How urban-rural life experiences shape rural landscape preferences: Group differences and nonlinear threshold effects via VR UAV and interpretable machine learning

Urban-rural living experience can shape how individuals perceive and evaluate rural landscapes, yet it has rarely been treated as a central explanatory factor in landscape preference research. Moreover, the lack of systematic analyses integrating living experience with refined quantitative landscape assessment limits understanding of the scope, magnitude, and underlying linear and nonlinear mechanisms shaping experience-based differences in preferences. This study develops a multi-source analytical framework integrating computer vision (CV), drone aerial photography, virtual reality (VR) experiments, and interpretable machine learning to examine how urban-rural living experience influences rural landscape visual perception satisfaction (RLVPS). Based on 4428 valid responses, we constructed a dual-indicator framework linking subjective perceptual indicators (SPIs) and objective environmental indicators (OEIs), and applied OLS regression and XGBoost-SHAP to compare group differences and identify nonlinear threshold effects. The results showed that urban-rural living experiences significantly shaped landscape preferences. Seven SPIs—color satisfaction, cleanliness, uniqueness, authenticity, livability, belonging, and naturalness—and four OEIs—public activity space area, plant species richness, road surface material, and road area proportion—were consistent predictors of RLVPS. However, rural participants placed greater emphasis on practical and functional attributes, including fences, signboards, buildings, bridges and road accessibility, whereas urban participants were more sensitive to aesthetic and ecological features, including GVI, water landscape quality, vegetation coverage, and tree proportion. In both groups, SPIs exerted stronger effects on RLVPS than OEIs. XGBoost-SHAP further revealed widespread nonlinear and threshold-dependent effects. For example, rural participants' RLVPS increased with SVF and public activity space area only within limited ranges ([0.55, 1.05–1.10] and [3, 7], respectively), while urban participants' RLVPS increased with vegetation coverage within the range of [0, 0.2] but declined beyond [0.2, 0.85]. These findings demonstrate that rural landscape preference is jointly shaped by experiential background, subjective perception, and environmental configuration, with effects that are often nonlinear rather than simply additive. This study provides a transferable ecological informatics workflow for linking multi-source environmental sensing, immersive perception experiments, and interpretable machine learning to support evidence-based, group-sensitive, and sustainable rural landscape planning. • A multi-source, cross-scale framework integrating computer vision, UAV observation, VR-based experiments, and interpretable machine learning was developed to analyse rural landscape preference. • Urban-rural living experience emerged as a substantive source of perceptual heterogeneity in rural landscape preference. • Rural participants prioritized practical and functional elements, whereas urban participants favored aesthetic and ecological features. • Subjective perceptual indicators and objective environmental features shaped preference, but the former consistently showed stronger explanatory power. • Widespread nonlinear relationships and threshold effects suggest that rural landscape optimization should be guided by calibrated intervention rather than a simple “more is better” logic.