffic Crashes and Crimes than the 90th percentile . Combined crashes and crimes is long- tailed to the right; calls for evaluating models at different points in the distribution

. What factors predict crashes and crime (“events”)? – final goal:

Eventsii f() X . Interested in quantiles:  = .25, .50, .75, .95 . Significant spatial dependence – Spatial Autoregressive version of quantile regression Model-based Underreporting Correction for Traffic Crashes . In the OLS model, Model I Model I crashes tended to be Poisson with Heterogeneity Poisson with Exogenous Underreporting Variable Marginal Effect Marginal Effect overpredicted in Intercept -4.21*** -2.13*** suburban locations EJ_TRACT (1=”Yes”) 0.65*** 0.33*** Environmental Factors and underpredicted in TAI 2 2.01*** 1.01*** the Chicago PED_LOW 1.61*** 0.59*** SUM_AADT 2 0.48e-06*** 0.24e-06*** downtown business SUM_LENGTH 2 -0.28e-03*** -0.14e-03*** district (the “Loop”) NO_SCHOOLS 0.19** 0.09** POP_SQMILE 9.10E-06 4.60E-04 and in southern areas PERCRIME 0.24** 0.12** PED 0.09*** 0.05*** of the City of Chicago Behavioral Factors WLKTOWRK 0.0008*** 0.0009 MEDHHINC99 -2.20E-07 -1.10E-06 PERNOCAR 2.60** 1.31** . Crashes modeled PER_COMM 1.37 0.69 with Poisson count PERCHILDREN -2.09 -1.05 PERLOWENGLISH 0.21 -0.1 data model with Probit Reporting Equation Intercept 5.40E-08 heterogeneity which COUNTY (1=”Cook”) 0.018** accounts for R 2 0.58# 0.61# Log-Likelihood -1763.25 -1511.36 exogenous /df 2 136.8 93.76 underreporting – Vuong Statistic - -60.75 s 0.13 (p< 0.0001) 0.18 (p< 0.517) acknowledging that r - 0 only a subset of the * Significant at 0.10 ** Significant at 0.05 *** Significant at 0.01 actual number of Cottrill, C., and Thakuriah, P. (2010) Evaluating pedestrian crashes in areas with high low- crashes that occurred income or minority populations. Accident Analysis and Prevention, 42(6), pp. 1718-1728. are reported (Twitter) data is useful in detecting events but very sparse

Geotagged Tweets Geolocalized Tweets Twitter users are not representative of the population; locations of those who choose to geotag are further not representative of the locations of all Twitter users – but we get a much larger sample allowing us to detect more events, and see activities in more places Bloustein School/Rutgers University Using our methods, we have discovered traffic-related tweets that are not in incident databases – in disadvantaged areas as well as in outlying areas;

This has significant potential for filling in underreporting and for more accurate understanding of risky areas and hazard spaces in cities

Davide-Paule, J. G., Y. Sun and P. Thakuriah. Beyond Geo-Tagged Tweets: Exploring the Geo-Localization of Tweets for Transportation Applications. Forthcoming in Big Data and Transportation, edited volume to be published by Springer. Paule, J. D. G., Y. Moshfeghi, J. Jose and P. Thakuriah (2017). On Fine-Grained Geo-Localization of Tweets. Proc ACM SIGIR conference, Amsterdam, Netherlands, 2017 (ICTIR’17). Bloustein School/Rutgers University The Reality – unintended consequences – or algorithmic bias? . Developing location-based micro-place operational strategies helps to reduce crime as well as hazards from traffic crashes. . Yet, huge problems with predictive policing and bias - “The City of Chicago has its own secretive [predictive policing] called the Strategic Subject Lists (SSL)….. 56 percent of black men in the city [between] the ages of 20 and 29 have an SSL score,” . “involves racial profiling, deconcentration of crime, and perpetuating corrupt policing practices” . Gunshot detection technology – eavesdropping on personal conversations? . How do you make trade-offs between technology, hazards and these complex social problems? UrbanBloustein Big Data School/Rutgers Centre University

High-fidelity understanding of behaviors and how we live, work and play – Links to health and economic and social wellbeing and externalities

