Relating 2D isovists to audiovisual assessments of two urban spaces in a neighbourhood

2.1 Soundscape and architectural morphology

Soundscape is defined as the acoustic environment as perceived or experienced and/or understood by a person or people, in context [6]. The term was coined by Schafer [7] as a reaction to the emerging field of environmental noise research that manages sound as waste, rather than as a resource. Following the European Landscape Convention [8], soundscape is defined as the acoustic environment of a place, as perceived by people, whose character is the result of the action and interaction of natural and/or human factors. A review on applications to soundscape planning was published by Brown [9]. A comprehensive book on soundscape in built environment by Kang and Schulte-Fortkamp [10] systematically collects concepts, measurement procedures and different applications to the soundscape approach in planning, design and assessment. Recently, a handbook collects the recent research and appilcations in the field of soundscapes, providing discussion of how it can enhance the quality of life [11]. The soundscape approach is standarized by ISO 12913 [6].

The field of soundscapes overlaps, to various degrees, with fields of acoustics such as sound quality and acoustic comfort in buildings. In this regard, urban sound propagation can be assumed to be a special case of outdoor sound propagation: in addition to the consideration of physical properties of the propagation medium, i.e., of air, it must account for reflection properties (absorption and scattering) and diffraction. All are material dependent with respect to energy loss, and they vary with geometry in respect to frequency, time and direction of sound incidence. In the latest, architectural morphology plays a crucial role.

2.2 Soundscape standards and nonexpert’ attributes

The current ISO 12913 standard framework presents different methods for data collection [12]. The questionnaire-based soundscape assessment (method A) is frequently used in because it provides a mathematical relation among the items of the questionnaire that compose it [13]. The core of this questionnaire is the set of eight perceptual attributes originally derived in the study by Axelsson et al. [14], explaining the biggest variance in a principal components analysis. Aletta et al. conceptually considered the attributes to form a two-dimensional circumplex on x and y axis, where all regions of the space are equally likely to accommodate a given soundscape assessment [15]. Mitchell et al. [13], the authors explain why the circumplex model has several aspects that make it useful for representing the soundscape perception of a space as a whole. The application of these methods to represent the soundscape of a place and its relation to social contexts and individual perception are crucial aspects for which this approach would probably not be sufficient [16]. Recent advances in the soundscape approach since the development of the standards has shifted some focus from individual soundscapes to characterizing the overall soundscape of the public spaces [17] and to making comparisons between different groups of people [18]. For this reason, researchers regard that considering natural variation in people’s perception, environmental variations (like weather changes) of a soundscape [19] or view field properties [20] must be a core feature of the soundscape approach. Mixed-method studies can provide additional insight into how people experience their environment and should be considered alongside or preceding the ISO standard soundscape, which focuses on how a space is likely to be perceived on a larger scale [13]. For these reasons, to bridge the gap between nonexperts attributes and their descriptions considering sound and vision, the individual vocabulary profiling (IVP) technique can be used, thus extracting the main perceptual dimensions regarding sound and vision, and correlating them with morphological measures. This study is in line of a previous proof of concept study [21].

2.3 Isovists

In geometry, an isovist refers to the spatial volume visible from a specific point in space, accompanied by precise information about the location of that point. Coined by Tandy in 1967 [22] and further developed by architect Benedikt [5], isovists are inherently three dimensional, although they can also be studied in two dimensions, either in a horizontal section (referred to as “plan”) or in various vertical sections intersecting the three-dimensional isovist. Each point in physical space is associated with its own isovist.

