To assess the level of visual naturalness in urban green spaces, we calculated the average percentage of visually natural elements based on multiple 180-degree panoramic photographs taken at each location. To ensure statistical robustness and capture the inherent heterogeneity of the environments, we selected 5–10 random sampling points within each green space type (Urban Parks, Forests, and Wetlands) rather than relying on a single viewpoint. This assessment combined computer vision techniques with manual evaluation to classify the proportion of natural and man-made features within each scene. First, we applied a semantic segmentation algorithm to analyze the scene captured in the view frame. Semantic segmentation is a technique that combines semantic and instance segmentation by identifying countable structures (such as trees and people) and uncountable structures (such as the sky), labeling each pixel to categorize the scene (29).

We used Fully Convolutional Networks (FCNs), a widely adopted method for semantic image segmentation, capable of identifying up to 150 site elements (29). In this study, spatial images from all sampling points were analyzed using a fully trained FCN model, with an average accuracy of over 90% (29, 30). Afterward, the results were manually reviewed and corrected to rectify any segmentation errors and to systematically aggregate the identified elements into the study’s target categories. Specifically, precise pixel-level adjustments were made to ensure that all landscape features were accurately assigned to either ‘natural’ or ‘man-made’ classes, prior to the exclusion of non-relevant elements (e.g., sky and pedestrians). The site features were then categorized into two groups: natural and man-made elements. Sky and human pixels were excluded from the analysis. The proportion of natural features was summed and averaged across all sampling points to calculate the overall visual naturalness of the green space.

This study utilizes the PHE threshold model to quantitatively evaluate the impact of urban green space exposure on human health. The model is constructed based on the dose–response relationship and the economic principle of diminishing marginal utility. It provides a theoretical framework to identify the non-linear dynamics between exposure duration (dose) and physiological health outcomes.

In this framework (Figure 1), the x-axis represents the duration of exposure to green spaces (e.g., remnant forests, wetland parks, or urban parks), while the y-axis represents the standardized physiological health indicators (e.g., EEG). The curve illustrates how health benefits accumulate over time and eventually plateau. Based on this trajectory, the model identifies two critical thresholds: (1) Efficiency Threshold (ET): This point marks the onset of rapid recovery, where the marginal health benefit of additional exposure begins to increase significantly. (2) Benefit Threshold (BT): This point marks where the exposure benefits are maximized and begin to plateau. It indicates the optimal level of green space exposure for health improvement, beyond which additional exposure yields diminishing returns.

To comprehensively evaluate the cumulative positive impact of green space exposure, we utilized the Area Under the Curve (AUC) method. AUC is widely used to evaluate dose–response relationships, particularly in studies exploring the link between natural exposure and health benefits (11, 12).

In this study, the AUC represents the total volume of physiological recovery accumulating over the exposure period. We calculated the AUC using the trapezoidal rule, which allows for the accurate quantification of cumulative effects for continuously differentiable functions within a specified interval. The calculation formula (1) and (2) are as follows:

Where S represents the accumulated health benefit (AUC), and the functions f(x) and d(x) represent the fitted dose–response curves for the respective physiological indicators over the exposure time x.

This study was conducted at the Xishanyang National City Wetland Park in China (30°N, 120°E). The experiment took place over 3 days in May 2024, from 8:30 a.m. to 5:30 p.m. each day. The study involved a total of 82 participants (detailed in Section 2.4.2). To ensure rigorous control over the experimental conditions, the study was conducted in multiple sessions.

In each session, participants were divided into two experimental conditions: the Open Eyes (OE) group and the Blindfold (BF) group. This quasi-experimental design exposed participants to different types of urban green spaces (UGS) along walking routes that included a park, a forest, and a wetland, providing three distinct natural environments for comparison (Figure 2). To assess the restorative effects, we used the Perceived Recovery Scale (PRS), which evaluates the restorative impact of visual stimuli. Participants were asked to undergo fatigue assessments before and after the exposure to minimize any potential “artificial” effects, ensuring the experiment reflected real-life conditions. The study aimed to quantitatively compare the physiological health effects of exposure to various UGS types on urban residents.

The EEG signals were acquired using the Emotiv EPOC X wireless headset (Emotiv Systems, San Francisco, CA, United States). The device consists of 14 saline-soaked felt sensors located at AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, and AF4, adhering to the international 10–20 system. Two references (CMS/DRL) were placed at the P3 and P4 positions. The sampling rate was set to 256 Hz (internally down-sampled to 128 Hz), with a bandwidth of 0.2–43 Hz. The impedance of all electrodes was monitored in real-time and maintained below 20 kΩ to ensure high signal quality.

