Understanding visitor flow and behaviour in developing tourism-service-oriented villages by space syntax methodologies: a case study of Tabian Rural Section of Qingshan Village, Hangzhou

ABSTRACT

In the built environment (BE) of tourism-service-oriented villages, understanding visitor flow volumes and behaviours is crucial for guiding space revitalization and tourism management. This research uncovers quantitative relationships between visitor flow volume, spatial visual exposure, and visitors’ behaviour at 12 sites in the Tabian Rural Section of Qingshan Village, a typical tourism service-oriented village in Hangzhou. The methodologies employed in this study are based on two space syntax algorithms: 3-D isovist analysis and agent-based models (ABMs). Visual Exposure Rates (VERs) in 3-D isovists along visitors’ routes were calculated using parametric programming, and visitor flow simulation was conducted using two ABMs. Real-world visitor volumes (RWFVs) and visitors’ behaviors were also recorded during an in-situ survey. A comprehensive correlation analysis revealed multiple associations. The results demonstrated that combined simulations from ABM I, and VERs explained 50.5% of the overall variability in RWFVs. Notably, VERs were found to be associated with visitors’ behaviours of photography and seating, while the outputs from ABM I showed a moderate correlation with RWFVs. This study innovatively combines space syntax models for extended application in the research of rural BEs, providing guidance for optimizing strategies to improve spatial configuration in homogeneous villages.

Graphical Abstract

KEYWORDS:

• Rural tourism
• visitor flow volume
• visitor behaviour
• 3-D isovist analysis
• agent-based model

1. Introduction

1.1. Space syntax methodologies for villages in Southern Yangtze region

In many rural tourism-oriented villages in Southern Yangtze region (“Jiangnan”) of China, built environment (BE) in villages often coexist with landscape elements and tourism facilities, forming multifunctional spaces with complex spatial configuration. It is believed that visitor flows and behaviours are useful for guiding revitalisation and tourism management in these areas (Chen 2022; MacDonald and Jolliffe 2003). Quantitative techniques, e.g., Geodesign tools, space syntax, etc., are useful in guiding spatial reuse, redesign and revitalisation in rural villages.

Space syntax is a set of spatial topological methods for the estimation and description of spatial form-function correspondence in BEs (Hillier et al. 1993; Turner and Penn 2002). Some researchers (Hosseini, Daneshjoo, and Yeganeh 2022; Turner et al. 2001) have proposed mathematical “spatial configuration representation models” within the space syntax system, e.g., axial analysis, isovist analysis, visibility graph analysis (VGA), agent-based models (ABM), etc. Space syntax has been applied in numerous sites in the Southern Yangtze region of China in related fields, such as geography, archaeology, and ancient architecture (Chen 2022; Huang et al. 2023; J. Lee, Ostwald, and Zhou 2023; Xu et al. 2020; Zhai, Baran, and Wu 2018; Zhou et al. 2023).

Rural-tourism-related studies conducted with space syntax methodology are still limited. Although there are few studies (Alfathani and Nurdini 2022; Fujii and Uchikawa 2019; Xu, Rollo, and Esteban 2021) attempted to quantify the relationships between visitors’ spatial perceptions and spatial configuration of rural BEs, only few case studies in Southern Yangtze region of China (Chu and Wu 2022; Zhou et al. 2023; Zhou et al. 2023b) have primarily applied space syntax to guide the spatial analysis or tourism management of rural villages. Almost no quantitative study to date has put forward the feasible methodologies for helping rural planners, architects and tourism managers to evaluate visitors’ behaviour preferences for those villages.

1.2. Review of 3-D isovist analysis and ABM analysis

1.2.1. The 3-D isovist analysis

The “isovist” refers to the optical fields visible from in the space syntax system, representing visibility from a given point of view at the “eye-level” (Batty 2001). In the landscape planning context, the morphology of the “isovist” is known as “viewshed” (Krukar et al. 2021). The embodied isovist analysis has been applied to quantify spatial visibility (Moonkham, Srinurak, and Duff 2023; Morello and Ratti 2009). The “3-D isovist is defined as 3-D visible space fields from spatial vantage points. Visitors” visual awareness and 3-D isovist volumetric metrics are strongly correlated in the form of spaces (Al-Jameel and Al-Moula 2023; Mustafa and Ali 2023). Since the common-used space syntax analysis software “DepthMap” can only solve in 2-D isovist instead of 3-D real-world space, some researchers tried to use 3-D modelling and parametric programmes to analysis spatial visual exposure and openness in 3-D isovists and calculate parameters related to visibility (Khanjari and Shakibamanesh 2023; Krukar et al. 2021; Mustafa and Ali 2023). There is still a lack of work concerning on 3-D isovist analysis in studies for rural village BEs (Hosseini, Daneshjoo, and Yeganeh 2022).

1.2.2. The ABM analysis

The agent-based model (ABM) is a 2-D pedestrian flow simulation model (Turner and Penn 2002) available for simulating the visitors’ preferences in different sites and can predict their spatial movements (Hacini and Bada 2022). The DepthMap software provides the ABM algorithm which allows virtual “agents” to walk in a 2-D knee-level accessible space, based on simulated visual perceptions with a set of parameters, and the simulated traces can simulate the walking patterns of visitors. Some studies (J. Kim and Kim 2023) have confirmed that the spatial activity patterns of “agents” in the model are generally consistent with those observed in real-world scenarios. Esposito et al. (2020) suggests that it is reasonable to use the ABM-based visitor flow simulation to support urban planning decisions. However, it remains unknown whether ABMs with specific parameter settings would be able to explain visitors’ tour route patterns and visitor flow variations at specific sites in rural villages.

