Affective Restoration in Bamboo Green Spaces: A Controlled Photo-Based Experiment Linking Place Structure, Visual Attention, and Electroencephalography (EEG) Responses

Abstract

Urban mental health burdens are increasing, prompting interest in how nearby green spaces aid emotional restoration. Bamboo-dominant green spaces are widespread in East Asia, but evidence connecting their management and structural features to restorative experiences is limited. This study conducted a controlled photo-exposure experiment in Ya’an, China, to examine how bamboo space typology and structural attributes relate to visual attention, affective responses, and short-term physiological recovery. One hundred and twenty participants viewed 50 photographs representing five bamboo space types (ecological conservation, productive–economic, protective–greenbelt, landscape–recreational, and understory–composite). Each image was linked to a matched field plot, enabling integration of structural indicators with eye tracking, EEG β/α, and repeated ratings of relaxation, pleasure, and preference. Results showed that landscape–recreational spaces received the highest affective ratings, while understory–composite spaces had longer fixations, indicating higher visual processing demands. Vertical stratification and groundcover coverage were robust predictors of affect beyond typology. Eye-movement metrics did not mediate structure–affect associations, and EEG β/α, as an auxiliary and context-dependent indicator under brief photo-based exposure, showed limited sensitivity. These findings offer insights into structural elements that can inform the design and management of bamboo green spaces for improved emotional restoration.

1. Introduction

Rapid urbanization has coincided with a growing mental-health burden, with anxiety-, depression- and stress-related symptoms increasingly reported in high-density living contexts and accompanied by substantial social costs [1,2]. Against this backdrop, evidence from environmental psychology and public health indicates that contact with natural environments can support stress recovery and psychological restoration [3,4,5], and participation in green exercise in wild or urban green spaces yields measurable psychological benefits [6]. In parallel, research on health-supportive environments has moved beyond the “amount of nature” to ask how and through which cues everyday scenes shape affective responses, preference, and perceived restoration [5,7,8,9,10,11]. This shift places the visual pathway at the center of inquiry: what is seen (element types, proportions, naturalness, and spatial organization) and how it is processed (fixations, saccades, and pupil responses) jointly contribute to emotional appraisal and restoration-related judgements [12,13,14,15].

Syntheses and multimodal evidence suggest that higher visible greenness (e.g., Green View Index, GVI) and more legible scene organization are often associated with greater relaxation, lower perceived stress, and stronger preference [5,11,16,17,18,19,20]. Eye tracking provides behavioral markers of attentional allocation and processing demands (e.g., fixation duration, search dynamics, and pupil responses), allowing the mechanism “visual exposure to attention to affect” to be tested with observable viewing behavior rather than inferred [12,14,15,21,22,23,24,25,26]. Physiologically, EEG offers a complementary window into the tension–relaxation continuum; exposure to greener or natural views has been linked to changes in α/β activity patterns (including findings based on passive viewing and video and VR exposure), although sensitivity can depend on exposure conditions and stimulus heterogeneity [27,28,29,30,31,32,33]. Together, these approaches enable a mechanism-oriented evaluation of the pathway from visual structure to visual behavior to affective and physiological responses, which can be translated into actionable cues for everyday nature design and management [9,10,34,35,36]. Bamboo landscapes are a particularly informative testbed in this context. Compared with many mixed forests, bamboo spaces often exhibit strong orderliness and controllable stand structure, and they are widely experienced in East Asian cities and peri-urban settings [37,38,39,40]. Importantly, different management goals can systematically reconfigure openness and permeability, vertical layering, and ground-layer conditions, producing repeatable structural contrasts that are meaningful to perception and use [39,40]. These properties make bamboo environments suitable for linking measurable structure to viewing behavior and restorative experience in a way that is potentially more design-translatable than broad “green vs. non-green” contrasts [9,10,11,37,41].

1.1. Evidence Synthesis: Perception and Structure Pathways

Prior work suggests two complementary ways through which everyday green space encounters may support affective restoration [42,43,44]. The perception pathway emphasizes visual exposure and attentional processing: what is visible and how it is inspected can shape momentary affect, preference, and restoration-related judgements, with eye tracking providing observable fixation- and pupil-based markers of viewing dynamics [45,46,47,48,49]. The structure pathway emphasizes how measurable vegetation configuration shapes scene legibility and related qualities (e.g., permeability and complexity), which are repeatedly linked to affective appraisal and restorative potential [41,50,51,52,53,54]. These pathways converge because vegetation structure constrains what is visually available and can guide how attention unfolds [55,56,57,58]. Bamboo green spaces offer a tractable context for testing this convergence, as management practices generate repeatable variation in density, vertical layering, and ground-layer conditions that are meaningful for both use and perception [59,60].

1.2. Unresolved Gaps and Guiding Questions

Despite increasing evidence that green space quality matters beyond quantity, three gaps remain for bamboo-rich urban contexts. First, bamboo places are managed for distinct functions, yet an operational typology that can be compared alongside objective structural indicators remains limited [37,41]. Second, it is unclear which stand and understory attributes act as the most influential and interpretable “levers” for affective outcomes once typology is accounted for [53,54,61]. Third, while attention is frequently proposed as a mechanism linking visual structure to affect, few studies test the structure–attention–affect chain using designs that simultaneously address repeated measures and stimulus heterogeneity [37]. Addressing these gaps is necessary to translate bamboo place characteristics into design- and management-relevant evidence.