The isovist serves as one of the two representations of spatial structure, alongside the spatial-envelope representation. This concept offers an approach to describing space from the perspective of an individual within an environment. It involves a polygonal representation covering the visible or reachable area when walking in a straight line from a specific position. The boundary shape of an isovist may remain consistent or vary depending on the location within a room. In convex rooms (e.g., rectangles or circles), the boundary shape and volume (or area in plan) of every isovist are uniform, even though the viewpoint’s location relative to the boundary may differ. In contrast, nonconvex rooms (e.g., L-shaped or partitioned spaces) can result in multiple isovists with varying volumes, some potentially covering the entire room, each with unique shapes such as large, small, narrow, wide, centric, or eccentric. Alternatively, one can conceptualize the isovist as the volume of space illuminated by a point source of light, extending its application to the 3D digital environment by representing the area not in the shadow cast by a single point light source. In this framework, it can be said that area and perimeter deserve a special attention among the isovist metrics, since they are commonly known as the basic geometric properties. In addition, they have been found out to be the most significant metrics in this study. However, other isovist metrics are also relevant; they have been included in the study (compactness, circularity, convex deficiency, occlusivity, minimum radial, maximum radial, standard deviation, variance, skewness, and disperssion). Widely used in architecture, isovists play a crucial role in the analysis of buildings and urban areas, and is one of the basic conceptions of other spatial analysis theories, such as space syntax, which aims to forecast likely effects of architectural and urban space on users [23], prediction of perceived landscape openness [24], or predictors of evacuation route choice [25], among other applications.

3.1 Urban locations

The study focuses on eight public spaces located in the area called Driescher Hof in Aachen, Germany. The public spaces are distributed across two distinct areas, a neighborhood shopping square and a neighborhood park, see Figure 1 (https://maps.app.goo.gl/rZZRpUN4n6sz7ggDA). The square (recording positions 1 to 4) provides services for the basic needs of the neighborhood residents, including a supermarket, pharmacy, a kiosk and a restaurant/café, all open during the recording times. The square is well organized and maintained, ornamented with greenery in its main entrances. In addition, the residents characterize the shopping square as a safe and comfortable place, offering opportunities to meet or to come across with neighbors during the day. The role of the shopping square in the neighborhood social life has been intensified in the lockdown due to the COVID-19 pandemic.

Figure 1

Recording positions 1 to 4 belong to the market place and 5 to 8 to the park.

The neighborhood park (positions 5 to 8) provides various recreational facilities. These include a playground, picnic tables and a basketball field. The park lacks physical and visual permeability to the rest of the district, especially in its north entrance, due to the wide facades of the four-storey residential buildings and high trees. According to informal questionnaires the residents related the feeling of unsafety and discomfort in this park to the lack of permeability with adjacent streets. Despite its pitfalls, the park has many positive characteristics, an attractive gathering place for the neighborhood, and many quiet areas, and it creates a sense of oasis within the residential area. The span on perceptually opposite features condensed in a relatively small area makes Driescher Hof an attractive and functional study case.

3.2 Individual visual and soundscape attributes

In the form of an online experiment, an IVP method was adapted and modified from the Flash profile [26] for the subjective evaluation of urban soundscapes. In IVP, each participant develops his own set of attributes for the evaluation. The implemented urban soundscape evaluation method consisted of one session with three phases for each participant. Each session lasted a maximum of 1.5 hours depending on the individual processing speed of the participant. The first phase was designed for the attribute development and elicitation process. The participants were asked to provide attributes that describe differences between the presented stimuli in the format of nouns. The second phase was designed to revise the attributes, with the option to discard or add nouns. The participants were asked to detail each noun in the format of two adjectives as opposites of a Likert scale. In the third phase, participants used their attributes and scales to rate the stimuli on a scale of 0 to 100. See Figure 2 for an example of the participant’s user interfaces. The sensory nature of the test implies participants comparing instantaneously all scenes for a single attribute, which is set to a maximum of eight scenes as previous literature suggested [26–28].

Figure 2

Online interface with 360-degree videos in YouTube used in first and third phases (up); form for the elicitation and discussion of attributes in the second phase, in red the agreed attributes (left-down); and rating questionnaire for the third phase (right-down).

The setup therefore consists of eight recorded positions (Section 3.1) reproduced via headphones and screen. The audio information is recorded in Ambisonics B format by using the Zoom H3-VR portable battery recorder, with the built-in microphone for first-order Ambisonics recording. All scenes were recorded with adjusted gains to avoid clipping. Afterward, the reproduction levels were compensated accordingly to ensure that the level differences between scenes are kept as in the real scenarios. The video information was recorded in 36 0 ∘ video in a panoramic monoscopic format with a resolution of 7,680 × 4,320 pixels, using the Insta360 Pro2 camera. Both formats are reproduced using the YouTube 360 audiovisual platform, thus allowing the participants to perform the test at their own homes under COVID restrictions. The recorded length for each stimulus is approximately 30 s. The recorded acoustic salient events at each position are described in the previous work [21]. The videos are accessible via the YouTube channel [29].