1.3. Visitor flow and behaviour in tourism-service-oriented villages

Within the built environment (BE) of tourism-service-oriented villages, visitors’ movement patterns and behavioral preferences are significantly influenced by a multitude of factors, include spatial attributes such as “routes characterized by minimum Euclidean distances” and “trajectories involving the fewest directional changes,” the functionality of different sites, and the presence of cultural events (Shatu, Yigitcanlar, and Bunker 2019). The synergistic application of visitor flow tracking methodologies combined with space syntax algorithms is considerable potential in quantifying visitor behaviours in a rigorous manner (Cheliotis 2020; Zhou et al., 2023a).

Environmental psychology studies affirm that the substantial impact of media representation on destination image perception and its influence on prospective tourists’ travel inclinations. Both favourable and unfavourable portrayals correlate with the propensity to visit. The arrangement of attractions and routes, physical and visual connectivity, accessibility, and the location of tourism activities can influence tourists’ spatial behaviours (Jamhawi, Zidan, and Sherzad 2023). Yang et al. (2020) present a holistic framework to explicate the multifaceted relationship variables affecting tourist behaviors. Further investigation underscores direct associations among destination image, destination persona, self-congruence, and the likelihood of repeat visits, and further reveals differential effects of destination personality on self-congruence for Chinese tourists which shapes their ideal self-concept and revisits intentions.

Therefore, the integrated use of spatial configuration and human dynamics plays a critical role in understanding the complex mechanisms affecting tourists’ path-finding and other behaviours within rural villages. By integrating empirical measurements of visitor flows with computational methods based on space syntax, researchers can derive nuanced insights into the quantitative relationships between environmental features and visitors’ behaviours. This contributes to evidence-based planning and management strategies for tourism-service-oriented rural settlements.

1.4. Research objectives and questions

It is noticed that the existing literature has not integrated space syntax methodologies to reveal visitor flows and behaviours in tourist-service-oriented villages. We believe that the combination of in-situ visitor flow observation and two space syntax algorithms, i.e., the ABM and the isovist analysis, can be of help to in-depth comprehending the relationship between visitor flows, behaviours and perceptions in these villages. Correlation between factors, i.e., real-world flow volumes (RWFVs), visitor behaviours, results of ABM-based flow simulation and visual exposure analysis in 3-D isovists, should be further studied.

It is noteworthy that an integrated methodology combining in-situ surveys and space syntax methodologies has been applied to understand the relationship between visitors’ behaviors and spatial characteristics of scenic areas, such as visitors’ tour routes in Chinese traditional gardens (T. Zhang, Lian, and Xu 2020) and ancient temple heritage spaces (Wu et al. 2024). These studies indicated that space syntax methodologies can effectively explain visitors’ walking and path-finding behaviours, while tourists’ stay duration can be predicted by visual exposures in the isovist.

However, it is found that homogeneous empirical studies have not been conducted in rural BEs, especially for those spaces in tourism-service-oriented villages. Therefore, understanding the relationships between environmental behaviors and spatial configuration will enhance the accuracy and feasibility of rural landscape and tourism studies. Thus, the goal of this research is to improve the understanding of environmental behaviors and spatial configuration, thereby enhancing rural landscape and tourism studies and guiding the design and management of rural tourism.

2. Materials and methods

2.1. The study area

Qingshan Village (119° 535’E, 25° 903’ N) is an eco-tourism-oriented village locates in Huanghu Town, Yuhang District, Hangzhou City. Since 2015, the local community has promoted the revitalisation of rural eco-tourism, and promoted more than 20 tourism projects, e.g., rural visitors’ libraries, eco-cultural centres and nature education camp, etc. These projects have significantly improved the rural landscape and attracted more than 30,000 tourists per year.

This study focuses on the Tabian Rural Section of Qingshan Village, situated approximately 1 km away from the Dongwu Rural Section, the central tourism hub of Qingshan Village (as depicted in Figure 1). The Tabian Rural Section is a nascent, tourism-service-oriented village that experiences negligible tourist influx during non-festival seasons. In recent times, concurrent with the burgeoning development of the Dongwu Rural Section, there has been a noticeable increase in visitor traffic flowing into the Tabian Rural Section. Consequently, it becomes imperative to investigate strategies for reviving and invigorating this emerging tourism-service-oriented village. Such efforts can be of help to offer suggestions for enhancing tourism amenities to the Dongwu Rural Section amidst a growing tourist population in the future.

Figure 1. The location of Qingshan Village.

During the course of our on-site survey, an appealing cultural and tourism fair took place in the Dongwu Rural Section, providing us with an opportune scenario to observe and analyse the actual flow volumes and visitor behaviours within the study area under conditions of heightened tourist activity. This event served as a practical and realistic backdrop against which to gauge visitor dynamics in the Tabian Rural Section.

2.2. Research process

The key objectives of this research are to reveal the quantitative relationships among visitor flow volume, visual exposure of the space, and visitor behaviour by digital landscape tools and programs with space-syntax-based methodologies. Two space syntax tools, i.e., 3-D isovist and ABM analysis, are adopted to calculate the correlation among 3-D visual exposure of the space, the simulated visitor flow, and the real-word visitor flow volume and visitor’ behaviours in a typical village. The framework of the research process is presented in Figure 2.

Figure 2. The framework of the research process.

