Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Landscape and Urban Planning 141 (2015) 100–111

Contents lists available at ScienceDirect

Landscape and Urban Planning


journal homepage: www.elsevier.com/locate/landurbplan

Research Paper

Classification of urban park soundscapes through perceptions of the


acoustical environments
Jin Yong Jeon 1 , Joo Young Hong ∗
Department of Architectural Engineering, Hanyang University, Seoul 133-791, Republic of Korea

h i g h l i g h t s

• Soundscapes in urban parks could be characterized by dominant sound sources.


• Classifications of soundscape in urban parks could be identified by using acoustic indicators.
• Human-made sounds play an important role in soundscape perception in urban parks.

a r t i c l e i n f o a b s t r a c t

Article history: The aim of this study is to classify the soundscape characteristics in urban parks based on perceptions of
Received 14 July 2014 acoustic environments and to examine useful acoustic indicators to identify soundscape classifications.
Received in revised form 22 April 2015 Both acoustic measurements and subjective evaluations of the soundscape were conducted at various
Accepted 5 May 2015
locations in three urban parks. Using hierarchical cluster analysis (HCA), the soundscape perceptions in
Available online 22 May 2015
urban parks were classified into three clusters characterized by the dominance of sound sources, such as
traffic noise, natural and human-made sounds. Both discriminant function analysis (DFA) and artificial
Keywords:
neural network (ANN) analysis were performed using various acoustic parameters to discern which of
Soundscape
Urban park them best differentiate the soundscape classifications in urban parks. It was found that the indicators
Soundscape classification representing sound strength (LAeq ), perceived pitch sensation ( 1 ) and strength of the pitch components
Soundscape indicators (˚1 ) of sound scenes could effectively identify the clusters. In particular, autocorrelation function (ACF)
parameters  1 and ˚1 were significantly correlated with identification of traffic noise and human sounds.
It was revealed that sounds caused by various human activities in parks play an important role to influ-
ence eventfulness of soundscape perception. Soundscape perceptions were also closely correlated with
esthetic quality, simplicity and sense of enclosure of landscape. Despite the fact that the results might
be limited to the local conditions in Seoul, the findings could provide useful information for designing
appropriate soundscapes in urban parks.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction integration by providing recreational spaces for urban dwellers.


City parks not only help improve the physical and mental health of
Urban parks play a vital role in sustainable urban environ- citizens, but also increase the vitality of cities (Brown, Schebella,
ments. City parks provide important environmental benefits, such & Weber, 2014; Chiesura, 2004; Coley, Sullivan, & Kuo, 1997;
as air and water purification and reduction of urban heat islands. Tzoulas et al., 2007). Recently, several landscape studies on urban
Urban green spaces also encourage physical activity and social parks have been conducted using the concept of soundscape
suggested by Schafer (1977) because landscape and soundscape
are closely associated and significantly influence the experience
∗ Corresponding author at: Architectural Acoustics Lab (Room 601), Department of of park users (Brambilla, Gallo, Asdrubali, & D’Alessandro, 2013;
Architectural Engineering, Hanyang University, 17 Haengdang-dong, Seongdong-gu, Brambilla, Gallo, & Zambon, 2013; Liu, Kang, Behm, & Luo, 2014;
Seoul 133-791, Republic of Korea. Tel.: +82 2 2220 1795; fax: +82 2 2220 4794. Liu, Kang, Luo, & Behm, 2013; Liu, Kang, Luo, Behm, & Coppack,
E-mail addresses: jyjeon@hanyang.ac.kr (J.Y. Jeon), st200045@hanmail.net 2013; Tse, Chau, & Choy, 2012). Carles, Barrio, & Lucio (1999) found
(J.Y. Hong).
1
that congruence between sound and visual images influence the
Address: Architectural Acoustics Lab (Room 605-1), Department of Architectural
Engineering, Hanyang University, 17 Haengdang-dong, Seongdong-gu, Seoul 133-
perception of environmental quality in urban green space; the
791, Republic of Korea. Tel.: +82 2 2220 1795; fax: +82 2 2220 4794. coherent combination of sound and visual elements leads to higher

http://dx.doi.org/10.1016/j.landurbplan.2015.05.005
0169-2046/© 2015 Elsevier B.V. All rights reserved.
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 101

