CDRC Data Partner Forum.
University of Leeds, 31st October 2023

Virtual Mirrors or Smoke and Mirrors?
Can Urban Digital Twins Make Cities Better?

Nick Malleson, University of Leeds, UK

Slides available at:
https://urban-analytics.github.io/dust/presentations.html

Critical (urban?) social problems

Children in the most deprived neighbourhoods have 10 years shorter life expectancy than those born in affluent neighbourhoods.

The use of food banks has increase by 81% in five years

Rape has effectively been decriminalised (in 2021 only 5% of recorded rapes resulted in a charge).

A third of English children leaving primary school are overweight or living with obesity

Can urban digital twins help?

What are Digital Twins?

A synthesis of computer models, sensor networks, visualisations, etc., that mirror a real-world system, product or process

Make More Digital Twins nature article
Tao and Qi (2019)

"Precise, virtual copies of machines or systems"

Examples:

Machines / products;
Manufacturing (factories);
Health / hospitals;
Smart Cities (e.g. Singapore, Victoria, Bradford);
... many others ...

But lots of 3D models. Few twins

What might an (urban) digital twin look like?

Different layers of data and models that may constitute a digital twin (from individual to macro)

Challenges for Urban Digital Twins

Challenges

Simulating Complex Urban Systems

Non-linearity, emergence

Behaviour and interactions of people

Difficult to abstract, risk of "smoothing out"

Need models that can account for these complexities

(Role for ABM?)

Challenges

Uncertainty

Many sources of uncertainty (Ghahramani, 2015; Edeling, et al., 2021):

Need to understand and be honest about uncertainties

Uncertainty Quantification can help (but relevant to DT models?)

Data

Even in the 'Age of Data', there are huge unknowns

Models (DTs) can be extremely detailed, but typically only coarse data are available

Challenges

Network server room

Computation and Model Synthesis

DTs are a "synthesis" of computer models

BUT: (Very!) computationally expensive models

Innovations in (e.g.) meteorology not necessarily applicable to urban DTs

Also technical challenges coupling models

Individual models are fragile, when combined ... ⚠

Progress towards urban digital twins

'Big' Data and Urban DTs

Pedestrian Mobility

Growth in availability of data for quantifying the 'ambient population'

Census, travel surveys, mobile phone activity, card payments, smart phone apps, social media, pedestrian counters (WiFi, CCTV)

Map and graph of model predictions
Model estimates of the success of public events compared to a 'normal' day.
Diagram of the ramp model components described in the slide text
Spooner et al. (2021) A dynamic microsimulation model for epidemics. Social Science & Medicine 291, 114461. DOI: 10.1016/j.socscimed.2021.114461

Model synthesis

DyME: Dynamic Model for Epidemics

COVID transmission model with components including:

dynamic spatial microsimulation, spatial interaction model, data linkage (PSM), ...

Represents all individuals in a study area with activities: home, shopping, working, schooling

Daily timestep

DyME Validation Drawback: Data

Incredible detailed model!

BUT only data available for validation: COVID cases and hospital deaths

Only quantify a tiny part of the transmission dynamics

Huge uncertainties

Building the model was the easy part ...

Different disease stages: susceptible, exposed (pre)(a)symptomatic, removed

Progress: Uncertainty Quantification

Model runs with large and small uncertainty. Other areas have well developed approaches to this

Many sources of uncertainty. Leverage uncertainty quantification?

history matching results: most likely parameter combinations
McCulloch, J., J. Ge, J.A. Ward, A. Heppenstall, J.G. Polhill, N. Malleson (2022) Calibrating agent-based models using uncertainty quantification methods. Journal of Artificial Societies and Social Simulation 25, 1. DOI: 10.18564/jasss.4791

Dynamic Calibration for DyME

Uncertain predictions made using the DyME parameter posteriors. The certainty increases as more data become available.
Source: M. Asher, N. Lomax, K. Morrissey, F. Spooner, N. Malleson (2023) Dynamic Calibration with Approximate Bayesian Computation for a Microsimulation of Disease Spread. Nature Scientific Reports 13:8637. DOI: 10.1038/s41598-023-35580-z.

Why we need Data Assimilation

Complex models will always diverge

(due to inherent uncertainties in inputs, parameter values, model structure, etc.)

Possible Solution: Data Assimilation

Used in meteorology and hydrology to bring models closer to reality. Combines:

Noisy, real-world observations

Model estimates of the system state

Data assimilation v.s. calibration

Example of optimising the model state using observations and data assimilation

For more information: urban-analytics.github.io/dust/.

Are they worth it?

Are they worth it?

Can urban digital twins help? Or should we focus on lower-hanging fruit?

Who will use them?

Will they help with ...

Life expectancy (can vary by 10 years)

Food banks

Pitiful rape (and others) prosecution and conviction rate

Obesity

CDRC Data Partner Forum.
University of Leeds, 31st October 2023

Virtual Mirrors or Smoke and Mirrors?
Can Urban Digital Twins Make Cities Better?

Nick Malleson, University of Leeds, UK

Slides available at:
https://urban-analytics.github.io/dust/presentations.html