The experiment was performed online with the support of an instructor in presence or in an online call. This allowed the instructor to check with the participant for a quiet and nondistracting surrounding, as well as for the correct use of headphones and the experience of the full 360 degree range of the videos. To ensure that all participants were exposed to similar acoustic levels, the calibration method offered by https://hearingtest.onlinewas used.

3.3 Isovist metrics

The calculation of isovist metrics has been performed using the DeCodingSpaces Toolbox for Grasshopper [36]. All metrics have been evaluated using 180 radials. The following isovist measure definitions have been calculated for each of the urban locations (Figure 3):

• Area

• Distance-weighted area

• Perimeter

• Compactness, property relative to a circle [37].

• Circularity, comparison of the area to the perimeter of the isovist [38].

• Convex deficiency: Difference of the areas of the convex hull and the shape divided by the hull area [37].

• Occlusivity: the proportion of edges of an isovist that are not physically defined [39]. The greater the level of occlusion, the greater the sense of mystery [38].

• Minimum radial. Minimum length of isovist radials.

• Maximum radial. Maximum length of isovist radials.

• Mean radial. Mean length of isovist radials.

• Standard deviation, of the length of all radials of an isovist.

• Variance of the length of all radials of an isovist.

• Skewness, third moment of the mean of the radials. Associated to the factor of compactness [5,40].

• Dispersion, equivalent to skewness [5,41].

Figure 3

Isovists for each of the studied urban locations.

3.4 Correlation between clustered vocabulary and isovist metrics

Between the rating values from the individual attributes and the isovist metrics, a Spearman correlation analysis has been performed. The low number of observations (eight locations in our case) suggests the use of this type of correlation analysis, rather than a Pearson correlation [42].

4.1 Clustered attributes

In total, 14 different clusters are extracted. Some clusters contain auditory attributes, some contain visual attributes, others contain audiovisual attributes, and others contain mixtures of any. Figure 4 displays the dendogram of the clustering result. The attributes are linked by branches of the tree. The x -axis displays the Euclidean distance between clusters.

Figure 4

Dendogram representing the Euclidean distances in the X -axis between each attribute value. The blue lines represent the clustered attributes.

The centroid is calculated per cluster and is represented by a vector. In the upper row of Figure 5, the superposition of the clustered attributes and the vectors are represented in the first three dimensions of the global PCA. It can be noted that the vectors of the clusters “peacefulness” and “comfort” point in opposite directions to clusters “loudness” and “crowdedness” in PC dimensions 1 and 2. This clearly defines two axes of perception with two opposing poles. Those axes can be seen as perceptual dimensions which have been commonly used by the participants to asses the environments. Furthermore, there are cluster vectors that are very close together, like “chill” and “comfort.” This suggests that both clusters contain similar perceptual dimensions, but it was decided to keep them separate due to possible differences in meaning. Figure 5 (lower row) shows the confidence ellipses grouping the park and market based on the perceptual data. There is also a significant difference between both places. This is clearly visible in the confidence ellipses in the PC dimensions 1–2 and 1–3, as they do not overlap. When represented in the PC dimensions 2–3, the ellipses show an overlap. This is assumed to be due to the variance within PC dimensions 2 and 3, which appears to be higher than for PC dimensions 1–2 and 1–3. A more distinct result would probably require a higher number of participants. The size of the ellipses indicates that the extreme positions (01, 02, 06, 07) have more variability in their perceptual assessment than the rest.

Figure 5

Clustered data points, vector representations of groups, and confidence ellipse of the three first PC dimensions explaining 79% of the variance, for the normalized data. The labels “market” and “park” are added to the lower plots to represent the centroids for the two environments.

4.2 Resulting isovist metrics

In Table 1, the isovist values for the eight recording positions are displayed (Figure 6).

Figure 6

Isovist metrics for the five highest correlating isovist metrics. For each metric, the range of values is displayed.

4.3 Correlation between attributes and isovist metrics

In this section, we discuss the results of the correlations. To facilitate the relation between them, some description of the isovist metrics and methods are included alongside the discussion.