environmental preference scores in the study. Jabben, Weber and objectives of this study are to classify the acoustic environments
Verheijen (2015) evaluated the restoration levels of urban parks in urban parks, based on perceptions of soundscapes and to iden-
and showed that their variation was large, mainly due to the tify the important acoustic indicators that describe the soundscape
differences in size of parks and noise levels. These demonstrate characteristics. To achieve these goals, acoustic measurement and
that soundscape, as well as landscape, contribute to provide a subjective assessments of sound and visual environments were
better environment for visitors in urban parks. conducted in three urban parks using soundwalk methodology.
Soundscapes in urban parks have been investigated using phys- Soundscapes in the urban parks were categorized using HCA, and
ical and psychological descriptors. The acoustic indicators used in the contribution of acoustic indicators in identifying soundscape
soundscape studies can be summarized in three aspects (Coensel characteristics was examined using DFA and artificial neural net-
& Botteldooren, 2006): sound strength, spectral content and tem- works (ANN). In addition, the relationship between sound and
poral structure. In previous studies (Brambilla, Gallo, Asdrubali, visual environments in urban parks was analyzed to investigate
et al., 2013; Brambilla, Gallo, & Zambon, 2013; Nilsson & Berglund, interaction between landscape and soundscape.
2006; Tse et al., 2012), A-weighted sound pressure levels (LAeq )
and loudness have been widely used to describe sound strength
2. Methodology
in urban parks. The spectral characteristics of sound environments
in urban parks have been quantified using physical indicators such
2.1. Selection of urban parks
as spectral centroid, sharpness, and LCeq − LAeq (Brambilla, Gallo,
Asdrubali, et al., 2013; Brambilla, Gallo, & Zambon, 2013; Hong
Seoul Forest Park (172,000 m2 ), Olympic Park (1,200,000 m2 ),
& Jeon, 2013). Acoustic indicators such as fluctuation strength,
and Sunyudo Park (110,000 m2 ), all located in Seoul, Korea, were
LA10 − LA90 , music-likeness, and crest factor have been applied
selected because the sizes of the parks are not only large enough
to explain temporal structures (Coensel & Botteldooren, 2006;
to consist of diverse spaces, landscapes, and uses, but also because
Hong & Jeon, 2013; Torija, Ruiz, & Ramos-Ridao, 2013; Yang &
they are popular urban parks frequently visited by citizens in Seoul
Kang, 2013). Regarding assessment for perceptual dimensions of
(see Fig. 1). Seoul Forest Park (SF) is surrounded by roads and
soundscapes, the semantic differential method employing var-
includes various facilities, such as a water fountain, grassy area, out-
ious adjective attributes has been commonly used to describe
door theater, forest glade, children’s playground, waterside shelter
soundscape characteristics and extract the principal components
and forest path. Olympic Park (OP) is situated in the city center and
of soundscape (Axelsson, Nilsson, & Berglund, 2010; Kang & Zhang,
provides sports facilities, culture, and leisure. OP contains a wide
2010). In addition, perceived sound sources have been surveyed by
grass area, a city lake, and outdoor squares of various sizes. Sun-
interviewing park users (Brambilla, Gallo, Asdrubali, et al., 2013;
yudo Park (SY), located on Sunyudo Island in the middle of the Han
Brambilla, Gallo, & Zambon, 2013; Liu et al., 2014; Liu, Kang, Luo,
River. SY provides leisure and cultural facilities with various water
& Behm, 2013; Liu, Kang, Luo, Behm, & Coppack, 2013; Nilsson &
features. A soundwalk method has been widely adopted to evaluate
Berglund, 2006).
soundscape in urban environments because it allows for the quan-
Several approaches have attempted to classify urban sound-
titative and qualitative evaluation of soundscapes as multimodal
scapes based on the physical and perceptual characteristics of
experiences (Jeon et al., 2013; Jeon, Lee, Hong, & Cabrera, 2011;
acoustic environments. Yang and Kang (2013) conducted cluster
Semidor, 2006). Both subjective responses and objective measure-
analysis for audio-recordings from urban environments to explore
ments of soundscapes can be obtained through conducting the
the differences between natural and urban sounds using various
soundwalk. In addition, the soundwalk can be employed with a
acoustic parameters, and they found that it is necessary to use a
relatively small sample size compared to survey methods because
combination of parameters, such as loudness, sharpness and fluc-
participants experience and assess the evaluation locations along
tuation, to classify types of sound sources. Torija et al. (2013)
all routes. Thus, in this study, a soundwalk method was employed
suggested acoustic indicators based on discriminant function anal-
to assess the soundscape quality in urban parks. Fig. 2 shows the
ysis (DFA) to clearly describe perceptual soundscape categories
locations selected for soundwalks at each park.
determined by hierarchical cluster analysis (HCA). Jeon, Hong,
The soundwalk route in each urban park was designed to include
and Lee (2013) clustered urban soundscapes into four categories
the evaluation locations with distinct visual and/or auditory qual-
based on subjective evaluations of acoustic and non-acoustic fac-
ities, such as open squares, playgrounds, water features, green
tors: designed soundscape, noisy soundscape, natural landmark
spaces, and an outdoor theater. In SY and OP, eight different loca-
and urban open space. Coensel and Botteldooren (2006) suggested
tions were chosen, and nine different locations were selected in SF,
criteria for tranquility in rural soundscapes based on acoustic
along the soundwalk routes.
parameters.
Compared to studies on urban and rural soundscapes, only a
few studies have been conducted on soundscape classification in 2.2. Questionnaire
urban parks. An exploration of soundscape classifications in urban
parks is important because their acoustic environments consist of The questionnaire for soundscape evaluations contained three
various types of sounds that may positively or negatively influ- sections. The first section included demographic information,
ence the soundscape (Tse et al., 2012). Even though Brambilla, including gender, age, and noise sensitivity. The second section
Gallo, Asdrubali, et al. (2013) and Brambilla, Gallo, and Zambon asked participants how they perceived the soundscape. Partici-
(2013) have categorized acoustic environments in urban parks into pants were asked to assess the identification of sound sources at
three clusters based on acoustic indicators including percentile each location using a 5-point scale with the response choices: −2
levels, loudness, and spectrum center of gravity, only a prelimi- Do not hear at all, −1 Hear a little, 0 Hear moderately, +1 Hear a lot,
nary investigation has been conducted regarding the relationship and +2 Dominates completely. The sound sources were classified
between the clusters and soundscape perceptions. Soundscape is a as traffic noise, human sounds (talking, footsteps, and sounds from
perceptual construct related to a physical environment (ISO 12913- human activities), natural sounds (water, bird song and wind) and
1, 2014). Therefore, it is necessary to investigate the soundscape other noises (construction noise, mechanical noise and music, etc.).
classification in an urban park based on perceptual dimensions, Perception of the acoustic environment at each location was also
and to examine objective soundscape indicators that differenti- evaluated using eight adjective attributes (see Table 1); these adjec-
ate between perceptual soundscape categories. In this sense, the tives were chosen based on the previous studies (Axelsson et al.,
102 J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111

Fig. 1. Aerial photos of (a) Seoul Forest Park (SF), (b) Olympic Park (OP) and (c) Sunyudo Park (SY).

2010; Hong & Jeon, 2013; Rådsten-Ekman, Axelsson, & Nilsson, 2.3. Procedure
2013).
In the third part, visual environments were evaluated using The soundwalks were conducted to evaluate soundscape quali-
ten adjective attributes (Table 2) selected from previous studies ties in each of the three urban parks on weekdays and on Saturdays
(Hong & Jeon, 2013; Oh, 1994; Shibata & Kato, 2001). The adjec- with clear weather between 13:00 and 17:00 in September and
tives describing soundscapes and landscapes in the questionnaire October 2012. The participants were asked to evaluate their percep-
were Korean, and correspond to the English adjectives listed in tions of soundscape and landscape at the selected locations along
Tables 1 and 2. The same questionnaire protocol was used in a the soundwalking route in the urban parks, using the questionnaire.
previous study (Jeon, Hwang, & Hong, 2014), which confirms that They usually stayed at each location for 3–10 min to complete the
the questionnaire protocol could provide highly comparable and questionnaire. It took approximately 1 h to complete all eight (OP
valid results. Participants assessed the degree to which an adjective and SY) or nine (SF) soundwalks within a park.
applied to their perception of the sound and visual environments Thirty participants, mostly university students, conducted the
using a 5-point scale with the response alternatives: −2 Strongly soundwalk evaluations at each of the three parks so that in total,
disagree, −1 Slightly disagree, 0 Neither disagree nor agree, +1 90 participants (58 males and 32 females, mean age = 24.7, standard
Slightly agree, and +2 Strongly agree. deviation (SD) of age = 4.2) took part in the soundwalks. Each partic-
ipant assessed the evaluation locations in the same order, and one
subjective response per location was obtained from each partici-
Table 1 pant, giving a total of 750 subjective responses (30 participants × 25
Factor loadings of the principal component analysis in perceptions of soundscapes.a
locations).
Attributes Component 1 (42.6%) Component 2 (25.8%) The use of binaural measurement technologies is essential for
Pleasant 0.28 0.80 recording sounds for soundscape study because humans possess a
Annoying 0.09 −0.87 binaural auditory system. Binaural recording techniques are copies
Eventful 0.88 −0.06 of the acoustical environment, which are designed to be as similar
Uneventful −0.78 0.11 as possible to the auditory perception of humans. It is emphasized
Exciting 0.88 0.08
Monotonous −0.65 0.18
in the literature that binaural recording is more useful than other
Calm −0.53 0.65 methods because, with this method, it is possible to recreate the
Chaotic 0.30 −0.83 spatial characteristics of the sonic environment (Genuit & Fiebig,
a
750 responses and 90 respondents. 2006; Semidor, 2006).
In this study, while the participants were evaluating the sound-
scape at the evaluation locations, binaural signals at the left and
Table 2
the right ear positions were recorded for a 3-min duration using
Factor loadings of the principal component analysis in perceptions of visual
environments.a a binaural microphone (Type 4101, B&K, Denmark) and a digital
recorder (Zoom, H4n, Japan).
Attributes Component 1 Component 2 Component 3
(41.2%) (14.3%) (12.4%)