1.3. Study Objectives

This study aims to strengthen place-based evidence on affective restoration in bamboo-rich urban spaces by linking management-defined bamboo typology with fine-scale, field-measured structural indicators and psychophysiological responses under standardized exposure conditions [37,41]. Using synchronized eye tracking, mobile EEG, and trial-level crossed mixed-effects models (participants × images), this study provides: (i) a replicable approach to specifying green space quality beyond area- or proximity-based metrics, (ii) empirical estimates of interpretable structural “levers” for affective restoration that can inform routine design and maintenance, and (iii) a mechanism-screening test of whether visual attention plausibly connects place structure to affective outcomes (Figure 1). By moving from coarse greenness measures to actionable place-quality indicators in a bamboo-dominated urban context, the findings offer practical insights on how to manage nearby nature to better support affective restoration in everyday spaces and settings with limited access to restorative environments [62,63,64,65,66].

Based on environmental psychology and restorative theory, we hypothesize that vertical stratification (VS) and groundcover coverage (GC) in bamboo spaces are positively correlated with affective restoration, where higher VS and greater GC are expected to enhance relaxation, pleasure, and preference. This hypothesis is grounded in the understanding that well-organized, layered structures provide clearer depth cues, reduce perceptual ambiguity, and facilitate positive emotional responses and preference formation.

2. Materials and Methods

2.1. Study Design and Overall Procedure

We conducted a laboratory-based photo-exposure experiment to test how management-defined bamboo space typology and measurable stand and understory attributes relate to visual attention, affective appraisal, and short-term physiological recovery. Five bamboo space types in Ya’an were defined as a priori by management and use context: ecological conservation (EC), productive–economic (PE), protective–greenbelt (PG), landscape–recreational (LR), and understory–composite (UC). We retained 10 representative photographs per type (50 stimuli total) and linked each photograph one-to-one to a matched field plot with corresponding structural indicators [21,67,68]. Sessions were conducted under controlled indoor conditions (10–30 May 2025). After informed consent and device setup, participants completed: (i) a preparation phase (resting and calibration) [13,69] and (ii) a brief traffic-noise exposure to standardize pre-viewing arousal followed by recording a post-stress EEG baseline used for baseline-corrected β/α [70]. Stress induction was achieved through the presentation of traffic noise, based on established protocols for noise-based stress induction in environmental psychology research. Previous studies have shown the effectiveness of this procedure in eliciting physiological stress responses. We treated β/α as a complementary physiological marker whose sensitivity may depend on exposure duration and stimulus heterogeneity; therefore, inference relies on trial-level models with appropriate covariate control. Participants also completed (iii) photo viewing (50 images; 15 s per image; 2 s inter-stimulus interval) with counterbalanced order [71,72], and (iv) scene-level ratings of relaxation, pleasure, and preference for each image. Eye movements and EEG were recorded continuously during viewing; each view-and-rate cycle was treated as one trial. The 15 s viewing duration was selected based on common practice in photo-based eye-tracking studies. Brief exposures in the 10–20 s range are typically used to capture stable fixation and scanning behaviors without inducing fatigue, making it ideal for repeated-measures designs. This study protocol was reviewed and approved by the Academic Ethics Committee of Sichuan Agricultural University. All procedures were conducted in accordance with research ethics principles and relevant institutional regulations (Approval No.: H20260004). The procedure is summarized in Figure 2.

2.2. Participants

Participants were recruited via campus advertisements and social media. To reduce bias associated with disciplinary training, we enrolled adults without formal education in landscape architecture or forestry. Eligibility required an age of 18–30 years, normal (or corrected-to-normal) vision and color vision, and no self-reported psychiatric history, substance dependence, or current use of psychotropic or hormonal medication. Participants were asked to avoid alcohol, strenuous exercise, and self-medication within 24 h before testing. Additional restrictions were applied to maximize eye-tracking quality (e.g., avoiding contact lenses and prominent eye makeup). In total, 120 participants completed the experiment with an approximately balanced distribution of age and sex [73,74].

2.3. Classification of Bamboo Forest Space Types

Ya’an is in the transition zone between the western Sichuan Basin and the eastern margin of the Qinghai–Tibet Plateau, characterized by pronounced topographic gradients and widespread bamboo habitats (Figure 3). Field surveys have reported more than 20 bamboo species (e.g., Bambusa emeiensis, Phyllostachys edulis, and Chimonobambusa quadrangularis), which contribute to carbon storage, microclimate regulation, and urban ecological patterning.

Conventional forest typologies defined by ecological function or management objectives (e.g., protective, economic, and landscape forests) offer limited resolution for distinguishing visual form and perceptual differences [52,64,65]. In line with recent work emphasizing the joint role of ecological function, spatial organization, and visual cues in restorative outcomes and environmental preference [10,28], we developed a perception-oriented typology based on site function and management and use context documented during field investigation. Type assignment was completed before any quantitative extraction and did not rely on the structural indicators analyzed below. We then collected stand-structure measurements and the green view index (GVI) to (i) characterize sites objectively and (ii) test whether continuous structural variation explains additional outcome variance beyond type-level contrasts.