The experiment was approved by the researcher’s institutional ethics committee. The GIS mapping geodata of the study area (in “.dxf” format, geodetic datum: WGS84, sampling time: 2023) is provided by the Geographic Information Public Service Platform of Zhejiang Province, P. R. China (https://zhejiang.tianditu.gov.cn/, access in July 2023). Necessary amendments were conducted based on in-situ survey. A landscape information model (LIM) of the study area is created by Rhinoceros 7 (a 3-D parametric modelling software developed by Robert McNeel & Associates). The parametric programme is developed by Grasshopper v1.0 (the visual programming platform of Rhinoceros 7) and UrbanXTool (a Grasshopper plug-in). The programme will be applied to 3-D visual exposure calculation. Meanwhile, the 2-D detailed plan map is drawn within Rhinoceros software and be input into DepthMap X (a free software developed by University College London, UK) to generate the simulated visitor flow distribution by two models set with different parameters.

Meanwhile, in-situ survey was conducted at 12 sites in Tabian Rural Section of Qingshan Village during 14:00–15:00 on 15 July 2023. The most sites are mainly the intersections of roads and tourist service facilities (Table 1). On that day, the Dongwu Rural Section hosted the official opening of the bazaar, which attracted a large number of tourists to participate in tourism activities. Thus, it provided us a unique and ideal opportunity for us to observe the visitor flows and behaviours in the study area.

Table 1. Photographs and detailed descriptions of 12 sites in tabian rural section of Qingshan Village.

Site No.	Photograph	Landscape elements	Functionalities
1		A small shop near the southern road.	Sale daily necessities and food.
2		Wasteland surrounded by bamboo.	Its surrounding area is used as a small temporary parking lot.
3		Surrounded by residences and courtyards, with a small transportation centre for tourism supplies nearby.	A small crossroad.
4		Surrounded by huts and courtyards.	A small crossroad.
5		Surrounded by huts.	A three-way traffic intersection.
6		Service centre.	A health service stations and a visitors’ centre.
7		A toilet.	A toilet near small crossroad.
8		A tourist bus stop surrounded by residences, with a signpost.	A main three-way traffic intersection that is used frequently by tourists.
9		A hotel surrounded by residences and small lawns.	Provide accommodation and catering services for visitors to the village.
10		A small pond surrounded by a low cobble stone wall and small lawns.	Some villagers often do laundry and wash vegetables. Some visitors also enjoy the waterscape here.
11		A barren park and some sport equipment.	Almost been abandoned.
12		Near a large bamboo grove.	Almost been abandoned.

The range of each site is determined as a circular area of 7.5 m radius, which enables researchers to observe the number and behaviour of visitors. The centres and outlines of these sites range are tagged in Figure 3.

Figure 3. The distribution of 12 sites and 7 observer volunteers.

Seven observer volunteers were asked to stay at the fixed standing position, respectively. Each volunteer was responsible for observing one or two site(s). Observer volunteers counted the visitor flow volumes and recorded behaviour of those visitors who have been entered the range (the fixed circular area) of each the site anonymously. Observers who were assigned to observe two sites used binoculars and kept switching view directions per 30 s, taking turns observing at both sites. Multiple flow volumes will be counted repeatedly if the same visitor enters the same range of site more than once during the survey. Visitor behaviour is classified into seven types, i.e., “chatting, quickly passing, eating, shopping, photographing, splashing, sitting”. Finally, correlation analysis between factors is conducted, which will be introduced later.

2.3. Digital analytic techniques

2.3.1. Visual exposure calculation by 3-D isovist analysis

The “isovist” is a spatial measure for illustrating visibility and visual characteristics from viewpoints (vantage points) in BE, quantitively presenting the amount of the visible boundary (Benedikt 1979). Some studies indicate that the distribution of the isovists is closely related to visitor behaviours when they perceive visual features in both natural and built environments (Batty 2001; Turner et al. 2001; Wiener and Franz 2005). The isovist analysis also believed to be of potential in studies on landscape architecture, in which the radius of the viewshed is set as “15 m” (G. Kim, Kim, and Kim 2019; Weitkamp 2011).

For those spaces with more complex spatial morphology, the 2-D isovists cannot reflect the condition of visitors’ visions when they walk along the tour routes; therefore, some geographers further proposed the “3-D isovist” (Hosseini, Daneshjoo, and Yeganeh 2022; Morello and Ratti 2009), extending the concept of the 2-D isovist to 3-D spaces. The 3-D isovists of the space are often calculated from multiple viewpoints along the tour route, whereas the distance between viewpoints is commonly set as 5 m for public spaces (G. Kim, Kim, and Kim 2019; Krukar et al. 2021). In this research, the 3-D visual exposure analysis was applied to quantitatively reflect the possibility that visitors would see 3-D objects while visiting the village along accessible tour routes.

The Grasshopper and the UrbanXTool plug-in are utilised to develop the parametric programme. The UrbanXtool plug-in provides 3-D isovist computing functions (Figure 4) according to the mathematical theories presented in the existing research (G. Kim, Kim, and Kim 2019). In the program, input parameters include visitors’ viewpoints and the 3-D mesh of the eye-level geo-feature elements. To meet the requirements of data format, the geometry objects were transformed into 3-D meshes with grid size of 0.2 m *0.2 m. The program will automatically generate continuous viewpoints at each 5 m intervals along the tour routes, with the height of the viewpoints at eye level of a typical person (1.5 m). The 3-D isovists from viewpoints are calculated, whereas the radius of the viewshed is set to 15 m. The isovist from each viewpoint are superimposed to obtain the distribution of 3-D visual exposure rates (VERs), which are visualised in the Rhino software in symbolised pseudo-colours, allowing us to observe how the values for 3-D visual exposure rates distributed throughout the space (Figure 5).

