The Characteristics of Visitor Behavior and Driving Factors in Urban Mountain Parks: A Case Study of Fuzhou, China

Abstract

Previous studies have focused on the linear relationship between recreation behavior and environmental variables. However, to inform the planning and design of recreational spaces, it is essential to understand the factors that contribute to differences in the spatial distribution of recreation behavior. This study investigates the characteristics of visitor behavior in urban mountain parks in Fuzhou City, Fujian Province, China. It describes the distribution of tourist numbers and the diversity of behaviors in these parks and explores the landscape driving factors of visitor behavior, as well as the interaction effects between the factors from the perspective of spatial driving forces. The results indicate that (1) The observed behaviors in the three parks are primarily access behaviors. The number of visitors and the diversity of behaviors show a high level in the morning and evening and a low level in the midday. (2) There was minimal variation in behavioral composition and behavioral diversity among the study plots of different elevation gradients in the three parks. However, the contrasts between different landscape types were more pronounced, with impermeable plazas exhibiting the highest behavioral diversity and park roads demonstrating the most homogeneous behavioral diversity. (3) The impact of environmental factors was more pronounced than that of landscape pattern factors. The environmental factors that most strongly influenced passing, dynamic, and static behaviors were spatial connectivity value, hard space proportion, and number of recreational facilities, respectively. In contrast, the hard space proportion was the strongest driver of behavioral diversity. Moreover, the interaction between the hard space proportion and spatial connectivity value was more pronounced in driving behavioral diversity, as well as the three behaviors.

1. Introduction

The process of urbanization has resulted in elevated stress levels among residents, which has led to an increased demand for natural landscapes [1]. Studies have shown that urban green spaces have restorative effects on public health [2]. In general, visitor behavior reflects external environmental stimuli [3,4,5]. Analyzing public recreational behavior to improve the green space quality effectively enhances the public’s recreational experience [6,7]. Accurately reflecting visitors’ behavioral characteristics and their environmental relationships is crucial [8].

The relationship between visitors and the environment has been a subject of considerable interest to researchers. Regarding the environmental factors that influence visitors’ behavior, previous studies have indicated that both the macro landscape pattern and the micro environmental element composition exert a significant impact on visitors’ environmental perception, with varying degrees of influence [9,10,11,12]. This may, therefore, exert an additional influence on tourists’ behavioral intentions [13,14]. Macro-level studies have elucidated the relationship between the spatial pattern of green space landscapes and public activities from the perspective of urban planning [15,16]. However, there has been less attention paid to the influence of landscape patches within smaller green spaces [17]. Consequently, the relationship between the composition of landscape patches within a park and the behaviors of tourists remains unclear. At the micro level, existing studies have discussed the impact of different visual elements or spatial morphology elements on the behaviors of visitors [18,19,20,21,22]. This type of research primarily employs multiple regression model analysis to verify the numerical relationship between behavior and various landscape elements. However, the distribution of various environmental elements in space may also drive the distribution of behavior [23,24]. Regression models, which are commonly used in this context, are unable to reflect this spatial relationship [25]. Consequently, an investigation into the spatial influence of micro-scale environmental elements and macro-scale landscape pattern composition on park visitors’ behavior would serve to enhance our understanding of the relationship between the environment and behavior. This, in turn, would facilitate the improvement of the urban green space landscape quality.

At the level of analytical tools, scholars are paying more and more attention to analyzing the activity characteristics of urban residents and their influencing elements based on GIS [26,27]. Previous studies have more accurately reflected the environmental factors affecting the spatial distribution of tourists through big data collection and spatial analysis. Compared to the direct observation of tourist distribution, this approach offers the advantages of larger data volumes and more comprehensive analysis results [28]. However, due to data accuracy limitations, most previous GIS analyses have quantified only the relationship between the spatial distribution and density of tourists and the environment, making it difficult to accurately reflect factors such as tourist attributes and behavior types. Therefore, exploring the spatial distribution of different behaviors and their environmental drivers is essential for understanding the relationship between behaviors and the environment. In the past, a variety of methods have been employed to collect behavioral data, including the use of unmanned aerial vehicles (UAVs), GPS positioning, accelerometers, and panoramic cameras. Although UAVs offer convenience, they are susceptible to errors due to factors such as tree cover. While GPS and accelerometers can accurately reflect indicators such as tourists’ activity trajectories and length of stay, they are costly. Panoramic cameras, which offer unobstructed observation and are cost-effective, are highly reliable for the collection of behavioral data [29].

Studies have characterized visitor behavior by examining distribution and preference differences across landscape spaces and elevations [16,25,29,30]. These differences are often evident in larger parks [31,32]. Most studies have employed indicators such as visitor numbers and density as a criterion for quantifying behavioral characteristics [33,34]. This approach can effectively visualize the spatial and temporal distribution characteristics of visitors’ behaviors in green spaces [35,36]. However, the composition of behaviors is often complex [37,38], with behaviors in the same space often composed of many different categories [39]. Furthermore, the number of behaviors makes it difficult to reflect the complexity between the distribution of different behaviors, as well as the uniformity of the distribution [40]. Thus, quantitative measures alone cannot fully capture the complexity of visitor behavior.