Appealing 0.84 0.07 0.02 2.4. Acoustic parameters


Uninteresting −0.84 0.16 0.12
Harmonious 0.71 0.30 0.03 Acoustic indicators describing the sound strength, spectral con-
Complex −0.39 −0.70 −0.01 tents, and temporal structure of soundscapes were used in this
Open 0.16 0.05 −0.86
study. A-weighted equivalent sound pressure level (LAeq ) and per-
Repulsive −0.70 −0.37 0.01
Interesting 0.77 −0.26 −0.07 centile sound pressure levels, including LA10 , LA50 and LA90 , were
Enclosed 0.09 −0.06 0.88 calculated to describe sound strength. With regard to the spec-
Simple −0.11 0.76 −0.14 tral contents of sounds, LCeq − LAeq and the center of spectral
Disharmonious −0.74 −0.28 0.07
gravity (log G) were used. LCeq − LAeq , indicating a relatively low
a
750 responses and 90 respondents. frequency content of sounds, was obtained by calculating the
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 103

Fig. 2. Pictures of the locations selected for soundscape evaluation.

difference between A- and C-weighted SPL levels. Urban inter- perception have been used to describe loudness, pitch, timbre, and
aural level difference (uILD2 ) values also were calculated to duration sensation in previous studies; ACF parameters were used
describe the spatial sound impression based on the previous study to describe single environmental noises such as road traffic noise,
(Rychtáriková & Vermeir, 2013). The spectral centroid was calcu- aircraft noise and floor impact sounds (Fujii, Soeta, & Ando, 2001;
lated according to Eq. (1), which was applied in previous studies Jeon & Sato, 2008; Sakai, Hotehama, Ando, Prodi, & Pompoli, 2002).
(Brambilla, Gallo, Asdrubali, et al., 2013; Coensel & Botteldooren, Recently, Valero and Alías (2013) investigated automatic identifi-
2006; Hong & Jeon, 2013). cation of various indoor and outdoor sound scenes recorded in real
 L /10 environments using narrow-band ACF analysis, and they demon-
10 i × Bi strated that ACF parameters could provide useful information to
G= 
i
(1)
L /10
10 i describe sound environments containing combined sound sources.
i
The ACF of a given source signal p(t) is defined by
where Li is the unweighted sound level in dB, measured for each  +T
one-third octave band with the center frequency Bi ranging from 1
˚() = p(t)p(t + )dt (2)
80 Hz to 8 kHz. LA10 − LA90 was analyzed to indicate the temporal 2T −T
variability of sound environments. In addition, the autocorrela-
tion function (ACF) parameters were adopted to describe both the where p(t) = p(t)*s(t), s(t) being the ear sensitivity, which consists
spectral and temporal characteristics of sounds. ACF parameters of the head-related, ear canal, and middle-ear transfer functions.
developed as spectra-temporal signal indicators based on human For convenience, s(t) may be chosen as the impulse response of an
104 J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111

A-weighted network. T is the interval in which p(t) is integrated and visual environment were calculated for each soundwalk par-
and  is the delay time. The normalized ACF is defined as Eq. (3) ticipant, and these scores were used in the following statistical
analyses.
˚()
() = (3)
˚(0)
Ando (2001) suggested four ACF parameters: ˚(0),  1 , ˚1 , and 3.2. Classification of soundscapes in urban parks
 e . ˚(0) is the sound energy at the origin of the delay ( = 0), which
is related to loudness;  1 [ms] represents the first peak of the ACF Various clustering algorithms including HCA, K-means, parti-
associated with perception of pitch that increases as  1 becomes tioning around medoids (PAM), and model-based clustering were
shorter; ˚1 [dB] is defined as the delay time and amplitude of the employed to classify the soundscapes in urban parks (Maffei,
first dominant peak of the ACF representing the strength of the Romero, & Brambilla, 2014; Zambon, Benocci, Angelini, Brambilla,
dominant perceived pitch;  e [ms] is the 10-percentile delay, the & Gallo, 2014). As independent variables for the clustering algo-
effective duration of the envelope of normalized ACF. In this study, rithms, the perceived sound sources and principal component
 1 and ˚1 were used among the ACF parameters to describe char- scores for soundscape factors were used to classify the soundscape
acteristics related to perceived pitch sensation. 2 T was set at 0.5 s, in urban parks. In total, 100 mean subjective responses to acoustic
and the running step was 10 ms (Jeon & Sato, 2008). ACF analyses environments were used, obtained from 25 locations on different
of the sounds were conducted using a Matlab-based analysis pro- days, because the soundwalks were carried out over a period of sev-
gram. All the acoustic indicators were calculated based on a 3-min eral days. The Euclidean distance among the subjective responses
audio recording made during the soundwalks. was chosen to be the metric of the distance among the variables.
Validation measures representing connectedness, compactness,
2.5. Statistical analyses and cluster separation were calculated to decide on the most appro-
priate method and the most appropriate number of clusters using
Principal components analysis (PCA) based on rating data for the “clValid” package (Brock, Pihur, & Datta, 2011) in R. Connect-
the adjective attributes was conducted to extract dominant factors edness, measured by the connectivity, indicates to what extent
regarding soundscape and landscape perception in urban parks. the observations are placed in the same cluster as their near-
Varimax rotation was applied to determine orthogonal compo- est neighbors in the dataset (Handl, Knowles, & Kell, 2005). The
nents and the number of components was obtained only those connectivity has a value between zero and ∞ and should be mini-
having eigenvalues greater than 1. Soundscapes in urban parks mized. Compactness is a measure of cluster homogeneity such that
were classified using cluster analyses based on perceived sound ideal clusters should minimize intra-cluster variance. Separation
sources and principal soundscape components from PCA. is a measure of the degree of distance between cluster centroids.
DFA and ANN were then applied to explore the indicator relating As compactness increases with the number of clusters, separa-
soundscape classification DFA is a useful statistical analysis to pre- tion decreases. The Dunn index (Dunn, 1974) and silhouette width
dict categorical dependent variables using a linear combination of (Rousseeuw, 1987) widely used indicators for compactness and
independent variables that best differentiate the dependent vari- separation. The Dunn index has a value between zero and ∞. The sil-
ables. A multi-layer perceptron (MLP) neural network model with houette width possesses a value between −1 and 1. A higher Dunn
a back propagation learning algorithm based on physical acoustic index and a higher silhouette width indicate better clustering.
indicators was also developed to describe the soundscape classi- Fig. 3 illustrates the internal validity measures for the four clus-
fications for comparison with the DFA results. SPSS version 19.0 tering algorithms. HCA showed lower connectivity values than the
software and R (R Development Core Team, 2014) were used to other clustering algorithms as shown in Fig. 3(a). HCA also pro-
conduct the statistical analyses. duced greater Dunn index values for all of the values of the number
of clusters; the highest Dunn index was found for three clusters
3. Results (see Fig. 3(b)). In addition, HCA with three clusters shows the high-
est values in terms of silhouette width (see Fig. 3(c)). These values
3.1. Principal components of soundscape and landscape indicate that the HCA performed significantly better than the rest
of the techniques, particularly for three clusters. Therefore, HCA
PCA was conducted using the 750 subjective responses from all was selected to be the preferred cluster algorithm, and three was
the participants. Tables 1 and 2 show the results of PCA for percep- determined to be the optimum number of clusters.
tions of soundscape and landscape, respectively. As listed Table 1, Based on the results of HCA, three clusters were classified; Clus-
the eight attributes describing soundscape were summarized as ters A, B, and C contained 71%, 17% and 12% of the perception
two components: Component 1 explains 42.6% of the variance of acoustic environments, respectively. Analysis of variance tests
with high positive loading for ‘eventful,’ ‘uneventful,’ ‘exciting,’ and were conducted to examine the significant mean differences of the
‘monotonous’ that could be interpreted as Eventfulness; Component variables among clusters to validate the classifications of urban
2 explains 25.8% of the variance and represents Pleasantness with parks. The results indicated that all the variables used in the HCA
high factor loading for the adjectives ‘pleasant,’ ‘calm,’ ‘chaotic,’ and were statistically significant (p < 0.01) according to the classifica-
‘annoying.’ The results of the PCA in the present study were sim- tions.
ilar to those in previous studies (Axelsson et al., 2010; Coensel & The composition of perceived sound sources was significantly
Botteldooren, 2006; Raimbault, Lavandier, & Bérengier, 2003). different in Clusters A–C (see Fig. 4). The sound from human activ-
Regarding landscape, three main components were obtained ities (34.7%) and natural sounds (38.6%) such as water sounds and
(Table 2): Component 1 (Attractiveness) is associated with the birdsongs were mainly perceived in Cluster A. In Cluster B, sounds
esthetic quality of the landscape, showing high factor loading with from humans were the most dominant, accounting for 48.6% of per-
the adjectives ‘appealing,’ ‘repulsive,’ ‘harmonious,’ ‘disharmo- ceived sound sources, while traffic noise and natural sounds were
nious,’ ‘interesting’ and ‘uninteresting’; Component 2 (Simplicity) relatively lower than those in Cluster A. In Cluster C, traffic noise
includes the adjectives ‘simple’ and ‘complex’; Component 3 (Enclo- was one of the most frequently perceived sound sources. These
sure) includes ‘open’ and ‘enclosed’ (Jeon, Hwang, & Hong, 2014). results imply that the characteristics of Clusters A–C can also be
Explained variances of Components 1–3 were 41.2%, 14.3% and explained by the dominance of sound sources; the environments
12.4%, respectively. Principal component scores for soundscape, classified into Cluster A could be interpreted as tranquil and calm
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 105