Figure 4. Visual programming for calculating visual exposure in 3-D isovist with Grasshopper.

Figure 5. Computational principles for the distribution of visual exposure and the VERs.

2.3.2. Visitor flow simulation by the ABM analysis

The agent-based model (ABM) is developed for simulating 2-D pedestrian movements using virtual “agents” with defined fields of view (Turner and Penn 2002). The ABM algorithm follows movement rules based on the spatial configuration, rather than relying on learned paths or destinations (Hillier et al. 1993). There are certain correlations between the ABM simulation and pedestrian patterns (Tang and Hu 2017; Turner and Penn 2002). In the ABM algorithm for “standard motion” mode, a simulation method that guides agents to walk referring to the geometric scale of the 2-D isovist, is applied in this research. To verify the effect of different parameter settings on results of visitor flow simulation, two ABM models, i.e., ABM I and ABM II, set with different parameters are utilised for comparison (Table 2). These parameters are listed as follows.

• Grid: According to the scale of the study area, the grid size of both ABMs is set to 1.0 m *1.0 m.

• “The Field of View”: determine the range of horizontal viewing angles within the view fields of virtual “agents”. “The Field of View” parameter of both ABMs is set to 170°, for most studies (Turner and Penn 2002; J. Kim and Kim 2023) verifies that this setting is close to the horizontal viewshed of visitors in natural state of motion.

• “The Steps Before Turn Decision”: the number of steps agents takes before they randomly change direction. The ABM II applies the default value (“3 steps”). However, to date, it has not been proven whether this default value would be validated in the condition of BE in tourism-service-oriented villages. A study on public spaces (J. Kim and Kim 2023) indicates that this parameter should be set as “12 steps” to achieve better simulation effect, and the ABM I adopt this parameter.

• “Timesteps in System”: the number of timesteps that “agents” move before ending the simulation. For the ABM I, this parameter is set to “1,000”, the same value used in most architectural and geographical studies. This parameter adopted for the ABM II is set to “200”, referring to a study of human behaviour in typical urban BE (Sutkaitytė 2020).

Table 2. The parameter settings of two ABMs.

Item	ABM I	ABM II
The Field of View Angle	170° (15 degrees)
Steps Before Turn Decision	12	3
Timesteps in System	1,000	200

2.4. Pre-processing of geodata

During the survey, the geo-feature elements and the roads and paths accessible to visitors in the study area were recorded. The topography of the study area is very flat, with almost no differences in altitude. Therefore, it is not necessary to consider the effect of the terrain on the visitor’s viewsheds at eye level and the accessibility conditions at knee level. Furthermore, there is no need to consider the effect of vegetation at both levels, as there are no tall arbours or shrubs (height ≥1.5 m) to obstruct the visitor’s view and prevent them from accessing the space.

A 2-D knee-level detailed plan drawing and a 3-D eye-level Landscape Information Model (LIM) are created according to the principles (Zhou et al. 2023) for drawing spatial elements in space syntax analysis (Table 3). The 2D drawing (Figure 6) is used to simulate visitor flow by ABMs, while the 3-D LIM (Figure 7) is modelled appropriately to meet the requirements of calculating visual exposure in 3-D isovists. Thus, geo-features which affect visitor flow are drawn in knee-level 2D drawing, while those that affect visitors’ viewshed are modelled in eye-level 3-D LIM. The ABM model does not require the manual input of fixed viewpoints, whereas the VER calculation requires the input of viewpoints.

Figure 6. Geo-feature elements at knee-level in the 2-D detailed plan drawing.

Figure 7. Geo-feature elements and generated viewpoints at eye-level in 3-D landscape information model. (a) Top view; (b) Isometric south-eastern view.

Table 3. Principles for drawing spatial elements.

Geo-feature elements	Hight above ground level	In the 2-D detailed plan drawing at the knee-level (Yes/No)	In the 3-D LIM at the eye-level (Yes/No)
Architecture	≥1.5m	Y, presented as closed polygons	Y, presented as 3-D boxes with real-world length, width, and height
Bamboo grove	≥1.5m		Y, presented as 3-D prisms
Wall	≥1.5m
Gate	≥1.5m
Fence/Road-block	<1.5m		N
Lawn/Shrubs	<1.5m
Water body	0m
Field	<1.5m
Inaccessible barren land	0m
Viewpoint along visitors’ tour route	1.5m	N	Y, presented as a group of viewpoints at 1.5m above ground

3. Results

3.1. Results of the visual exposure calculation

The distribution of the global visual exposure is calculated and visualised by Grasshopper and Rhinoceros software as described above (Figure 8). It is noticed that the spaces with higher VERs often locates at the some frequent-used crossroads and relatively open spaces. The VERs at the places with higher densities of residences were found lower, due to the fact that walls surround most residences obstructed the visitors’ viewsheds. The mean VERs within the range of each site (defined in Section 2.2) are calculated, respectively (Table 4). The highest visual exposure effects occur at the surroundings of the service centre near the Site No. 6 (777.54), followed by Site No. 12 (709.68), No.7 (686.04) and No.5 (704.58).

Figure 8. Distribution of the 3-D visual exposure in the study area at top view (the greenish hue represents lower amount of flow volume, and vice versa for reddish hues).

Table 4. The mean visual exposure rates within the range of each site.

