Context dependent sensory integration modulates landscape preference in urban parks Authors: Yonghong Gan, Jiayi Li, Jiaqi Huang, Zhengbin Chen, Zhibiao Chen Published: 2026 ABSTRACT Rapid urbanization positions urban parks as critical sensory infrastructure for psychological restoration. However, traditional landscape planning remains constrained by a visually dominant paradigm, often overlooking the non-linear, multisensory dynamics of human perception in complex environments. This study proposes and tests a Context-Dependent Attribute-based Holistic Perception Model (Path B) to investigate how auditory perception and Audiovisual Harmony (AV-Harmony) drive landscape preferences across functionally distinct zones. Utilizing a pooled analysis of three representative park typologies in China (N = 472), we compared the model explanatory efficacy of the traditional Element-based Aggregation Model (Path A) versus the Attribute-based Holistic Perception Model (Path B). Unlike the static visual-centric approach, our results—verified by robust GEE estimates—reveal a dynamic, context-adaptive integration mechanism. The Attribute-based Holistic Perception Model (Path B, which assumes heuristic aggregation of perceived attributes into higher-order holistic properties) demonstrated superior efficiency, explaining significantly more variance than element-based models (Path A) (e.g., Adj. R² = 0.526 vs. 0.098) by capturing holistic atmospheres rather than discrete objects. Our empirical findings support a model of functional perceptual modulation (conceptually linked to sensory gating), characterized by a context-adaptive shift in sensory weighting. In restorative Resting Zones, the results reveal an Acoustic Filtering priority (B = 0.343, p < 0.001), where soundscape quality is strongly associated with overall satisfaction; conversely, in social Activity Zones, Audiovisual Harmony (AV-Harmony) emerges as a particularly strong predictor (B = 0.566), prioritizing semantic congruence over absolute sensory quality via processing fluency. Furthermore, soundscape quality appears to act as a sensory aesthetic leveler, contributing to the attenuation of visual disparities across park tiers and potentially promoting sensory equity. These findings suggest the value of shifting emphasis from conventional Greening to sensory Tuning, offering a cost-effective and evidence-based framework for designing more restorative, inclusive, and resilient urban health infrastructure.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-52401-1. Introduction The 21st century is fundamentally characterized by rapid urbanization, with over 68% of the global population projected to reside in cities by 20501,2. This densification concentrates environmental stressors—such as chronic noise, heat island, and visual disorder—which are increasingly linked to rising prevalence of psychological and cardiovascular disorders3,4. In this context, urban parks have become critical sensory infrastructure for psychological restoration and public well-being, directly relevant to the achievement of the UN Sustainable Development Goals (SDGs), particularly SDG 3 and SDG 115. Extensive research has demonstrated that urban parks provide important Cultural Ecosystem Services (CES), especially stress reduction and restorative benefits6–8. However, as urban land becomes scarcer and more expensive, the planning challenge has shifted from merely increasing green space quantity to optimizing its restorative efficiency9. In high-density cities, sensory chaos—characterized by chronic noise pollution or visual clutter—often compromises the functional benefits of green spaces, leading to a sensory mismatch10. This phenomenon is particularly acute in Asian urban contexts, where parks are frequently encircled by arterial roads, creating a constant conflict between visual greenery and “acoustic grayness”11,12. Addressing these inequities requires cost-effective, inclusive design strategies that enhance well-being without prohibitive capital investment13. Despite the multisensory nature of human experience, landscape research remains constrained by a visual-centric paradigm14. Traditional evaluations prioritize visual metrics—such as the Green View Index (GVI) or vegetation coverage—to predict preference15,16. Conversely, the Soundscape—defined by ISO 12913 as the acoustic environment as perceived or experienced by people, in context—emphasizes the subjective human response over physical decibel levels17. The standard highlights that soundscape perception is multidimensional, involving qualitative attributes such as tranquility and vibrancy that complement or even negate visual benefits11,18. A critical theoretical gap exists in how sensory inputs are integrated. Prevailing Element-based Aggregation Models (EAM, hereafter Path A) typically assume that landscape preference is a linear sum of discrete physical objects, such as the count of trees or water features19,20. However, Path A fails to account for Attribute Perception—the process by which the human brain uses heuristic processing to aggregate sensory data into emergent, holistic properties (e.g., Harmony or Tranquility) to reduce cognitive load21,22. To address these complexities, this study proposes the Context-Dependent Sensory Integration Framework (CDSIF), integrating Attention Restoration Theory (ART) and Cognitive Load Theory (CLT). Following a Purposeful Stratified Sampling strategy validated in urban morphology research23,24, we selected three park typologies in Zhangzhou, China, representing global archetypes of contemporary green spaces. As illustrated in Fig. 1, this framework posits a mechanism of Functional Perceptual Modulation, where the functional context acts as a moderator between environmental input and overall preference. Additionally, the framework anticipates a potential pattern of sensory aesthetic leveling, whereby soundscape quality may contribute to stabilizing overall landscape preference despite variations in visual quality across park typologies. Based on this framework, we specifically hypothesize that: (1) In Restorative Contexts (Resting Zones), the auditory system is hypothesized to act as a Primary Acoustic Filter, contributing to an Acoustic-Dominance Pattern where soundscape quality is prioritized in association with the “soft fascination” required for restoration25,26; (2) In Social Context (Activity Zones), Audiovisual Harmony (AV-Harmony) is hypothesized to optimize processing fluency through a Cross-modal Pattern, supporting semantic congruence between sight and sound27,28; (3) Furthermore, we argue that the Attribute-based Holistic Perception Model (Path B), which assumes heuristic aggregation of perceived qualities into higher-order atmospheric attributes, offers superior explanatory efficiency compared to the traditional Element-based Aggregation Model (Path A), which relies on the linear summation of discrete features29,30. Audiovisual Harmony (AV-Harmony) is operationalized herein as a specific cross-modal attribute within the Path B framework, rather than the model itself. By employing Generalized Estimating