The five types were defined as follows: EC (ecological conservation), conservation-priority stands with minimal disturbance and restricted human use; PE (productive–economic), stands primarily managed for bamboo timber or shoot production with routine silvicultural operations; PG (protective–greenbelt), belt-shaped stands along roads, rivers, or settlement edges serving buffering and boundary functions; LR (landscape–recreational), park-oriented stands managed for public access and recreation (e.g., walking, resting, viewing) [55]; and UC (understory–composite), stands integrating a bamboo overstory with systematic understory use (e.g., crop cultivation or ecological breeding) within a composite livelihood and management system. This typology provides an operational basis for comparing type-specific mechanisms linking bamboo-space attributes to visual behavior and restorative responses, and for informing health-oriented green space planning and management [55,56]. For clarity and comparability, we summarize the five bamboo space types, their primary management goals, and their expected structural characteristics in Table S1.

2.4. Visual Structural Features of Bamboo Spaces

To quantify the visual structure of bamboo spaces and relate it to visual behavior and affective and physiological responses, nine indicators were selected [41,50] (Figure 4). Stand structure was described by stand density (CDen) [75], diameter at breast height (DBH), culm height (CH) [75], height under branch (CBH), vertical stratification (VS) [50], and crown width (CDW) [76]. Understory conditions were characterized by groundcover coverage (GC) and groundcover height (UH; operationalized as the mean height of understory or groundcover vegetation within quadrats) [77]. Visual exposure was captured by the green view index (GVI) [78]. Together, these indicators represent key visual cues of spatial structure, vegetation morphology, and visible greenness that are commonly linked to perceived scene complexity and legibility [9,10,11].

Field measurements followed a standardized plot–quadrat protocol (Figure 4). At each site, a 30 m × 30 m plot was established with five 3 m × 3 m quadrats (four corners and the center) [77]. Quadrat-level GC and UH were measured in all five quadrats and averaged to derive plot-level understory descriptors. Stand-level metrics (CDen, DBH, CH, CBH, VS, and CDW) were measured at the plot scale using a consistent field procedure. GVI was computed from the final stimulus photographs as the proportion of visible green pixels within the image. All structural and visual indicators were linked one-to-one to the corresponding stimuli and were integrated with eye-tracking, EEG, and self-report outcomes in subsequent analyses.

2.5. Measurement of Behavioral, Physiological and Subjective Indices

2.5.1. Eye-Tracking Metrics

Eye movements were recorded using a wearable eye tracker ((Tobii Pro Glasses 2; Tobii Pro AB, Stockholm, Sweden)) at 100 Hz while stimuli were presented on a 17.3-inch monitor at a fixed viewing distance. We extracted five image-level metrics commonly used to characterize free-viewing behavior: fixation count (FC), total fixation time (TFT), average fixation time (AFT), saccade count (SC), and average pupil diameter (APD). To ensure accurate measures of visual attention, we applied standard outlier handling procedures to remove extreme values and performed blink-related exclusions, particularly for pupil diameter (APD). To reduce bias from low-level image properties, stimulus photometric characteristics were quantified and included as covariates where relevant [79,80,81,82,83].

2.5.2. Electroencephalography Metrics

EEG was recorded with a mobile 16-channel system ((MB8600; BoYing BI, China)) following the international 10–20 layout. For each trial, spectral power was computed over the 15 s viewing epoch; alpha and beta bands were defined as 8–13 Hz and 13–30 Hz, and β/α was baseline-corrected using the post-noise segment recorded immediately after stress induction [84,85,86]. β/α values were averaged across prefrontal and occipital channels (Fp1, Fp2, F3, F4, O1, O2). Standard preprocessing and quality control procedures were applied [85,87]. Participants were retained for EEG analyses only when a valid baseline and all 50 viewing epochs met quality criteria, yielding an analytic EEG sample of 65 participants (3250 trials). Because mobile EEG is sensitive to motion artifacts and impedance fluctuations, stringent participant-level quality criteria were applied; consequently, the analyzable EEG sample was reduced, which may decrease sensitivity to detect small-type effects under brief, static photo exposure (15 s).

2.5.3. Subjective Perception Indices

Subjective responses were assessed with three scene-level ratings: relaxation, pleasure, and preference, reflecting immediate affective appraisal during viewing [7]. Participants were instructed to imagine arriving at and staying in the depicted bamboo setting after mental fatigue and to report their immediate feelings. After the viewing phase, each image was rated individually on 7-point Likert scales: relaxation (“I feel relaxed when viewing this scene”; 1 = very tense, 7 = very relaxed), pleasure (“This environment makes me feel pleasant”; 1 = very unpleasant, 7 = very pleasant), and preference (“I like this scene and would like to stay here”; 1 = strongly dislike or unwilling to stay, 7 = strongly like or very willing to stay). These ratings were analyzed jointly with eye-tracking and EEG measures in the subsequent models.

2.6. Statistical Analysis

All analyses were conducted in Python (version 3.11), using pandas and numpy for data handling and statsmodels and SciPy for statistical modeling. Trial-level linear mixed-effects models were fitted in statsmodels, and all tests were two-tailed with α = 0.05. For RQ1, analyses were conducted at the stimulus level (N = 50 photographs). Participant responses were averaged within photographs to obtain photo-wise means. One-way ANOVAs compared the five bamboo space types (EC, PE, PG, LR, UC) for structural and visual parameters, eye-tracking metrics (TFT, AFT, FC, SC, APD), affective ratings, and the EEG β/α. Tukey’s HSD was used for post hoc comparisons. Effect sizes are reported as ω² with 95% confidence intervals [88]. For RQ2, Spearman correlations based on photo-wise means (N = 50) were computed descriptively. Primary inference used trial-level linear mixed-effects models with crossed random intercepts for participant and image [63,64,89]. Fixed effects included bamboo space type (EC as reference) and trial order; additional structural predictors were specified as a priori and z standardized. Mean luminance and RMS contrast were included in APD and β/α models. We treated baseline-corrected β/α as a complementary physiological marker; therefore, inference and interpretation primarily rely on trial-level mixed-effects estimates rather than stimulus-level ANOVA contrasts. For FC and SC, robustness was assessed using log(1 + count) LMMs and Poisson mixed models [90]. For RQ3, mediation was tested at the trial level using crossed mixed-effects models with random intercepts for participant and image and fixed effects controlling for bamboo space type (EC as reference) and z-standardized trial order; indirect effects were estimated as a × b and evaluated using 95% Monte Carlo confidence intervals based on 5000 draws, with mediation inferred when the interval excluded zero [66] (

Table 1. Measures, data levels, and model roles across RQ1–RQ3.