Site No.	1	2	3	4
3-D visual exposure analysis
Mean VERs	621.22	677.51	634.23	688.41
Site No.	5	6	7	8
3-D visual exposure analysis
Mean VERs	674.58	777.54	696.04	646.98
Site No.	9	10	11	12
3-D visual exposure analysis
Mean VERs	654.03	694.36	618.33	709.68

3.2. Results of visitor flow simulation

The results of visitor flow simulation are visualised in the DepthMap software in the symbolised pseudo-colours of “equal range” (Figures 9, 10). The mean flow volume of each grid within each range of sites in the simulation results of ABM I and ABM II was calculated, respectively, excluding the null-value grids in the range of each site (Table 5). The efficacy and feasibility of two types of ABMs will be discussed later.

Figure 9. Results of visitor flow simulation by ABM No. 1 (the bluish hue represents lower amount of flow volume, and vice versa for reddish hues).

Figure 10. Results of visitor flow simulation by ABM II (the bluish hue represents lower amount of flow volume, and vice versa for reddish hues).

Table 5. The flow volume of each grid within each range of sites calculated by ABMs.

Site No.	Simulated Visitor flow volumes
	ABM I	ABM II
1	87 ± 11.3	68 ± 9.8
2	61 ± 10.9	51 ± 10.1
3	34 ± 5.3	18 ± 3.5
4	31 ± 5.5	29 ± 3.7
5	109 ± 21.1	80 ± 20.8
6	91 ± 12.3	57 ± 9.2
7	86 ± 8.9	61 ± 6.6
8	96 ± 15.3	51 ± 12.4
9	115 ± 16.7	70 ± 10.1
10	86 ± 8.5	66 ± 6.9
11	128 ± 21.6	117 ± 15.2
12	56 ± 8.8	51 ± 6.4

3.3. Real-world visitor flow volume and behaviour

During the in-situ survey, in accordance with the method in Section 2.2, the real-world flow volume (RWFV) and victors’ behaviours at each site are observed, as listed in Table 6. To provide a clearer visualization of these observations, Figure 11 depicts the percentage of visitors engaging in each behaviour type relative to the total number of visitors per site.

Figure 11. Percentage of the number of visitors who perform each type of behaviour out of the total number of visitors.

Table 6. The in-situ visitor flow volumes and data of visitor behaviours (unit: number of visits).

Site No.	RWFV	Chatting	Quickly passing	Eating	Shopping	Photographing	Splashing	Sitting
1	44	3	25	5	6			5
2	12		11			2		1
3	20	4	14					2
4	17	2	15
5	82	3	79
6	87	4	65			12		6
7	45		44			1
8	75	2	73			2
9	43	5	28	10
10	39	8	19				7	5
11	39	4	35
12	18		15			3

The majority of locations showed that over half of the visitors transited through the observational areas swiftly. Conversely, at those sites that were nearly abandoned (e.g., Sites 3, 9, 11, and 12), there was a significant lack of engagement in any recreational activities. Notably, at Sites 4 and 5, the majority of passersby originated from the Dongwu Rural Section of Qingshan Village, where their predominant tendency was to traverse the Tabian Rural section quickly. However, relatively diverse behaviours were observed at Sites 1 and 10. At Site 10, some visitors engaged in leisurely activities such as sitting and splashing near a small pond, while at Site 1, visitors’ behaviours include eating and socializing. Additionally, at Site 9, i.e., a hotel offering accommodation and dining services, visitors were also observed partaking in taking meal and talking. This comprehensive data collection thus reveals distinctive patterns in visitor behaviour across various sites within the study area, providing insights into the functional utilization and potential improvements for each site based on visitor preferences and activities.

3.4. Correlation analysis

To reveal the internal correlations among VERs, the simulated visitor flow volumes by ABM I/II, and real-world flow volumes (RWFV), the Pearson correlation test was proceeded according to the formula (1) by a programme developed within Python 3.0 with the Matplotlib and NumPy packages. The results are listed in Table 7.

Table 7. Results of the primary-turn Pearson correlation test.

	RWFV	ABM I	ABM II	VER
RWFV	1
ABM I	0.628**	1
ABM II	0.339*	0.886**	1
VER	0.305*	−0.045	−0.066	1

***r>0.80 - High linear correlation.

**0.50≤r<0.80 - Moderate linear correlation.

*0.30≤r<0.50 - Low linear correlation.

(1) $\mathit{R}={\displaystyle {\sum }_{i=1}^n{\left(x-\overline{x}\right)}^2}\mathit{y}-\mathit{y}2{\displaystyle {\sum }_{i=1}^n{\left(x-\overline{x}\right)}^2}\mathit{y}-\mathit{y}2$(1)

where $\mathit{R}$ is the Pearson correlation coefficients, $\mathit{n}$ is the total amount of samples, $\mathit{x}$ is each sample, and $\bar{x}$ is the average value of samples.’

Moderate linear correlations between “RWFV – ABM I” (R = 0.628), “ABM I – ABM II” (R = 0.886) are found, whereas low linear correlations between “RWFV – ABM II” (R = 0.339), and “RWFV – VER” (R = 0.305) are also noticed.

Then, a statistical programme was developed using Python with the Matplotlib and Scikit-learn packages to perform the linear regression analysis among three variables at all sites. The results are shown in Table 8. Equation of the regression model is calculated as the formula (2).

