1 Introduction

Urban regions are subject to constant transformation processes. To ensure social acceptance of these transformations, modern urban planning needs to include stakeholders such as environmentalists and residents with different ages and genders in the planning processes. However, for effective participation of stakeholders in urban planning processes, visualizations are required to communicate different planning scenarios (Postert et al. 2022). Such visualizations can provide stakeholders with insights into planning details as well as broader overviews that cannot be communicated verbally or based on textual information (Al-Kodmany 1999; Gottwald et al. 2021; Jamei et al. 2017). Consequentially, choosing the most desirable or acceptable planning scenario can be based on more informed decision making.

The urban planning processes of developing, representing, communicating and exploring different planning scenarios are supported by cartographic visualization methods. The range of these cartographic visualization methods evolves with the availability of new technologies (Dickmann and Dunker 2014). Examples are the introduction of digital maps, web maps and 3D visualization techniques, e.g., based on methods from the video and gaming industries (Edler et al. 2018; Kolbe et al. 2005; Schmidt & Weiser 2012; Vosselman 2003).

In recent years, advances in virtual reality (VR) hardware and software have further expanded the possibilities of cartographic visualization and representation. Game engines such as Unity or the Unreal Engine provide ways to quickly and easily create interactive 3D spatial representations of real-world spaces, historical reproductions or planning scenarios (Chandler et al. 2022; Edler et al. 2019a, b; Ferworn et al. 2013; Indraprastha & Shinozaki 2009; Keil et al. 2021). Such 3D spaces can have a high realism, due to the availability of naturalistic physics in these engines, including simulated gravity, real-time lighting, collision calculations and 3D soundscapes (Edler et al. 2019a, b; Hruby 2019; Kersten et al. 2018). Furthermore, 3D models created in game engines can be easily adapted for use with VR hardware. Weber et al. (2022) demonstrated that providing people with information displayed in VR can affect decision making in political processes. This illustrates the potential of using VR visualizations during urban planning processes. Furthermore, Parsons (2014) argued that using VR environments not only provides experimental control of perceptual stimuli, is also supports ecological validity due to the representativeness of real-life situations.

VR models can be explored in an immersive way from an egocentric, stereoscopic perspective in a scale of 1:1 (Baños et al. 2004; Hruby et al. 2020; Slater et al. 1996), and thus more similar to the perception of real-world spaces compared to 2D and pseudo-3D maps or screen-based 3D representations. Consequentially, using VR for planning scenarios can provide stakeholders with a more engaging and realistic impression of specific planning scenarios compared to other visualization approaches (Meenar and Kitson 2020). Thus, potential conflicts with specific stakeholder groups when selecting certain planning scenarios can be better identified and the most suitable scenario is more likely to be selected.

Regardless of the selected type of visual representation, the informative value of most stakeholder participation processes can be limited due to the way in which responses are acquired. Preferences, expectations, fears and objections of stakeholders regarding specific planning scenarios are usually captured with explicit response methods, for example self-reports or questionnaires (e.g., Ballantyne et al. 2013; Ghorbanzadeh et al. 2019; Tress and Tress 2003). However, stakeholders can be unaware of certain environmental parameters that affect their responses to these planning scenarios. For example, studies demonstrate that people can be unaware of how air pollution and noise levels affect their physical comfort and their stress level (Hébert et al. 2005; Schnell et al. 2012). Thus, using merely explicit response methods during stakeholder participation in urban planning processes cannot uncover all potential comfort or health issues related to specific urban planning scenarios.

In psychological studies, physiological measures such as electrodermal activity (EDA), heart rate, blood pressure, skin temperature or pupil dilation are commonly used to assess implicit (unconscious) responses to stimuli. Such physiological measures can be used as indicators for stress, cognitive load and both positive and negative affects (emotional reactions) such as happiness, anger and fear or disgust (Chen et al. 2017; Ekman et al. 1983; Hyönä et al. 1995; Reinhardt et al. 2012; Sinha et al. 1992). In spatial research, physiological measures have been suggested to address the effects of stress on the formation of mental spatial models (Weibel et al. 2018). Applying physiological measures during the presentation of urban planning scenarios to stakeholders could not only provide helpful information about unconscious responses to these scenarios. Implicit measures have additional advantages over explicit ones. Firstly, implicit measures are not susceptible to the social desirability bias (participants giving responses that makes them look good to others, cf. Jo 2000; Nederhof 1985) and are less likely to produce an observer expectancy bias (e.g., participants trying to confirm the experimenters expectations, cf. Barber & Silver 1968; Meehan et al. 2005). Secondly, it is difficult to temporally relate explicit responses (e.g., with a stress rating scale) to specific stimuli of a planning scenario. The explicit responses can be gathered during the exploration of the planning scenario. In this case, the distraction created by the response can interfere with the perception of the scenario. Alternatively, explicit responses can be gathered after exploring the scenario. In this case, it is difficult to relate the responses to specific stimuli or a combination of stimuli in the scenario (Meehan et al. 2005). Implicit measures like EDA can be continuously recorded during planning scenario exploration with non-invasive hardware. Therefore, if short physiological response latencies are considered (cf. Benedek and Kaernbach 2010; Kappeler-Setz et al. 2013), physiological responses and the stress and affects associated with these responses can be temporally directly related to specific stimuli of a planning scenario.