Domain	Variables	Data Level and Source	Role in Analyses
Typology	Type (EC/PE/PG/LR/UC)	Stimulus-level (Site-assigned a priori)	Group (RQ1) Fixed (RQ2, RQ3)
Structural	CDen, DBH, CH, CBH, VS, CDW, GC, UH, GVI	Image-level (plot measures)	Describe (RQ1) Predict (RQ2, RQ3)
Outcomes	Eye-tracking, Affective, β/α	Trial-level (participant × image)	Endpoints
Controls	Trial order (z)	Trial-level	Covariate

Note: RQ1 uses photo-wise means (N = 50); RQ2–RQ3 use trial-level LMMs (participant × image).

).

3. Results

3.1. Group Differences in Structural, Visual and Affective and Physiological Indices (RQ1)

For RQ1, outcomes were analyzed at the stimulus level (N = 50 photographs). Eye-tracking metrics, affective ratings, and EEG indices were averaged across participants within each photograph to obtain photo-wise means. Photo-wise values were compared across bamboo space types using one-way ANOVA with Tukey HSD post hoc tests (Figure 5 and Figure 6).

3.1.1. Structural Characteristics of the Five Bamboo Space Types

Because the typology was assigned a priori based on management and use context, we first examined whether the five bamboo space types could be differentiated by the measured structural and visual parameters (Figure 5). CDen, DBH, CH, CBH, VS, GC, and CDW differed among types (all p < 0.001), whereas UH and GVI did not (p = 0.506 and 0.592). Effect sizes were large, particularly for GC (ω² = 0.658, 95% CI [0.435, 0.700]), VS (ω² = 0.430, 95% CI [0.155, 0.541]), and CDen (ω² = 0.373, 95% CI [0.096, 0.498]); full ω² estimates with 95% CIs are reported in Table S2. Descriptively, PE showed the lowest density (CDen = 19.7) but the largest culm size and canopy development (DBH = 10.26 cm; CH = 14.76 m; CDW = 3.42 m). LR, UC, and PG were much denser (CDen = 183.2/152.8/147.0), whereas EC showed the highest stratification and the most continuous ground layer (VS = 3.0; GC = 80.6%). UC combined high density with sparse groundcover (GC = 10.5%). The non-significant UH and GVI differences indicate broadly comparable understory height and visible greenness across the stimulus set.

3.1.2. Eye-Tracking Indices

At the stimulus level, bamboo space type differed for fixation-based indices (Figure 6). TFT and AFT varied across types (TFT: F = 4.043, p = 0.007; AFT: F = 4.837, p = 0.002), and FC showed a smaller but significant effect (F = 2.656, p = 0.045). Tukey HSD comparisons indicated the longest TFT in UC (with LR also relatively high) and the longest AFT in LR; PE and EC tended to be lower, with PG intermediate (Figure 6). APD and SC did not differ across types (both p > 0.10). Importantly, higher TFT and AFT should not be equated with a more restorative experience; under free viewing, these indices can reflect either effortful parsing of occluded and information-dense scenes or sustained, fluent viewing of coherent scenes, and therefore should be interpreted alongside affective ratings and structural context.

3.1.3. Affective Ratings and β/α Ratio

Stimulus-level type differences in EEG β/α are reported descriptively; inferential conclusions are based on trial-level crossed mixed-effects models that account for participant- and image-level heterogeneity. Affective ratings showed a consistent typological gradient (Figure 6): relaxation, pleasure, and preference were highest in LR and lowest in UC (Tukey HSD). At the stimulus level, β/α also differed among types (F = 63.507, p < 0.001), with higher photo-wise means in UC and lower means in LR.

To evaluate whether this contrast persisted once trial-level variability was modeled, we fitted crossed random-intercepts mixed-effects models (participants × images). Adding bamboo space type (M1) and additional covariates (M2) provided no evidence of improved fit relative to the intercept-only model (M0), as indicated by higher AIC and BIC (

Table 2. Model comparison for trial-level mixed-effects models predicting baseline-corrected EEG β/α (participants × images).

Model	Fixed Effects (Added Sequentially)	n_obs	AIC	BIC
M0	Intercept only	3250	18,607.57	18,656.26
M1	M0 + Type	3250	18,609.26	18,664.04
M2	M1 + Trial order + Mean luminance + RMS contrast	3250	18,610.70	18,677.65

Notes: M0 included an intercept only; M1 added bamboo space type; M2 further added trial order and photometric covariates (mean luminance and RMS contrast). All models included crossed random intercepts for participants and images and were fitted using maximum likelihood. Lower AIC and BIC values indicate better fitness. n_obs denotes the number of trials included in model fitting.

