This study recruited a sample of 633 undergraduate students from a university. Recruitment notices were disseminated through campus communication groups. The primary purpose of the study and the intended use of the data were clearly communicated in the recruitment message. Participation was entirely voluntary and anonymous. All participants were required to provide informed consent prior to completing the questionnaire. The study was reviewed and approved by the Ethics Committee of Southwest University.

Given that the Perceived Restorativeness Scale (PRS) assesses subjective perceptions of specific environments, we developed visual stimulus materials for this study.

Drawing on previously validated stimuli (31) and theoretical assumptions from existing literature (32, 33), we created three images depicting varying levels of vegetation—from none to high. To ensure control over confounding variables, we used AI-generated, photorealistic images, following the approach adopted in recent research (34, 35). All images depicted open, boundary-free outdoor spaces. The three images shared a common base of a concrete ground surface, with the amount of visible vegetation progressively increasing and the presence of built/artificial elements decreasing accordingly. Using a concrete base ensured consistency, walkability, and relevance to urban settings.

Participants were then asked to complete an electronic questionnaire on a tablet device. Each participant was randomly assigned one of the three images and instructed to respond based on their imagined experience in that environment.

In the final sample, male participants slightly outnumbered female participants, comprising 59.4% of the total (Appendix Table 2). More than half of the students were in their second year of undergraduate study. The majority reported a household monthly income within the range of 0–10,000 RMB (approximately 0–1,400 USD as of June 2025). Due to the randomized image assignment design, the number of participants exposed to each of the three vegetation-level scenarios was approximately equal.

When all items were loaded onto the theoretical structure, the model demonstrated suboptimal fit indices (SRMR = 0.100; NFI = 0.874; TLI = 0.875; CFI = 0.887; RMSEA = 0.108). We identified that Items 10 and 13 had factor loadings below 0.50 (see Appendix Figure 1), and therefore, these items were removed from further analysis. After their removal, model fit improved and met acceptable thresholds (SRMR = 0.724; NFI = 0.910; TLI = 0.913; CFI = 0.922; RMSEA = 0.095). In addition, both Average Variance Extracted (AVE) and Composite Reliability (CR) values met the recommended criteria (see Appendix Figure 2).

It is worth noting that the inter-factor correlation between “being away” and “fascination” were relatively high, approaching the liberal threshold of concern (r = 0.9) (Appendix Figure 2), as suggested by others (51, 52). We further tested discriminant validity. By the Fornell–Larcker criterion (53), AVE values were lower than squared inter-factor correlations, indicating limited discriminant validity. Item-level checks showed no major cross-loadings; only two items slightly exceeded 0.30 on non-theorized factors, well below the 0.40 threshold (54–56) and much lower than their intended loadings (Appendix Figure 3).

Kruskal–Wallis tests showed significant differences across vegetation conditions for Being Away (H = 61.29, p < 0.001, η² = 0.094), Fascination (H = 68.86, p < 0.001, η² = 0.106), Compatibility (H = 71.60, p < 0.001, η² = 0.110), PRS total (H = 84.08, p < 0.001, η² = 0.130), and physical activity intention (H = 54.40, p < 0.001, η² = 0.081). Coherence showed only a small effect (H = 14.59, p = 0.001, η² = 0.020), with significant differences limited to the comparison between the no-vegetation and high-vegetation conditions (Figure 2).

Mediation analyses showed significant indirect effects for Being Away (β = 0.201), Fascination (β = 0.235), and Compatibility (β = 0.246; all p < 0.001), explaining 67–83% of the total effect (Table 1). Coherence had only a marginal effect (β = 0.015, p = 0.044; 5.03%). Using the PRS total score, the indirect effect absorbed nearly all of the association (β = 0.285, p < 0.001; 95.64%), while the direct path was non-significant (β = 0.013, p = 0.600) (as indicated by the pathways between variables in Figure 3), indicating full mediation.

Sensitivity analysis using all PRS items yielded nearly identical results (Appendix Figure 4): total effect β = 0.298 (95% CI = 0.224–0.367, p < 0.001), indirect effect β = 0.288 (95% CI = 0.235–0.340, p < 0.001), with mediation efficiency of 96.64%.

In this study, we validated the reliability and validity of the Chinese version of the Perceived Restorativeness Scale (PRS) in a sample of Chinese university students. Using this tool, we examined the pathway linking vegetation levels to intention to engage in physical activity through perceived restorativeness. To minimize confounding factors inherent in observational studies, we employed an image-based experimental design to collect data. Our main findings indicate that higher vegetation levels are associated with greater perceived restorativeness and increased intention to engage in physical activity. Furthermore, perceived restorativeness plays a significant mediating role in the relationship between vegetation level and physical activity intention. Notably, when using the total PRS score as the mediator, we observed a full mediation effect, as evidenced by the near-zero direct effect after including the mediator (57). Although this does not completely exclude the possibility of other mediating factors (58), it does provide strong evidence supporting the critical role of perceived restorativeness in linking vegetation levels and physical activity intention.

