Rethinking Urban Heat Risk through Objective–Subjective Heat Exposure Mismatch: An LCZ-Based Framework for Spatial Analysis and Associated Factors

Increasing extreme heat events have highlighted the need to identify urban heat risks more accurately. However, existing assessments rarely examine whether such objective–subjective divergence is systematically associated with specific urban spatial contexts, limiting the identification of heat-risk blind spots that are not visible on conventional temperature maps.This study uses Xi'an's main urban area as a case, adopting the Local Climate Zone (LCZ) framework as a unified spatial analytical basis. By integrating Landsat-derived land surface temperature, population density, Weibo check-in texts, urban morphology, and public facility data, we construct population-weighted objective heat exposure (OHE) and BERT-based subjective heat exposure (SHE) indicators, the latter capturing observable public thermal perception from social media texts. The interpretable machine learning method CatBoost-SHAP is then applied to examine differences in the factors associated with OHE, SHE, and their deviation. Results show significant spatial mismatch between OHE and SHE across LCZ types: LCZ1, LCZ2, LCZ3, LCZ6, and LCZ7 are mainly characterized by High OHE–Low SHE, while LCZ4, LCZA, and LCZB show Low OHE–High SHE, indicating systematic deviation between objective and subjective heat exposure. OHE is primarily associated with physical factors such as building height, building density, and vegetation coverage, whereas SHE and its deviation from OHE show stronger associations with socioeconomic conditions and the proximity of potential heat-adaptation resources. Impervious surface fraction and vegetation coverage also exhibit divergent response patterns between the OHE and SHE models, indicating that physical cooling effects do not necessarily translate directly into perceived heat relief. By treating objective–subjective heat mismatch as an explicit diagnostic dimension and using LCZs to organize the spatial contexts in which such divergence occurs, this study extends conventional temperature-hotspot assessment toward a multi-signal framework integrating physical exposure, observable thermal perception, and adaptation-related conditions. The proposed analytical logic provides a transferable approach for identifying locally specific heat-risk blind spots in other cities.