Fig. 3. Internal validation measures: (a) connectivity, (b) Dunn index, and (c) silhouette width (100 of mean subjective responses for 25 locations on different days).

Table 3 into Cluster A obtained high values of Pleasantness, ranged from


Pearson’s correlation coefficients between soundscape components and sound
−0.91 to 1.14, while the variation of Eventfulness in Cluster A was
sources.a
greater than that of Pleasantness, ranging from −1.70 to 1.13. It
Traffic noise Human sounds Natural sounds can be said that a high percentage of natural sounds and human
Pleasantness −0.65** 0.04 0.55** sounds increased Pleasantness scores and caused a wide variation
Eventfulness −0.34** 0.77** −0.19 in component scores of Eventfulness in Cluster A. The mean values
a
750 responses and 90 respondents. of Pleasantness and Eventfulness in Cluster A were 0.34 (SD = 0.40)
**
p < 0.01. and −0.16 (SD = 0.67), respectively. Perception of acoustic environ-
ments in Cluster B could be characterized as a wide range of Pleas-
antness (mean = −0.63 and SD = 0.66) and highly positive values of
soundscapes, Cluster B represents sound environments with vari-
Eventfulness (mean = 0.92 and SD = 0.44), due to the human sounds
ous human activities and Cluster C represents noisy soundscapes.
being dominantly identified in this cluster. Cluster C showed
As listed in Table 3, Pleasantness was negatively correlated with
negative component scores for Pleasantness (mean = −1.51 and
traffic noise, but had positive relationships with natural sounds.
SD = 0.58) and Eventfulness (mean = −0.33 and SD = 0.35) because
It was found that the dominance of human sounds increased the
traffic noise was the most frequently perceived sound source.
perception of Eventfulness (r = 0.77, p < 0.01), whereas Eventfulness
showed negative correlations with identification of traffic noise
(r = −0.34, p < 0.01). 3.3. Physical acoustic environments
These findings are in good agreement with previous studies
(Axelsson et al., 2010; Hong & Jeon, 2013) and clearly explain the The calculated acoustic parameters in Clusters A–C are listed in
soundscape characteristics in Clusters A–C. Principal component Table 4. In terms of the sound strength, Cluster A showed lower
scores of soundscape, Pleasantness and Eventfulness in term of Clus- sound pressure levels at 59.4 dBA, whereas sound levels in Clusters
ters A–C, were plotted in Fig. 5. Acoustic environments classified B and C were greater than 66 dBA of LAeq . The largest value of LA50
106 J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111

Table 4
Mean acoustic parameters in Clusters A–C.a

Cluster LAeq LA10 LA50 LA90 LA10 − LA90 LCeq − LAeq Log G 1 ˚1 uILD2

59.4 61.0 57.1 54.8 6.2 9.9 2.72 6.77 0.18 2.4
A
(6.0) (6.2) (6.4) (6.6) (3.5) (4.6) (0.33) (3.1) (0.07) (2.6)
66.8 69.3 61.8 57.7 11.6 7.0 2.83 5.35 0.30 1.7
B
(7.22) (8.0) (5.6) (5.1) (6.2) (4.1) (0.22) (2.3) (0.07) (1.2)
69.0 71.1 68.2 65.0 6.0 8.91 2.62 10.00 0.12 2.1
C
(4.8) (4.6) (5.2) (5.1) (0.83) (3.7) (0.24) (3.4) (0.02) (1.6)
a
The numbers in parentheses represent standard deviation.