Equations (GEE) to account for spatial clustering, this study answers three core questions: (1) Do the drivers of landscape preference shift dynamically between distinct functional zones? (2) Does the proposed Attribute-based Model (Path B) significantly outperform the element-based model (Path A)? (3) What are the relative, context-dependent contributions of Visual Landscape Preference (VLP), Soundscape Preference (SP), and Audiovisual Harmony (AV-Harmony) to Overall Landscape Preference (OLP)? Fig. 1The conceptual framework of context-dependent sensory integration. This framework contrasts the traditional Element-based Aggregation Model (Path A: linear summation of discrete physical features) with the proposed attribute-based holistic perception model (Path B: heuristic aggregation of perceived attributes into emergent holistic properties, such as harmony and tranquility). Within Path B, audiovisual harmony (AV-Harmony) is incorporated as a specific cross-modal integrative attribute rather than the model itself. The functional perceptual moderator (diamond) modulates the integration of Perceived Visual Landscape Attributes (PVLA), perceived soundscape attributes (PSA), and AV-Harmony into a context-adaptive pattern (acoustic-dominance pattern in resting zones versus cross-modal congruence pattern in activity zones). The conceptual framework of context-dependent sensory integration. This framework contrasts the traditional Element-based Aggregation Model (Path A: linear summation of discrete physical features) with the proposed attribute-based holistic perception model (Path B: heuristic aggregation of perceived attributes into emergent holistic properties, such as harmony and tranquility). Within Path B, audiovisual harmony (AV-Harmony) is incorporated as a specific cross-modal integrative attribute rather than the model itself. The functional perceptual moderator (diamond) modulates the integration of Perceived Visual Landscape Attributes (PVLA), perceived soundscape attributes (PSA), and AV-Harmony into a context-adaptive pattern (acoustic-dominance pattern in resting zones versus cross-modal congruence pattern in activity zones). Methods Study areas and typological site selection strategy To ensure generalizability, this research employed a Purposeful Stratified Sampling Strategy23,24 across three distinct urban parks in Zhangzhou, China, reflecting global typologies of contemporary green spaces (Fig. 2): (1) Zhongshan Park (Historical-Cultural): A 4.33-ha traditional civic hub. Visually defined by a century-old canopy of Camphor trees (Cinnamomum camphora); acoustically characterized by a complex, human-dominated soundscape (45–70 dB); (2) Jiulong Park (Comprehensive Water-Feature): A 9.9-ha modern recreational space with a 40% water surface ratio. Features open vistas and a composite soundscape (50–75 dB) where geophony (fountains) blends with anthropogenic activity; (3) Zhishan Park (Ecological-Mountainous): A 7.72-ha Urban Forest refuge on the urban fringe. Defined by topographic elevation and native vegetation, offering a “Hi-Fi” natural sanctuary (40–60 dB) dominated by biophony (cicadas, birdsong). Concurrent with the questionnaire surveys, investigators recorded the occurrence frequencies of soundscape elements during five-minute observation periods and simultaneously measured sound pressure levels using an AWA6218A noise meter positioned 1.5 m above ground level at each sampling site31. These dB ranges and qualitative characterizations serve only as contextual background to illustrate the acoustic environments of the three park typologies and were not included as quantitative predictors in any statistical models. Within these parks, 18 survey points were established and categorized into Resting Zones (n = 10) (secluded pavilions, forest trails) and Activity Zones (n = 8) (entrance plazas, fitness corners) based on intended design function and environmental affordances (Fig. 2). This setup allows for the analysis of “sensory mismatch” which is particularly acute in high-density Asian cities, where parks are often encircled by arterial roads, creating conflict between visual greenery and acoustic grayness11,12. Fig. 2Park locations and sampling point distribution. Park locations and sampling point distribution. Measurement variables and instrument development The measurement instrument was developed through a three-stage process, including a systematic review, expert consultation, and a pilot study (N = 50). Exploratory Factor Analysis (EFA) with Varimax rotation refined the scales, resulting in the removal of two items (“Color Richness” and “Vegetation Layers”) due to low communality or cross-loading (Table S1). The questionnaire comprised four parts (Table 1): (1) Element-based Aggregation (Path A): Participants rated 14 specific features (7 visual, 7 acoustic) using a 5-point Likert scale. This model assumes preference is a linear summation of discrete physical objects19,20; (2) Attribute-based Holistic Perception (Path B): This part evaluated Perceived Visual Landscape Attributes (PVLA) and Perceived Soundscape Attributes (PSA). PVLA (9 items) assessed holistic qualities (e.g., naturalness, visual harmony)31,32, demonstrating high consistency (α = 0.863). PSA (5 items: tranquility, vividness, richness, harmony, and disturbance) assessed qualitative dimensions. These items were adapted from the subjective soundscape indicators outlined in ISO 12913-233, with tranquility and harmony primarily corresponding to the pleasantness component, vividness and richness reflecting eventfulness, and disturbance capturing negative valence. This operationalization aligns with the standard’s emphasis on context-dependent human perception rather than purely physical acoustic metrics(α = 0.842); (3) Audiovisual Harmony (AV-Harmony): Measured via a single-item scale (1 = very discordant, 5 = very harmonious) as the primary proxy for multisensory congruence within the functional perceptual modulation analysis; (4) Overall Preferences: Outcome measures included VLP, SP and OLP, rated on a 10-point scale to mitigate the “ceiling effect”. Table 1Evaluation items for landscape element preference and perceived environmental attributes.CategoryDimensionPreference itemsScaleLandscape element preferenceVisual elementsLake water, lawn, trees/shrubs, rocks, sculptures, buildings, pavilions/corridors1–5 LikertSound elementsWater sounds, bird songs, singing, talking, playing sounds, traffic sounds, broadcast sounds1–5 LikertPerceived environmental attributesPerceived visual landscape attributes (PVLA)Naturalness, spatial openness, plant diversity, color contrast, plant coverage, architectural density, topographic undulation, visual harmony, tidiness1–5 LikertPerceived soundscape attributes (PSA)Tranquility, vividness, richness, harmony, disturbance (inverse coded)1–5 LikertAudiovisual harmonyAV-harmony“How well does the sound match the visual scenery?”1–5 LikertOverall preferenceOutcome measuresVisual landscape preference (VLP), soundscape preference (SP), overall landscape preference (OLP)1–10 ScaleNote: PSA dimensions were selected and adapted to