As the capital of Fujian Province, Fuzhou has a population of over five million and features more than 50 mountains of varying sizes. Fuzhou has a subtropical monsoon climate, marked by hot, extended summers and mild, brief winters. Many mountain parks have been constructed as rest and recreation areas for urban residents [41]. Exploring the characteristics of urban residents’ recreational behaviors in mountain parks and their driving factors is therefore significant. To this end, the three mountains in Fuzhou (Wushan Park, Yushan Park, and Pingshan Park) were selected as the study area to investigate the behavioral characteristics of visitors. The geographical detector model (GDM) was employed to examine the driving and interacting effects of various landscape factors on recreational behavior [42,43]. The study is comprised of two primary aspects. The study aimed to describe the behavioral characteristics of tourists in the three mountains in Fuzhou from the perspectives of behavioral quantity, diversity, and compositional differences. Additionally, it sought to explore the driving force and interaction of various environmental factors on public recreational behaviors.

2. Materials and Methods

2.1. Research Site

This research focuses on Yushan Park, Wushan Park, and Pingshan Park, all located within the second ring road of Fuzhou, as the subjects of the study. These three mountain parks are nestled within the hills of the ancient city of Fuzhou, with their relative elevation ranging between 40 and 90 m and covering areas ranging from 10 to 60 hectares. Although the relative elevation of the three parks is within the hilly range, Fuzhou residents have long called the three locations “three mountains”. Therefore, many scholars in research have classified them as mountain parks. The history of the three mountains dates to the Tang Dynasty. Following the Tang Dynasty, the construction of the Fuzhou City wall was primarily centered around these three mountains (Figure 1). As a result, these mountains have preserved numerous historical and cultural relics, including ancient temples and stone monuments, thereby attracting many visitors. This provides ample opportunities for the observation of relational behavior in the study. A total of 35 study plots were selected based on an analysis of the landscape characteristics of the three parks.

2.2. Spatial Typology

Previous studies have indicated that visitors’ recreational preferences change depending on various landscape scales, which may also lead to a different composition of recreational behaviors [44]. Compared to general parks, the three mountains parks consist of numerous classical gardens with more specific landscape space types. The study was conducted to investigate differences in the composition of behaviors at different landscape space scales. According to the internal environment of the parks, the landscape space was categorized into six intermediate types: Type A—Lawn and waterfront area; Type B—Park roads; Type C—Impermeable plazas; Type D—Pavilions; Type E—Built structures; Type F—Forest viewpoints (Figure 2). In addition, altitude may also influence the behavioral perceptions and preferences of recreationists. Previous surveys have been based on peaks with heights of 100 m or more, so whether elevation makes a difference in parks below 100 m, such as the three mountains parks, is also a concern of the study. Therefore, the study clustered the elevation of each study plot of each park into three categories by the K-means algorithm and divided each study plot into a “Low Elevation Area (LEA)”, “Medium Elevation Area (MEA)”, or “High Elevation Area (HEA)”.

2.3. Data Acquisition

2.3.1. Acquisition of Visitor Behavior Data

The study conducted monitoring across three consecutive weekends with clear weather to observe the 35 designated study plots within the three mountain parks. The research employed the instantaneous sampling methodology, utilizing an Insta 360° One X panoramic camera to document imagery of visitors’ recreational behaviors at each of the 35 study plots. Data collection took place every two hours, from 6:00 a.m. to 6:00 p.m. The information captured encompassed critical demographic and behavioral aspects, including the gender, age, nature of the activity, and the precise location of the individuals engaged in recreational pursuits. Two research assistants, trained in the observational techniques of the System for Observing Play and Recreation in Communities (SOPARC) [45], were tasked with systematically recording visitor behaviors. They meticulously categorized these behaviors into three distinct modes: static, dynamic, and passing behaviors, used as the basis for calculating behavioral richness. The types of behavior are illustrated in the table below (

Table 1. Behavioral classification.

Major Behavioral Categories	Sub-Behavioral Categories	Specific Behavioral Categories	Behavioral Schematic Diagram
Static Behavior (SB)	SB1. Leisure and Relaxation Activities (LRA)	LRA1. Sitting
		LRA2. Stationary Standing
		LRA3. Short Sleep
		LRA4. Smartphone Usage
	SB2. Nature Engagement Activities (ENA)	ENA1. Observing Flora or Fauna
		ENA2. Viewing Scenery
		ENA3. Photograph
	SB3. Social Interaction Activities (SIA)	SIA1. Communicating
		SIA2. Phone Conversation
		SIA3. Playing Chess and Cards.
		SIA4. Tea Gatherings
		SIA5. Set Up a Stall
Dynamic Behavior (DB)	DB1. Site- related Activities (SRA)	SRA1. Dancing
		SRA2. Physical Fitness
		SRA3. Choral Singing.
		SRA4. Ball Games.
	DB2. Free Activities (FA)	FA1. Group Photo Session
		FA2. Visiting an Exhibition
		FA3. Frolicsome Play
		FA4. Childcare
Passing Behavior (PB)		PB1. Walking
		PB2. Running
		PB3. Ridding

). In addition, the observers roughly divided the visitors into four groups based on their physical characteristics: children (under 18), young adults (18–40), adults (40–60), and seniors (over 60).