A paradigm shift from theoretical model- based approaches to AI – need an “optimal” mix of the two Lifelogging Autographer - Still pictures Research possibilities: . Lifelogging through every 5 seconds both outdoors . Travel behaviour wearable sensors – a and indoors research multimedia personal . Driving styles and archive eco-friendly behaviour . Image data on citizens’ ° A custom 136 eye view lens, . Fine-grained data on everyday living an ultra small GPS unit, . Digital image processing to quality of built Bluetooth, and 5 in-built environment retrieve data on multiple sensors - ambient light / factors on which it is . Social networks accelerometer / magnetometer . Many others difficult to survey people / PIR / temperature

Outdoors Indoors Data Preparation: Multi-sensor wearable device data Movement analysis to annotate movement data with the contextual information and to discover new insights into indoor mobility patterns among different people. Image + sensors = multi-sensor data analysis

GPS Acceleration Magnetometer Light sensor Luminance Temperature Orientation Exposure Identifying complete movement profiles and social interactions Temperature Luminosity Indoor Indoor/outdoor classification - identify on the basis of temperature and luminosity values whether person is indoors or outdoors. Results show that we Outdoor can classify images into outdoor and indoor locations with 93.24 % correctly classified instances. Indicators possible: Activity detection - Differences in . Time-varying indicators of waste generation, acceleration patterns can be used energy and water usage for annotation of various activities, . Total (indoor + outdoor) activity levels as well indoor as outdoor ones. . Independence in daily living Various acceleration values for 1- . Degree of uneasiness and disturbance in standing; 2-sitting; 3-walking and mobility 4-driving. . Degree of isolation in everyday living

Co-detection problem – find out the extent to which people have interactions with others, how much time they spend with others, how often they are in meetings etc Bloustein School/Rutgers University Development of traffic disturbance index . Driver inattention is a leading cause of crashes . Pedestrian uncertainty at key locations (looking for cars, conflicts etc) affect quality of travel . Can we use lifelogging data to sense areas of conflict – disturbance index . By disturbance we mean here looking (turns and reorientation – and extent of reorientation - of an individual’s body into a direction different to the one the individual is heading)

Images showing heading of a Individual disturbance Using multiple sources of driving/ri can be defined as a personal sensor information, we ding difference between GPS can index the street network individual /Road network heading with the degree of uncertainty and Life-logging data and perceived conflict from orientation image and related data Bloustein School/Rutgers University

Indoor and outdoor walking Contrary to our expectations - . How much do people walk indoors? . Do people who walk a lot indoors walk less outdoors (eg – people who walk more indoors may live in larger houses, hence have higher incomes and own cars, and hence may walk less outdoors due to car travel) . Estimation of outdoor walking . . Could propensity for possible due to mode detection People who walk more outdoors also physical activity be more from GPS data walk more indoors intrinsic; . Estimation of both outdoor and . People who walk less . How do we ensure indoor indoor walking is possible due to outdoors tend to design and (outdoor) built mode detection with lifelogging stand or sit more environment to offer data (input features - acceleration, indoors physical activity magnetic field readings and possibilities for “low orientation) volume” users Bloustein School/Rutgers University Co-detection - Developing a social isolation index – using machine vision to count people and distance/depth and orientation from images – work in progress Face Detection . dlib library . Pretrained model on 3 Precision 0.996 million faces from Recall 0.869 various datasets . ResNet network with 27 convolutional layers

Person Detection . Tensorflow Deep Learning library Precision 0.944 . Pretrained model on Microsoft COCO Recall 0.851 (Common Objects in Context) dataset . Faster R-CNN with ResNet Bloustein School/Rutgers University Social Isolation and Worker Wellbeing and Mental Health Work Status . Most workers work in “moderately” social environments . Those who are unemployed and seeking work tend to be quite socially isolated

Occupations . Managerial and professional positions are exposed to interactions with others . Greater share of clerical and semi-routine and manual jobs are exposed to social isolation, compared to other occupations UrbanBloustein Big Data School/Rutgers Centre University

Public Transport Availability and Housing including Rental Housing Price Data Private Sector Data

Sales data Link – Land to registry

What is the role of transportation services and infrastructure in Advertisements for property increasing or falling sales prices? – Implications for economic benefits analysis . Sentiment mining of real estate Where are new agent language developments (create thesaurus) occurring or where are . Linkage to wider areas losing set of urban population? – indicators Implications for service development Labour Market Accessibility – Access to jobs by public transport

Monitoring transit performance for Transit Availability Index every train, bus and ferry stop – London Bus Stops Transit GTFS data Transit Availability Index – Manchester Bus Stops

