urbanLab Magazin 2021 - Transformation

The revitalization project of Deutzer Hafen
aims to transform the old industrial
harbour district of Cologne into a new,
vibrant neighbourhood. Designed by
the Danish architecture firm COBE, the
project was chosen in a competition organized
by the city in 2016.
The mixed-use buildings are supposed
to house 5.000 people and serve
as workspace for another 4.500, who
will be able to access the rest of Cologne
through new bicycle and pedestrian
paths, including a bridge that spans
directly to the city centre. New public
transportation routes, such as water
buses and a train (S-Bahn) station are
also planned, as well as mobility stations
with bike and car sharing offers.
All these possibilities and easy access
aim to encourage people to leave their
cars at home.
The future district is used as case study
in building a decision support system
for urban planning, that is composed
of three parts: (1) a synthetic population,
(2) an agent-based model and (3)
a tangible user interface. This system
enables users to get in touch with an
agent-based simulation without any
knowledge in coding or even interacting
with computers. It connects physical
objects to digital information. The aim
of the simulation is to measure urban
vitality, based on theories developed by
Jane Jacobs and Jan Gehl.
1 SYNTHETIC
To simulate the activity and travelling
patterns, the model needs a synthetic
population, which is a virtual representation
of the community of the modelled
area. It is commonly built by combining
census and travel or time use survey
data, that may not always be up to date,
because of the amount of time and resources
taken to make such surveys.
That is why an experimentation with a
new approach is made: building the synthetic
population from a combination of
census and social media data. We are in
an age with a constant flow of user generated
content coming from location
based social networks (LBSN) such as
Twitter, Facebook, and Instagram, where
people share where they are, when
they are and what they are doing. Mining
this data can allow us to produce
a sort of digital census, that is cheaper
and fresher than traditional surveys.
Over 100.000 social media posts inside
the city of Cologne are collected from
Twitter and Instagram. The users collected
are anonymously profiled and
have their activity patterns inferred,
resulting in a population that reflects a
sample of the city. Their home and work
location are inferred by finding the locations
from where they post the most,
following a common approach like done
by Swier et al. (2015). The coordinates
from where they post from are categorized
in activities with the help of the
Open Street Maps API, which returns
information about locations near the
coordinates, like the building use or
if there are amenities there. With the
location points categorized, they are
now split into a timetable with the 24
hours of a day. Each user has their own
table that shows which activities they
usually do in each hour, from being at
home to eating out or going to school.
But it is still unknown who these people
are - students? Stay-at-home parents?
Or maybe executives? This is an important
information for the model because
it influences, for example, which type
of transportation one uses to move
around the city. This information is
predicted with the help of a Bayesian
Network, commonly used in population
synthesising to predict missing values
in the data (Medium from 25.05.2017).
It consists of a probabilistic graphical
model that represents the conditional
dependencies between variables. For
example, given a simple graph of three
variables: rain, sprinkler, and grass wet.
The grass being wet depends on if it
is raining or if the sprinkler is on. The
sprinkler being on, on the other hand,
depends on the rain since it will not be
turned on when it is raining. Given a dataset
that contains for example number
of people in a household and location,
but no number of cars, a Bayes Net that
was previously trained on a dataset
STUDENT IDEAS
131