reflect core perceptual attributes in ISO 12913-2, focusing on subjective qualities experienced in context. Table 1Evaluation items for landscape element preference and perceived environmental attributes.CategoryDimensionPreference itemsScaleLandscape element preferenceVisual elementsLake water, lawn, trees/shrubs, rocks, sculptures, buildings, pavilions/corridors1–5 LikertSound elementsWater sounds, bird songs, singing, talking, playing sounds, traffic sounds, broadcast sounds1–5 LikertPerceived environmental attributesPerceived visual landscape attributes (PVLA)Naturalness, spatial openness, plant diversity, color contrast, plant coverage, architectural density, topographic undulation, visual harmony, tidiness1–5 LikertPerceived soundscape attributes (PSA)Tranquility, vividness, richness, harmony, disturbance (inverse coded)1–5 LikertAudiovisual harmonyAV-harmony“How well does the sound match the visual scenery?”1–5 LikertOverall preferenceOutcome measuresVisual landscape preference (VLP), soundscape preference (SP), overall landscape preference (OLP)1–10 ScaleNote: PSA dimensions were selected and adapted to reflect core perceptual attributes in ISO 12913-2, focusing on subjective qualities experienced in context. Evaluation items for landscape element preference and perceived environmental attributes. Note: PSA dimensions were selected and adapted to reflect core perceptual attributes in ISO 12913-2, focusing on subjective qualities experienced in context. Data collection and statistical analysis All protocols were approved by the Ethics Committee of Minnan Normal University and conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all participants prior to their participation. Data were collected in Spring 2017 during clear weather. To capture the situated sensory experience and the parks as “living places,” surveys were conducted during peak social hours (9:00–11:00 and 15:00–18:00) on both weekdays and weekends34,35. To examine potential temporal heterogeneity, sensitivity analyses were performed by stratifying the sample by day type (weekday vs. weekend). The main patterns of context-dependent sensory weighting remained qualitatively stable across strata. Researchers conducted face-to-face interviews to situate the data in specific space-time contexts, acknowledging the challenges of situated research such as episodic acoustic events and social interactions36. To minimize Common Method Bias (CMB), the order of questions was randomized, and respondents were assured of anonymity. A total of 540 questionnaires were distributed (30 per point), yielding 472 valid responses (Response rate: 87.4%). To evaluate statistical power, a post-hoc analysis was conducted using G*Power 3.137 for multiple regression (fixed model, R2 increase), assuming a medium effect size (f2 = 0.15) at α = 0.05. The achieved power exceeded 0.95 for both subsamples (N_Resting = 261; N_Activity = 211). Additionally, to justify the use of Generalized Estimating Equations (GEE) with cluster-robust standard errors, the intra-class correlation coefficient (ICC) was calculated for the nested data structure (respondents within 18 survey points). The ICC values ranged from 0.12 to 0.28 across key outcome variables, indicating moderate clustering and supporting the necessity of accounting for within-cluster dependence. To ensure comparability across the three independent samples collected at different parks, demographic profiles were compared using chi-square tests (Table 2). No significant differences were observed in gender (χ2 = 4.499, p = 0.105), age group (χ2 = 2.85, p = 0.82), education level (χ2 = 1.56, p = 0.92), occupation (χ2 = 3.12, p = 0.85), or visit frequency (χ2 = 1.24, p = 0.94). This sample homogeneity suggests that observed differences in landscape perceptions are attributable to park-specific environmental characteristics rather than respondent composition biases. Table 2Comparison of respondents’ demographic characteristics by park type (N = 472).VariableCategoryZhongshan Park (n = 151) n(%)Jiulong Park (n = 154) n(%)Zhishan Park (n = 167) n(%)Total (N = 472) n(%)χ²/p-valueGenderMale62 (41.1)76 (49.4)88 (52.7)226 (47.9)4.499/0.105Female89 (58.9)78 (50.6)79 (47.3)246 (52.1)–Age group18–30 years38 (25.2)40 (26.0)47 (28.1)125 (26.5)2.85/0.8231–45 years50 (33.1)52 (33.8)55 (32.9)157 (33.3)–46–60 years40 (26.5)37 (24.0)39 (23.4)116 (24.6)–≥ 60 years23 (15.2)25 (16.2)26 (15.6)74 (15.7)–Education levelHigh school or below28 (18.5)25 (16.2)32 (19.2)85 (18.0)1.56/0.92Bachelor/college90 (59.6)95 (61.7)98 (58.7)283 (59.9)–Master or above33 (21.9)34 (22.1)37 (22.2)104 (22.0)–OccupationStudent25 (16.6)30 (19.5)28 (16.8)83 (17.6)3.12 / 0.85Employed75 (49.7)72 (46.8)82 (49.1)229 (48.5)–Retired35 (23.2)37 (24.0)40 (24.0)112 (23.7)–Other16 (10.6)15 (9.7)17 (10.2)48 (10.2)–Visit frequencyWeekly or more70 (46.4)72 (46.8)75 (44.9)217 (46.0)1.24/ 0.94Monthly42 (27.8)45 (29.2)48 (28.7)135 (28.6)–A few times per year28 (18.5)26 (16.9)30 (18.0)84 (17.8)–First time/ rarely11 (7.3)11 (7.1)14 (8.4)36 (7.6)–Note: Values are presented as n (%). Chi-square (p χ ²) tests were conducted to compare the distribution of each demographic variable across the three park-based respondent groups (Zhongshan Park, Jiulong Park, and Zhishan Park). Reported p-values refer to between-group differences among the three parks. No statistically significant differences were observed for any demographic variable (p > 0.05), indicating acceptable sample comparability across sites. Table 2Comparison of respondents’ demographic characteristics by park type (N = 472).VariableCategoryZhongshan Park (n = 151) n(%)Jiulong Park (n = 154) n(%)Zhishan Park (n = 167) n(%)Total (N = 472) n(%)χ²/p-valueGenderMale62 (41.1)76 (49.4)88 (52.7)226 (47.9)4.499/0.105Female89 (58.9)78 (50.6)79 (47.3)246 (52.1)–Age group18–30 years38 (25.2)40 (26.0)47 (28.1)125 (26.5)2.85/0.8231–45 years50 (33.1)52 (33.8)55 (32.9)157 (33.3)–46–60 years40 (26.5)37 (24.0)39 (23.4)116 (24.6)–≥ 60 years23 (15.2)25 (16.2)26 (15.6)74 (15.7)–Education levelHigh school or below28 (18.5)25 (16.2)32 (19.2)85 (18.0)1.56/0.92Bachelor/college90 (59.6)95 (61.7)98 (58.7)283 (59.9)–Master or above33 (21.9)34 (22.1)37 (22.2)104 (22.0)–OccupationStudent25 (16.6)30 (19.5)28 (16.8)83 (17.6)3.12 / 0.85Employed75 (49.7)72 (46.8)82 (49.1)229 (48.5)–Retired35 (23.2)37 (24.0)40 (24.0)112 (23.7)–Other16 (10.6)15 (9.7)17 (10.2)48 (10.2)–Visit frequencyWeekly or more70 (46.4)72 (46.8)75 (44.9)217 (46.0)1.24/ 0.94Monthly42 (27.8)45 (29.2)48 (28.7)135 (28.6)–A few times per year28 (18.5)26 (16.9)30 (18.0)84 (17.8)–First time/ rarely11 (7.3)11 (7.1)14 (8.4)36 (7.6)–Note: Values are presented as n (%). Chi-square (p χ ²) tests were conducted to compare the distribution of each demographic variable across the three park-based respondent groups (Zhongshan Park, Jiulong Park, and Zhishan Park). Reported p-values refer to between-group differences among the three parks. No statistically significant differences were observed for any demographic variable (p > 0.05), indicating acceptable sample comparability across sites. Comparison of respondents’ demographic characteristics by