Good transit availability – 24 hour service & small headways Poor transit availability – specific service hours and longer headways between vehicles Identifying areas at high risk Temporal, not just spatial mismatch of transport poverty Labour Force Survey, 2011 and 2016

New project looks at the spatial distribution of jobs estimated to be lost due to massive automation Will draw links to future infrastructure policy

What is the role of transportation systems on joblessness and employment outcomes? 1.78 million people at risk of transport poverty in England and . By tracking UK-wide public transport and roads Wales performance, our results show that UK public transport schedules and operations need to be re-evaluated to Results highlights relationships match the changing nature and location of jobs and between spatial economy, urban locations of workers. form and changing nature of . An increase in traffic congestion is positively associated jobs with rise in unemployment benefits claimants. Bloustein School/Rutgers University Build complex person-level microsimulation models to forecast impacts of urban transport policy Potential User Work-life Index Forecasts

70,000

60,000

50,000 Cost Scenario 3 40,000 Cost Scenario 1 30,000 Cost Scenario 2

(2002 dollars) (2002 20,000

10,000

0

EstimatedNet Benefits over Worklife 20 25 30 35 40 45 50 55 60 65 Age Cohort during Base Year (yrs)

Average lifecycle economic return on $1 investment in smart mobility for low-wage workers is estimated to be $15 Bloustein School/Rutgers University Agent-Based Models of Social Exclusion

Types of agents Life choices Global variables Juveniles - Job 0-15 year old Continue Education Determine wage From salaries of available stock/Economy Make some life decisions occupations based on Start new job • Highest completed level of - Retirement Leave/lose job education • Age Age • Ex-convict status - Safety-net Working age adults Involvement in crime 16-64 years old Explore influence of other - Era: To modify factors on wages Family decisions Make life choices • Gender parameters that • Begin cohabitation • Race • End cohabitation capture • Have children legal/cultural Retirees Move/change neighborhood 65+ years old changes (at least retirement eligible, for simplicity) Removed from workforce, but can be part of the networks of others

High incidence of childhood poverty is a strong predictor of adult poverty as is living in deprived neighbourhood during childhood; Higher likelihood for escape from a life of poverty for those who turned 16 in the period from 1990-1999. Least successful were the ones who turned 16 prior to 1980. Bloustein School/Rutgers University Implications

. New forms of data allow previously unobserved behaviours to be analysed . Applications – transport and mobility, energy consumption, public health, assistive living, use in economic studies, time use assessment . Travel behaviour and health research (examples) . Driving styles and eco-friendly behaviour . Fine-grained data on quality of built and social environment . Social networks . Good part of our lives indoors – and alone without interaction with others – implications for mental health and social strategies . New ways of being and increasing digitalisation of our daily lives have implications for use of resources, ways of learning and education, social and political behaviours and other aspects with implications for planning and policy . AI and the Future of Work and Infrastructure Bloustein School/Rutgers University The Process and Impact

Biggest Challenge of all – IMPACT How do we go from data and technology to impact and “good” Adoption/ Implementation

societal and economic Engagement

outcomes? Value-Proposition Governance,

Public and Private Private andPublic Leadership, Citizen Citizen Leadership, and Actionable Activism, Advocacy, Strategies Knowledge Discovery Data Analytics

Urban Data Infrastructure Informatics Urban What does it take for data-driven public and civic systems to work? . Data infrastructure – the technical, methodological and the “soft” aspects . Domain knowledge and understanding paradoxes and redundancies . Value networks and leadership and champions . Skills – disciplines, techniques and teams . Communications strategies – decision-making on the basis of scientific evidence, public engagement strategies, prepared citizenry Urban Big Data Centre (1) Data Infrastructure - Context Driving the Work Aspects Characteristics Information management: 1) Information generation and capture 2) Management 3) Processing 4) Archiving, curation and storage

5) Dissemination and discovery Technological Data Preparation 1) Information retrieval and extraction 2) Data linkage/information integration 3) Data cleaning, anonymization and quality assessment