The spectral characteristics of the sound environments differed


significantly. In Clusters A and C, the higher values of LCeq − LAeq
and the lower values of log G were calculated compared with those
in Cluster B. This is because the sounds from human beings, such as
talking and sounds from children’s playgrounds, perceived in Clus-
ter B contain relatively high SPL at high frequencies. Cluster C was
dominated by traffic noise, which contains higher low-frequency
levels and shows the lowest spectral centroid of sounds among the
clusters.
Regarding the ACF parameters, the mean value of  1 in Cluster
B was 5.35 ms, which was shorter than in other clusters; the sound
environments classified in Cluster C show  1 value of 10.00 ms. As
noted above, these results indicate that the sound scenes in Cluster
B are composed of human-made sounds with a high frequency of
perceived pitch, whereas Cluster C includes traffic sounds with a
low frequency of perceived pitch. With respect to pitch strength,
Cluster B shows stronger perceived pitch sensation with the largest
mean value of ˚1 (0.30). The differences of  1 and ˚1 among the
clusters were larger than the barely noticeable differences found in
previous studies (You & Jeon, 2008).
Fig. 4. Perceived sound sources refer to the percentage of respondents who heard a Table 5 shows the correlation coefficients between the sub-
given source in Clusters A–C (100 of mean subjective responses for 25 locations on jective measures and the physical acoustic indicators. Similar
different days). to previous studies (Axelsson et al., 2010; Hong & Jeon, 2013),
Pleasantness component scores were negatively correlated with
acoustic indicators representing the loudness of sounds such as
was found in Cluster C at 68.2 dBA due to the dominance of traffic LAeq , percentile levels, and ˚(0). It was also found that Evenful-
noise, followed by Clusters B and A. Cluster B included sound envi- ness component scores were highly correlated with not only sound
ronments with high temporal variability (the mean LA10 − LA90 of loudness, but also time variability and spectral content. Eventfulness
Cluster B was 11.6 dBA) because Cluster B contained mainly sounds was positively correlated with LA10 − LA90 and negatively correlated
from various human activities, whereas the temporal variability with  1 and ˚1 . Regarding the relationship between the perceived
in Clusters A and C was approximately 6 dBA. The uILD2 values sound sources and acoustic indicators, LA50 and LA90 represent
for clusters indicating an interaural level difference were less than background sound levels associated with the perception of traf-
2.5 dBA. fic noise, whereas spectra-temporal indicators were significantly
correlated with the identification of sounds from human activities
and nature. Among the ACF parameters,  1 and ˚1 show strong
relationships with human-made and natural sounds, but a signifi-
cant correlation was not found between  e and any type of sound
source.

3.4. Discriminant function analysis

In the previous sections, soundscape qualities in urban parks


were classified into three groups based on perception of acoustic
environments. In this section, critical physical indicators explain-
ing the characteristics of Clusters A–C were investigated using
DFA. Independent variables for acoustic parameters, including LAeq ,
LA10 − LA90 , log G,  1 and ˚1 to represent sound strength and tem-
poral and spectral characteristics of acoustic environments, were
selected based on the correlation coefficients between the acoustic
parameters and the subjective measures. Clusters A–C were chosen
as the categorical dependent variables.
Two discriminant functions were generated through DFA (see
Fig. 5. Principal component scores in terms of soundscape components; the circle
(), triangle (), and square () indicate Clusters A, B and C, respectively (100 of Table 6). The first and second functions explain 68.0% and 32.0%
mean subjective responses for 25 locations on different days). of the variance, respectively. Wilks’ lambda values, defined as the
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 107

Table 5
Pearson’s correlation coefficients between acoustic parameters and subjective measures.a

LAeq LA10 LA50 LA90 LA10 − LA90 LCeq − LAeq Log G 1 ˚1

Pleasantness −0.44** −0.45** −0.47** −0.41** −0.12 0.13 0.01 −0.15 0.02
Eventfulness 0.38** 0.39** 0.27** 0.21* 0.34** −0.31** 0.21* −0.41** 0.56**
Traffic noise 0.18 0.17 0.31** 0.32** −0.22* 0.13 −0.24* 0.37** −0.44**
Human sounds 0.26** 0.27** 0.17 0.12 0.27** −0.30** 0.16 −0.38** 0.59**
Natural sounds −0.33** −0.35** −0.23* −0.13 −0.41** −0.06 0.35** −0.21* −0.33**
a
100 of mean subjective responses for 25 locations on different days.
*
p < 0.05.
**
p < 0.01.

Table 6
Summary of discriminant functions analysis.a

Function Eigenvalue Variance (%) Canonical correlation Wilks’  2 df p

1 1.10 64.9 0.72 0.30 115.1 10 0.00


2 0.60 35.1 0.61 0.63 44.4 4 0.00
a
100 of mean subjective responses for 25 locations on different days.

ratio of error variance to total variance, were statistically significant


for Functions 1 and 2 (p < 0.01).
The relative contribution of each variable to the functions is pre-
sented in Table 7(a). LAeq (1.31) and  1 (0.87) were found to be
more significant contributors to Function 1 than other variables,
whereas the contribution of ˚1 (0.91) was the largest to Function
2. Similar results were found in the correlation coefficients between
the independent variables and the two discriminant functions (see
Table 7(b)); LAeq and  1 were highly correlated with Function 1.
Significant positive correlations were found between Function 2
and LA10 − LA90 (r = 0.68) and ˚1 (r = 0.90). These findings imply that
Function 1 represents loudness of sound and perceived pitch sensa-
tion, whereas Function 2 is associated with perceived pitch strength
and the temporal variability of sound environments.
As shown in Fig. 6, the acoustic environments in Cluster C
obtained highly positive discriminant scores (2.65) for Function
1 (loudness & perceived pitch sensation) than compared to Clus-
ters A (−0.43) and B (−0.09). The average discriminant scores for
Function 2 (perceived pitch strength & the temporal variability) in
Cluster B was 1.68, and those in Clusters A and C were −0.36 and Fig. 6. Discriminant function scores in terms of the soundscape classifications; the
circle (), triangle (), and square () indicate Clusters A, B and C, respectively (100
0.26, respectively. The discriminant functions evaluated at group of mean subjective responses for 25 locations on different days).
centroids clearly show that Function 1 (perceived pitch strength
& the temporal variability) plays a critical role in differentiating
Cluster C from Clusters A and B, and Function 2 (perceived pitch cross-validation was performed only for the cases in the DFA. In
strength and the temporal variability) distinguishes Cluster B from cross-validation, each case is classified using the functions derived
Clusters A and C. from all cases other than that that case. Table 8(b) presents the
Table 8(a) shows the classification results from the DFA. Over- cross-validation results which show that 82.0% of cross-validated
all, a correct classification rate of 84.0% was achieved. Clusters A grouped cases are correctly classified, which is almost the same
and B were classified with 84.5% and 70.6% accuracy, respectively.
All acoustic environments in Cluster C were correctly identi- Table 8
fied by the discriminant functions. Leave-one-out cross-validation Classification results from DFA.a
was performed in order to examine the validity of the DFA. The
Actual cluster Predicted group Total
membership (%)