2.3.2. Acquisition of Behavioral Driving Factors

The behavior and perception of visitors to green spaces are influenced by a variety of factors, which can be categorized into micro-level environmental factors and macro-level landscape pattern factors [46,47]. Drawing from this framework, and considering the unique context of these three parks, the study categorizes the factors influencing visitors’ behavior into environmental factors and landscape pattern factors.

At the micro level, drawing from a comprehensive literature review and previous studies, environmental factors influencing the quality of green space landscapes are classified into four main groups: spatial factors, natural factors, visual factors, and facility factors [48]. Following field research conducted at the three mountain parks, the study has organized the following five landscape factors based on the classification: visual factors, hardscape factors, facility factors, spatial factors, and natural factors. Details regarding the content of each element and the experimental methods employed are provided in the table below (

Table 2. Meaning and calculation of various landscape factors.

Landscape Factors Types	Indicators of Factors	Indicators Calculation Content	Quantitative Methods
Visual Factors (VF)	VF1. Sky Visibility	The proportion of sky in the visible range	Image Semantic Segmentation [49]
	VF2. Green View Ratio	The proportion of all vegetation in the visible range	Image Semantic Segmentation
	VF3. Arboreal Proportions	The respective proportions of arboreal within the visible range	Image Semantic Segmentation
	VF4. Shrub Proportions	The respective proportions of shrub within the visible range	Image Semantic Segmentation
	VF5. Herbaceous Plants Proportions	The respective proportions of herbaceous plants within the visible range	Image Semantic Segmentation
	VF6. Bare Ground Proportion	The proportion of bare ground and dead wood within the visible range	Image Semantic Segmentation
Hardscape Factors (HF)	HF1. Hard Space Proportion	The proportion of buildings and roads in the landscape view or visual field	Image Semantic Segmentation
Facility Factors (FF)	FF1. Number of Leisure Facility	The number of benches, seats, and pavilions specifically designed for visitor relaxation	Counting [50]
Spatial Factors (SF)	SF1. Harmonic Mean Depth	Refers to a measure used to describe the depth or distance of spatial relations within linguistic structures	Spatial Syntax [51]
	SF2. Spatial Connectivity	Number of nodes in the system that are directly connected to a particular node	Spatial Syntax
	SF3. Holistic Integration	Indicates the degree of connectivity between this space and all other spaces within the entire system	Spatial Syntax
	SF4. Localized Integration	Indicates the degree of connectivity between this space and the surrounding areas	Spatial Syntax
Natural Factors (NF)	NF1. Average Temperature	The temperature recordings in Celsius	Field Measurement
	NF2. Average Humidity	The amount of moisture present in the air	Field Measurement
	NF3. Gradient	The ratio of the vertical height h to the horizontal width l of the slope	Dem Data Analysis
	NF4. Elevation	-	Dem Data Analysis

).

Based on lessons learned from previous research, landscape pattern indices are deployed as a method to measure macroscopic landscape factors. Commonly used quantitative indicators in landscape pattern elements include the percentage of landscape (PLAND), largest patch index (LPI), landscape shape index (LSI), and patch cohesion index (COHESION) [9,52].

The study utilized GF-2 satellite remote sensing images captured in June 2022, with a spatial resolution and accuracy of 1 m. Following the pre-processing of the remote sensing images with atmospheric correction and radiometric calibration, the remote sensing images were subjected to supervised and classified analysis using the Random Forest plug-in in ENVI5.6 software. The characteristics of the spatial vegetation within the urban green space in Fuzhou were combined with the land types within the study site to create a classification system comprising five categories: water body, construction land, bare land, forest land, and grassland. The land use classification results were then examined. The test results demonstrated high consistency (KAPPA = 0.87), which met the accuracy requirements. Finally, the landscape pattern index of the buffer zone with a radius of 100 m around each sample point was calculated [53] (Figure 3).

2.4. Methodology

2.4.1. Visitor Density Calculator

This study employs kernel density estimation (KDE) to illustrate the comprehensive spatial distribution and density of behaviors across the three mountains and to initiate an investigation into the distributional characteristics of different behaviors. KDE can visually reflect the density distribution of visitors to different entertainment venues. KDE is calculated with the following formula [54]:

$$f\left(x\right)={\sum }_{i=1}^n{\displaystyle \frac{k}{\pi x^2}}\left({\displaystyle \frac{d_{ix}}{r}}\right)$$

(1)

f(x) represents the density of the behavioral sampling points, r denotes the search radius, d_ix is the distance from the behavioral sampling point to position x, and k is the weight value.

2.4.2. Tourist Behavioral Diversity Calculator

The study employed the Shannon–Weiner and Simpson indices, which are frequently utilized in soundscape diversity and biodiversity studies [55], to quantify behavioral diversity indices (BDIs) and behavioral dominance indices (BDOs), respectively [56,57]. BDIs reflect distributional homogeneity, whereas BDOs indicate the degree of dominance of a specific behavior over a designated period. In general, higher BDI and BDO values indicate a greater degree of behavioral richness. The BDI and BDO indices are calculated as follows:

$$BDI=-{\sum }_i^S\left(P_i\ln P_i\right)$$

(2)

$$BDO=1-{\sum }_{i=1}^S{P_i}^2$$

(3)

In this formula, S represents the number of types of behaviors in the space, P_i represents the number of individuals performing the ith behavior in the space as a proportion of the total number of individuals, and P_i² can represent the probability that both types of behaviors taken after random sampling are of the ith behavior.

