Spatial pattern variations in AI-generated urban park layouts under retained urban constraints
Year: 2026
Type: article
Abstract: How generated urban park layouts respond to different conditional spatial constraints remains unclear. There are three research gaps in current research: (1) Most studies examine individual spatial elements separately and provide limited comparisons among different retained constraint types; (2) Existing park cases differ in site conditions, making it difficult to compare constraint-related spatial responses under comparable conditions; (3) How combined retained constraints modify the spatial responses associated with individual constraints remains unclear. To address these problems, we construct a constraint-conditioned comparative generative framework for urban park layouts: (1) First, establish a common environmental baseline condition and construct single- and multi-constraint generation conditions by progressively introducing internal roads, building footprints, and water bodies; (2) Secondly, use conditional GANs to generate park layouts under different constraint conditions; (3) Finally, evaluate the generated layouts using 15 Spatial Pattern Indicators and use statistical comparisons and effect-size estimation to characterize spatial responses across constraint conditions. The research results show that: (1) Differences among single-constraint configurations were most pronounced in the connectivity of walkable spaces and landscape pattern, particularly in RCC, ED, and SHEI. (2) Indicator sensitivity varied across constraint types, with some indicators showing limited responses under individual constraints but greater responses under selected combined conditions. (3) Responses to combined constraints depended on the baseline configuration, with some additional constraints producing limited changes under already constrained conditions.
Source: Scientific Reports
Institutions: Beijing Forestry University, Michigan State University
Language: English
Subfield: Urban Studies
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $2,690
Award: G5884435596
OpenAlex ID: W7227419199