High correlation values can be found between pairs of perceptual attributes and isovist parameters in Figure 7. It is worth noting that correlation can be positive or negative. It can also be derived that some clusters of attributes do correlate better than others. The isovist indicators have been grouped accordingly to highlight the highest correlating clusters. The first family of metric isovists is based on the basic property area ( A ). However, while area is relatively straightforward to calculate, it provides no information about the shape of the isovist. For this reason, the perimeter ( P ) length was soon identified as one means of differentiating shapes with similar areas but different boundary conditions [40]. In our study, both area and distance-weighted area correlate highly with C9 (Peacefulness and Naturalness) and C4 (Liveliness). Both clusters do not limit to neither visual nor acoustic assessments. They group assessments on rather multimodal experience. It could be concluded that the bigger the visible area, the bigger the sensation of being in a natural space. Curiously enough, the perceived liveliness is also directly proportional to area. Furthermore, perimeter is positively correlated with C14 (loudness activity), which is a clearly audio-related cluster of attributes. The correlation tells us that, in this study case, the higher the perimeter of the isovist is, the higher is the perceived loudness due to activity. This relation could originate from the number and the kind of sound sources and their distribution. But this must be further explored before general conclusions can be drawn.

Figure 7

Correlation values between individual attributes and isovist metrics, for AudioVisual attributes. In brackets, the attributes are clustered and a cluster name is given representing the most used word in the cluster or the main meaning.

Area and perimeter are frequently combined as ratio (A:P) [43]. In an isovist, the perimeter has both a number of edges (Pol) and a number of edge changes in direction, known as “jaggedness” (J). Moreover, the perimeter can be made up of two types of edges, “boundaries” (P), being edges which correspond to solid surfaces, and “occluded radials” (O), which are edges that result from part of the environment being obscured by a closer object. These radials are also not congruent with a physical barrier and therefore change as observation points change. From these measures, a range of secondary indices have been developed for architectural analysis, including occluded/perimeter ratio (O:P) and average occluded length/area ratio (AvO:A). [37] argues that the mathematical measures concavity (Con) – which he also calls convex deficiency – and circularity (Circ) are especially useful for comparing isovists. Convex deficiency is the difference of the areas of the convex hull and the shape divided by the hull area [37], while circularity is a comparison of the area of the isovist to the perimeter of the isovist. In the correlation, both circularity and convex deficiency show almost the same behavior, meaning that they are redundant for the isovist analysis, which confirms study by Arthur [37]. Both metrics correlate negatively with C1 (Urbanity light) and C8 (Level annoyance). This has practical implications: the higher the convex deficiency is, the bigger is the area of dents over the convex hull in the isovist. This means that more disturbing objects are placed around the observer and more hidden geometries are visible, so direct view of the sky and the horizon is blocked. Convex deficiency also implies that there are places where enemies could be hiding [37], which could result in an uncomfortable feeling. This observation cannot be confirmed. In contrast, the negative correlation with C1, C3, C7, and C8 shows that higher convex deficiency correlates with a reduction in luminosity and annoying noise, as sound sources are possibly heard at a lower level without being seen. C9 is mostly correlating positively, meaning that convex deficiency is related to the feeling of relaxation and naturalness.

Three additional measures are associated with the visual “enticement” embodied in a space. Drift magnitude (DrM) is the distance from the observation point, d, to the “center of gravity,” c, of the isovist polygon, where the center of the isovist is calculated as a “polygonal lamina” [39]. The distances between observation point and center of gravity in the x and y planes individually and the square root of the sum of the squared planar differences is the magnitude of the isovist drift. In our dataset, this indicator does not show clear correlation with specific clusters.

Finally, the “statistic” isovist metrics give information on the radials starting in the observer point and arriving to the boundaries of the isovist [41]. Benedikt [5] proposes that the statistical measures variance (variance of the radials’ length) and skewness can also be used to quantify the dispersion of the perimeter around the observation point and the asymmetry of the isovist polygon. Variance (M2) is the second moment of the mean of the radials, and skewness (M3) is the third moment of the mean of the radials. In our dataset, mean radial and standard deviation correlate very similar to area. This is logical since mean radial gives an idea of the size of the perimeter, as well as area does. Variance and max radial, on the other hand, correlates similarly as perimeter.