2.4.3. Tourist Behavioral Variability Calculator

In the field of ecology, the Bray–Curtis distance is a method used to compare the differences in the abundance and distribution of each species between communities on a case-by-case basis. In this study, it was employed to quantify the dissimilarities in the diversity of recreational behaviors observed across different landscape spaces and parks. Based on this, the variability of the behavioral distribution was calculated using non-metric multidimensional scaling analysis (NMDS). The Bray–Curtis distance was calculated as follows [58]:

$$D_{BC}\left(S_i,S_j\right)={\displaystyle \frac{\sum \left|f_{ik}-f_{jk}\right|}{\sum \left(f_{ik}+f_{jk}\right)}}$$

(4)

D represents the Bray–Curtis distance between spaces S_i and S_j. This distance is a quantitative measure of the difference in behavioral composition between the two spaces. The relative share of the kth behavior in spaces S_i and S_j is denoted by f_ik and f_jk, respectively.

2.4.4. Geographical Detector Model (GDM) Calculation Method

Geographical detector is a model that explores the spatial heterogeneity of geographic phenomena and quantifies the driving effects between various factors [59]. The basic principle is that, if the independent variables of a factor in a study area affect the dependent variable, then their spatial distribution is similar [60]. This study uses the factor probes and interaction probes in the model to analyze the level of influence of the factors affecting the spatial distribution of visitors, as well as the degree of interaction [61].

Factor detection mainly explores the influence level of the factors affecting the spatial distribution of visitors on the behavior, and the q-value represents the explanatory power of each factor, and the larger the q-value is, the stronger the corresponding factor’s explanatory power of the behavior is and vice versa. The q-value is calculated by the following formula [24]:

$$q=1-{\displaystyle \frac{{\sum }_{t=1}^m{N}_t{\sigma }_t^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}$$

(5)

The formula includes q, which represents the factor’s influence (q ∈ [0, 1]), t, which is the stratification number of the independent variable, and m, which represents the stratification of the dependent variable Y or factor X. N and N_t represent the number of sample units in stratification t and the whole region, respectively. The variance of the dependent variable Y in stratification t and the whole region are denoted by σ_t² and σ², respectively. SSW and SST represent the sum of the intra-stratum variance and the total variance of the whole region, respectively.

Interaction detection is used to explore the q-value of any two factors (X₁ and X₂) superimposed on each other, which investigates the degree of influence of multi-factor interactions on visitor behavior. The table below shows the relationship between the two factors (

Table 3. The type of interaction and the interaction relationship between the two factors.

Interaction	Presentation Formula
Weakened, nonlinear	$q\left(x_1\cap x_2\right)<\min [q\left(x_1\right),q\left(x_2\right)]$
Weakened, unique	$\min [q\left(x_1\right),q\left(x_2\right)]<q(x_1\cap x_2)<\max [q(x_1),q(x_2\left)\right]$
Enhanced, bilinear	$q(x_1\cap x_2)>\max [q\left(x_1\right),q\left(x_2\right)]$
Independent	$q(x_1\cap x_2)=q\left(x_1\right)+q\left(x_2\right)$
Enhanced, nonlinear	$q(x_1\cap x_2)>q\left(x_1\right)+q\left(x_2\right)$

).

2.5. Data Analysis

In this study, we employed kernel density estimation (KDE) and other methods to describe the quantitative distribution characteristics of tourists’ behaviors. We then used the BDI and BDO indices to study the diversity of behaviors and analyze the differences in the composition of behaviors in different spaces through non-metric multidimensional scale analysis (NMDS) based on the Bray–Curtis distance and permutational multivariate analysis of variance (PERMANOVA). After initially filtering the various environmental factors through Spearman correlation and a multiple covariate test, we used the geographical detector model (GDM) to explore the driving and interactive effects between various environmental factors and behaviors (Figure 4).

3. Results

3.1. Characteristics of Visitors’ Distribution in the Three Mountain Parks

3.1.1. Characteristics of the Overall Number of Visitors’ Behavior

A total of 4248 visitors were recorded at the three mountains, of whom 502 were from Yushan Park, 1575 were from Wushan Park, and 1167 were from Pingshan Park. The overall gender distribution was relatively even, with a slight majority of male visitors. The specific demographics are shown in

Table 4. Characteristics of visitors to the three parks.

	Male	Female	Children (Age < 18 Years)	Young Adults (Age 18–40 Years)	Adults (Age 40–60 Years)	Seniors (Age > 60 Years)
Yushan Park	765	741	96	75	393	942
Wushan Park	852	864	171	252	864	288
Pingshan park	558	609	63	108	354	642

. The distribution of behavioral types demonstrates that many observed behaviors are passage behaviors, followed by dynamic and static behaviors. Passage behaviors account for 50.5% of the total, dynamic behaviors for 22%, and static behaviors for 27.5%. A total of 24 behavioral subtypes were recorded, with the highest total number of dispersal types in the passage behaviors, accounting for 50.4% of the total number of behavioral types and representing a total of 2130 individuals. The next most prevalent categories were viewing exhibits in dynamic activities and talking in static behaviors, which accounted for 12.9% and 7.7%, respectively.