Signal Preprocessing: Raw EEG data were exported and processed using MATLAB R2022b (MathWorks, Natick, MA, United States) and the EEGLAB toolbox. To remove noise and artifacts, the following preprocessing steps were applied: (1) Filtering: A high-pass filter at 0.16 Hz and a low-pass filter at 43 Hz were applied to remove baseline drift and high-frequency noise. A notch filter at 50 Hz was strictly applied to eliminate power line interference. (2) Artifact Removal: Eye blink and muscle movement artifacts were identified and corrected using Independent Component Analysis (ICA). Segments with voltage amplitudes exceeding ±100 μV were automatically rejected.

Feature Extraction: Fast Fourier Transform (FFT) was performed to convert the time-domain signals into the frequency domain (Power Spectral Density, PSD). The specific frequency bands were defined as follows: θ (Theta, 4–8 Hz), α (Alpha, 8–13 Hz), and β (Beta, 13–32 Hz). The relative power of each band was calculated as the ratio of the specific band’s power to the total power (4–32 Hz). Physiological indicators, including the Relative α index (relaxation), Relative β index (stress), and β/α ratio (alertness), were then computed for the “Stress” (0–4 min) and “Exposure” (4–20 min) phases. Data points were fitted every 30 s to generate trend curves. Specifically, the EEG data from all participants in each group were aggregated at each time point to construct the group-level dose–response trajectory.

A total of 82 participants completed the experiment. The study population consisted of 45 males and 37 females. The participants included university students, park visitors, and staff members, ensuring a diverse representation. All participants reported good general physical and mental health status prior to the experiment.

Meteorological data monitored during the experiment (temperature, humidity, wind speed, and noise levels) were analyzed to ensure environmental consistency. The average temperature was 24 °C, and the relative humidity was 65.4%, and wind speed was 1.2 m/s. ANOVA results indicated no significant differences in meteorological conditions across the three experimental days (p > 0.05), ruling out weather as a confounding variable.

The visual naturalness of each location was assessed using a well-trained Fully Convolutional Network (FCN) model to identify site characteristics, specifically the percentage of natural elements in the environment. After manual inspection and re-classification, the results revealed that city parks (UP) had a significantly lower visual naturalness compared to forests (RF) and wetlands (WP). The visual naturalness scores were 85% for UP, 99% for RF, and 95% for WP.

For plant species richness, data were collected within a 100 m² area around the experimental stopping points. The results showed that the wetlands had the highest plant species richness, followed by forests, with urban green spaces having the lowest richness.

Exposure to different urban green spaces (UP, RF, and WP) had significant effects on all selected EEG metrics (p < 0.05). To accurately interpret the temporal dynamics shown in Figures 3–5, it is essential to distinguish between the two experimental phases represented on the x-axis (0–20 min): (1) Stress Induction Phase (0–4 min): During this initial period, participants performed stress tasks. As shown in the figures, stress-related indicators (e.g., Relative β index) exhibited an upward trend or remained at a high level, confirming the successful induction of physiological stress prior to exposure. (2) Green Space Exposure Phase (4–20 min): This phase corresponds to the actual viewing of the green spaces. Our analysis of health restoration specifically focuses on this window.

Analyzing the Exposure Phase (4–20 min), participants in the OE group exposed to the UP environment exhibited significant improvements. Relaxation occurred rapidly after the onset of exposure (i.e., immediately after the 4-min mark), showing a more pronounced reduction in stress, as indicated by positive EEG changes (Figure 3). In comparison, the BF group also experienced relaxation and stress relief in the UP environment, but the temporal trajectory indicates that effects were slower and weaker than those observed in the OE group (Figure 3). Interestingly, neither the OE nor the BF group was able to engage in deep thinking while exposed to the urban park environment. This lack of effective deep thinking persisted throughout the exposure time, which is likely due to external factors such as noise, a fixed sitting posture for extended periods, and other environmental conditions.

Exposure to the RF environment, as indicated by the EEG results, shows that the forest environment contributes to an overall improvement in participants’ mental and physical health, promoting relaxation and stress relief (Figure 4). However, it is noteworthy that the relative α index and the relative (α + θ) index were negatively affected in the OE group over the course of the experiment (Figures 4b,h). These negative effects may be attributed to the fear and anxiety experienced by participants with open eyes in the natural forest, suggesting that visual exposure to dense forests may trigger complex psychological fluctuations over time rather than a linear recovery. In contrast, blindfolded participants in the RF environment appeared more relaxed. However, their heightened alertness and anxiety might have prevented them from achieving a state of deep thinking or full immersion in the experience. Overall, the OE group exhibited more significant physical health benefits than the BF group.