In this article, we explore to what extent physiological measurements can be used to assess implicit responses of stakeholders during the exploration of VR-based urban planning scenarios. Using a low-cost EDA wristband, we focus on the development of an approach that is easily replicable with any urban planning model and generates data that is easy to evaluate. Opposed to other physiological measures (e.g., wired EDA or EEG), the used wristband does not restrict the mobility of the user and does not interfere with the exploration of a VR-based urban model. In addition to the low costs and ease of implementation, EDA measurement is characterized by the ability to be used as a measure of both stress and fear responses (Faghih et al. 2015; Reinhardt et al. 2012). Skin conductance, as measured with EDA measures, reflects activation of the sympathetic nervous system and the resulting stimulation of sweat glands (Faghih et al. 2015). In an exploratory study, we test to what extent VR-based urban planning scenarios can induce stress and fear responses that can be identified in the EDA data. Furthermore, we investigate to what extent the developed approach can be used to temporally relate such stress and fear responses to specific stimuli of a planning scenario. With our findings, we aim to support urban planners in harnessing implicit response information of stakeholders in addition to the usually merely explicit responses gathered during participation processes. Thus, the knowledge gained from letting stakeholders explore different planning scenarios could be expanded and decision making in urban planning processes could be improved.

2 Methods

Given that the focus of this manuscript was the development of an approach to easily assess and evaluate implicit responses to urban planning scenarios, we used an exploratory sample of six geography students from the Ruhr University Bochum (4 females, 2 males) for sample data acquisition. The study has been controlled and approved by the ethics committee of the Faculty of Geosciences at the Ruhr University Bochum (13 October 2022).

2.1 Scenario

Using Unity (version 2019.4.28f1), we created a VR capable 3D city model designed to provide an example of a typical German city (see Fig. 1). The model was roughly based on the 3D model created by Weißmann et al. (2022). It consisted of an urban scene with an intersection composed of a four-lane main road with two tram tracks in the middle that was crossed by a two-lane side road. Large residential buildings were located alongside the roads. Furthermore, we implemented a traffic system consisting of pedestrians, cars, buses and trams. Three scenes were derived from this 3D city model and adjustments in terms of the modeling and the soundscape were made.

Fig. 1.
figure 1

3D city model. The developed model can be explored with VR hardware from an immersive egocentric perspective

The purpose of the first scene was to asses to what extent a VR-based scenario can produce a fear reaction that can be identified in the EDA data and temporally related to a specific stimulus. In this scene, the location of the VR player was located in the middle of the two-lane road (see Fig. 2). Aside from birds flying around and subtle background traffic noises (Unity volume: 0.036), the scene started out quiet and static. This calm start was used to assess an EDA baseline. 60 s after the start key was pressed, two trucks appeared on opposite directions of the road, drove toward the participants, passed them 19 s later and disappeared in the distance after 19 more seconds. Diesel engine sounds (Unity volume: 1) were attached to the trucks to generate a realistic 3D soundscape (cf. Edler et al. 2019a, b). Colliders were placed on the road in front of the VR player location. These colliders were invisible to the observers and were used to trigger actions based on calculated object collisions. As soon as the trucks collided with them, the recording of a honking sound was played (Unity volume: 1). This scenario was meant to simulate a fear inducing crossing of a road where no traffic island is available to separate pedestrians from heavy traffic. After the trucks disappeared, the scene was then presented for 60 more seconds to assess the duration of the assumed peak in the EDA data.