3.1.2. Characteristics of the Spatial and Temporal Distribution of Visitors’ Behavior

The spatial distribution of visitors in the three mountain parks is shown in Figure 5 below, in which the passing behavior has the widest distribution, covering almost all the study plots; the static behavior is mainly distributed in the study plots with sitting benches, pavilions, and other facilities such as YS2, WS5, and PS3; and the dynamic behavior is mainly distributed in the study plots with a certain area of impermeable plazas, such as YS1, WS5, and PS6.

The time change trend of visitors’ behavior in the three mountains is illustrated in the curve (Figure 6), which demonstrates that the number of visitors in the three mountains exhibits a V-shaped trend over time. Specifically, there is a high concentration of visitors in the morning and evening and a low concentration in the afternoon. The data on the percentage of visitors’ behavior in each time period in the three mountains has been collated, and the results are presented in the following table. Excluding instances of necessary passage, the data indicate that, overall, visitors engage in a variety of activities. The activities that visitors to the three mountains most frequently engage in are viewing the scenery, engaging in fitness activities, observing flora and fauna, and reading books and using mobile devices (Figure 6).

3.2. Characteristics of Behavioral Diversity

3.2.1. Distribution of Behavioral Diversity

To examine the time-varying behavior of visitors in the three parks, we plotted the change curves of the BDI and BDO indices for a single day (6:00–6:00 p.m.) (Figure 7). Higher indices indicate a greater abundance of visitor behavior. The results are presented below and are comparable to the V-distribution of the number of visitors. The data suggest that visitor behavior is more diverse in the morning and evening, with activities such as fitness and dance being more common during those times compared to the afternoon. In contrast, visitor behavior during midday (12:00–2:00 p.m.) is more uniform.

The study categorized visitor behaviors at all study plots and calculated the behavioral BDI and BDO indices for each landscape space. The diversity of each landscape space was compared horizontally, and the results are shown below. The diversity index of the study plots within the impermeable plazas is generally higher, indicating a greater variety of behaviors assumed by the study plots in this space. In contrast, the study plots on the park road exhibited the most homogeneous activities (Figure 8).

Furthermore, the study calculates the behavioral diversity index under three elevation spaces to investigate the discrepancies in the behavioral diversity index at elevation spans below 100 m. The BDI and BDO indices for the three elevation spaces are calculated separately, and the results are presented below. The overall results demonstrate that the discrepancies between the BDI and the BDO at the three elevation scales are minimal (Figure 8).

3.2.2. Differences in Behavioral Composition between Spaces

The Bray–Curtis distance equation was employed to quantify the extent of behavioral divergence between the six types of landscape spaces. Each study plot’s behavioral profile was subjected to a non-metric multidimensional scaling (NMDS) analysis, which indicated that the model fit was enhanced following dimensionality reduction (stress = 0.171 < 0.2) (Figure 9). The results of the permutation multivariate analysis of variance (PERMANOVA) demonstrated that there were notable differences between the various landscapes. The PERMANOVA results indicated that there were significant differences between the different landscapes (p = 0.003 < 0.05, R² = 0.254). Further analysis of the behavioral composition of each landscape space by the PERMANOVA revealed that there was a significant difference between the behavioral composition of the impermeable plazas and that of the structure landscape (R² = 0.350, p = 0.035 *), while there was also a significant difference between the behavioral composition of the structure landscape and that of the resting platform (R² = 0.228, p = 0.035 *). Of particular note is the significant difference between the behavioral composition of the road space and the hard plaza. Furthermore, a more significant difference was observed in the behavioral composition between the road space and the impermeable plazas (R² = 0.374, p = 0.035 *).

The results of the non-metric multidimensional scaling analysis for the three elevation spaces are presented in the following section. The results demonstrate that the model exhibits a high degree of fit following dimensionality reduction (stress = 0.171 < 0.2) (Figure 7). However, the results of the replacement multivariate variance indicate that the divergence between the sample points is not statistically significant (p = 0.78 > 0.05, R² = 0.0457). The aforementioned results suggest that the behavioral composition of the three mountains exhibits minimal spatial differentiation at the elevation scale. This may be attributed to the relatively narrow elevation range of the three mountain parks, which limits the impact of elevation on visitor behavior and the distribution of behavioral choices.

3.3. Visitor Behavior Relevance Exploration

3.3.1. Environmental Factors That Influence Recreational Behavior

To identify the key drivers of visitor behavior, we first assessed the significance of each environmental factor on various visitor behaviors and behavioral diversity using Spearman correlation tests and visualized the results as shown below (Figure 10). The hard space proportion and the sky visibility showed a significant positive correlation with both the total number of visitors and the frequency of dynamic and static behaviors (q > 0, p < 0.01), indicating that visitors prefer open, impermeable plazas.

In terms of behavioral diversity, we used two indicators, BDI and BDO, to quantify the range of behaviors (Figure 10). Overall, behavioral diversity exhibited a significant positive correlation with the hard space proportion (q > 0, p < 0.05), suggesting that an adequate amount of hard surface area may not only influence the occurrence of diverse behaviors but also facilitate a broader range of behavior types.

