Decoding urban thermal well-being: bridging the gap between physical cooling and social perception in wetlands via AI-enhanced sensing

Urban wetland parks serve as vital infrastructure for climate resilience, yet a critical epistemological gap remains between their objective thermodynamic performance and the subjective thermal well-being of visitors. To bridge this divide, this study establishes an integrated computational social science framework that couples remote sensing metrics with AI-decoded digital narratives. Focusing on the high-density context of Shenzhen, we synthesized Landsat-derived environmental data with large-scale crowd-sourced digital footprints, employing Large Language Models (LLMs) to extract in-situ thermal sentiments from 11,109 online reviews. The analysis suggests that human thermal experience may follow non-linear trajectories relative to physical indicators. Patch Richness Density (PRD) was strongly associated with cooling efficiency, suggesting that landscape heterogeneity may enhance thermal well-being through both physical air exchange and the psychological benefits of visual complexity. The study further identifies a counter-intuitive “accessibility paradox”: a moderate density of impervious-surface patches (PD-I) is associated with higher “cool” perception, with an exploratory empirical turning point around 32.3 n/ha in the Shenzhen wetland-park sample. This pattern suggests that well-distributed access infrastructure may act as a behavioral mediator, facilitating visitor access to cooling cores while remaining subject to site-specific ecological and urban-morphological conditions. Additionally, comfortable perception shows an empirical turning point around a Land Surface Temperature (LST) of 28.8 °C, indicating a context-dependent adaptation range rather than a universal comfort standard. These findings challenge the conventional “minimize-pavement” assumption and suggest that thermal satisfaction depends on a strategic balance between ecological integrity and recreational accessibility. Rather than providing prescriptive design benchmarks, the identified non-linear patterns offer hypothesis-generating evidence and indicative design references for human-centric urban governance. This study contributes a replicable, AI-enhanced paradigm for aligning climate resilience research with public thermal perception.