Table 7 A B C
Contribution of independent variables to discriminant Functions 1 and 2: (a) stan- 60 9 0
dardized canonical discriminant function coefficients and (b) structure matrix for A 71
(84.5) (12.7) (0)
pooled within-groups correlations between the discriminating variables and the 5 12 0
standardized canonical discriminant functions.a (a) Original B 17
(29.4) (70.6) (0)
0 0 12
(a) (b) C 12
(0) (0) (100)
Function 1 Function 2 Function 1 Function 2
60 9 2
A 71
LAeq 1.31 0.35 0.47 0.54 (84.5) (12.7) (2.8)
LA10 − LA90 −0.46 0.06 −0.04 0.68 (b) 6 10 1
B 17
Log G −0.38 0.30 −0.11 0.19 Cross-validated (35.3) (58.8) (5.9)
1 0.87 0.38 0.35 −0.28. 0 0 12
C 12
˚1 0.08 0.91 −0.28 0.90 (0) (0) (100)
a a
100 of mean subjective responses for 25 locations on different days. 100 of mean subjective responses for 25 locations on different days.
108 J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111

Table 9 4. Discussion
Classification results from ANN.a

Actual Cluster Predicted group Percent 4.1. Indicators for soundscape classification in urban parks
membership correct (%)

A B C Acoustic indicators for the soundscape classifications in urban


parks were investigated through DFA and ANN. As shown in Table 7,
A 40 2 2 90.9
B 4 4 0 50.0 increments of LAeq and  1 increase discriminant scores for Function
Training 1. This indicates that as the loudness and low-frequency content of
C 1 0 9 90.0
Overall percent 85.5 sound environments increase, discriminant scores for Function 1
A 24 3 0 88.9 increase; therefore sound scenes dominated by traffic noise could
B 4 5 0 55.6 be identified based on loudness and pitch sensation. Cluster C
Testing
C 0 0 2 100.0 is clearly differentiated from Cluster A by both LAeq and  1 (see
Overall percent 81.6
Fig. 8(a)). In addition, increasing perceived pitch strength (˚1 ) and
a
100 of mean subjective responses for 25 locations on different days. temporal variation (LA10 − LA90 ) increases the discriminant scores
for Function 2; therefore sounds from people talking and human
activities could be identified based on high pitch strength with large
temporal variation. Cluster B, dominated by human-made sounds,
percentage as that produced using the original functions. These
obtains relatively higher values of ˚1 and LA10 − LA90 than Clusters
results indicate that the original discriminant functions are reliable
A and C, but the discriminating effect of ˚1 was relatively greater
and stable for the current data set.
than LA10 − LA90 as shown in Fig. 8(b).
Similarly, among the acoustic indicators describing character-
istics of sound environments, LAeq ,  1 , and ˚1 have the greatest
3.5. Artificial neural network analysis effect on the developed ANN model. The results of DFA and the
ANN model demonstrate that loudness and spectra-temporal char-
The same acoustic indicators used in DFA were selected as acteristics are important factors in soundscape, and those three
the input variables in ANN. The network inputs were normalized parameters could be useful indicators to classify the soundscape in
between 0 and 1 to reduce the effect of scale on the input vari- urban parks. In particular, among ACF parameters,  1 and ˚1 could
ables, and one hidden layer was chosen for four networks, similar be useful indicators to identify types of sound sources incorporating
to previous studies (Yang & Kang, 2013; Yu & Kang, 2009). The sam- both spectral and temporal characteristics of sounds. These findings
ples were randomly divided into a training group (62 cases, 62%) are in agreement with Valero and Alías (2013), who also found that
and a test group (38 cases, 38%). Table 9 summarizes the classi-  1 is useful in detecting a particular noise source with frequencies
fication results of the estimated ANN model for the soundscape below 200 Hz, and ˚1 could be used to identify strongly pitched
Clusters A–C. The accuracy of the training and test groups were components in complex sounds such as people cheering.
85.5% and 81.6%, respectively. Overall, the ANN model in this study Classifications of soundscapes in urban parks and rural area
could accurately classify 84.0% of the acoustic scenes, similar to the have been made using physical parameters in previous studies
results from DFA. Approximately 90.1% and 91.7% of Clusters A and (Brambilla, Gallo, Asdrubali, et al., 2013; Coensel & Botteldooren,
C were correctly identified, but the ANN model demonstrated lower 2006). Specifically, Coensel and Botteldooren (2006) applied log G
predictive ability in classifying Cluster B with 52.9% accuracy. The and LA50 as acoustic indicators to describe soundscapes in rural
importance of each acoustic indicator in determining the output areas, and they proposed that values of log G > 2.8 would indicate
value is shown in Fig. 7. It was found that LAeq was the most impor- tranquil rural soundscapes. Brambilla, Gallo, Asdrubali, et al. (2013)
tant contributor for building the network, followed by  1 , ˚1 , log G classified three clusters using various physical parameters, and
and LA10 − LA90 . they concluded that LA50 = 50 dBA and log G = 2.1 could be values to
clearly indicate the clusters. However, log G values in Clusters A–C
are not significantly different when the classification is made based
on soundscape perceptions as shown in Fig. 8(c). This indicates that
log G values are not an appropriate indicator to describe the sound-
scapes in urban parks. Therefore,  1 and ˚1 are better indicators
than log G to describe soundscape categorizations in urban parks.
Regarding the accuracy of automatic soundscape classifications,
discriminant functions and the ANN model correctly classified
approximately 84% of the sound samples into original clusters sim-
ilar to a previous study (Yang & Kang, 2013). Even though Cluster C
was almost perfectly identified by DFA and the ANN model, lower
prediction accuracy in identifying Clusters A and B was found. In
particular, the correct classification rate in Cluster B was low. This
is probably because 34.7% of human-made sounds such as peo-
ple’s voices were perceived as secondary sound sources in Cluster
A even though their sound level is lower than that found in Cluster
B (Fig. 4).

4.2. Relationship between acoustic and visual perceptions

Correlation coefficients between the sound and visual compo-


nents are presented in Table 10. The Attractiveness of a landscape
Fig. 7. Sensitivity analysis of the independent variables (100 of mean subjective was significantly correlated with Pleasantness, whereas a signifi-
responses for 25 locations on different days). cant correlation with Eventfulness was not found. Among types of
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 109

Fig. 8. Acoustic indicators in terms of soundscape classification (100 of mean subjective responses for 25 locations on different days).

