Recurrent visual behavior and viewpoint-specific spatial features in a heritage environment: immersive eye-tracking evidence from Jichang Garden

Abstract

How multiple dimensions of viewpoint-specific spatial structure relate directionally to recurrent visual behavior in heritage settings remains insufficiently specified at the level of matched scenes. This paper examines this relationship in Jichang Garden, China, by combining immersive eye-tracking with viewpoint-matched spatial features derived from three-dimensional visibility analysis. Gaze data from 51 participants across 22 useable viewpoints were used to identify four recurrent visual behavior patterns, exploratory scanning, hotspot attention, near-far alternation, and route-oriented attention, together with three broader exploration strategies. These behavioral patterns were then related to viewpoint-level spatial features, including openness, depth, complexity, orientation, viewpoint geometry, and scene composition. In the mixed-effects models, higher complexity was associated with more exploratory scanning and less hotspot attention, whereas greater openness was independently associated with less exploratory scanning. Viewpoint geometry showed differentiated behavioral profiles: greater upward visibility was associated with more exploratory scanning but less hotspot attention and near-far alternation, whereas greater eye-level visibility was associated with less route-oriented attention. Greater vegetation visibility was associated with more exploratory scanning, while greater water visibility was associated with more route-oriented attention. Random Forest diagnostics indicated moderate predictive performance and range-dependent relationships, particularly for openness. Brief post-session debriefs provided additional context for selected scenes in which specific focal elements, including a particular tree, koi, and inscriptions, were not represented as separate predictors in the spatial models. Overall, the study provides case-based, scene-level evidence from one heritage setting on how viewpoint-matched spatial features relate to recurrent visual behavior and where additional interpretive context remains useful.