). Consistently, in the trial-level model, fixed effects of type, trial order, and photometric covariates (mean luminance and RMS contrast) were not statistically reliable for β/α (all p > 0.14;

Table 3. Trial-level crossed random-intercepts mixed-effects model predicting EEG β/α ratio.

Predictor	Estimate (B)	SE	95% CI (Lower)	95% CI (Upper)	z	p
Intercept	1.244	0.169	0.914	1.575	7.38	<0.001
LR (vs. EC)	0.163	0.250	−0.326	0.653	0.65	0.513
PE (vs. EC)	0.253	0.245	−0.227	0.734	1.03	0.302
PG (vs. EC)	0.203	0.235	−0.258	0.663	0.86	0.388
UC (vs. EC)	0.043	0.238	−0.424	0.510	0.18	0.858
Trial order (z)	0.066	0.081	−0.092	0.225	0.82	0.412
Mean luminance (z)	0.129	0.105	−0.078	0.335	1.22	0.221
RMS contrast (z)	−0.005	0.101	−0.204	0.193	−0.05	0.957

Notes: Values are fixed-effect estimates (B) with standard errors (SE), Wald z statistics, p values, and 95% confidence intervals. Bamboo space type was dummy coded with EC as the reference category. Trial order, meaning luminance, and RMS contrast were z standardized. The model included crossed random intercepts for participants and images and was fitted by maximum likelihood.

). This null pattern is consistent with (i) the context-dependent sensitivity of β/α under brief (15 s) static photo exposure, (ii) substantial image-level heterogeneity (see variance partitioning below), and (iii) reduced statistical sensitivity to small-type effects due to the reduced analyzable EEG sample after stringent quality control. Figure 7 shows substantial within-type dispersion in image-wise β/α, including a small number of extreme photographs. Variance partitioning indicated that image-level heterogeneity accounted for 67.6% of total β/α variance, between-participant variance for 0.03%, and residual trial-level variance for 32.4% (Table S3). Accordingly, the stimulus-level between-type differences in β/α are treated as descriptive patterns shaped by within-type, image-specific variability, whereas inference in subsequent analyses involving β/α relies on the mixed-effects estimates.

3.2. Associations Between Visual Structure, Visual Behavior and Affective and Physiological Responses (RQ2)

3.2.1. Descriptive Associations and Redundancy Diagnostics

Stimulus-level Spearman correlations based on photo-wise means (N = 50) were used to summarize bivariate patterns and screen for redundancy among structural indicators (Figure 8). Several size-related stand metrics were strongly intercorrelated, supporting parsimonious trial-level specifications. In the outcome domain, CDen correlated with TFT (ρ = 0.355) and AFT (ρ = 0.304), DBH correlated with APD (ρ = 0.295), and UH correlated negatively with AFT (ρ = −0.338) (Figure 8). VS and GC showed the clearest affective links (ρ = 0.43–0.45 with relaxation, pleasure, and preference) and were inversely related to β/α (VS: ρ = −0.455; GC: ρ = −0.515). These correlations are reported descriptively; inference for RQ2 is based on the trial-level mixed-effects models below.

3.2.2. Trial-Level Mixed-Effects Models for Visual Behavior and Physiological Responses

RQ2 was evaluated using trial-level linear mixed-effects models (LMMs) with crossed random intercepts for participant and image, and fixed effects for type (EC as reference), trial order, and pre-specified structural predictors (

Table 4. Trial-level mixed-effects models predicting eye-tracking and physiological outcomes.

Outcome	Predictor	β	95% CI	p
TFT	Type: LR vs. EC	12.095	[−336.273, 360.463]	0.946
TFT	Type: PE vs. EC	45.866	[−324.321, 416.053]	0.808
TFT	Type: PG vs. EC	−166.773	[−499.691, 166.144]	0.326
TFT	Type: UC vs. EC	209.473	[−125.867, 544.813]	0.221
TFT	CDen (z)	68.296	[−69.066, 205.659]	0.33
TFT	Trial order (z)	−23.074	[−134.01, 87.863]	0.684
AFT	Type: LR vs. EC	68.331	[10.293, 126.369]	0.021
AFT	Type: PE vs. EC	18.582	[−43.078, 80.241]	0.555
AFT	Type: PG vs. EC	40.45	[−14.995, 95.894]	0.153
AFT	Type: UC vs. EC	67.701	[10.224, 125.177]	0.021
AFT	CDen (z)	10.075	[−12.871, 33.02]	0.389
AFT	UH (z)	−3.643	[−21.719, 14.433]	0.693
AFT	Trial order (z)	−5.959	[−24.672, 12.755]	0.533
FC	Type: LR vs. EC	−0.719	[−1.412, −0.025]	0.042
FC	Type: PE vs. EC	−0.123	[−0.863, 0.616]	0.744
FC	Type: PG vs. EC	0.097	[−0.568, 0.762]	0.775
FC	Type: UC vs. EC	−0.104	[−0.773, 0.565]	0.761
FC	CDen (z)	0.143	[−0.134, 0.42]	0.312
FC	Trial order (z)	−0.01	[−0.035, 0.014]	0.411
SC	Type: LR vs. EC	−0.014	[−0.072, 0.043]	0.632
SC	Type: PE vs. EC	−0.023	[−0.084, 0.038]	0.46
SC	Type: PG vs. EC	−0.003	[−0.058, 0.052]	0.917
SC	Type: UC vs. EC	0.007	[−0.049, 0.064]	0.8
SC	CDen (z)	−0.001	[−0.024, 0.022]	0.916
SC	Trial order (z)	−0.079	[−0.097, −0.061]	<0.001
APD	Type: LR vs. EC	11.261	[−11.229, 33.75]	0.326
APD	Type: PE vs. EC	−14.939	[−41.841, 11.964]	0.276
APD	Type: PG vs. EC	−2.428	[−21.808, 16.952]	0.806
APD	Type: UC vs. EC	−7.486	[−30.062, 15.091]	0.516
APD	DBH (z)	4.613	[−4.017, 13.242]	0.295
APD	Trial order (z)	−5.521	[−12.093, 1.051]	0.1
APD	Mean luminance (z)	1.899	[−6.703, 10.501]	0.665
APD	RMS contrast (z)	−0.744	[−9.039, 7.552]	0.861
β/α	Type: LR vs. EC	0.229	[−0.306, 0.764]	0.402
β/α	Type: PE vs. EC	0.422	[−0.159, 1.002]	0.154
β/α	Type: PG vs. EC	−0.006	[−0.659, 0.646]	0.986
β/α	Type: UC vs. EC	−0.104	[−0.906, 0.699]	0.8
β/α	VS (z)	0.087	[−0.241, 0.414]	0.604
β/α	GC (z)	−0.172	[−0.61, 0.266]	0.442
β/α	CDW (z)	−0.148	[−0.353, 0.056]	0.156
β/α	Trial order (z)	0.064	[−0.102, 0.23]	0.449
β/α	Mean luminance (z)	0.158	[−0.053, 0.368]	0.142
β/α	RMS contrast (z)	0.024	[−0.18, 0.228]	0.815