Fig. 2
figure 2

Scene 1. Two trucks coming from opposed directions and sharply passing the participants were meant to induce a physiological fear reaction

The second and third scene were designed to induce different stress levels based on traffic density and traffic noise. In both scenes, the VR player was located next to where the two-lane road crossed the four-lane main road. The second scene (see Fig. 3) contained 20 motorized vehicles and a subtle background traffic noise was played (Unity volume: 0.036). Scene 3 (see Fig. 4) contained 52 motorized vehicles equipped with a combustion engine 3D sound (Unity volume: 1) and the background traffic noise was increased (Unity volume: 0.525). Furthermore, the model of a parked bus was placed in the side street, which was equipped with a 3D sound of a running diesel engine (Unity volume: 1). Both the second and the third scene were presented for 60 s. The vehicles in both scenes passed along the main road in four waves to simulate a traffic light-controlled traffic situation. The scenarios presented in these scenes aimed to compare a typical current traffic situation in a German city (Fig. 4) with a potential future traffic scenario where combustion engine cars are replaced by electric cars and the general traffic density is reduced (Fig. 3, e.g., by improving public transport options that replace private transport options).

Fig. 3
figure 3

Scene 2. The traffic system consisted of 20 motorized vehicles and subtle background traffic noise

Fig. 4
figure 4

Scene 3. The number of motorized vehicles was increased to 52. Furthermore, the volume of the background traffic noise was increased and 3D combustion engine sounds were attached to the vehicles

2.2 Procedure

Participants sat in a swivel chair and were equipped with an HTC Vive Pro VR head mounted display (HMD). Additionally, an Empatica E4 wristband used for EDA recording was wrapped around their non-dominant wrist. This watch-shaped wristband was selected, because it should produce less interference with the task compared to other EDA measures, e.g., wired EDA sensors on the fingertips. To prevent undesired data artifacts caused by arm movements (Kappeler-Setz et al. 2013), participants were asked to rest their non-dominant arm on the armrest of the chair and were instructed to look around and experience the environments displayed in the HMD. As it was not possible to temporally synchronize the EDA wristband with the VR model code based, we were required to set a timestamp for identifying the start of the experiment and the presentation of specific stimuli in the EDA recordings. For this purpose, the VR model started with the first scene in an idle state and no dynamic stimuli. The dynamic scenes were programmed to start with a key press. By pressing this key and a timestamp button on the EDA wristband simultaneously, we were able to record a timestamp for the EDA data that corresponded to the start of the experiment. After the start key was pressed, no additional input from the experimenter or the participants was necessary. At the end of each scene, the next scene was started automatically.

2.3 Measures

The used Empatica E4 wristband records EDA data with a temporal resolution of 4 Hz. In our approach, we investigate the unfiltered skin conductance level (SCL) in Microsiemens as provided by the wristband. The visualized data is investigated for phasic activations (sudden peaks in the SCL) that are assumed to reflect fear-induced stress and other stress reactions to the visual and auditory stimuli provided by the VR-based urban model (cf. Kappeler-Setz et al. 2013; Trujillo et al. 2019). The use of unfiltered EDA data not only allows for a simple and easily replicable approach. It also provides the potential to monitor and evaluate real-time data in future approaches.

3 Results

All data processing and visualization was carried out using R Studio (Version 1.4.1717). To temporally align the EDA data of different participants and to identify the stimulus events within the data, we cropped the data of each participant using the recorded timestamp as the start time and 398 s (experiment duration) after the timestamp as the end time. Additionally, an adjusted timeline was applied to each recording beginning with zero seconds at the recorded timestamp. This allowed us to plot multiple recordings of different participants in a single graph and to visually compare their EDA responses to specific stimuli (see Fig. 5). R Scripts for the cropping and visualization together with sample data can be found in the supplementary material.

Fig. 5
figure 5

Electrodermal activity in Microsiemens. Each colored line represents a different recording. Scene switches and the truck stimulus are marked in the data with vertical lines

In Fig. 5, we can observe distinct phasic activation shortly after the trucks passed by the participants in the first scene and after the switch from the second to the third scene. Furthermore, moderate phasic activation occurred shortly after the switch from the first to the second scene. General SCLs of most participants were higher in the third scene compared to the second scene. Moreover, SCLs slowly dropped in the first scene after the trucks passed and after the start of the second and third scene. In the data of one participant (orange line), we can observe a few unique moderate phasic activations during the second scene. Furthermore, EDA values of some participants (blue, yellow and green line) are extremely low and seem to have no distinct temporal relation to the presented stimuli. However, if we set the dimensions of the EDA scale separately for each participant based on their individual minimum and maximum SCL values, we can observe similar patterns as visible in the EDA data of other participants (see Fig. 6).

Fig. 6
figure 6

Electrodermal activity in Microsiemens scaled on individual participant levels. Setting the plotted EDA scale based on individual response levels can help to visualize otherwise missed events in the EDA data

4 Discussion

With the described workflow, we intend to provide urban planners with a simple and affordable approach to assess implicit responses of stakeholders to specific urban planning scenarios. In our experiment, we investigated to what extent VR-based urban models can produce fear-induced stress and other stress reactions that are reflected in EDA data.