(2) $\mathrm{RWFV}=-99.204+\mathit{\hspace{0.17em}}0.542\ast \mathrm{ABM}\mathrm{I}+\mathit{\hspace{0.17em}}0.148\ast \mathrm{VER}$(2)

“ABM I” and “VER” can explain 50.5% of the overall variation in “RWFV” (R² = 0.505). The model satisfies the F-test (F = 4.599, p = 0.042 < 0.05). The regression coefficient of “ABM I” is 0.542 (t = 2.739, p = 0.023 < 0.05), which indicates that simulation of the ABM I has significant positive relationship with “RWFV”, while “VER” has relatively inconspicuous influence on “RWFV” (t = 1.424, p = 0.188 > 0.05).

Table 8. Statistical coefficients of the linear regression model.

Item	Unstandardized Coefficients		Standardized Coefficients	t	p	Collinearity diagnosis
	B	Std. Error	Beta			VIF	Tolerance
Constant	−99.204	71.898	-	−1.380	0.201	-	-
‘ABM I’	0.542	0.198	0.643	2.739	0.023*	1.002	0.998
‘VER’	0.148	0.104	0.334	1.424	0.188	1.002	0.998
R^2	0.505
Adj R^2	0.396
F	F (2,9) = 4.599, p = 0.042
D-W	1.269

n = 12; Dependent Variable: Real-world flow volume (RWFV).

*p<0.05 **p<0.01.

To calculate the correlation between RWFV, ABM I, AMB II, VER and the percentage of the number of visitors who perform each type of behaviour, the Pearson correlation test is conducted again, as listed in Table 9. We found low linear correlations between “Eating – ABM I” (R = 0.345) and “Sitting – ABM I” (R=-0.304), as well as a moderate linear correlation between “Photographing – VER” (R = 0.515). No valid correlation is found among RWFV and all types of visitor behaviours.

Table 9. Results of the secondary-turn Pearson correlation test.

Type of visitor behaviour	RWFV	ABM1	ABM2	VER
Chatting	−0.237	−0.155	−0.142	−0.092
Quickly passing	0.133	0.027	0.059	−0.144
Eating	−0.001	0.345*	0.165	−0.128
Shopping	0.007	0.055	0.102	−0.175
Photographing	−0.131	−0.254	−0.188	0.515**
Splashing	−0.054	0.045	0.077	0.167
Sitting	−0.164	−0.304*	−0.237	0.099

**0.50 ≤ R < 0.80 - Moderate linear correlation.

*0.30 ≤ R < 0.50 - Low linear correlation.

4. Discussion

4.1. Relationships between VERs and RWFVs

Low linear correlation between the visual exposure rates (VERs) and real-world visitor flow volumes (RWVFs) is found. Visitors’ photographing behaviour is moderate linearly correlated with VERs. Relevant phenomena are listed and discussed as follows:

1. Visitor behaviour is found purposeful and not entirely influenced by visibility. It is noted that visitors were not willing to walk through alleys with lower VERs. Some of the road crossings have a high degree of visual obstruction (e.g., site no. 3, 8), resulting in relatively low VERs, which makes these spaces less visible. Some of the almost abandoned spaces (e.g., sites 11, 12) obtained higher VERs, however, RWFVs are much lower, because they were not surrounded by sufficient attractive elements. Although relatively higher VERs and certain landscape effects were observed at sites 2 and 12, visitor behaviour there is rather limited. Most of the visitors passed by quickly, while some of them took photographs. This can be explained by the fact that these sites seem almost abandoned, lacking the necessary facilities and landscape elements. These phenomena can be explained as follow. Tourists usually prefer to walk in public open traffic spaces in places with complex spatial arrangements, with the amount of traffic being the main criterion influencing the pleasantness of BE landscapes, which corresponds to several existed literature (Wright and Curtis 2002; Foster and Giles-Corti 2008). Thus, visitors’ continuous movement along tour routes transforms their walking experience into a continuous sensory stimulation and aesthetic cognitive activity (Taylor 2003; Ahmed and Mushref 2021). On the contrary, the willingness of the visitors to stay in the non-public places with lower VERs was relatively low.

2. The visitor flow in the study area is not entirely influenced by visual exposure factors. It appears that most visitors entering the study area were purpose-driven, either en route to or returning from the Dongwu Rural Section of Qingshan Village. They tended to quickly pass by, engage in activities such as having meals or snacks, and shopping. The majority did not seem inclined to spend more time in the Tabian Rural Section of Qingshan Village. This can be explained as follows: The characteristics of each geo-feature are generally thought to correlate with the frequency and concentration of visitors, and the adjacent view is one of the main factors affecting visual quality (Hillier and Hanson 1984). The short duration of visual exposure of visitors can also significantly impact their aesthetic response, as indicated by several studies (Lyubomirsky 2011; Wilson and Dyke 2016; Mullin et al. 2017). Numerous studies (Weinstoerffer and Girardin 2000; De la Fuentet et al., 2006; Germino et al. 2001) confirmed that farmland with deeper horizons is often perceived as influenced by neighboring views with high VERs. Manning and Freimund (2004) also found that spaces with higher VERs can increase the stay duration of visitors, and higher RWFVs can reduce visitors’ sensitivity to visual impacts.

3. Overall proportion of “photographing” behaviour is in moderate linear correlation with the VERs (R = 0.515). This also confirms that the majority of visitors prefer taking photographs in those areas with higher VERs. This phenomenon can be explained as follows. When visitors are in these areas with higher VERs, the visual exposure of the landscape elements in their view fields is more concentrated, enabling them to find attractive views. Some spatial elements, e.g., waterscapes, field and architecture, can encourage visitors to stop and take photographs, which is consistent with the findings of Gao et al. (2023).

