Contents lists available at ScienceDirect
Landscape and Urban Planning
journal homepage: www.elsevier.com/locate/landurbplan
Research Note
Explaining subjective perceptions of public spaces as a function of the built
environment: A massive data approach
Tomás Rossetti
a,b,1
, Hans Lobel
a,c
, Víctor Rocco
a,b
, Ricardo Hurtubia
a,b,d
a
Department of Transport Engineering and Logistics, Pontificia Universidad Católica de Chile, Chile
b
Centre for Sustainable Urban Development (CEDEUS), Chile
c
Department of Computer Science, Pontificia Universidad Católica de Chile, Chile
d
School of Architecture, Pontificia Universidad Católica de Chile, Chile
ARTICLEINFO
Keywords:
Perceptions
Discrete choice models
Machine learning
Public spaces
ABSTRACT
People’s perceptions of the built environment influence the way they use and navigate it. Understanding these
perceptions may be useful to inform the design, management and planning process of public spaces. Recently,
several studies have used data collected at a massive scale and machine learning methods to quantify these
perceptions, showing promising results in terms of predictive performance. Nevertheless, most of these models
can be of little help in understanding users’ perceptions due to the difficulty associated with identifying the
importanceofeachattributeoflandscapes.Inthiswork,weproposeanovelapproachtoquantifyperceptionsof
landscapes through discrete choice models, using semantic segmentations of images of public spaces, generated
through machine learning algorithms, as explanatory variables. The proposed models are estimated using the
Place Pulse dataset, with over 1.2 million perceptual indicators, and are able to provide useful insights into how
users perceive the built environment as a function of its features. The models obtained are used to infer per-
ceptual variables in the city of Santiago, Chile, and show they have a significant correlation with socioeconomic
indicators.
1. Introduction
Urban landscapes are experienced by their users in great part
through visual perceptions. This has been explored in the literature,
showing that perceptions can influence the intensity of use of public
spaces (Khisty, 1994; Shriver, 1997) or encourage the use of certain
transportation modes (Antonakos, 1995; Hunt & Abraham, 2007;
Hyodo, Suzuki, & Takahashi, 2000; Jiang, Christopher Zegras, &
Mehndiratta, 2012; Tilahun and Li, 2015; Zacharias, 2001). Given this,
it is relevant to understand how public spaces are perceived, which
allows to identify possible interventions that can nudge users’ behavior
towards more sustainable practices, such as preferring denser neigh-
borhoods or active transportation.
The literature has reported multiple efforts to quantify these per-
ceptions, dealing with costly data-collection processes. Nevertheless, a
recent body of literature has made use of massive data-collection
techniques to quantify these perceptions using machine learning
models.Inspiteoftheadvantagesthesemodelspresentintermsoftheir
data processing and predictive abilities, they do not provide
information that directly explains the drivers behind respondents’ de-
cisions. Because they work as “black boxes,” there is no straightforward
way of verifying if certain attributes, such as the presence of vegetation
or pedestrians, have positive or negative impacts over respondents’
perceptions. Moreover, they cannot allow to infer elasticities between
the presence of different features, which could help understand how
people are willing to trade off between them.
In this work, we propose a novel methodology to quantify percep-
tions of images of public spaces using a massive dataset: Place Pulse
(Salesses,Schechtner,&Hidalgo,2013).Thisprojecttakesadvantageof
the opportunity provided by platforms like Google Street View that
freely make public a virtually global coverage of two-dimensional
images of public spaces in several cities. While two-dimensional images
are not capable of conveying the whole complexity of a public space,
they provide a good proxy for this. With this dataset, we argue in favor
of the methodological usefulness of discrete choice models whose pri-
mary inputs are semantic segmentations of images of public spaces.
These not only are theoretically better suited for many datasets and
efficient enough for massive amounts of data, but have interpretable
https://doi.org/10.1016/j.landurbplan.2018.09.020
Received 21 March 2018; Received in revised form 21 September 2018; Accepted 24 September 2018
1
Address: Vicuña Mackenna 4860, Macul, Santiago, Chile.
E-mail address: terosset@uc.cl (T. Rossetti).
Landscape and Urban Planning 181 (2019) 169–178
Available online 11 October 2018
0169-2046/ © 2018 Elsevier B.V. All rights reserved.
T