Greenery Alone Is Not Enough: Developing a Quantified Model of the Built Environment for Restorative Campuses Using AI-Driven VR and EEG

Abstract While greenery is often associated with cognitive restoration, this study demonstrates that it alone may not be sufficient to optimize neurocognitive recovery in academic environments. Using ecologically validated virtual reality (VR) environments developed through artificial intelligence (AI)–driven generative modeling (Skybox AI Model 3.1; procedural and diffusion-based rendering) and electroencephalography (EEG)–based neurophysiological assessments, we explore how the composition of built and natural elements alongside higher-order spatial constructs like prospect, refuge, and novelty influences restoration. Cognitive restoration was indexed using normalized alpha brainwave (NAB) activity, a relative measure of alpha power commonly associated with relaxed wakefulness and mental recovery. Stepwise multiple regression models were employed to test two hypotheses. Model 1 examined the influence of macro built environment variables (BEVs), operationalized as visual percentage compositions of buildings, greenery, water, and visible sky within 360° panoramic views on NAB activity. These BEVs were manipulated in VR individually and in combination, and their intensification was quantified using visual area coverage. The model explained 64.5% of the variance in NAB (adjusted R 2 = 0.645, p < 0.001), with greenery (standardized β = 9.009) contributing most positively, although its effects plateaued with intensification. Built structures (standardized β = 8.738), typically considered detrimental, were found to enhance NAB when spatially balanced with sky openness ( β = 4.193), water features ( β = 3.641), and environmental variation, suggesting a design-dependent restorative effect. Model 2 assessed the role of higher-order spatial constructs: prospect, refuge, environmental novelty, and overall environmental category quantified through expert evaluations of perceptual cues, spatial enclosure, and uncommon design patterns. This model accounted for 67.7% of the variance in NAB (adjusted R 2 = 0.677, p < 0.001), with novelty (standardized β = 0.517) outweighing the influence of individual physical variables. Rather than offering a binary green versus built narrative, the study proposes a predictive framework for restorative academic design based on an optimized interplay of physical and perceptual variables. By integrating EEG data with AI-driven spatial simulations, this research contributes to evidence-based campus planning, suggesting that cognitive restoration emerges not solely from greenery, but from a compositional harmony of built and natural features, balanced prospect–refuge dynamics, and perceptual novelty.