park type (N = 472). Note: Values are presented as n (%). Chi-square (p χ ²) tests were conducted to compare the distribution of each demographic variable across the three park-based respondent groups (Zhongshan Park, Jiulong Park, and Zhishan Park). Reported p-values refer to between-group differences among the three parks. No statistically significant differences were observed for any demographic variable (p > 0.05), indicating acceptable sample comparability across sites. Data were analyzed using IBM SPSS Statistics 27.0. The analytical strategy followed a progressive logic to systematically test the Context-Dependent Sensory Integration Framework (CDSIF): (1) Descriptive Analysis & ANOVA: To assess macroscopic patterns and test for Sensory Aesthetic Leveling (the phenomenon where soundscape quality equilibrates overall landscape preference despite visual disparities). Effect sizes (partial η2) were calculated to contextualize the practical significance of differences beyond p-values. (2) Hierarchical Multiple Regression: To compare the Model Explanatory Efficiency (Adj. R²) between the Element-based Aggregation Model (Path A)and the Attribute-based Holistic Perception Model (Path B). This step evaluates whether perceived holistic attributes (e.g., harmony, tranquility) offer superior predictive power over discrete physical features. (3) Generalized Estimating Equations (GEE): To verify the Functional Perceptual Modulation hypothesis while accounting for the nested structure of the data (respondents clustered within 18 survey points). GEE with a Huber-White sandwich estimator was employed to generate cluster-robust standard errors, ensuring the reliability of parameter estimates (B) and p-values despite potential within-cluster correlation. Day type was not included as a covariate in the primary models. Before interpreting the models, comprehensive diagnostics—including Variance Inflation Factors (VIF < 2.2), Durbin-Watson statistics (≈ 2.0), and Harman’s Single Factor Test—were performed to ensure the absence of multicollinearity, residual dependency, and common method bias. Results Observed pattern of sensory aesthetic leveling: descriptive discrepancy between visual and overall preferences To understand baseline user experiences, we examined macroscopic differences in landscape preferences across three distinct park typologies. As shown in Table 3, one-way ANOVA results revealed a significant disparity in Visual Landscape Preference (VLP) (F (2, 469) = 10.13, p < 0.001, partial η² = 0.041). Post-hoc tests confirmed that Zhishan Park (Ecological) received significantly higher visual ratings compared to Zhongshan (Historical) and Jiulong (Comprehensive) parks (p < 0.05). This corresponds to a small effect size according to Cohen’s (1988) conventions. While the difference in visual preference is statistically detectable, its practical magnitude remains modest, consistent with Zhishan Park’s superior vegetation density. Table 3ANOVA results for landscape preferences across park types.VariableZhongshan (historical)Jiulong (comprehensive)Zhishan (ecological)F-valuep-valuePartial η²Post-Hoc (Tukey HSD)Visual preference (VLP)7.60 ± 1.417.38 ± 1.187.96 ± 1.1910.13< 0.001***0.041Zhishan > OthersSoundscape preference (SP)7.10 ± 1.717.21 ± 1.117.15 ± 1.340.260.768 (ns)0.001–Overall preference (OLP)7.40 ± 1.557.50 ± 1.047.51 ± 1.261.210.301 (ns)0.005–Note: N = 472. ***p < 0.001, ns = not significant. Table 3ANOVA results for landscape preferences across park types.VariableZhongshan (historical)Jiulong (comprehensive)Zhishan (ecological)F-valuep-valuePartial η²Post-Hoc (Tukey HSD)Visual preference (VLP)7.60 ± 1.417.38 ± 1.187.96 ± 1.1910.13< 0.001***0.041Zhishan > OthersSoundscape preference (SP)7.10 ± 1.717.21 ± 1.117.15 ± 1.340.260.768 (ns)0.001–Overall preference (OLP)7.40 ± 1.557.50 ± 1.047.51 ± 1.261.210.301 (ns)0.005–Note: N = 472. ***p < 0.001, ns = not significant. ANOVA results for landscape preferences across park types. Note: N = 472. ***p < 0.001, ns = not significant. In contrast, no significant differences were observed for Soundscape Preference (SP) or Overall Landscape Preference (OLP) (p > 0.05, partial η2 < 0.006) across sites. This descriptive pattern is consistent with a potential sensory aesthetic leveling effect, whereby soundscape quality appears to stabilize overall landscape preference despite visual disparities. However, this observation derives primarily from descriptive ANOVA comparisons and would benefit from future confirmatory tests such as moderation analysis. Model comparison: explanatory power of attributes (path B) vs. elements (path A) Hierarchical multiple regression analysis was conducted separately for each functional zone to compare the explanatory efficiency (Adj. R²) of the Element-based Aggregation Model (Path A) and the Attribute-based Holistic Perception Model (Path B) (Table 4). In Step 1, demographic covariates were entered; in Step 2, the respective predictors (discrete elements for Path A or perceived attributes for Path B) were added. Adj. R² values reported here represent the final model fit after Step 2, demonstrating the incremental contribution of each modeling approach. The robustness of parameter estimates and context-dependent predictor effects was subsequently verified using Generalized Estimating Equations (GEE) with cluster-robust standard errors (see Table 5). In the visual domain, Path B consistently outperformed Path A. In Resting Zones, while Path A explained 19.8% of the variance through discrete elements such as Lake Water, Path B more than doubled the explanatory power (Adj. R² = 0.450; Table 4) and identified Visual Harmony as the primary driver. Similarly, in Activity Zones the attribute Tidiness surpassed individual element in predictive weight. The discrepancy was most pronounced in the auditory domain. For Activity Zones, Path A explained only 9.8% of variance, while Path B reached 52.6% (Adj. R² = 0.526), with Tranquility and Harmony emerging as the strongest predictors. To formally test the superiority of Path B over Path A, Vuong’s closeness test was performed; results indicated that Path B provided a significantly better fit (p < 0.001) in all comparisons. This consistent trend supports the hypothesis that attribute-based heuristic processing (Path B) offers superior cognitive efficiency in complex urban environments. Table 4Comparison of predictive efficiency: element aggregation (Path A) vs. attribute perception (Path B).Dependent variableFunctional zoneModel typeAdj. R²F-ValueSig.Key predictors (β)Visual preference (VLP)Resting zonePath A: Elements0.19828.31< 0.001Lake water (0.269),Path B: Attributes 0.450 53.88< 0.001Visual harmony (0.300)Activity zonePath A: Elements0.1219.91< 0.001Lake water (0.238)Path B: Attributes 0.379 32.87< 0.001Tidiness (0.254)Soundscape preference (SP)Resting zonePath A: Elements0.11010.23< 0.001Broadcast (0.454))Path B: Attributes 0.218 25.16< 0.001Harmony (0.220)Activity zonePath A: Elements0.09810.080.002Water sounds (-0.329)Path B: Attributes 0.526 117.58< 0.001Tranquility (0.437)Note: N_Resting = 261; N_Activity = 211. Adj.R2 values