sounds, Attractiveness was negatively correlated with perception of variation of soundscapes in urban spaces is driven mainly by human
traffic noise, whereas a positive correlation was found with iden- activities (Liu, Kang, Luo, & Behm, 2013; Liu, Kang, Luo, Behm, &
tification of natural sounds. Simplicity of visual environments was Coppack, 2013). In this sense, the soundscape classification of a cer-
negatively associated with the Eventfulness of sound environments tain place can change according to the presence of human activity.
(r = −0.29) and human sounds (r = −0.21), but a strong correlation Fig. 9 shows variation of soundscape classification at each selected
was found with natural sounds. Various leisure activities, such as location as the results of the soundwalk evaluations. The evaluation
riding bikes and children playing, are usually carried out in urban locations, SF1–3, SF9, OP4, and SY6 were classified into Cluster A or
parks with complex visual elements such as playgrounds and sports B depending on the presence of human activities during the sound-
facilities, which increase the time variability of sounds. walks; SF1, SF9 and SY6 are open squares in the parks. SF2 and SF4
The sense of Enclosure was significantly related to Pleasantness are an outdoor theater and children’s playground, respectively.
(0.38) and Eventfulness (−0.41) of sounds (p < 0.01). This may be Due to the close interaction between landscape and soundscape,
because park visitors usually use large and flat spaces in urban both landscape and soundscape should be taken into account when
parks for active recreational activities with small or large groups, designing urban parks. For instance, installing earth berms could
whereas passive uses such as lying down, sitting on the grass, and be one way to design an urban park adjacent to heavy traffic that
sitting on a bench are more likely to occur in enclosed spaces would provide both noise reduction by controlling sound propaga-
surrounded by trees (Goličnik & Ward Thompson, 2010). This tion and improvement of esthetic quality by increasing greenery.
implies that the use of space and visual components influencing Another possible example is that landscape architects could cre-
the pattern of human behaviors might affect soundscape percep- ate acoustic buffer zones or enclosed spaces to form tranquil sound
tions in urban parks. It has been established that a spatiotemporal environments in urban parks.

Table 10
Pearson’s correlation coefficients between soundscape and landscape components.a

Pleasantness Eventfulness Traffic noise Human sounds Natural sounds

Attractiveness 0.85** 0.16 −0.60** 0.14 0.39**


Simplicity 0.16 −0.29** −0.22* −0.21* 0.45**
Enclosure 0.38** −0.41** −0.11 −0.28** 0.21*
a
100 of mean subjective responses for 25 locations on different days.
*
p < 0.05.
**
p < 0.01.
110 J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111

Despite these findings, there are inherent limitations in this


research. One limitation is that subjective responses in this study
may be influenced by sequential biases because the soundwalk
evaluations were conducted in the same order along the sound-
walk routes. In addition, the findings might be limited to local
conditions because the evaluations of urban park soundscapes
were conducted only in Seoul, Korea. The classifications of urban
park soundscapes might vary in different cities and countries due
to socio-cultural backgrounds. Thus, further study is needed to
perform cross-cultural comparisons of urban park soundscapes
for soundscape classifications and indicators. In addition, future
research will be required to explore appropriate sound and visual
elements corresponding to different functional spaces within urban
parks.

References

Ando, Y. (2001). A theory of primary sensations and spatial sensations measuring


environmental noise. Journal of Sound and Vibration, 241(1), 3–18. http://dx.doi.
org/10.1006/jsvi.2000.3272
Axelsson, Ö., Nilsson, M. E., & Berglund, B. (2010). A principal components model of
soundscape perception. The Journal of the Acoustical Society of America, 128(5),
2836–2846. http://dx.doi.org/10.1121/1.3493436
Brambilla, G., Gallo, V., Asdrubali, F., & D’Alessandro, F. (2013). The perceived quality
of soundscape in three urban parks in Rome. The Journal of the Acoustical Society
of America, 134(1), 832–839. http://dx.doi.org/10.1121/1.4807811
Brambilla, G., Gallo, V., & Zambon, G. (2013). The soundscape quality in some urban
parks in Milan, Italy. International Journal of Environmental Research and Public
Health, 10(6), 2348–2369. http://dx.doi.org/10.3390/ijerph10062348
Brock, G., Pihur, V., & Datta, S. (2011). clValid, an R package for cluster validation.
Journal of Statistical Software, 25(4).
Brown, G., Schebella, M. F., & Weber, D. (2014). Using participatory GIS to measure
physical activity and urban park benefits. Landscape and Urban Planning, 121,
34–44. http://dx.doi.org/10.1016/j.landurbplan.2013.09.006
Carles, J., Barrio, I., & Lucio, J. D. (1999). Sound influence on landscape values. Land-
scape and Urban Planning, 43, 191–200.
Chiesura, A. (2004). The role of urban parks for the sustainable city. Landscape and
Urban Planning, 68(1), 129–138. http://dx.doi.org/10.1016/j.landurbplan.2003.
08.003
Coensel, B. D., & Botteldooren, D. (2006, August). The quiet rural soundscape and
how to characterize it. Acta Acustica United With Acustica, 92, 887L 897.
Coley, R. L., Sullivan, W. C., & Kuo, F. E. (1997). Where does community grow?
Fig. 9. Variation of soundscape classification at evaluation locations for (a) SF (b) The social context created by nature in urban public housing. Environment and
OP and (c) SY (100 of mean subjective responses for 25 locations on different days). Behavior, 29(4), 468–494. http://dx.doi.org/10.1177/001391659702900402
Dunn, J. C. (1974). Well separated clusters and fuzzy partitions. Journal on Cybernet-
ics, 4, 95–104.
5. Conclusion Fujii, K., Soeta, Y., & Ando, Y. (2001). Acoustical properties of aircraft noise measured
by temporal and spatial factors. Journal of Sound and Vibration, 241(1), 69–78.
http://dx.doi.org/10.1006/jsvi.2000.3278
Classification-based acoustic indicators are not sufficient to
Genuit, K., & Fiebig, A. (2006). Psychoacoustics and its benefit for the soundscape
describe the acoustic quality of urban parks due to the complicated approach. Acta Acustica United with Acustica, 92, 952–958.
interaction between different sounds and human auditory percep- Goličnik, B., & Ward Thompson, C. (2010). Emerging relationships between design
and use of urban park spaces. Landscape and Urban Planning, 94(1), 38–53. http://
tion. In this study, various acoustic environments in urban parks
dx.doi.org/10.1016/j.landurbplan.2009.07.016
were assessed by performing soundwalks. The soundscapes were Handl, J., Knowles, J., & Kell, D. B. (2005). Computational cluster validation in
then classified into three groups based on the perceptual evalua- post-genomic data analysis. Bioinformatics (Oxford, England), 21(15), 3201–3212.
tion of these acoustic environments. The soundscape classification http://dx.doi.org/10.1093/bioinformatics/bti517
Hong, J. Y., & Jeon, J. Y. (2013). Designing sound and visual components for enhance-
can be explained by the dominant sound sources and perceptual ment of urban soundscapes. The Journal of the Acoustical Society of America,
components of the soundscape: Cluster A includes tranquil envi- 134(3), 2026–2036.
ronments with high pleasantness of sounds, Cluster B includes ISO. (2014). TC 43/SC 1/WG 54, 12913-1 Acoustics—Soundscape—Part 1: Definition and
conceptual framework. Geneva, Switzerland: International Standard Organiza-
acoustic scenes dominated by eventful sounds from human activ- tion.
ities, and Cluster C includes traffic noise with less pleasantness Jabben, J., Weber, M., & Verheijen, E. (2015). A framework for rating environmental
and less eventfulness. The three soundscapes can be differenti- value of urban parks. The Science of the Total Environment, 508, 395–401. http://
dx.doi.org/10.1016/j.scitotenv.2014.12.007
ated using acoustic parameters, particularly the combination of Jeon, J. Y., & Sato, S. (2008). Annoyance caused by heavyweight floor impact sounds
sound strength and spectra-temporal characteristics of sounds. LAeq in relation to the autocorrelation function and sound quality metrics. Journal of
is a useful indicator for the evaluation of environmental quiet- Sound and Vibration, 311, 767–785.
Jeon, J. Y., Hong, J. Y., & Lee, P. J. (2013). Soundwalk approach to identify urban
ness.  1 and ˚1 satisfactorily discriminated between sound types,
soundscapes individually. The Journal of the Acoustical Society of America, 134(1),
particularly traffic noise and sounds made by humans. Acoustic 803–812. http://dx.doi.org/10.1121/1.4807801
environments dominated by traffic noise have relatively higher Jeon, J. Y., Lee, P. J., Hong, J. Y., & Cabrera, D. (2011). Non-auditory factors affecting
urban soundscape evaluation. The Journal of the Acoustical Society of America,
 1 and lower ˚1 values. In contrast, auditory scenes consisting
130(6), 3761–3770.
mainly of human-made sounds show lower  1 and greater ˚1 val- Jeon, J. Y., Hwang, I. H., & Hong, J. Y. (2014). Soundscape evaluation in a catholic cathe-
ues than the other clusters. In addition, human activities in urban dral and buddhist temple precincts through social surveys and soundwalks. The
parks significantly affect soundscape, and that visual components Journal of the Acoustical Society of America, 135(4), 1863–1874.
Kang, J., & Zhang, M. (2010). Semantic differential analysis of the soundscape in
influencing the park users’ behaviors are closely related to the per- urban open public spaces. Building and Environment, 45(1), 150–157. http://dx.
ception of soundscapes. doi.org/10.1016/j.buildenv.2009.05.014
J.Y. Jeon, J.Y. Hong / Landscape and Urban Planning 141 (2015) 100–111 111