4.2. Validity of the ABM for simulating visitor flow

Although the high linear correlation between the simulation results of ABM I and ABM II (R = 0.886) is found, the visitor flow simulation conducted by the ABM I performed much better than which conducted by the ABM II. The simulation results of ABM I have a high correlation with the RWFV (R = 0.628), while the simulation results of ABM II have a much lower correlation with RWFV (R = 0.339). These phenomena can be explained as follows.

1. Visitors to the Tabian Rural Section of Qingshan Village walked along the long and narrow south-north main road connected with some inconspicuous alleys. The landscape elements along these alleys were nonattractive for tourists. Therefore, setting the “Steps Before Turn Decision” parameter to “12 steps” can more accurately reflect people’s turning decisions in the village spaces, and the simulation results are relatively closer to the real-world visitor flow volumes. In addition, compared to typical urban BEs, visitors often tend to be forced to turn back and try to proceed path-finding behaviours while moving through the village spaces. Therefore, if the “Time-steps in System” parameter is set to “1,000 steps”, the “agents” will move for a longer period of time during the simulation. Such a setting is indicated to be more appropriate for imitating how visitors walk.

2. Sites with no specific function (e.g., Site Nos. 2, 3, 8, 11, 12) or pretty limited functions (e.g., Site Nos. 4, 9) resulted in a propensity for visitors to quickly pass by these spaces. However, the impact of site functionality on visitors is not reflected in the ABM. In fact, the validity of the ABM I for simulating visitor flow condition is not extremely high (R = 0.628), indicating that differences in functionality among sites present non-negligible effects on the RWFV.

3. It is observed that some visitors frequently made trade-offs between two criteria when planning their itineraries, i.e., “the shortest route” (geographically) and “the route with the least directional changes” (topologically). Most visitors tended to minimise both criteria, but they usually preferred the route with the “least directional change”, which has been further found on the basis of previous studies (Dalton 2003; Shatu, Yigitcanlar, and Bunker 2019).

4.3. Relationships among visitor flow, behaviour and spatial pattern

Several factors, including spatial morphology, walking speed and flow volume–capacity relationships, are often thought to correlate with visitors’ flow patterns (Boukelouha and Gauthier 2020). Visitor behaviour and distribution of RWFVs are influenced by spatial patterns to a certain extent as follows.

1. We found that the tour route network was distributed in a “fishbone-like spatial pattern”, with a variety of side roads intersecting with main roads, many of which are relatively tortuous and lack sufficient connectivity to facilitate a circular flow through the space. Similar spatial pattern is often formed by almost spontaneous succession of the rural landscape. In spaces with such complex pattern and intricate spatial configurations, isovist analysis is appropriate for the explanation of spatial experience and behaviour (Wiener and Franz 2005). Additionally, homogeneous studies on traditional Huizhou Villages in China have revealed that the spatial typology and layout attributes of streets significantly influence regional social interaction preferences within rural villages. Social activities in the rural BEs are predominantly concentrated on branch streets, with pedestrian traffic accounting for more than half of the visitor flow in all street networks (Ding, Gao, and Ma 2022), which is explicitly higher than the findings in this research. This discrepancy can be explained by the fact that streets with lower accessibility and lower VERs have a greater impact on pedestrian traffic in the Tabian Rural Section.

2. The ABM I with a higher value of “Steps Before Turn Decision” exhibited some correlations with static behaviors of visitors, such as eating (R = 0.345) and sitting (R = −0.304). This suggests that the less accessible and connected the internal space configuration, the more steps are required before visitors can decide their walking behaviors, and the less likely they are to find facilities or appropriate spaces to rest. Similar explanations for visitors’ walking and pathfinding behaviors in tourism attraction spaces are also indicated in the space syntax analysis of large-scale ancient temple spaces (Zhou et al. 2023a). Research findings on subway station spaces indicate that while ABMs have acknowledged limitations in fully explaining pedestrian movement patterns, they do capture the tendency for individuals to adopt routes that ostensibly minimize travel distance. However, the present study, as reported by Kim and Kim (2023), reveals that tourists navigating rural spaces do not necessarily adhere to the shortest path principle when exploring such environments. In the context of rural landscapes, our recent investigation found a different phenomenon: visitors do not consistently select the most direct path during their explorations in such settings.

3. It is noteworthy that there are many disconnected roads and paths in the study area, and the information presented on signposts was over-simplified and confusing, making it difficult for visitors to process their path-finding behaviours. This is largely consistent with empirical studies (Boukelouha and Gauthier 2020; J. Lee, Ostwald, and Zhou 2023b), which also found a strong correlation between the spatial characteristics of the pedestrian environment along the tour routes and the morphology and connectivity of the pedestrian network. These findings correspond with Hillier and Hanson (1984) that circular paths through open spaces at the periphery of a building can create circuit tour routes, which is effective in reducing the difficulty of visitor pathfinding.

4. Bamboo groves at the south-west and north-east of the study area presented strong visual impacts, significantly obstructed visitors’ views and preventing them from passing through. There were also some fields on the outskirts of the village, whereas there was no path that was available for walking through and lacked landscape character surrounding those semi-natural fields. Zhai et al. (2018) suggests that the variations in the level of use of visitors’ tour routes in open spaces in scenic areas are often correlated with spatial patterns, and this seems to be generally in line with our findings. It is believed that there is a positive correlation between larger visible area and visitors’ visual perception of openness (Dosen and Ostwald 2016; Zhou et al., 2023a), assuming that people intuitively search for larger areas in which they are less likely to collide with objects or other evacuees.