are reported here solely to compare the relative explanatory efficiency between the Element-based Aggregation model (Path A) and the Attribute-based Holistic Perception model (Path B). These hierarchical regression results focus on model fit; the robustness and context-dependent significance of individual predictors under the nested sampling structure were further confirmed using Generalized Estimating Equations (GEE) (Table 5), from which the non-standardized coefficients (B) reported in the Abstract and Results section (Table 5) were derived. Table 4Comparison of predictive efficiency: element aggregation (Path A) vs. attribute perception (Path B).Dependent variableFunctional zoneModel typeAdj. R²F-ValueSig.Key predictors (β)Visual preference (VLP)Resting zonePath A: Elements0.19828.31< 0.001Lake water (0.269),Path B: Attributes 0.450 53.88< 0.001Visual harmony (0.300)Activity zonePath A: Elements0.1219.91< 0.001Lake water (0.238)Path B: Attributes 0.379 32.87< 0.001Tidiness (0.254)Soundscape preference (SP)Resting zonePath A: Elements0.11010.23< 0.001Broadcast (0.454))Path B: Attributes 0.218 25.16< 0.001Harmony (0.220)Activity zonePath A: Elements0.09810.080.002Water sounds (-0.329)Path B: Attributes 0.526 117.58< 0.001Tranquility (0.437)Note: N_Resting = 261; N_Activity = 211. Adj.R2 values are reported here solely to compare the relative explanatory efficiency between the Element-based Aggregation model (Path A) and the Attribute-based Holistic Perception model (Path B). These hierarchical regression results focus on model fit; the robustness and context-dependent significance of individual predictors under the nested sampling structure were further confirmed using Generalized Estimating Equations (GEE) (Table 5), from which the non-standardized coefficients (B) reported in the Abstract and Results section (Table 5) were derived. Comparison of predictive efficiency: element aggregation (Path A) vs. attribute perception (Path B). Note: N_Resting = 261; N_Activity = 211. Adj.R2 values are reported here solely to compare the relative explanatory efficiency between the Element-based Aggregation model (Path A) and the Attribute-based Holistic Perception model (Path B). These hierarchical regression results focus on model fit; the robustness and context-dependent significance of individual predictors under the nested sampling structure were further confirmed using Generalized Estimating Equations (GEE) (Table 5), from which the non-standardized coefficients (B) reported in the Abstract and Results section (Table 5) were derived. Functional perceptual modulation: weight-switching pattern in predictor contributions To test the Functional Perceptual Modulation hypothesis, Generalized Estimating Equations (GEE) with robust standard errors was conducted to predict OLP across different zones (Table 5). Diagnostics confirmed the absence of multicollinearity (all VIF < 2.2) and independent residuals (Durbin-Watson ≈ 2.0). Sensitivity analyses stratifying the data by day type (weekday vs. weekend) indicated that the observed numerical patterns in sensory weighting remained qualitatively consistent across strata. A distinct numerical pattern in sensory weighting was observed across functional contexts (Fig. 3). In Resting Zones (N = 261), Soundscape Preference (SP) exhibited a numerically higher predictive weight (B = 0.343, 95% CI [0.153, 0.534], p < 0.001) compared to Visual Preference (B = 0.307, 95% CI [0.170, 0.444], p < 0.001). This pattern is consistent with the hypothesized prioritization of acoustic filtering in restorative contexts. In contrast, within Activity Zones (N = 211), the relative importance of Visual Preference (VLP) numerically increased (B = 0.297, 95% CI [0.188,0.406], p < 0.001), exceeding the contribution of Soundscape Preference (B = 0.285, 95% CI [0.208,0.362], p < 0.001). Across both functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerged as a particularly strong predictor (B ≥ 0.566, p < 0.001), highlighting the prominent role of multisensory congruence and processing fluency in shaping landscape preference. However, direct comparisons of unstandardized coefficients (B) should be made with caution, as VLP, SP, and OLP were measured on a 1–10 scale, whereas AV-Harmony was measured on a 1–5 scale. The observed numerical differences between coefficients are descriptive in nature; formal statistical tests of equality between predictors (e.g., SP versus VLP within each zone) were not conducted in the present analysis. Table 5Parameter estimates of the GEE models for overall landscape preference (OLP).Functional contextPredictor variableBSE95% confidence intervalWald χ2p-valueResting zone (N = 261)(Constant)0.6210.355[-0.073, 1.316]3.0730.080Visualpreference (VLP)0.3070.070[0.170, 0.444]19.183< 0.001Soundscapepreference (SP) 0.343 0.097[0.153, 0.534]12.493< 0.001AV-Harmony 0.584 0.133[0.324, 0.845]19.318< 0.001Activity zone (N = 211)(Constant)1.0940.462[0.188, 2.001]5.6020.018Visualpreference (VLP) 0.297 0.056[0.188, 0.406]28.416< 0.001Soundscapepreference (SP)0.2850.039[0.208, 0.362]52.248< 0.001AV-Harmony 0.566 0.100[0.370, 0.763]31.948< 0.001Note: Dependent Variable: Overall Landscape Preference (OLP). Models were estimated using Generalized Estimating Equations (GEE) with a linear link function and an independent working correlation structure.Robust estimators (Huber-White sandwich estimator) were used to account for the nested data structure (respondents clustered within survey points: n_Resting = 10, n_Activity = 8). B represents the unstandardized regression coefficient; SE represents the robust standard error. The Wald χ² statistics test the null hypothesis that each individual coefficient equals zero. Direct comparisons of unstandardized B coefficients should be interpreted with caution, as VLP, SP, and OLP were measured on a 1–10 scale, whereas AV-Harmony was measured on a 1–5 scale. The observed numerical differences between coefficients (e.g., SP vs. VLP within zones) are descriptive; formal statistical tests of coefficient equality were not conducted in the present analysis. Table 5Parameter estimates of the GEE models for overall landscape preference (OLP).Functional contextPredictor variableBSE95% confidence intervalWald χ2p-valueResting zone (N = 261)(Constant)0.6210.355[-0.073, 1.316]3.0730.080Visualpreference (VLP)0.3070.070[0.170, 0.444]19.183< 0.001Soundscapepreference (SP) 0.343 0.097[0.153, 0.534]12.493< 0.001AV-Harmony 0.584 0.133[0.324, 0.845]19.318< 0.001Activity zone (N = 211)(Constant)1.0940.462[0.188, 2.001]5.6020.018Visualpreference (VLP) 0.297 0.056[0.188, 0.406]28.416< 0.001Soundscapepreference (SP)0.2850.039[0.208, 0.362]52.248< 0.001AV-Harmony 0.566 0.100[0.370, 0.763]31.948< 0.001Note: Dependent Variable: Overall Landscape Preference (OLP). Models were estimated using Generalized Estimating Equations (GEE) with a linear link function and an independent working correlation structure.Robust estimators (Huber-White sandwich estimator) were used to account for the nested data structure (respondents clustered within survey points: n_Resting = 10, n_Activity = 8). B represents the