It is noteworthy that when Coherence was modeled as the mediator, its effect was weak. Differences appeared only between no- and high-vegetation conditions, implying that order and legibility emerge mainly with abundant vegetation. We assume that coherence may also be less relevant for motivating activity than fascination or compatibility. Methodological factors—such as simple AI-generated images and high inter-factor correlations—may have further reduced its role. Future studies with richer, multi-sensory stimuli should test whether this reflects theory or method.

This study measured physical activity intention rather than behavior. Although intention is a key predictor, the intention–behavior gap means strong intentions may not translate into action (59, 60). Thus, our findings reflect motivation rather than behavior. Future work should include objective or self-reported measures (e.g., accelerometry, EMA, follow-ups) to test whether restorative environments increase actual activity.

In summary, these findings largely support our Hypotheses 1 and 2 and offer a theoretical foundation for understanding how green spaces promote or stimulate physical activity.

Although the validation of the PRS was not the primary focus of this study, it represents a necessary step to enhance internal validity and demonstrate the credibility of our evidence. The PRS is one of the earliest and most widely used instruments for measuring environmental restorativeness (61); however, its validation within Chinese settings remains very limited. Many studies have directly employed Chinese translations of the PRS without further examination of its reliability and validity (62–64). Through a literature review, we identified only one earlier Chinese study that used exploratory factor analysis to preliminarily examine an earlier version of the PRS with fewer items (65). The revised version employed in our study, however, had never been translated or psychometrically tested previously.

In this study, we observed satisfactory structural validity; nevertheless, the high inter-factor correlations suggest that the original theoretical structure might face challenges. Despite this, drawing on the concept of known-groups validity—which posits that measurement outcomes should reflect differences between theoretically distinct groups (66)—the observed differences in PRS subscale and total scores across different vegetation scenarios further support the scale’s measurement validity.

In conclusion, further validation work is warranted, such as conducting exploratory and confirmatory factor analyses in larger and more representative populations.

To date, numerous studies have examined the association between vegetation levels and physical activity (13, 16, 62). However, these studies often employed observational designs, which cannot rule out confounding factors or reverse causality—for example, individuals with more resources for physical activity might be more likely to live in greener neighborhoods. Moreover, as Markevych et al. (11) pointed out, previous research rarely distinguished the specific locations where physical activity occurred, potentially recording some activities unrelated to green spaces, thereby distorting findings. In contrast, this study used a design similar to a randomized controlled trial and specifically assessed participants’ intentions to engage in physical activity in defined scenes. These findings reinforce and refine the understanding that green spaces can promote physical activity. It should be noted, however, that since our study is still based on surveys rather than actual behavioral experiments, we measured physical activity intention rather than objectively measured physical activity levels. Therefore, these findings cannot fully explain the observed promotion effect of green spaces on physical activity.

Nevertheless, our findings provide an important mechanism to explain the relationship between green space and physical activity, especially as we found a full mediating effect. Prior to this study, numerous investigations confirmed a positive association between vegetation and perceived environmental restorativeness (67, 68). However, only a few studies attempted to further link environmental restorativeness with physical activity (27, 28). Our study further validates this indirect pathway and emphasizes that residents’ pursuit of psychological restoration opportunities is a key reason for their engagement in green space activities.

It should be noted that data were collected near final exams, when stress and fatigue are high. This may have heightened sensitivity to restorative environments and inflated restorativeness and intention scores, potentially overestimating the mediation effect. Future studies under more neutral conditions are needed to confirm robustness.

This study has two main contributions:

Beyond the university context, our findings suggest that urban planners, public health practitioners, and campus designers could leverage vegetation to create environments that simultaneously foster restoration and physical activity. Investing in greener spaces is not merely aesthetic but represents an evidence-based strategy to support healthier, more active communities.

A limitation concerns the ecological validity of our experimental design. Although the use of images as stimuli is common in this type of research, it may reduce overall perceived restorativeness judgments and neglect other sensory modalities. Prior work has shown that image-based methods can underestimate restorativeness compared to real-world environments (69). Moreover, Grahn has emphasized that sensory inputs beyond vision—such as sound—are essential for stress restoration (70). As a result, our findings should be interpreted as reflecting responses to visual aspects of vegetation rather than the full spectrum of restorative experiences. Future research employing field experiments or immersive technologies (e.g., virtual or augmented reality) could enhance ecological validity and more accurately inform health promotion and urban design practices.

Second, to control environmental variables, we used simulated rather than real-scene photographs, which may affect the authenticity of the experience. Third, we used a convenience sample from only one university, limiting representativeness. Future research should consider broader populations. Finally, some deficiencies exist in the reliability and validity of the scales used, which restrict the internal validity of the study. Another limitation relates to sample composition. Male students accounted for nearly 60% of participants, creating a gender imbalance that may have influenced PRS responses and physical activity intentions.