4.4. Optimising strategies for improving spatial configuration

It is widely believed that visitor behaviour is often related to their environmental perception in typical tourism-oriented villages (Andrades, Dimanche, and Ilkevich 2015). According to these findings on visitor flows and behaviours, we propose some optimization strategies to improve the spatial configuration of the Tabian Rural Section.

Firstly, different spatial scales and forms can influence tourists’ tour paths. It is observed that the unreasonable distribution of commercial sites in tourism villages often leads to an uneven distribution of village vitality (Gao, Li, and Sun 2023). Therefore, commercial shops and landscape facilities are more likely to cause tourists to exhibit a variety of behaviors. The functional zones should be renovated by integrating obsolete and abandoned facilities to meet the needs and preferences of residents and tourists. Landscaping elements, such as public squares and pavilions, will be provided to meet the needs for gathering and resting. Additional public toilets will be provided, along with improved sanitation facilities and shops.

Secondly, to improve accessibility within the villages, it is essential to avoid the inefficient wandering of visitors. The road network should be re-planned to form a complete road circulation. More signposts should be installed, and the guidance system will be improved at nodes such as three-way intersections with high RWFVs. Spatial optimization design strategies can enhance the spatial layout and accessibility of tourist villages, creating more integrated and multifunctional spaces that cater to diverse needs and activities (Wang, Zhu, and Che 2022).

4.5. Limitations and suggestions

In research utilizing space syntax methodologies, trade-offs between data and algorithmic aspects often exist. The methodologies should be based on geodata that capture essential aspects of space configuration, which can influence visitors’ perceptions and behaviors. Meanwhile, the space syntax model should be simple enough to be created and adapted in research and applied design (Hosseini, Daneshjoo, and Yeganeh 2022; Wu et al. 2024). Digital tools in this research are based on isovist analysis and Agent-Based Model (ABM) analysis within the space syntax framework, with limitations discussed as follows.

1. Although there are various algorithms applied for the 3-D isovist analysis, we only apply the “visual exposure” algorithms, which cannot thoroughly explain the effect of small-scale landscape elements on visitors’ spatial visibility and perception.

2. We performed visitor flow simulation using two ABMs with the DepthMap software. However, we identified some points for improvement. Firstly, both ABMs with different parameters incompletely account for spatial functionalities’ effects on visitors’ perceptions. Secondly, geodesign techniques and workflows for landscape architects and geographers may further integrate multiple algorithms (T. Li et al. 2022). We suggest that interconnecting BIM, AI, eye-tracking and other emerging technologies could facilitate the creation of more precise parameters (Penn & Turner, 2002; F. Zhang et al. 2018).

3. The in-situ survey was conducted on a festival day in the neighboring village of the study area, namely, the Dongwu Rural Section of Qingshan Village. Therefore, the results obtained are relatively specific and may not be representative of those obtained under regular conditions in common village spaces. Further empirical studies in other villages are required to verify the applicability of the methodologies and tools used in this research.

5. Conclusion

The case study revealed quantitative relationships between visitor flow and behavior at 12 sites in a tourism-service-oriented village. Space syntax methodologies, including 3-D isovist analysis and two Agent-Based Models (ABMs) set with different parameters, were applied to calculate Visual Exposure Rates (VERs) along tour routes and simulate visitor flows. Real-world visitor volumes (RWFVs) and visitor behavior were recorded at 12 sites during an in-situ survey. Several phenomena were revealed through correlation analysis and linear regression models. The main findings are summarized as follows.

1. VERs had a low linear correlation with RWFVs. Visitor behaviour was purposeful and not entirely influenced by visual exposure factors. There is a positive relationship between higher VERs and specific visitor behaviour, i.e., photographing and sitting.

2. It shows that RWFV, visitor behaviour and VER are correlated at some sites and that ABM I (The Field of View Angle = 15 degrees, Steps Before Turn Decision = 12, Timesteps in System = 1000) has a moderate degree of validity and relatively high feasibility for simulating visitor flows in the study area. The simulation results of ABM I and VERs are shown to explain about half of the total variation in RWFVs. However, the non-negligible effect of the functional differences of the different attractions on visitor flows is not taken into account by the ABMs.

3. The route network of the village is distributed in a “fishbone-like spatial pattern” tend to form almost spontaneously in the rural BEs. Main roads are intersected by a variety of lanes, many of which are relatively curvilinear and do not provide sufficient connectivity to promote the spatial flow circulation.

Empirical findings from this research contribute to an enhanced understanding of environmental behaviors and spatial configuration, offering valuable insights for the planning, design, and management of tourism-service-oriented villages.

Ethics statement

This study was approved by the Committee of the authors’ institution.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

All relevant data are included in the paper.

Additional information

Funding

The research was supported by the Scientific Research Foundation for the High-level Talent Introduction, Zhejiang University of Technology (Project #118001929); National Natural Science Foundation of China (Project #51208467).

Notes on contributors

Shuqin Wang

Shuqin Wang She is a postgraduate student of design at Zhejiang University of Technology.

Yan Huang

Yan Huang He is the Ph.D. from Nanjing Forestry University, the post-doctoral fellow at Zhejiang University, P.R. China, and the visiting scholar at the University of Sheffield, UK. He is currently the head teacher of the Department of Environmental Design, School of Design and Architecture, Zhejiang University of Technology, P. R. China. His main research interests are rural landscape, rural development, and human geography.

Tianjie Li

Tianjie Li He is a Ph. D. candidate of design from Zhejiang Sci-Tech University. His main research interests are rural landscape, GIS, geoecology, digital design techniques and heritage geography.