3.3.2. Landscape Pattern Factors That Influence Recreational Behavior

Spearman’s correlation was used to analyze the relationships between the landscape form indices of five landscape patches and the diversity and frequency of different recreational behaviors. The numbers 1 to 5 denote water body, construction land, bare land, forest, and lawn, respectively. The results, shown in the figure below (Figure 11), indicate that, except for the forest form index (LSI4), which has a significant positive correlation with dynamic behaviors and total visitation (q > 0, p < 0.05), the associations between the other indices and the diversity and frequency of recreational behaviors are relatively weak. This may be attributed to the fact that visitor behavior is not directly influenced by factors such as landscape patch composition.

3.3.3. Extraction of Major Landscape Factors

Visitor behavior is influenced by many factors, but it is difficult to quantify the influence of each factor on different behaviors and the interaction between them, so we selected the landscape factors that have a greater correlation with visitor behavior, as mentioned above, i.e., the landscape factors that satisfy q < 0.05. The landscape factors considered are the number of leisure facilities (NLF), spatial connectivity value (SC), green visibility ratio (GVR), hard space proportion (HS), sky visibility (SV), bare ground proportion (BP), arboreal proportion (AP), and shrubs proportion (SP). To visualize each landscape factor, the study used the inverse distance weighting method to visualize the landscape factor indicators of each study plot to show the effect of spatial differentiation, and the spatial distribution of each factor is shown in the figure below (Figure 12). To ensure that there is no multicollinearity among the factors, it is necessary to diagnose the multicollinearity of the above factors, and we use the variance inflation factor diagnostic method of SPSS 26.0 software to verify the results, which are shown in the following table; the results show that the variance inflation factor (VIF) of each variable is less than 5 (

Table 5. Diagnosis of landscape factor covariance.

Factors	Covariance Statistics
	Tolerances	VIF
Number of leisure facilities (NLF)	0.796	1.256
Spatial Connectivity (SC)	0.516	1.936
Green view ratio (GVR)	0.332	3.011
Hard space proportion (HP)	0.516	1.938
Sky visibility (SV)	0.274	3.651
Bare ground proportion (BP)	0.740	1.351
Arboreal proportion (AP)	0.265	3.774
Shrubs proportion (SP)	0.624	1.603

), which indicates that there is no covariance problem among the selected independent variables. Therefore, the above variables can be used as drivers for the study.

3.4. Driving Factor Analysis

3.4.1. Analysis of Visitors’ Behavior Drivers

Table 6. Behavioral factor detection results.

Factors	BDI	BDO	Total Number	Static Behavior	Dynamic Behavior	Passing Behavior
NLF	0.305223	0.19159	0.241322	0.383413	0.203616	0.156245
SC	0.167488	0.128352	0.279607	0.295082	0.050766	0.408536
GVR	0.166849	0.132109	0.046521	0.132047	0.080607	0.000581
HP	0.323279	0.279227	0.1598	0.192634	0.224013	0.041819
SV	0.151975	0.140865	0.178616	0.110659	0.061234	0.157544
BP	0.012177	0.02718	0.047045	0.05449	0.049493	0.038515
AP	0.198535	0.168921	0.202913	0.120521	0.191272	0.186338
SP	0.032529	0.012226	0.078083	0.051718	0.075686	0.102993

shows the results of the factor detector of various factors affecting different types of behaviors, and the figure shows the degree of influence of various factors (Figure 13). The above results explain the role of each factor in influencing various types of behaviors. The specific results are as follows.

In terms of various types of behaviors, the factor with the greatest influence on static behavior was NLF (p = 0.38), followed by SC (p = 0.30) and HP (p = 0.19), which suggests that a certain size and number of leisure facilities are the most important drivers of static behaviors such as sitting and chatting, and the factor with the greatest influence on dynamic behavior was HP (p = 0.22), followed by NLF (p = 0.20) and AP (p = 0.19), which suggests that the hard proportion of the space is the most important driver for visitors to engage in dynamic behaviors such as square dancing and sports activities; the factor that had the greatest influence on access behavior was SC (p = 0.41), followed by AP (p = 0.19) and SV (p = 0.16), suggesting that the spatial amount of site connectivity was the most significant driver of pass-by behavior compared to the other factors.

Additionally, the most significant driver of each of the influences on the total number of visitors was SC (p = 0.28), suggesting that the amount of space directly connected to the event site was the most significant driver of the total number of incoming and outgoing visitors.

We examined how individual factors contribute to driving behavior and the synergies between them. We analyzed individual environmental factors using GDM’s interaction detector. As shown in the figure (Figure 14), the results of the interaction showed that, among the eight factors affecting visitors’ different types of visitor behaviors (static, dynamic, and passing), the interaction between any two factors was greater than that of a single factor independently, and the types of interactions were either two-factor enhancement or nonlinear enhancement (A: two-factor enhancement, and B: nonlinear enhancement). This also indicates that the factors driving visitor behavior do not exist independently but are the result of interactions among multiple factors.

For specific behavioral activities, the total number of visitors was most affected by the interaction of SC ∩ HP (q = 0.84), showing that both had a significant driving effect on the number of visitors. For dynamic behavior, the interaction of SC ∩ HP (q = 0.61) and the interaction of NLF ∩ HP (q = 0.55) had a stronger driving effect. Static behavior, on the other hand, was driven more by the interaction of NLF ∩ HP (q = 0.53). Passing behavior was primarily driven by the interaction of SC ∩ HP (q = 0.72) and, to a lesser extent, by SC ∩ SV (q = 0.71).

