Assessing the impact of waterfront trail aesthetics on psychological restoration in urban environments: a deep learning and random forest approach

Urban waterfront trails can promote psychological restoration, yet how specific landscape attributes shape scenic beauty and multidimensional restorative perception across different revetment types remains unclear. This study quantified 15 landscape variables (F1–F15) across 30 urban waterfront trail sites and examined their associations with scenic beauty estimation (SBE) and four restorative dimensions (emotional, cognitive, physiological, and behavioral). Spearman correlation analysis, non-parametric group comparisons, and Random Forest modeling were applied to identify key predictors, with variable importance interpreted as predictive contributions rather than causal mechanisms. Partial-mediation structural equation models (SEM) were then constructed for the overall sample and for artificial, mixed, and natural revetment types. Results indicated that F4 was the most influential predictor of SBE, followed by F7, F13, F2, and F5. SEM analyses showed that SBE functioned as a central mediator linking landscape attributes to restorative outcomes: F4 was positively associated with SBE ( β = 0.460), and SBE positively predicted all four restorative dimensions ( β = 0.802–0.917), with additional direct paths indicating partial mediation. Grouped SEMs further revealed revetment-type-specific pathway structures, suggesting that revetment context conditions which landscape elements most strongly relate to perceived beauty and psychological restoration.