Notes: Trial-level LMMs with cross random intercepts for participant and image. Type was dummy coded with EC as the reference. Continuous predictors were z standardized. Photometric covariates (meaning luminance; RMS contrast) were included for APD and β/α. n = 3250 trials for all outcomes.

; Figure 9a). Mean luminance and RMS contrast were additionally included for APD and β/α. Across outcomes, models were fitted to n = 3250 trials. After adjustment, CDen was not associated with TFT, FC, or SC (all p > 0.22), whereas AFT was higher in LR and UC than in EC and FC was slightly lower in LR (Table 4; Figure 9a). As noted above, higher AFT and TFT should not be interpreted as inherently more restorative and is therefore discussed jointly with affective ratings. SC decreased across trials (p < 0.001). Sensitivity analyses for FC and SC using log(1 + count) LMMs and Poisson mixed models yielded consistent conclusions (Table S4). No fixed effects were detected for APD (all p ≥ 0.10) or β/α (all p > 0.14) once covariates were accounted for.

3.2.3. Trial-Level Mixed-Effects Models for Affective Responses

Affective ratings showed robust type contrasts and additional structure-related associations (

Table 5. Trial-level mixed-effects models predicting affective ratings.

Outcome	Predictor	β	95% CI	p
Relaxation	Type: LR vs. EC	1.02	[0.88, 1.15]	<0.001
	Type: PE vs. EC	0.25	[0.11, 0.39]	<0.001
	Type: PG vs. EC	0.26	[0.09, 0.43]	0.003
	Type: UC vs. EC	0.09	[−0.13, 0.31]	0.444
	VS (z)	0.17	[0.08, 0.25]	<0.001
	GC (z)	0.02	[−0.09, 0.13]	0.730
	Trial order (z)	−0.10	[−0.15, −0.06]	<0.001
Pleasure	Type: LR vs. EC	1.15	[1.01, 1.28]	<0.001
	Type: PE vs. EC	0.33	[0.19, 0.47]	<0.001
	Type: PG vs. EC	0.34	[0.17, 0.51]	<0.001
	Type: UC vs. EC	0.30	[0.08, 0.52]	0.007
	VS (z)	0.15	[0.07, 0.24]	<0.001
	GC (z)	0.11	[−0.00, 0.22]	0.058
	Trial order (z)	−0.09	[−0.13, −0.04]	<0.001
Preference	Type: LR vs. EC	1.29	[1.14, 1.45]	<0.001
	Type: PE vs. EC	0.29	[0.13, 0.45]	<0.001
	Type: PG vs. EC	0.37	[0.18, 0.57]	<0.001
	Type: UC vs. EC	0.39	[0.14, 0.64]	0.002
	VS (z)	0.11	[0.02, 0.21]	0.023
	GC (z)	0.20	[0.08, 0.33]	0.002
	Trial order (z)	−0.14	[−0.19, −0.09]	<0.001

; Figure 9b). Relative to EC, LR showed the largest positive contrasts across outcomes. VS was a positive predictor of relaxation, pleasure, and preference after controlling for type and trial order, while GC contributed primarily to preference. Trial order was negatively associated with all affective ratings (all p < 0.001). Model-based EMMs are shown in Figure 10 and partial effects in Figure 11.

3.3. Mediation Analyses (RQ3)

Mediation was tested at the trial level using crossed mixed-effects models. Across the primary structural cues (VS and GC) and affective outcomes, indirect effects via AFT were small and their 95% Monte Carlo confidence intervals (5000 draws) included zero (

Table 6. Indirect effects (ab) for AFT mediating associations between structural cues (VS, GC) and affective ratings (trial-level mixed-effects mediation).