Regarding the temporal trends in the WP environment, the results revealed that both the OE and BF groups experienced a reduction in urban stress, fostering a relaxed and happy state throughout the exposure session (Figure 5). However, a negative effect was observed on the β/α ratio in the BF group (Figure 5e). This could be attributed to participants feeling panic and anxiety in the dark environment, leading to sustained heightened vigilance and preventing them from achieving a calm state.

In summary, when comparing the EEG indicators of the OE groups exposed to three natural environments—UP, RF, and WP (Table 1)—it is evident that all three environments contributed to improvements in participants’ psychological and physical health, promoting relaxation and reducing stress. Notably, the relative α index showed a negative effect during exposure to the RF environment compared to the other two green spaces (Table 1), which could be due to the panic experienced in the forest. Comparison of the three green spaces in terms of the relative β/α ratio revealed that all three environments helped reduce alertness, alleviate pressure, and foster inner peace (Table 1). AUC analysis also confirmed that all three environments contributed to better physical and mental health and helped ease urban stress. The relative (α + θ) index indicated that deep thinking was not effectively engaged in either the UP or RF environments (Table 1). This was attributed to the noise and artificial structures in the UP environment, which hindered deep thinking, while exposure to the RF environment led to fear and anxiety, further preventing participants from engaging in deep thought.

Subjective assessments corroborated the physiological EEG findings, showing significant psychological benefits. (1) Perceived Recovery Scale (PRS): Participants perceived the Remnant Forest (RF) and Wetland Park (WP) as significantly more restorative than the Urban Park (UP). The average PRS scores were 4.2 (RF) and 4.1 (WP), compared to 3.6 (UP) (p < 0.05), aligning with the higher visual naturalness of these sites. (2) WHO-5 Well-being Index: The WHO-5 scores significantly improved after exposure. The average score increased from 52.1 (Pre-exposure) to 68.4 (Post-exposure) (p < 0.01), confirming that short-term green space exposure effectively enhances immediate subjective well-being.

Exposure to green spaces has become a central focus in the integration of ecology and public health, shifting the focus from passive “downstream” health outcomes to active “upstream” health interventions (13, 31–35). By proposing the PHE threshold model, this study systematically assesses both the positive and negative effects of green space exposure on human health. The model provides a framework that not only quantifies the health benefits of green space exposure but also addresses potential adverse effects, such as the degradation of urban greenery, offering a more comprehensive understanding of the health impacts associated with green space.

The PHE threshold model plays a crucial role in guiding more scientifically informed green space exposure practices. By quantifying the PHE thresholds, the model helps maximize health benefits within a short exposure time, ensuring that residents gain the most from their time spent in green spaces (11, 12, 36, 37). Additionally, understanding the factors that influence these thresholds will assist urban planners and decision-makers in creating and maintaining effective green spaces. Overall, the model contributes to both theoretical and practical aspects of green space exposure research, providing valuable insights into the relationship between green space exposure (GE) and its associated health outcomes.

The primary goal of this study was to assess the physiological health effects of different types of urban green spaces, including remnant forests, wetland parks, and urban parks. Our results confirm that the three green space types align well with the PHE threshold model, with distinct thresholds for efficiency, benefit, and harm. In line with previous research and supporting our hypothesis (33, 34), we found that green spaces with higher naturalness, such as forests and wetlands, produce stronger physiological health benefits compared to urban parks, which show lower restorative effects. The comparison of time to reach the threshold and AUC (Table 1) revealed that participants achieved the most significant physical health recovery in the RF environment, followed by the WP environment, with the UP environment showing the lowest recovery effect. This finding further supports the notion that more natural environments are more beneficial for urban residents, especially for weekend or leisure activities aimed at stress relief (38–40).

Furthermore, consistent with other studies (36, 37), our results suggest that “blue spaces” (water-related green areas) have the highest potential for physical and psycho-logical recovery (26, 41, 42). Notably, forest environments, with their higher biological recovery potential, outperform urban green spaces in promoting health (12). This indicates that different natural environments induce varying psychophysiological recovery mechanisms, emphasizing the importance of environmental complexity. According to Attention Restoration Theory, environments with richer biodiversity and more natural stimuli are more likely to attract attention effortlessly, leading to enhanced recovery (26, 40). Additionally, the microclimates created by biodiversity—such as the sounds of water, wind, and bird calls—contribute to faster and more significant health benefits, making these environments more conducive to physical and psychological rejuvenation (9, 11, 43).