unstandardized regression coefficient; SE represents the robust standard error. The Wald χ² statistics test the null hypothesis that each individual coefficient equals zero. Direct comparisons of unstandardized B coefficients should be interpreted with caution, as VLP, SP, and OLP were measured on a 1–10 scale, whereas AV-Harmony was measured on a 1–5 scale. The observed numerical differences between coefficients (e.g., SP vs. VLP within zones) are descriptive; formal statistical tests of coefficient equality were not conducted in the present analysis. Parameter estimates of the GEE models for overall landscape preference (OLP). Visual preference (VLP) Soundscape preference (SP) Visual preference (VLP) Soundscape preference (SP) Note: Dependent Variable: Overall Landscape Preference (OLP). Models were estimated using Generalized Estimating Equations (GEE) with a linear link function and an independent working correlation structure.Robust estimators (Huber-White sandwich estimator) were used to account for the nested data structure (respondents clustered within survey points: n_Resting = 10, n_Activity = 8). B represents the unstandardized regression coefficient; SE represents the robust standard error. The Wald χ² statistics test the null hypothesis that each individual coefficient equals zero. Direct comparisons of unstandardized B coefficients should be interpreted with caution, as VLP, SP, and OLP were measured on a 1–10 scale, whereas AV-Harmony was measured on a 1–5 scale. The observed numerical differences between coefficients (e.g., SP vs. VLP within zones) are descriptive; formal statistical tests of coefficient equality were not conducted in the present analysis. Fig. 3The Context-Dependent Modulation of Sensory Weighting: Comparison of Unstandardized Coefficients (B) for Visual Landscape Preference (VLP), Soundscape Preference (SP), and Audiovisual Harmony (AV-Harmony) in predicting Overall Landscape Preference (OLP). The chart, derived from robust GEE estimates, illustrates a numerical pattern across functional contexts: Soundscape Preference (SP) shows a numerically higher predictive weight than Visual Landscape Preference (VLP) in Resting Zones, while the relative importance of VLP numerically increases in Activity Zones. Across both functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerges as a particularly strong predictor. It should be noted that direct comparisons of unstandardized coefficients (B) are limited by differences in measurement scales (VLP/SP/OLP on a 1–10 scale versus AV-Harmony on a 1–5 scale).The observed numerical differences between coefficients are descriptive in nature; formal statistical tests of equality between predictors (e.g., SP versus VLP within each zone) were not conducted in the present analysis. The Context-Dependent Modulation of Sensory Weighting: Comparison of Unstandardized Coefficients (B) for Visual Landscape Preference (VLP), Soundscape Preference (SP), and Audiovisual Harmony (AV-Harmony) in predicting Overall Landscape Preference (OLP). The chart, derived from robust GEE estimates, illustrates a numerical pattern across functional contexts: Soundscape Preference (SP) shows a numerically higher predictive weight than Visual Landscape Preference (VLP) in Resting Zones, while the relative importance of VLP numerically increases in Activity Zones. Across both functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerges as a particularly strong predictor. It should be noted that direct comparisons of unstandardized coefficients (B) are limited by differences in measurement scales (VLP/SP/OLP on a 1–10 scale versus AV-Harmony on a 1–5 scale).The observed numerical differences between coefficients are descriptive in nature; formal statistical tests of equality between predictors (e.g., SP versus VLP within each zone) were not conducted in the present analysis. Demographic correlations Spearman correlation analysis (Table 6) revealed that in Activity Zones, preference significantly correlated with age (r ≈ 0.21), suggesting that social use patterns may be demographic-dependent. Conversely, in Resting Zones, Soundscape Preference showed no significant correlation with gender, age or education (p > 0.05), providing no evidence of demographic heterogeneity in this sample for restorative acoustic environments. Table 6Spearman correlation coefficients between demographic characteristics and landscape preferences.Functional zonePreference typeGenderAgeEducationVisit frequencyStay durationResting zone (N = 261)Visual (VLP)− 0.129*0.233**− 0.1000.0970.095Soundscape (SP)0.0100.054− 0.0090.0680.058Overall (OLP)− 0.0030.127*− 0.076− 0.0050.106Activity zone (N = 211)Visual (VLP)− 0.0610.200**− 0.1280.209**0.061Soundscape (SP)0.0230.206**− 0.193**0.203**0.177*Overall (OLP)− 0.0050.206**− 0.0990.214**0.060Note: N_Resting = 261, N_Activity = 211. Values are Spearman rho (ρ) correlation coefficients. *p < 0.05, **p < 0.01. Table 6Spearman correlation coefficients between demographic characteristics and landscape preferences.Functional zonePreference typeGenderAgeEducationVisit frequencyStay durationResting zone (N = 261)Visual (VLP)− 0.129*0.233**− 0.1000.0970.095Soundscape (SP)0.0100.054− 0.0090.0680.058Overall (OLP)− 0.0030.127*− 0.076− 0.0050.106Activity zone (N = 211)Visual (VLP)− 0.0610.200**− 0.1280.209**0.061Soundscape (SP)0.0230.206**− 0.193**0.203**0.177*Overall (OLP)− 0.0050.206**− 0.0990.214**0.060Note: N_Resting = 261, N_Activity = 211. Values are Spearman rho (ρ) correlation coefficients. *p < 0.05, **p < 0.01. Spearman correlation coefficients between demographic characteristics and landscape preferences. Note: N_Resting = 261, N_Activity = 211. Values are Spearman rho (ρ) correlation coefficients. *p < 0.05, **p < 0.01. Discussion Attribute perception and the logic of cognitive efficiency A central finding of this study is that the Attribute-based Holistic Perception Model (Path B) explained 2–5 times more variance in visitor preference than the Element-based Aggregation Model (Path A) (Table 4). This is consistent with the idea that in information-rich urban environments, the human brain relies on heuristic processing to aggregate perceived attributes into higher-order holistic properties (e.g., Visual Harmony, Soundscape Tranquility) rather than piecewise evaluation of discrete elements37,38. Importantly, within this Path B framework, Audiovisual Harmony (AV-Harmony) serves as a key integrative attribute reflecting cross-modal congruence, rather than being synonymous with the model itself. The limitations of the element-based approach were particularly evident in the negative association found for Water Sounds in Activity Zones. While earlier element-centric studies focused on cataloging discrete sound “objects”39, our results suggest that in a bustling context, the brain prefers to perceive sound as a continuous background attribute. Transitioning from “listening to objects” to “perceiving atmospheres” mirrors the brain’s evolutionary quest for processing fluency; It simplifies complex multisensory streams into holistic judgments, thereby staying within the limits of human working