The visualized SCLs demonstrate some clear trends across the participants. Firstly, the trucks passing the participants in the first scene seem to have induced a pronounced fear response that is reflected in high phasic activations of the SCLs. Secondly, distinct phasic activations occurring shortly after the switch from the second to the third scene support our assumption that the increase of traffic noise levels and general traffic density in a VR-based city model produces a physical stress reaction that is reflected in SCLs. Thus, in agreement with previous studies (e.g., Faghih et al. 2015; Reinhardt et al. 2012; Tashjian et al. 2022), we were able to use EDA as a measure for fear and stress responses. Thirdly, SCLs of most participants showed a recovery trend after the trucks in the first scene disappeared and during the second and third scene. This is in agreement with statements of Kappeler-Setz et al. (2013) and Greco et al. (2016) that the peak of phasic EDA activation is usually followed by a recovery phase. The slow but steady and continuous drop of SCLs observed in the EDA data of several participants during the second and third scene is interpreted as a habituation to the auditory and visual stimuli. Similar habituation effect of skin conductance responses (SCRs) to auditory and visual stimuli have been reported by Lykken et al. (1988), Bacigalupo and Luck (2018) and Barry and Sokolov (1993). Fourth, we can observe short latencies of the phasic activations after the trucks passed in the first scene and after the switch to the second and third scene. This is in agreement with Benedek and Kaernbach (2010), Leiner et al. (2012) and Sjouwerman and Lonsdorf (2019), who report varying latencies between almost one and three seconds between the stimulus presentation and the incline of the SCL and a latency between 1.5 and 6.5 s after the stimulus presentation until the peak of the phasic activation is reached. These latency variations need to be considered when observed phasic activations in the EDA data are temporally matched to specific stimuli in the VR-based 3D model (Babaei et al. 2021). Otherwise, the wrong stimuli could be interpreted as stress or fear inducing, especially if many different stimuli are presented in quick succession. Thus, potentially conflicting parameters of a planning scenario should be presented separate from each other within sufficient time intervals (e.g., five to ten seconds) whenever possible. This can help to ensure that the physiological reactions can be clearly assigned to a specific stimulus. Furthermore, by temporally separating potentially relevant stimuli, potential overlaps of several phasic activations can be prevented (Greco et al. 2016).

In addition to the trends observed across participants, we also found some artifacts and differences between participants in the visualized data. For example, some peaks in the EDA data cannot be linked clearly to a specific stimulus. In the orange line in Fig. 5, we see some moderate phasic activations during the second scene that cannot be observed in the data of other participants. After the experiment, this participant reported to have looked barely to the left side during this scene. Therefore, due to the limited field of view of the HMD, this participant reported to have been surprised by the vehicles that appeared in waves in the field of view. This might explain the pattern of EDA peaks and demonstrates an issue related to VR-based stimulus presentation. As the user can freely choose the visual perspective of a VR scene by rotating and moving the head (Keil et al. 2021), experimenters have only limited control over the stimulus presentation. During stakeholder participation processes, this could make it difficult to match measured physiological responses to the stimuli that triggered these responses. Therefore, we propose to record the scene in the field of view that is dynamically selected by the stakeholders during the presentation of VR-based urban planning scenarios for later matching with SCL events. Even more relevant information on potentially stress or fear inducing stimuli can be obtained by using VR HMDs with eye tracking capabilities, for example the HTC Vive Pro Eye. As eye fixations are argued to reflect attentional processing (Henderson 2003; Just & Carpenter 1976; Poole & Ball 2006), investigating eye fixations can help to identify stimuli associated with a phasic activation in the SCL data. However, even if physiological measures, field of view recordings and eye tracking provide information that cannot be gathered with explicit response methods, applying subsequent interviews or think aloud techniques is highly recommended, as this can shed light on data artifacts that are otherwise difficult to explain.

Some general differences of SCLs between participants were also observed in the data. Some participants (e.g., the one represented by the orange line in Fig. 5) showed strong differences between tonic and phasic components in the EDA data whereas SCLs of others (the ones represented by the blue, yellow and green line in Fig. 5) appeared almost flat in comparison. This problem is amplified by the fact that certain (albeit small) groups of non-responders and hyper-responders have been identified in EDA experiments (Fung et al. 2005; Mirkin & Coppen 1980; Tran et al. 2023), as well as effects of caffeine consumption and medication on SCls and SCRs (Babaei et al. 2021). This illustrates the difficulty to compare SCL trends and events between participants when raw SCL data is investigated. Possible solutions for addressing different individual SCLs would be to standardize the data between participants using Z-scores, percentages based on individual maximum scores recorded after a startle stimulus, or to decompose the data into tonic (baseline levels) and phasic components (cf. Babaei et al. 2021; Benedek & Kaernbach 2010; Braithwaite et al. 2015; Leiner et al. 2012).