3.4.2. Analysis of Behavioral Diversity Driver

To further explain the diverse behaviors of visitors in the three mountain parks, we continued the previous approach to explore the drivers of behavioral diversity, The results are shown in

Table 7. Behavioral diversity factor detection results.

Factors	BDI	BDO
NLF	0.305223	0.19159
SC	0.167488	0.128352
GVR	0.166849	0.132109
HP	0.323279	0.279227
SV	0.151975	0.140865
BP	0.012177	0.02718
AP	0.198535	0.168921
SP	0.032529	0.012226

. Based on the results of the factor detector, the largest driver of the BDI and BDO indices was HP (q = 0.32, 0.28), followed by NLF (q = 0.31, 0.19) and AP (q = 0.19, 0.17), respectively. This result suggests that the hard proportion of space is a greater driver of behavioral diversity.

The results of the interactions are shown in the figure below (Figure 15): the interaction effect of HP ∩ SC is the strongest for both BDI and BDO (q = 0.71,0.63), followed by GVR ∩ SC (q = 0.72, 0.60); in addition, AP ∩ SC (q = 0.68, 0.55) and SV ∩ HP (q = 0.66, 0.59) are also more obvious in the driving effect of the BDI and BDO. The above results show that, among all the factors affecting behavioral diversity, the interaction between SC and HP has the most obvious driving effect on behavioral diversity, and in addition, the interaction between SC and GVR also has an obvious driving effect on visitor diversity.

4. Discussion

4.1. Characteristics of Visitors’ Recreational Behavior

Passing behavior was found to account for 50.5% of the study in the three mountain parks, indicating that passing is the most common behavior during the day, which is consistent with other studies [62,63,64]. Analysis of the visiting population reveals that individuals over 50 years old tend to visit the park more frequently and stay longer [65,66,67]. The growing number of middle-aged and elderly tourists may be associated with the increasing longevity of the Chinese population and the high proportion of individuals in these age groups within the overall demographic [68]. The influence of the health and wellness culture in China has resulted in a notable preference among the elderly for Tai Chi and exercise during the relatively cool morning hours, as well as for lunch breaks during the relatively hot midday hours [69]. Conversely, they tend to engage in less physical activity during the afternoon. This has a profound impact on the temporal distribution of dynamic versus static behaviors. From the perspective of activity preference, older individuals tend to gravitate towards a more tranquil ambience, whereas younger individuals often seek out environments with a more vibrant energy [70]. Consequently, the three mountains may hold a particular appeal for those in the older age group.

Behavioral distributions exhibit a striking divergence in spatial patterns. The study examines the differences in behavioral richness and composition across different spatial scales through the lens of behavioral diversity, illuminating the influence of distinct spatial features on visitor behavior. At the level of behavioral diversity, the BDI index and the BDO index offer a more nuanced understanding of the complexity and homogeneity of tourist behaviors in diverse landscape spaces than the mere enumeration of behavioral categories [71]. At the level of behavioral composition, previous studies tended to focus on the overall pattern of visitor behavior and the types of behaviors within a single space [72,73]. In contrast, NMDS analysis based on the Bray–Curtis distance quantifies and compares the variability of behavioral composition among different spaces one by one [74,75]. It has been demonstrated that differences in the external environment can constitute differences in behavior [76]. This study found that the differences in behavioral composition among the six types of landscape spaces were more pronounced, with hard spaces exhibiting the highest behavioral richness and road spaces exhibiting the thinnest behavioral composition. The rationale behind this phenomenon can be attributed to the fact that parkways are typically designed with the primary objective of optimizing transportation efficiency, rather than fostering social interaction [44]. However, this study found that behavioral differences among visitors to the three mountains were weakly differentiated across the three elevation scales. Previous studies have indicated that visitor preferences and perceptions in mountain parks may vary with the elevation [77]. This study suggests that, since the overall elevation of the three mountains is less than 100 m, visitor behavioral preferences vary less across this lower elevation span, making differences in the behavioral composition of the visitors less pronounced.

4.2. Environmental Factors That Influence the Behavior of Tourists

The visual landscape elements have a greater impact on the recreational behavior of visitors to the three mountain parks than the landscape pattern elements, and we hypothesize that, in smaller-scale green spaces, the landscape patches do not have a direct impact on the perception of visitors, while the impact of smaller-scale visual landscape elements can often have significant effects on the recreational perception of visitors [78]. It is noteworthy that the behavior of the three mountains exhibits a relatively weak correlation with the gradient. A review of China’s “Code for the Design of Public Parks, GB 51192-2016” was conducted [79], and based on on-site measurements, it was determined that the slopes of each activity space in the three parks are generally not higher than 10°. Furthermore, the planners have provided steps in the relatively steep areas, and in general, the activity spaces in the three mountains are very gentle. The gentle slopes may facilitate accessibility and reduce the amount of physical exertion required by visitors [80]. In general, an activity space with a slope of 5°–10° is optimal for a pleasant recreational experience [81]. Therefore, the slope of the three mountains’ space is suitable for most types of activities, which indicates that the slope of the internal space has a minimal effect on the behavioral distribution of visitors in the three mountains.