memory and cognitive capacity40,41. By empirically supporting this direction, our findings imply that sustainable urban design should transition from inventory-based planning (the cognitively taxing addition of specific features) to attribute-based curation, prioritizing the shaping of sensory coherence over discrete element accumulation. Functional perceptual modulation: dominance vs. congruence The “crossover” pattern observed in the robust GEE estimates (Fig. 3; Table 5) provides empirical support for the mechanism of Functional Perceptual Modulation. This pattern suggests that sensory weighting is not a fixed construct but appears to be dynamically associated with the user’s functional goals. In Resting Zones, the relative predictive strength of Soundscape Preference is consistent with an auditory corollary to Attention Restoration Theory (ART)42. Our findings position the auditory system as a Primary Perceptual Filter (conceptually acting as a “Security Gatekeeper”) that regulates the restorative process. Even in visually pristine settings, intrusive noise triggers an orienting response, compelling the engagement of directed attention and disrupting the “soft fascination” required for recovery43,44. This aligns with Ratcliffe’s (2021) assertion that sound serves as the “sensuous container” for restoration45; if the “container” is breached by anthropogenic noise, the restorative potential of the visual environment is effectively negated. The lack of demographic correlation in these zones (Table 6) further suggests that the requirement for acoustic safety may represent a pre-cognitive baseline consistent across diverse populations in this sample46,47. In Activity Zones, the ascent of Audiovisual Harmony (AV-Harmony) as the primary predictor supports Processing Fluency Theory48. In high-stimulation environments, visitors prioritize semantic alignment between sight and sound. This creates a “Sensory Compensation Effect”: users demonstrate a higher tolerance for absolute decibel levels provided the sound is congruent with the visual context49,50. Interestingly, the continued significance of Tranquility in our model suggests that users in social contexts interpret tranquility not as absolute silence, but as signal-to-noise clarity—a state where vibrancy is present but not chaotic. It should be noted, however, that direct magnitude comparisons with other predictors are limited due to differences in their response scales (VLP, SP, and OLP measured on a 1–10 scale versus AV-Harmony on a 1–5 scale). Nevertheless, across both functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerged as a particularly strong predictor, highlighting the prominent role of multisensory congruence and processing fluency in shaping landscape preference. Sensory aesthetic leveling: environmental justice and policy translation A notable observation in the results is a descriptive pattern consistent with sensory aesthetic leveling. As shown in Table 3, Zhishan Park possessed significantly higher visual quality (partial η² = 0.041; small effect size) than the other sites, yet its Overall Landscape Preference (OLP) was statistically “pulled down” to the level of visual inferior parks (partial η² = 0.005, negligible effect). This pattern suggests that the acoustic environment may serve as a stabilizing buffer, being associated with “capping” the potential satisfaction derived from visual amenities due to higher contextual expectations in pristine settings51. Although the visual disparity reached statistical significance, its small effect size implies limited practical impact on overall preference in the absence of acoustic optimization. This indicates that, without concomitant acoustic optimization, high-capital visual investment tends to yield diminishing returns in terms of perceived well-being52. Such findings may have relevance for environmental justice (SDG 10) and can be viewed as hypothesis-generating. Visual upgrades—such as mature trees and water features—are capital-intensive and often concentrated in affluent districts53. However, our results point to the potential of soundscape quality as one factor that could help bridge structural gaps between disparate green spaces, supporting more equitable delivery of restorative benefits across the urban fabric —pending dedicated analyses that incorporate socio-spatial variables such as neighborhood SES or differential accessibility. Bridging statistical abstraction and lived experience: the role of ambiance While our model identifies recurrent perceptual patterns, we acknowledge that sensory experience in urban parks cannot be fully reduced to a fixed set of predefined variables; rather, it unfolds as an embodied and situated ambiance shaped by context, temporality, social interaction, and lived memory36,54. In this sense, the attribute-based indicators employed in our Path B model—such as Harmony and Tranquility—should not be interpreted as exhaustive representations of experience, but as quantitative proxies for dimensions of atmosphere that emerge through multisensory engagement with place55. We further acknowledge that such indicators are reductionist by nature. Nevertheless, they serve as essential communicative bridges between the phenomenological richness of lived experience and the pragmatic requirements of urban planning, where transferable, operationalizable evidence is necessary for design and policy decisions. From this perspective, our results regarding Audiovisual Harmony (AV-Harmony) in Activity Zones appear consistent with the notion of situated listening35. Visitors do not perceive sound as an isolated acoustic stimulus, but as part of a broader sonic experience inseparable from spatial form, social practices, movement, and the architectural logic of the environment34,56. Accordingly, the present model should be understood not as a complete account of park experience, but as one analytical layer within a more complex experiential ecology. Future studies would therefore benefit from combining attribute-based quantitative assessment with soundwalks, sound ethnography, and other in situ phenomenological approaches to better capture how urban parks are inhabited, interpreted, and felt as living places over time. Methodological validity: field ecology vs. laboratory simulations Our field study offers support for the perceptual modulation mechanism in an in situ setting, capturing the ecological complexity that laboratory simulations (e.g., VR) often fail to replicate. As noted by Tarlao et al. (2022), lab participants may over-rationalize sensory judgments through task-rationalization57. In contrast, our finding—such as the higher tolerance for traffic noise in Activity Zones when Audiovisual Harmony (AV-Harmony) is high—directly challenges VR-based findings where traffic noise almost universally degrades preference. This suggests that in situ, the social context and physical affordances act as powerful psychological buffers that laboratory simulations fail to replicate. Limitations and future research directions First, the data were collected in 2017. Although this serves as a critical pr-pandemic