Predictor (X)	Outcome (Y)	Indirect Effect, ab	95% CI for ab	95% CI Includes 0
VS	Relaxation	6.0 × 10−5	[−1.34 × 10−3, 1.59 × 10−3]	Yes
VS	Pleasure	1.0 × 10−5	[−1.46 × 10−3, 1.49 × 10−3]	Yes
VS	Preference	2.0 × 10−4	[−1.31 × 10−3, 2.13 × 10−3]	Yes
GC	Relaxation	6.0 × 10−5	[−1.58 × 10−3, 1.85 × 10−3]	Yes
GC	Pleasure	2.0 × 10−5	[−1.70 × 10−3, 1.73 × 10−3]	Yes
GC	Preference	1.7 × 10−4	[−1.61 × 10−3, 2.43 × 10−3]	Yes

Notes: Indirect effects were estimated as the product of coefficients (a × b). The mediator model regressed AFT on X and covariates; the outcome model regressed Y on AFT and X, controlling bamboo space type (EC as reference) and z-standardized trial order, with crossed random intercepts for participant and image. Continuous predictors (VS, GC, AFT) were z standardized. Indirect effects were estimated as the product of coefficients (a × b). Confidence intervals for the indirect effect were computed using a Monte Carlo method with 5000 draws. Mediation was considered supported when the 95% confidence interval excluded zero.

), providing no evidence of mediation, and adding AFT did not materially change the corresponding structure–affect estimates. Extending the same tests to alternative mediators (TFT, FC, SC, and APD) and to type contrasts (each type vs. EC) yielded the same null pattern (Table S5). Importantly, this lack of statistically supported mediation should be interpreted as limited evidence for the specific hypothesized structure–attention–affect pathway under the present design and operationalization, rather than as evidence that eye-tracking measures are uninformative. Trial order showed systematic associations with several oculomotor indices and modest shifts in affective ratings; it was retained as a covariate to account for within-session adaptation.

4. Discussion

4.1. Differences in Visual Behavior and Affective Physiological Responses Across Bamboo Space Types

Clear typological contrasts emerged in affective experience. LR scenes consistently elicited higher relaxation, pleasure, and preference, whereas UC scenes tended to score lowest on these ratings. A mechanism consistent with both the typological definition and the measured structure is that LR settings typically support more fluent scene appraisal by offering clearer spatial organization and visual access—features that align with higher legibility and coherence cues [18,34,46,91,92]—whereas UC scenes often combine high stand density with understory-integrated use, which can introduce more occlusion, local heterogeneity, and fragmented depth cues that increase perceptual parsing demands and dampen affective evaluation [14,48,93,94,95].

Typological differences were also expressed in fixation-based eye-tracking metrics, but their direction did not map one-to-one onto affect. At the stimulus level, fixation duration measures (TFT and AFT) varied across types, indicating systematic differences in how viewers allocated attention among bamboo settings. Importantly, trial-level models clarified that the most reliable gaze patterns were selective: AFT was higher in LR and UC relative to EC, FC was slightly lower in LR, and SC showed a robust decline with trial order rather than stable type-specific contrasts. This configuration suggests that longer fixations can reflect at least two qualitatively different processes under free viewing: (i) effortful parsing in visually occluded or information-dense scenes (plausibly more common in UC), and (ii) sustained, stable viewing in scenes that are aesthetically coherent and easy to integrate (plausibly more common in LR). Consequently, fixation prolongation should not be interpreted as inherently “better” or “worse” without considering concurrent affective reports and the structural context that likely drives the processing mode [14,96,97].

Physiologically, EEG β/α did not show reliable space-type effects in trial-level mixed models, and variance partitioning indicated that β/α variability was dominated by image-level heterogeneity and residual trial-wise noise rather than typology. This pattern is consistent with the broader methodological insight that mobile EEG indices during brief exposures may be more responsive to coarse contextual contrasts than to fine-grained structural differentiation among natural scenes, particularly when stimulus-specific variance is strong [98,99,100,101]. Accordingly, in the present trial-level design β/α is most defensible as an auxiliary physiological signal interpreted alongside self-report and gaze. Notably, the co-occurrence of type differences in affect and gaze should not be taken as evidence of attentional mediation; this mechanism is evaluated directly in Section 4.3.

4.2. Structural Attributes as Design Levers Beyond Bamboo Space Type

Beyond broad contrasts between bamboo space types, trial-level models indicated that a small set of structural attributes—specifically vertical stratification (VS) and groundcover coverage (GC) explained incremental variance in affective outcomes after accounting for type and trial order. VS showed a robust positive association with relaxation, pleasure, and preference, whereas GC contributed most clearly to preference, with weaker and less consistent evidence for pleasure. These results imply that, within an already bamboo-dominated (high-greenness) stimulus set, how vegetation is organized vertically and at the ground layer is more informative for affective appraisal than further increasing visible greenness per se [19,20].

Mechanistically, higher VS likely provides clearer depth cues and more organized near–mid–far layering, supporting coherent scene interpretation and facilitating positive affect and preference. Consistent with information-processing accounts, preference benefits of richness and complexity tend to depend on perceived order and organization [36,44,53,98,99]. In contrast, GC operates via pedestrian-level cues: a continuous but visually controlled ground layer can signal care and maintenance while preserving visual access, whereas sparse or highly heterogeneous ground conditions may increase ambiguity in walkability and safety cues [18,19,102]. This distinction may help explain why GC effects concentrated on preference (an evaluative “like–dislike” judgment) while evidence for pleasure was comparatively weaker. After redundancy control and simultaneous adjustment for space type and key structural cues (especially VS and GC), geometric descriptors did not emerge as stable incremental levers. This does not imply these features are unimportant; rather, their informative content is partly shared with typology and with VS and GC, so unique contributions may be attenuated under multicollinearity and become specification-sensitive [103,104]. Accordingly, a non-significant increment should be interpreted as “not uniquely identifiable under the current specification,” rather than “no effect” [104].