While we hypothesized that more natural environments would exhibit greater recovery potential, an interesting finding emerged: regardless of environmental type, significant restorative effects were observed within a short period. EEG indices reached the efficiency threshold within approximately 9 min of exposure, and the benefit threshold was achieved within 14 min, particularly in forest environments (11–13). The slight variation in recovery times may be attributed to the relatively small sample size. Previous research on dose–response relationships has similarly indicated that even short exposure durations can promote physiological recovery, and that repeated short-term exposures may be more beneficial than prolonged single sessions (11, 12, 40, 41). Our findings extend this evidence by demonstrating measurable physiological changes across different types of natural environments during short exposures, thereby testing the applicability of the proposed PHE threshold model in quantifying both the positive and potential negative effects of green space exposure (Table 1).

This finding of rapid physiological restoration aligns with Ulrich’s Stress Reduction Theory (SRT), which posits that exposure to nature can trigger an immediate parasympathetic response to reduce arousal (41). Our results extend this theory by quantifying the specific time-dose required. Consistent with our observation of the 3–5 min efficiency threshold, previous laboratory studies have demonstrated that physiological stress markers (such as blood pressure and muscle tension) can drop significantly within just 3–4 min of viewing natural landscapes (7–9). Furthermore, recent multisensory experiments have confirmed that combining visual inputs with natural sounds (e.g., birdsong and water flow) and olfactory cues (e.g., phytoncides) significantly accelerates this recovery process compared to visual-only exposure (8, 42).

The superior performance of Remnant Forests (RF) and Wetland Parks (WP) in our study can be attributed to this multisensory synergy. Unlike urban parks, forests and wetlands provide a richer “soundscape” and distinct microclimatic benefits. For instance, wetlands generate higher concentrations of negative air ions and stable humidity, which have been shown to regulate autonomic nervous activity and enhance mood (26, 43). Similarly, the complex vegetation structure in forests not only offers high visual naturalness but also blocks urban noise and provides a sense of “being away,” facilitating deeper immersion (11, 12).

Therefore, our study validates the application of the PHE threshold model, confirming that while all green spaces reduce stress, the “dose” required to reach the benefit threshold varies by type. Short-term exposures (as brief as 10–15 min) in high-quality natural environments are sufficient to maximize physiological benefits, a finding that has significant implications for urban planning: specifically, that accessible “pocket” natural spaces may offer immediate relief for time-constrained urban residents (11, 12, 36, 44–46).

This study has several limitations that should be addressed in future research. First, the short duration of the experiment introduced minor irregularities in the EEG data curves, making it challenging to precisely determine threshold points within the PHE model. Future studies should extend exposure durations and incorporate additional physiological indicators to enhance model accuracy and robustness. Second, controlling for unexpected outdoor disturbances was inherently limited; however, such variations represent an authentic aspect of natural exposure and contribute to the ecological validity of the findings. Importantly, no major external disturbances were reported by researchers or participants during the trials.

Furthermore, it is important to note the limited range of visual naturalness represented in the selected experimental sites. The three green space types analyzed—Urban Parks (85%), Wetland Parks (95%), and Remnant Forests (99%)—all possess relatively high levels of visual naturalness. The absence of a “low-naturalness” control group (e.g., built environments or gray spaces with minimal vegetation) restricts our ability to observe the physiological responses at the lower end of the environmental spectrum. Consequently, the threshold model derived in this study may primarily reflect the dynamics within high-quality green spaces. Future research should aim to include a broader gradient of settings, ranging from heavily urbanized areas to pristine nature, to fully characterize the dose–response relationship and identify health thresholds across diverse environmental characteristics. Finally, while this study successfully established the PHE threshold model using EEG data, we acknowledge that it focused on a single physiological dimension. Future research will expand the validation of the PHE model by incorporating additional physiological indicators, such as Heart Rate Variability (HRV) and Blood Pressure (BP). Integrating these metrics will provide a more comprehensive multidimensional perspective on the health restoration thresholds of urban green spaces. Integrating these factors will help refine the PHE threshold model and provide stronger empirical evidence for planning forest and wetland environments that optimize both ecological sustainability and public health benefits.