However, concerning the use of data post-processing techniques, a tradeoff between knowledge gained and technical feasibility needs to be considered. The approach described in this article is meant to provide urban planners with simple means to investigate affects and stress associated with specific elements of urban planning scenarios. Detailed data filtering and analyses, as usually applied in psychophysiological studies, would be difficult to integrate into conventional urban planning processes due to the time and expertise required. Instead, we suggest to further develop the described approach into a real-time evaluation tool. This would allow urban planners to identify potentially conflicting issues of urban planning scenarios already during the stakeholder participation process and to further address these issues during interviews. Additional detailed investigation and evaluation of the identified issues could be carried out in subsequent studies.

Concerning the generalizability of the reported results, it needs to be considered that a small and relatively uniform study sample was used. Approaches to measure implicit responses to VR-based urban models should also consider potential age differences of stakeholders. This includes, for example, technology acceptance or the susceptibility to VR sickness (LaViola 2000; Lee et al. 2019; Saredakis et al. 2020). Therefore, to ensure inclusion of all relevant age groups, stakeholder participation processes using VR methods need to ensure ease of use and need to reduce the risk of VR sickness, e.g., by choosing appropriate VR locomotion methods or by reducing the field of view (Edler et al. 2019a, b; Langbehn et al. 2018; LaViola 2000).

Another limitation concerning the application of the described approach in urban planning scenarios is the limited freedom of movement in our study design. Asking participants to stay seated helped to reduce EDA data artifacts (Tronstad et al. 2022). However, during planning participation processes, stakeholders should be able to explore virtual scenarios as freely as possible. Therefore, other ways to handle EDA data artifacts should be explored. We suggest to either measure EDA data at more than one skin location (Tronstad et al. 2022) or to use accelerometer data, which is also available in the Empatica E4 wristband. This data could be used to identify sudden wrist movements and to either remove the resulting EDA data artifacts or to add a timed note to the corresponding EDA data.

With regard to the choice of assessing EDA at the wrist, a tradeoff between user comfort and data validity must be considered. We chose to use the Empatica E4 wristband, because it is less obtrusive compared to measurements at the palm or fingers. Thus, participants should be less distracted from the VR experience. However, EDA data recorded at the wrist has been argued to be less representative of stimulus related responses, because it is superimposed by thermoregulatory activity (Babaei et al. 2021). Therefore, to ensure construct validity, using EDA devices that record EDA data at the fingers with as little distraction as possible (e.g., wireless recordings) should be considered.

Concerning the use of EDA as a measure for fear, it is important to mention that phasic activation of EDA may be seen as an indirect response to fear stimuli, Fear arousal causes a stress response (Rodrigues et al. 2009). This stress response is in turn reflected in increased phasic activation of the EDA data. However, for urban planning decisions it can be important to differentiate between fear-based stress and other forms of stress, for example caused by continuous noise pollution. This may be difficult in VR environments created for urban planning processes, because, opposed to the stimuli in our experiment, the stimuli in urban planning scenarios are not necessarily chosen to elicit specific emotional responses. Therefore, it may be desirable to simultaneously utilize additional physiological measures as skin temperature or blood pressure to better differentiate between different emotions and sources of stress (cf. Ekman et al. 1983; Sinha et al. 1992).

The described approach is a first step toward providing urban planners with more extensive feedback from stakeholder participation processes. In addition to the assessment of stress and fear induced by traffic situations investigated in our study, the approach could also be applied to other urban planning issues. For example, modern game engines allow to create comparisons of different construction project scenarios and to simulate shadows, daylight and artificial lighting within these scenarios. By visualizing and exploring these urban planning scenarios in VR, stress and fear issues could be identified and resolved. Future research should focus on the implementation of additional implicit measures as pupil dilation, heart rate, blood pressure and skin temperature as measures for stress, cognitive load, fear, anger, disgust or happiness (Chen et al. 2017; Ekman et al. 1983; Hyönä et al. 1995; Reinhardt et al. 2012; Sinha et al. 1992). Integrating additional physiological measures into a real-time indicator for stress and affective responses can be a promising approach for urban planners to assess previously uncovered unconscious responses to urban planning scenarios and, consequentially, to develop scenarios with a higher acceptance of stakeholders.