Finally, trial order showed a consistent negative association with affective ratings, consistent with habituation and fatigue during rapid, repeated exposure. This suggests that even structurally high-quality bamboo scenes may yield diminishing marginal affective gains when experienced monotonously, highlighting the value of episodic variation in layering or ground-layer texture along routes or within sites.

4.3. Attention as a Mediator Linking Structure to Affective Outcomes

Given the robust associations of bamboo space type and key structural cues (VS and GC) with affective ratings, we tested the proposed attentional pathway using trial-level crossed mixed-effects mediation within the same framework as RQ2 (random intercepts for participant and image). Across VS and GC and all three affective outcomes, indirect effects were consistently small with confidence intervals spanning zero, and inclusion of gaze mediators produced negligible attenuation of the structure–affect estimates. This suggests that, once participant dependence and image-specific heterogeneity were modeled, the structure-related affective advantages did not operate through systematic shifts in global scene-level gaze duration, fixation frequency, or overall scanning intensity.

A parsimonious interpretation is that affective appraisal in bamboo scenes may rely primarily on rapid, holistic evaluation of scene organization and meaning, as emphasized in classic restorative and preference frameworks that foreground fluent processing and low effort [11,43,44]. Global metrics aggregated over a 15 s interval (e.g., mean fixation duration or total fixation time) are relatively coarse and can reflect both “soft fascination” and increased processing demand depending on context; in dense natural scenes, longer fixations may arise from occlusion, clutter, or reduced visual permeability rather than uniquely positive engagement [15,26,49,105]. Therefore, attention-related pathways may be stimulus-contingent and better captured by mechanism-matched mediators (e.g., AOI-based dwell time, early vs. late viewing phases, or scan-path organization) under longer or more immersive exposure [33,46].

4.4. Implications, Limitations, and Future Research Directions

These findings have practical implications for bamboo-space planning and management in bamboo-rich urban contexts. First, function- and management-based typologies provide a useful operational starting point for describing bamboo settings, but the results also motivate within-type structural tuning rather than a single-minded pursuit of visible greenness. In design terms, prioritizing clearer vertical organization (higher VS) and maintaining a continuous—but not visually overwhelming—ground layer (GC) appears more consistently linked to positive affective appraisal than simply maximizing overall green proportion within already green-dominant scenes. Second, the results underscore the potential value of everyday, accessible bamboo spaces for emotional well-being; along frequently used corridors, combining layered structure, coherent ground conditions, and clear sightlines may help transform routine exposure into more reliably positive experiences. Third, although global eye-tracking metrics did not mediate structural effects on affect, the integrated use of eye tracking, EEG, and self-reports provides a practical multi-method toolkit for evaluating bamboo scenes and flagging configurations that may induce visual fatigue or diminished comfort already at the design stage.

Several limitations temper interpretation and point to specific next steps. First, stimuli were static photographs rather than in situ experiences or immersive environments. While this enabled strict control over composition and a precise one-to-one linkage between measured structure and visual exposure, it cannot reproduce multisensory aspects of bamboo stands (microclimate, soundscapes, bodily movement) that are known to shape affective and physiological outcomes. Second, the sample consisted of non-specialist university students from a single institution, which reduces confounding by professional training but limits generalizability across age groups, residents with long-term familiarity, and professional users. Third, indicators were intentionally parsimonious: baseline-corrected β/α was used as the primary EEG outcome, and global image-level eye-tracking metrics were emphasized, which may be insufficiently sensitive to fine-grained attentional mechanisms [106]. Finally, the structural set focused on geometric configuration and visible greenness; biodiversity, seasonality, and cultural meaning—factors repeatedly linked to nature-related benefits—were not explicitly modeled. Moreover, the cultural context of the study (Ya’an, China) may limit the generalizability of the findings to other regions with different cultural perceptions of bamboo landscapes. Bamboo landscape perception may vary across different cultures, and further research is needed to validate these findings in other cultural contexts [7,93,107].

Future research should validate these findings under more naturalistic exposure (field or immersive VR) and in broader community samples. Mechanistic tests may benefit from combining time-resolved or ROI-based eye-tracking analyses with additional physiological markers, and from incorporating biodiversity, seasonality, and culturally salient cues. Longitudinal or intervention designs (e.g., before–after management changes along bamboo corridors) would further strengthen causal inference and increase relevance for place-based planning decisions.

5. Conclusions

Using a controlled photo-exposure design with field-to-image linkage, this study shows that management-defined bamboo space types differ reliably in affective appraisal, and that interpretable structural attributes help explain additional variance beyond typology. Landscape–recreational scenes elicited the highest relaxation, pleasure, and preference, whereas understory–composite scenes were consistently lowest and showed longer fixation durations, suggesting greater visual processing demands. Across models, vertical stratification and groundcover coverage emerged as robust “levers” associated with better affective outcomes. In contrast, baseline-corrected EEG β/α and global eye-movement summaries did not provide consistent mechanistic mediation, implying that affective appraisal may be driven more by holistic scene evaluation under brief, static exposure. Together, the findings translate bamboo place structure into actionable targets for routine design and management aimed at supporting everyday affective restoration.