Analysis 1) Develop and apply methods to analyse various domain challenges

Methodological 2) Ascertain uncertainty, biases and error propagation in the data Getting and using data is hard . Data acquisition . What are the data sources? . Making a case for data sharing and resolving: . Incompatibilities with business models . Concerns over reputational harm . Lack of resource to facilitate data sharing . Governance and ethical issues around data . Mix of established and fluid legal framework . Data protection and privacy . Commercial and other sensitivities . Licensing and partnership-building with data owners . Sustainable data sources . Responsibilities . Business model for the data access to continue . Risks to continued accessibility of data – technical, organizational, legal, political Urban Big Data Centre (2) Need for domain knowledge Aspects Characteristics 1) Having a theoretical or conceptual framework to guide the system 2) Understanding metrics, definitions, and changing ideologies and methods to solving domain problems 3) Determining validity of approaches and limits to knowledge from data-driven approach

4) Information paradoxes (Jevons paradox), user equilibrium versus

Theoretical and Theoretical epistemological system equilibrium 1) Data entrepreneurship, innovation networks and power structures 2) Value propositions and economic implications 3) Data acquisitions strategies, access and governance framework 4) Privacy, security and trust management

5) Responsible innovation and emergent ethics Political economy Political Planning organizations (3) Value Traditional Urban Data Operational agencies Networks Research organizations and universities and leaders Users Consulting firms General-purpose ICT Infomediaries Data Entrepreneurs for Smart Cities and . Smart City Companies Institutional . Multiple-service ICT Companies Transformations Urban Information Service Provider Infomediaries - Partnerships with . City Information Services academics, industry . Location-Based Services and local governments . Location-Based Social Networks Thakuriah, P., L. Dirks, and Y. Keita Mallon (2016). Urban Open and Civic Data Infomediaries Emerging Urban Digital . Open Data Organizations Infomediaries and Civic Hacking in Emerging Urban . Civic Hacking Organizations Data Initiatives. In Seeing . Cities through Big Data: Community-Based Information Service Organizations Research, Methods and Independent and Open Source Developer Infomediaries Applications in Urban Informatics, Springer, NY, . Independent App Developers pp. 189-207. . Open Source Developers (4) Skills – Backgrounds & Disciplines . Substantive knowledge of the field (, transport planning and engineering, criminology, social work, energy, etc) . Spatial sciences (GIScience, ) . Statistics (modelling uncertainty, mixed models and hierarchical data structures) . (information management, information retrieval, HCI) . Economics (4) Skills - Techniques . Specialist urban modelling and simulations . Data gathering: science of sensors, remote sensing, survey methods, core understanding of new forms of data and how they work . Data analytics: machine learning, advanced statistical analysis, urban and transport modelling and simulations, GIS, spatial analysis, visualisation . Information management: systems, databases, programming skills, machine learning, data structures, algorithms . Information governance: legal and economic aspects of data management, privacy and security . Business management: project management, business case development, monetisation and ROI analysis (4) Skills - Team Composition . Domain experts . Information management . Analysts . Experts on data acquisition, sharing, standards . Experts in governance, ethics, privacy . Consumer analysts – people who assess and understand users needs and market . Communications and outreach . Experts in commercialisation, business case development Successful teams learn from each other, listen to needs, are open to new ideas, and are constantly seeking to collaborate. (5) Scientific evidence – the crisis

“At a time when decision-makers too often ignore, misunderstand, or misuse relevant evidence, we need new ways to communicate policy-relevant scientific evidence to decision-makers and influencers in all areas of government and society,” said Rush Holt, chief executive officer at American Association for Advancement of Science (AAAS).

One of the biggest challenges “is to be as unbiased and neutral as possible” and to avoid any notion that scientists and researchers are “just another special interest group,” Public Communications & Prepared Citizenry . Civil infrastructure and planning have a long history of public engagement - tends to be somewhat top-down, to inform or to defuse tensions . Ideas behind Future Cities – long-term and sustained engagement with members of the public throughout, not just to discuss plans that have already been made . The other side of the coin – how can we support citizens to be diligent and receptive to new ideas and solutions? . Lifelong learning – and the role of persuasion for investment in lifelong learning due to economic benefits . Perhaps technology can play a bigger role - use of interactive and participatory tools, hackathons, town-hall meetings – but sustaining public interest is difficult . Incentive-based models? Tax policy? Personal learning environments? Many thanks to the following collaborators:

Yeran Sun Katarzyna Sila-Nowicka Caitlin Cottrill Jinhyun Hong Obinna Anejinou Andrew McHugh Nebiyou Tilahun Jorge Davide-Gonzalez Paule Christina Boididou Mesut Yucel

Please do not distribute without permission Urban Big Data Centre

[email protected] @vthakuriah

Thank you!