Liu, J., Kang, J., Behm, H., & Luo, T. (2014). Effects of landscape on soundscape per- Schafer, R. (1977). The soundscape: Our sonic environment and the tuning of the world.
ception: Soundwalks in city parks. Landscape and Urban Planning, 123, 30–40. New York: Destiny Books.
http://dx.doi.org/10.1016/j.landurbplan.2013.12.003 Semidor, C. (2006). Listening to a city with the soundwalk method. Acta Acustica
Liu, J., Kang, J., Luo, T., & Behm, H. (2013). Landscape effects on soundscape experi- United with Acustica, 92(August), 959–964.
ence in city parks. The Science of the Total Environment, 454–455, 474–481. http:// Shibata, T., & Kato, T. (2001). Image retrieval system for street-landscape images
dx.doi.org/10.1016/j.scitotenv.2013.03.038 using adjectives. Systems and Computers in Japan, 32(6), 62–71. http://dx.doi.
Liu, J., Kang, J., Luo, T., Behm, H., & Coppack, T. (2013). Spatiotemporal variability of org/10.1002/scj.1035
soundscapes in a multiple functional urban area. Landscape and Urban Planning, Torija, A. J., Ruiz, D. P., & Ramos-Ridao, A. F. (2013). Application of a methodology
115, 1–9. http://dx.doi.org/10.1016/j.landurbplan.2013.03.008 for categorizing and differentiating urban soundscapes using acoustical descrip-
Maffei, L., Romero, V., & Brambilla, G. (2014). Characterization of the soundscape of tors and semantic-differential attributes. The Journal of the Acoustical Society of
urban waterfronts. In Proceeding of Forum Acusticum 2014 Krakow, Poland. America, 134(1), 791–802. http://dx.doi.org/10.1121/1.4807804
Nilsson, M., & Berglund, B. (2006). Soundscape quality in suburban green areas and Tse, M., Chau, C., & Choy, Y. (2012). Perception of urban park soundscape. The Journal
city parks. Acta Acustica United with Acustica, 92, 903–911. of the Acoustical Society of America, 13(4), 2762–2771.
Oh, K. (1994). A perceptual evaluation of computer-based landscape simulations. Tzoulas, K., Korpela, K., Venn, S., Yli-Pelkonen, V., Kaźmierczak, A., Niemela, J.,
Landscape and Urban Planning, 28(93), 201–216. et al. (2007). Promoting ecosystem and human health in urban areas using
R Core Team. (2014). R. A language and environment for statistical com-puting. Vienna, green infrastructure: A literature review. Landscape and Urban Planning, 81(3),
Austria: R Foundation for Statistical Computing. http://www.R-project.org/ 167–178. http://dx.doi.org/10.1016/j.landurbplan.2007.02.001
Rådsten-Ekman, M., Axelsson, Ö., & Nilsson, M. E. (2013). Effects of sounds from Valero, X., & Alías, F. (2013). Narrow-band autocorrelation function features for the
water on perception of acoustic environments dominated by road-traffic noise. automatic recognition of acoustic environments. The Journal of the Acoustical
Acta Acustica United with Acustica, 99(2), 218–225. http://dx.doi.org/10.3813/ Society of America, 134(1), 880–890. http://dx.doi.org/10.1121/1.4807807
AAA.918605 Yang, M., & Kang, J. (2013). Psychoacoustical evaluation of natural and urban sounds
Raimbault, M., Lavandier, C., & Bérengier, M. (2003). Ambient sound assessment of in soundscapes. The Journal of the Acoustical Society of America, 134(1), 840–851.
urban environments: Field studies in two French cities. Applied Acoustics, 64(12), http://dx.doi.org/10.1121/1.4807800
1241–1256. http://dx.doi.org/10.1016/S0003-682X(03)00061-6 You, J., & Jeon, J. Y. (2008). Just noticeable differences in sound qual-
Rousseeuw, P. J. (1987). Silhouettes: A graphical aid to the interpretation and vali- ity metrics for refrigerator noise. Noise Control Engineering Journal, 56(6),
dation of cluster analysis. Journal of Computational and Applied Mathematics, 20, 414–424.
53–65. http://dx.doi.org/10.1016/0377-0427(87)90125-7 Yu, L., & Kang, J. (2009). Modeling subjective evaluation of soundscape quality in
Rychtáriková, M., & Vermeir, G. (2013). Soundscape categorization on the basis of urban open spaces: An artificial neural network approach. The Journal of the
objective acoustical parameters. Applied Acoustics, 74(2), 240–247. http://dx.doi. Acoustical Society of America, 126(3), 1163–1174. http://dx.doi.org/10.1121/1.
org/10.1016/j.apacoust.2011.01.004 3183377
Sakai, H., Hotehama, T., Ando, Y., Prodi, N., & Pompoli, R. (2002). Diagnostic sys- Zambon, G., Benocci, R., Angelini, F., Brambilla, G., & Gallo, V. (2014). Statistics-based
tem based on the human auditory–brain model for measuring environmental functional classification of roads in the urban area of Milan. In Proceeding of
noise—An application to railway noise. Journal of Sound and Vibration, 250(1), Forum Acusticum 2014 Krakow, Poland.
9–21. http://dx.doi.org/10.1006/jsvi.2001.3884

You might also like