normative baseline—unconfounded by the acute environmental anxieties of the COVID-19 era—future longitudinal research is required to verify the temporal stability of these functional perceptual modulations across different socio-economic and cultural contexts. Second, we recognize that reducing the “lived experience” to statistical variables is a reductive methodology58. Our model captures perceptual patterns but may not fully encompass the phenomenological complexity of parks as “living places” influenced by territorial memory and social practices34,59. While the ISO 12913 framework provides a standardized foundation for soundscape assessment, we acknowledge ongoing debates regarding its capacity to fully capture the emotional and situated richness of soundscape experiences in real-world contexts. Future studies should integrate qualitative methods like sound ethnography to complement these quantitative findings and better address the inherent subjectivity and variability of perceptual experiences56. Third, the use of a single-item proxy for Audiovisual Harmony (AV-Harmony) and the cross-sectional nature of the data limit inferences regarding direct physiological causality or underlying mechanisms. Although preliminary checks indicated reasonable alignment with related Perceived Soundscape Attributes (PSA) items, sensitivity analyses (re-running the GEE models with and without the AV-Harmony item) confirmed that the main patterns remained qualitatively stable. Nevertheless, identified relationships should be interpreted as stable statistical associations rather than established neurobiological processes. Future studies employing multi-item scales would help strengthen this component of the model. Fourth, the study was conducted in a single Chinese city (Zhangzhou) using three local park typologies selected to represent broader global archetypes of contemporary urban green spaces. While this approach enhances internal validity and control over cultural and climatic variables, it necessarily limits the generalizability of findings to other urban contexts with differing climatic conditions, cultural soundscape norms, or park management practices. Future multi-site, cross-cultural studies are therefore essential to test the robustness and boundary conditions of the proposed Functional Perceptual Modulation mechanism. Finally, the direct contribution to urban resilience requires further exploration, specifically regarding how these functional perceptual modulations adapt to extreme environmental stressors like urban decay or climate-induced heat stress. Conclusions This study bridges the gap between cognitive sensory science and sustainable urban design by showing that landscape preference is governed by a holistic, attribute-based cognitive process (Path B) rather than the simple linear accumulation of discrete physical elements (Path A). Through a robust pooled analysis of diverse urban park typologies, we empirically demonstrate that the human brain prioritizes emergent properties—such as Visual Harmony and Soundscape Tranquility—which offer significantly higher model explanatory efficiency (up to five times more variance explained) than isolated sensory objects. Fig. 4The empirical path model of perceptual modulation mechanism. This figure visualizes the context-adaptive weighting of sensory inputs verified by robust GEE estimates. The diamond indicates the Functional Perceptual Modulation logic, illustrating the transition from environmental inputs to restorative outcomes: (1) In Resting Zones, a prioritized Acoustic Filtering is observed (B = 0.343 > 0.307), where the soundscape quality is associated with the restorative process; (2) In Activity Zones, Context-Adaptive Visual Weighting (B = 0.297 > 0.285) is associated with semantic alignment. Across all functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerges as a particularly strong predictor (B ≥ 0.566), highlighting the prominent role of processing fluency and cognitive efficiency. Note: B values represent robust estimates; * p < 0.001. The empirical path model of perceptual modulation mechanism. This figure visualizes the context-adaptive weighting of sensory inputs verified by robust GEE estimates. The diamond indicates the Functional Perceptual Modulation logic, illustrating the transition from environmental inputs to restorative outcomes: (1) In Resting Zones, a prioritized Acoustic Filtering is observed (B = 0.343 > 0.307), where the soundscape quality is associated with the restorative process; (2) In Activity Zones, Context-Adaptive Visual Weighting (B = 0.297 > 0.285) is associated with semantic alignment. Across all functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerges as a particularly strong predictor (B ≥ 0.566), highlighting the prominent role of processing fluency and cognitive efficiency. Note: B values represent robust estimates; * p < 0.001. The core theoretical contribution is the empirical support provided for the Functional Perceptual Modulation model. As illustrated in the Empirical Path Model (Fig. 4), the hierarchy of multisensory perception is context-adaptive and verified by robust GEE estimates. The model identifies a numerical “weight-switching” pattern moderated by the functional context: (1) in restorative Resting Zones, an Acoustic Filtering priority is observed (B = 0.343 > 0.307), where the auditory system is associated with a primary perceptual filter linked to the restorative process; and (2) in social Activity Zones, a Context-Adaptive Visual Weighting (B = 0.297 > 0.285) is associated with semantic alignment with the social environment. Across all functional contexts, Audiovisual Harmony (AV-Harmony) consistently emerges as a particularly strong predictor (B ≥ 0.566), acting as a cognitive bridge to overall preference through processing fluency and cognitive efficiency. Furthermore, a descriptive pattern observed in the results, which we tentatively term Sensory Aesthetic Leveling (perceptual stabilization), suggests that soundscape quality may act as a stabilizing force associated with attenuating visual disparities across disparate urban environments. Such findings may have relevance for environmental justice (SDG 10) and can be viewed as hypothesis-generating. For policymakers and urban planners, these findings support the potential value of shifting emphasis from conventional “Greening” (mere element addition) to sensory “Tuning” (attribute optimization). This may warrant consideration of strategic acoustic zoning and context-specific acoustic planning, along with perceptual performance indicators to evaluate park performance beyond decibel-based metrics. By recognizing that urban parks are not merely aesthetic assets but resilient sensory infrastructure, cities could more effectively deliver Cultural Ecosystem Services (CES). Ultimately, prioritizing the harmonious integration of sight and sound offers a cost-effective and equitable pathway to fostering public well-being, advancing the United Nations Sustainable Development Goals for inclusive, healthy, and resilient cities. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1