The GDM analysis revealed that spatial connectivity (SC) emerged as the strongest driver of the number of visitors, mirroring its role as the primary driver of passing behavior, given its association with the highest frequency of passing behaviors. Consistent with the findings of a previous study in Bangkok [82], the single factor that drove static behaviors the strongest was the number of leisure facilities (NLF), which has been highlighted several times in previous studies [83,84]. Some studies have shown that the leisure activities of the senior population are closely related to the leisure facilities [85]. The prevalence of senior visitors in the three mountain parks is notable, with a tendency to gravitate towards facilities offering shade for resting. Additionally, the primary driver of dynamic behavior appears to be the presence of hard pavement (HP), as evidenced by the association between high physical activity and hard plaza conditions [86,87]. Past research has indicated that areas such as sport fields and children’s playgrounds often witness heightened physical activity [88,89], along with the influence of recreational amenities on such activity levels [90,91]. Nevertheless, it is worth noting that the three mountain parks lack recreational facilities conducive to high physical activity, potentially contributing to the limited engagement in dynamic behaviors among visitors. The interaction between spatial connectivity (SC) and hard pavement (HP) greatly influences these behaviors. This underscores the substantial impact of the interplay between spatial connectivity and hard pavement area on behavioral diversity.

For behavioral diversity, the diversity index can more accurately reflect the complexity and homogeneity of visitor behavior in different landscape spaces compared to previous counting research studies. We observed that impermeable plazas have the highest behavioral diversity and the park’s road has the thinnest behavioral composition. Further GDM analysis proved that impermeable plazas have a driving effect on the BDI and BDO, which is consistent with the results of previous studies [92,93]. Studies on spatial morphology have demonstrated that the diversity of physical activities will be higher in spaces with simpler shapes and more regular edges [94]. According to our observations, many impermeable plazas in these three mountain parks are regular rectangular-shaped, which may be one of the reasons for the higher behavioral diversity. This is because the configuration of the space enables the regulation of the movement of individuals throughout the space with a specific level of organization, thus facilitating the occurrence of various types of activities. Although the complex shape of the space exudes a stronger sense of security and a quieter atmosphere, it has a certain inhibiting effect on visitors’ willingness to engage in physical activity. In addition, the interaction results showed that the driving effect on the BDI and BDO would be more pronounced under the interaction of HP and SC, which may be due to the fact that spaces with higher connectivity values have a higher number of passing visitors [95], which increases the attractiveness of the space, suggesting that impermeable plazas that are most interconnected with other spaces are teeming with the most diverse activities. It is worth noting that the interactions of HP and CV all have a strong driving effect on the three behaviors, which further indicates that the interaction of spatial connection value and paved area has a non-negligible effect on the behavioral diversity. Therefore, it is important to focus on improving the landscape quality of areas with a high hard pavement proportion and high spatial connectivity values in the park for visitors’ recreational experience.

4.3. Limitations and Perspectives

The driving mechanism of behavior is a highly complex system. At the behavioral level, although the study employs a more mature behavioral observation method, it still has the disadvantage of strong subjectivity, which introduces a certain degree of error into the experimental data. Furthermore, the study did not sufficiently examine regional cultural factors, tourist sources, and other elements, thereby limiting the ability to construct a comprehensive behavioral driving mechanism. This makes the construction of behavioral driving mechanisms imperfect.

Considering the shortcomings, it is imperative that future research on the relationship between green space behavior and the environment considers the influence of green space users’ personal elements. Furthermore, the technology that can accurately identify behavioral categories and trajectories and collect environmental data in a comprehensive manner is of great importance for increasing the accuracy of experiments. It is anticipated that this technology will be introduced into relevant practice in the future.

5. Conclusions

Urban mountain parks play an important role in promoting the daily recreation of urban residents, and understanding the behavioral habits and influencing factors of mountain park users can explore the potential effects of landscape green spaces. By investigating the relationship between the objective environment and visitors’ behavior in the three mountains, this study provides an explanation of the influencing mechanism of recreational behavior in the three mountains and enriches the knowledge of mountain park behavior.

The results of the study show that (1) the visitors in the three mountains are mainly middle-aged and old-aged, and the visitors’ behavior has significant changes in the time dimension, and the morning (6:00–8:00 a.m.) is the most frequent time period for all kinds of behaviors; (2) the richness of visitors’ behaviors and behavioral compositions are obviously differentiated under different landscape spaces, in which the behavioral richness in the impermeable plaza is the highest and that of the park road is the lowest. Park managers should pay more attention to the age-friendly design of impermeable plazas. (3) The environmental factors play a greater role in visitors’ behavior than the landscape pattern factors, so designers should pay more attention to improving the visual landscape within the activity space. (4) Different types of behavioral driving factors are different: the spatial connectivity, the hard space proportion, and the number of leisure facilities have the most obvious driving effects on passing behavior, dynamic behavior, and static behavior, respectively. For high spatial connectivity, a high hard pavement area of the space occurs in a more diverse type of behavior; for this type of space, designers should consider the diverse recreational needs of recreationists and focus on the maintenance of recreational facilities to ensure that they meet the recreational needs of elderly tourists.